Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT,...

45
DEGREE PROJECT, IN , SECOND LEVEL MEDIA TECHNOLOGY STOCKHOLM, SWEDEN 2015 Photogrammetric software as an alternative to 3D laser scanning in an amateur environment MARKUS WARNE KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF COMPUTER SCIENCE AND COMMUNICATION (CSC)

Transcript of Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT,...

Page 1: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

DEGREE PROJECT, IN , SECOND LEVELMEDIA TECHNOLOGY

STOCKHOLM, SWEDEN 2015

Photogrammetric software as analternative to 3D laser scanning in anamateur environment

MARKUS WARNE

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF COMPUTER SCIENCE AND COMMUNICATION (CSC)

Page 2: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

EXAMENSARBETE  VID  CSC,  KTH  

 Fotogrammetrisk  programvara  som  alternativ  

till  laser  3D-­‐skanning  i  amatörmiljö  

 

Photogrammetric  software  as  an  alternative  to  

3D  laser  scanning  in  an  amateur  environment  

 Markus  Warne  

[email protected]  

Examensarbete  i  medieteknik  

Handledare  på  CSC  var  Vasiliki  Tsaknaki  

Handledare  på  CUT  var  Krzysztof  Skabek  

Examinator:  Haibo  Li  

Uppdragsgivare:  Politechnika  Krakowska  (Cracow  University  of  Technology)    

Page 3: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

Photogrammetric  software  

as  an  alternative  to  3D  laser  

scanning  in  an  amateur  

environment  

 Abstract  Photogrammetric  software  today  is  at  a  level  where  it  is  accessible  to  the  mainstream  

public  and  without  larger  effort  is  able  to  reconstruct  digital  3D  models  from  

photographic  input.    This  thesis  investigates  the  performance  of  photogrammetricly  

reconstructed  models  and  evaluates  them  by  comparing  the  results  to  their  

corresponding  reconstructed  models  from  a  3D  laser  scanner  with  a  focus  on  smaller  

objects  in  an  amateur  environment.  The  evaluation  is  performed  on  four  different  

objects,  which  are  all  individually  compared  to  their  scanned  counterpart.  They  are  

compared  both  with  a  subjective  judgment  of  quality  and  by  numerically  measuring  the  

point-­‐to-­‐point  distance  on  the  models.  From  the  results  conclusions  are  drawn  that  the  

methods  can  produce  similar  results  albeit  there  are  many  performance  factors  

discovered  for  a  good  reconstructions  with  photogrammetry.  The  properties  of  the  

physical  object  and  the  quality  of  the  visual  input  data  stand  out  as  the  most  important  

factors.  

   

Page 4: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

Fotogrammetrisk  

programvara  som  alternativ  

till  laser  3D-­‐skanning  i  

amatörmiljö  

 Abstrakt  Den  fotogrammetriska  programvaran  som  existerar  idag  är  tillgänglig  för  allmänheten  

men  framförallt  kapabel  att  återskapa  digitala  3D-­‐modeller  utan  större  ansträngning.  

Denna  rapport  utforskar  och  utvärderar  möjligheterna  att  återskapa  dessa  objekt  för  att  

sedan  jämföra  hur  dessa  står  sig  gentemot  motsvarande  återskapade  objekt  med  en  3D  

laser  skanner.  Fokus  ligger  på  att  se  hur  mindre  objekt  kan  återskapas  i  en  amatörmiljö.  

Testerna  genomförs  på  fyra  olika  objekt  genom  att  först  återskapa  dessa  digitalt  m.h.a.  

fotogrammetri  för  att  sedan  jämföra  dessa  inviduellt  med  motsvarande  modeller  

återskapade  m.h.a.  3D-­‐skanning.  Utvärderingen  sker  subjektivt  med  en  bedömning  av  

kvalité  men  även  genom  att  mäta  avstånden  från  punk  till  punkt  på  modellerna.  Från  

resultaten  kan  slutsatserna  dras  att  det  går  att  nå  likvärdiga  resultat  med  fotogrammetri  

som  3D-­‐skanning  men  dessa  beror  på  ett  antal  kritiska  faktorer.  Objektets  fysiska  

egenskaper  samt  kvalitén  av  den  visuella  data  som  används  framstår  som  nyckelfaktorer  

för  att  lyckas  med  en  bra  digitalt  återskapad  modell.    

 

     

Page 5: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

1   Introduction  ...................................................................................................................................................  1  

1.1   Goal  of  the  thesis  .................................................................................................................................  2  

1.2   Research  questions  ............................................................................................................................  2  

1.3   Limitations  .............................................................................................................................................  3  

2   Background  ....................................................................................................................................................  4  

2.1   Related  research  .................................................................................................................................  4  

2.2   Triangulation  ........................................................................................................................................  5  

2.3   3D  laser  scanning  ................................................................................................................................  5  

2.3.1   How  laser  triangulation  sensors  work  ..............................................................................  5  

2.4   Photogrammetry  .................................................................................................................................  5  

2.4.1   The  basics  of  Photogrammetry  ............................................................................................  6  

3   Method  .............................................................................................................................................................  8  

3.1   Literature  study  ...................................................................................................................................  8  

3.2   Quantitative  evaluation  ....................................................................................................................  8  

3.3   Qualitative  observations  ..................................................................................................................  8  

4   Evaluation  setup  ..........................................................................................................................................  9  

4.1   Data  generation  -­‐  Photogrammetric  approach  ......................................................................  9  

4.1.1   Photo  environment  setup  and  camera  parameters  ....................................................  9  

4.1.2   Photogrammetric  processing:  Agisoft’s  Photoscan  ..................................................  10  

4.1.2.1   Camera  alignment  ..........................................................................................................  10  

4.1.2.2   Dense  point  cloud  ...........................................................................................................  10  

4.1.2.3   Mesh  construction  ..........................................................................................................  11  

4.1.3   Photogrammetric  processing:  Autodesk’s  123D  Catch  ..........................................  11  

4.2   Data  generation  -­‐  laser  scanning  approach  ..........................................................................  12  

4.2.1   Environment  and  setup  ........................................................................................................  12  

4.2.2   Konica  Minolta  Vivid  9i  .........................................................................................................  12  

4.2.3   Model  reconstruction  from  point  cloud  ........................................................................  12  

4.3   Data  comparison  ..............................................................................................................................  13  

4.3.1   Alignment  ...................................................................................................................................  13  

4.3.2   Measurement  ............................................................................................................................  13  

5   Results  ...........................................................................................................................................................  15  

5.1   Case  1  –  Quadric  object  .................................................................................................................  15  

5.1.1   3D  print  reconstructed  with  PhotoScan  ........................................................................  16  

Page 6: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

5.1.2   3D  print  reconstructed  with  123D  Catch  ......................................................................  17  

5.2   Case  2  –  Angel  figure  ......................................................................................................................  18  

5.2.1   3D  model  reconstructed  with  PhotoScan  .....................................................................  19  

5.2.2   3D  model  reconstructed  with  123D  Catch  ...................................................................  20  

5.3   Case  3  –  Monkey  figure  .................................................................................................................  21  

5.3.1   3D  model  reconstructed  with  PhotoScan  .....................................................................  22  

5.3.2   3D  model  reconstructed  with  123D  Catch  ...................................................................  23  

5.4   Case  4  –  Wooden  cat  .......................................................................................................................  24  

5.4.1   3D  model  reconstructed  with  PhotoScan  .....................................................................  24  

5.4.2   3D  model  reconstructed  with  123D  Catch  ...................................................................  26  

6   Analysis  and  discussion  .........................................................................................................................  27  

6.1   Comparing  the  results  of  the  photogrammetric  reconstructions  and  the  laser  

scanned  reconstructions  ..........................................................................................................................  27  

6.2   Strengths  and  weaknesses  of  the  photogrammetric  reconstruction  software  

applications  ...................................................................................................................................................  28  

6.3   The  photogrammetric  reconstruction  process  as  a  whole  ............................................  29  

6.4   Can  photogrammetry  yield  similar  results  to  3D  laser  scanning  when  used  in  an  

amateur  home  setting  for  smaller  objects?  .....................................................................................  30  

7   Conclusion  ...................................................................................................................................................  31  

7.1   Future  research  ................................................................................................................................  32  

8   References  ...................................................................................................................................................  33  

9   Appendix  ......................................................................................................................................................  35  

9.1   Case  1  –  deviation  results  .............................................................................................................  35  

9.1.1   Photoscan  model  compared  to  3D  scanned  model  ..................................................  35  

9.1.2   123D  Catch  model  compared  to  3D  scanned  model  ................................................  35  

9.2   Case  2  –  deviation  results  .............................................................................................................  36  

9.2.1   Photoscan  model  compared  to  3D  scanned  model  ..................................................  36  

9.2.2   123D  Catch  model  compared  to  3D  scanned  model  ................................................  36  

9.3   Case  3  –  deviation  results  .............................................................................................................  37  

9.3.1   Photoscan  model  compared  to  3D  scanned  model  ..................................................  37  

9.3.2   123D  Catch  model  compared  to  3D  scanned  model  ................................................  37  

9.4   Case  4  –  deviation  results  .............................................................................................................  38  

9.4.1   Photoscan  model  compared  to  3D  scanned  model  ..................................................  38  

Page 7: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

 

9.4.2   123D  Catch  model  compared  to  3D  scanned  model  ................................................  38  

Page 8: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  1  

 

1    Introduction  Recently,  there  has  been  a  growing  interest  in  3D  printing  technologies,  with  a  number  

of  hardware  and  applications  with  a  focus  on  the  context  of  everyday  use  (Crum  2014).  

This  development  indicates  a  future  where  3D  printing  could  be  a  tool  for  personal  use  

for  everyone,  not  just  experts.  The  3D  printing  techniques  are  quickly  advancing  to  

become  more  accessible  to  the  general  public  with  new  models  created  specifically  for  

home  use  and  a  lower  budget  (Matter  and  Form  2014).  What  has  not  been  discussed  in  

depth  is  the  other  side  of  the  spectrum.  What  will  we  print?  The  models  and  data  must  

come  from  somewhere.  As  the  demand  and  ability  to  print  3D  models  increases,  the  

supply  of  models  must  also  follow  according  to  basic  economic  theory.  When  this  

happens,  the  market  will  desire  a  method  for  gathering  3D  model  data  that  is  accessible  

on  an  amateur  scale  to  as  many  individuals  as  possible,  at  a  low  cost.  

 

Methods  such  as  laser  scanning  are  not  new,  the  first  triangulation  laser  scanning  

technology  was  developed  already  in  1978  (Mayer  1999).  However,  these  methods  were  

and  are  still  inaccessible  to  the  mainstream  public,  at  least  to  some  degree,  as  the  cost  of  

acquiring  the  technology  or  the  knowledge  needed  to  operate  it  is  simply  too  high  for  

the  average  user.  A  relatively  cheap  alternative  to  laser  scanning  is  stereo  

photogrammetry.  With  the  help  of  specific  software  and  the  advanced  triangulation  

algorithms  that  are  available  today,  this  method  can  be  used,  similarly  to  a  3D  scanner,  

to  reconstruct  and  digitally  model  real  life  physical  objects  from  just  a  set  of  ordinary  

photos.  These  methods  now  open  up  new  possibilities  when  it  comes  to  creating  and  

sharing  3D  models  on  a  much  larger  scale,  by  amateur  users,  as  they  become  more  

accessible  from  an  economic  and  technological  viewpoint  to  the  general  public.  

 

This  is  a  very  important  area  as  these  technologies  are  on  the  brink  of  becoming  

mainstream  and  integrated  to  our  everyday  lives.  3D  printing  and  3D  reconstruction  

might  be  as  common  as  sharing  a  file  over  the  internet  is  today,  in  just  a  couple  of  years.  

It  is  important  to  analyze  these  different  methods,  where  they  are  in  their  current  state  

and  possibly  draw  conclusions  on  what  still  needs  to  and  can  be  improved  in  the  future.      

Page 9: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  2  

 

1.1 Goal  of  the  thesis  The  goal  of  this  thesis  is  to  take  a  closer  look  at  3D  reconstruction  of  objects,  by  using  

photogrammetry  software  as  an  alternative  to  laser  scanning.    Furthermore,  to  

investigate  if  this  is  a  viable  alternative  for  mainstream  users  and  available  to  generate  

models  of  a  comparable  quality  to  that  of  a  reconstructed  model  from  a  laser  scanner.    

The  main  focus  will  be  on  comparing  the  models  reconstructed  from  the  two  methods  

mentioned  above  (reconstruction  with  laser  scanning  and  reconstruction  with  stereo  

photogrammetry),  as  described  in  Figure  1.  Specifically,  I  will  investigate  the  deviations  

of  the  surfaces  between  the  two  methods  while  trying  to  assess  the  quality  of  the  

models,  by  comparing  them  to  each  other  numerically.

 Figure  1  –  Overview  of  reconstruction  methods  

 

1.2 Research  questions  •  Can  photogrammetry  yield  similar  results  to  3D  laser  scanning  when  used  in  an  

amateur  home  setting  for  small  objects?  

o Numerical  comparison  of  deviation  between  the  reconstructed  surfaces    

 

• How  do  these  two  methods  (laser  scanning  and  stereo  photogrammetry)  

reconstructing  models  compare  to  each  other?  

o Strengths  and  weaknesses  

Page 10: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  3  

1.3 Limitations  The  comparison  is  limited  to  two  photogrammetric  methods  and  the  laser  scanning  is  

used  just  as  reference  measurement.  The  quality  of  the  scanned  models  and  accuaracy  

for  the  specific  scanner  has  already  been  carried  out  in  another  publication  (Spytkowska  

2008).  Ideally  a  set  of  models  from  several  laser  scanners  would  be  used  as  to  be  able  to  

generalize  the  results  from  laser  scanning  and  identify  similarities.  In  addition,  only  two  

photogrammetric  software  applications  will  be  used  to  reconstruct  3D  models,  which  

could  also  limits  the  conclusions  drawn  from  comparing  the  two  techniques  and  

identifying  general  strengths  and  weaknesses.  

 

Furthermore,  the  quality  of  the  reconstructed  objects  will  be  assessed  subjectively,  due  

to  the  nature  of  such  qualitative  evaluations.  The  limitation  is  the  fact  that  there  is  no  

digital  original  object  to  compare  and  quantify  the  deviation  from  the  digital  

reconstructions,  as  in  most  cases  the  originals  are  small  physical  objects.    

The  reconstructed  models  will  be  compared  to  a  digital  representation  of  the  original,  in  

this  case,  a  3D  scanned  version  of  the  original  object,  which  in  turn  will  be  treated  as  

reference  for  deviation  measurements.  This  will  provide  quantifiable  results  of  deviation  

between  the  two  methods,  3D  scanning  and  photogrammetry,  but  there  is  no  way  to  

compare  these  to  that  of  the  original  object.  

 

One  of  the  big  limitations  of  this  thesis  is  that  in  order  to  stay  true  to  the  “amateur  

mainstream  user”  perspective  a  number  of  commercial  software  applications  have  been  

used  in  parts  of  the  process.  This  generates  some  “black  box”  parts  where  the  

transparency  of  the  process  is  limited,  as  we  only  know  what  we  put  in  and  what  comes  

out.  Assumptions  are  made  in  these  cases  based  upon  general  procedures  and  

algorithms  within  the  field  but  complete  certainty  can’t  be  achieved.  It  has  been  made  

clear  which  these  parts  are  and  when  assumptions  are  made  they  are  clearly  indicated.  

 

   

Page 11: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  4  

2 Background  2.1 Related  research  Laser  scanning  has  historically  been  used  for  scanning  small  objects  in  a  controlled  

environment,  where  there  is  a  possibility  of  scanning  the  object  in  360  degrees  angle.  

This  is  due  to  the  fact  that  3D  scanners  often  have  a  small  field  of  view  and  larger  objects  

often  have  to  be  scanned  in  several  iterations  and  then  combined  together  in  order  to  

reproduce  and  complete  the  model  (Chen  et  al.  2000;  Seitz  &  Curless  2006).  However  

there  are  some  exceptions  with  laser  scanners  such  as  LIDARs  which  are  laser  scanners  

optimized  for  capturing  large  geographic  areas  and  similar  scenarios.  On  the  other  hand,  

photogrammetry  has  mostly  been  used  in  these  types  of  situations,  when  there  is  a  need  

to  scan  large-­‐scale  areas,  such  as  air  photography,  cartography,  mapping  of  

archaeological  sites  and  other  situations  with  large  objects,  when  there  is  no  need  to  

capture  small  details  but  rather  focus  on  extracting  measurements  etc.  Photogrammetry  

is  rapidly  becoming  more  and  more  common,  for  reconstructing  high  detailed  3D  

models  (Blizard  2014;  Poznanski  2014).  This  can  be  seen  especially  on  the  Internet,  

where  its  accessibility  to  the  mainstream  public  has  inspired  many  hobbyists  and  so  

called  DIY  (do  it  yourself)  enthusiasts  (Blizard  2014)  but  also  in  the  entertainment  

industry  such  as  games  (Poznanski  2014)  and  movies  (Wolff  2004).  Previously  

photogrammetry  has  also  been  used  in  combination  with  laser  scans  as  a  compliment  to  

provide  accurate  textures  for  the  model  that  were  generated  by  laser  scanning.    

 

Attempts  to  reconstruct  digital  3D  models  with  the  help  of  photogrammetry  were  

already  done  in  1984  (Benard  1984)  and  there  are  even  some  cases  where  the  results  of  

the  reconstructions  have  been  compared  to  laser  scanning  techniques  (Baltsavias  1999;  

Fassi  &  Fregonese  2013).  However  most  of  them  are,  as  mentioned  earlier,  focused  on  

archaeological,  architectural  or  geo-­‐data  scenarios  where  the  objects  are  usually  very  

large  in  scale.  This  thesis  focuses  on  the  perspective  of  smaller  sized  objects,  up  to  one  

cubic  meter,  and  as  some  earlier  research  suggests  (Baltsavias  1999)  this  area  might  be  

more  challenging  for  photogrammetric  reconstruction  of  models.  In  addition,  3D  

scanners  have  largely  dominated  the  digital  reconstruction  of  small  objects  as  they  are  

usually  built  for  this  purpose  whereas  reconstruction  with  the  help  of  photogrammetry  

Page 12: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  5  

has  not  been  applied  in  these  scenarios  as  often  and  if  so  only  partly  or  complimentary  

for  smaller  objects.  

 

2.2 Triangulation  Triangulation  is  a  central  concept  for  both  of  the  techniques  applied  in  this  thesis,  as  

both  photogrammetry  and  most  laser  scanning  methods  are  based  around  this  principle.  

It  is  a  method  used  to  calculate  the  position  of  points  in  3D  space.  It  is  used  in  a  wide  

range  of  applications  and  scenarios  such  as  navigation,  astronomy  and  many  more  due  

to  its  broad  and  dynamic  origin.    Triangulation  works  by  mathematically  intersecting  

converging  lines  in  space  from  at  least  two  known  points  to  that  of  the  investigated  

point.  By  measuring  the  difference  in  the  angle  to  the  investigated  point  the  precise  

location  of  the  point  in  3D  space  can  be  determined.  (Spytkowska  2008)  

 

2.3 3D  laser  scanning  2.3.1 How  laser  triangulation  sensors  work  

A  triangulating  laser  scanner  simplified  consists  of  two  components,  a  transmitter  and  a  

receiver.  The  transmitter  usually  consists  of  a  laser  diode  that  projects  a  ray  of  light  on  

the  object.  A  charged  coupled  device  (CCD)  sensor  detects  the  reflection  and  due  to  

displacements  in  the  object  the  angle  of  reflection  varies  depending  on  form  and  

distance.  Thus  the  difference  can  be  measured  due  to  the  principles  of  triangulation.  

From  this  data,  a  point  cloud  is  generated  based  on  each  specific  measurement,  each  

point  with  a  specific  distance  from  the  scanner.  This  measurement  is  carried  out  

thousands  of  times  to  generate  a  large  point  cloud  representing  the  object.  This  discrete  

point  data  cloud  can  then  be  interpolated,  usually  with  the  help  of  some  complimentary  

software,  to  create  a  3D  surface  that  consists  of  not  just  points  but  polygons.  

(Spytkowska  2008;  Dold  &  Brenner  2006)  

 

 

2.4 Photogrammetry  Photogrammetry  is  the  science  of  acquiring  measurements  and  3D  coordinates  from  

photographs.  It  can  be  compared  to  the  reverse  engineering  of  photographs  as  it  aims  to  

turn  2D  information  into  3D  information,  the  opposite  of  a  camera  (Figure  2).  

Page 13: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  6  

Photogrammetry  is  often  applied  in  topography  scenarios  like  satellite  and  aerial  

photography  but  also  in  close  range  scenarios,  such  as  3D  reconstruction.  

 

 Figure  2  –  Input  and  output  of  optical  data  capturing  methods  

 

2.4.1 The  basics  of  Photogrammetry  

The  core  principle  for  photogrammetry  is  triangulation.  Due  to  several  photographs  (at  

least  two  or  more)  with  overlapping  information,  rays,  as  they  are  called,  can  be  

calculated  from  the  camera  position  to  the  object.    By  calculating  the  difference  in  angle  

between  the  different  rays  the  distance  and  position  of  the  camera  can  be  established.  

This  is  also  exactly  the  way  our  eyes  work  when  estimating  depth.  In  photogrammetry  

the  difference  in  the  (x,y)-­‐plane  is  used  to  triangulate  the  measured  point.  Furthermore  

Page 14: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  7  

photogrammetry  applies  this  principle  to  multiple  points  at  a  time  with  theoretically  no  

limit  to  the  number  of  points  measured  at  the  same  time  (Slama  et  al.  1980).  

 

To  be  able  to  compare  the  3D  laser  scanning  method  to  the  photogrammetric  approach  

we  have  to  go  from  physical  3D  domain  to  digital  3D  domain  as  seen  in  Figure  2.  

However  strictly  photogrammetry  consists  of  2D  input  in  the  form  of  photos,  thus  we  

have  to  include  the  photography  part  from  Figure  2  to  achieve  the  complete  flow  of  

physical  3D  domain  to  digital  3D  domain.    

 

As  mentioned  before  (Figure  2)  photogrammetry  is  in  a  way  the  reversed  process  of  

photography.  Unfortunately  the  photographic  process  is  not  perfect  as  information  is  

lost  when  taking  a  photo,  if  it  was  perfect,  as  in  no  information  lost,  just  two  photos  

would  be  more  than  enough  to  recreate  the  3D  scene.  So  to  compensate  for  this  missing  

information  several  photographs  (absolute  minimum  of  two)  have  to  be  used  to  aid  the  

calculation.  The  coordinates  acquired  from  these  calculations  are  the  final  result  from  

photogrammetry,  this  can  be  presented  in  the  form  of  a  point  cloud  or  some  other  data  

set  that  then  usually  is  used  further  to  extrapolate  a  3D  surface  (Greve  1996;  Schenk  

2005).  

   

For  best  results  the  images  used  as  input  for  photogrammetry  a  few  parameters  are  

important.  First  is  the  focus  of  the  camera,  since  photogrammetry  uses  pixel  for  points  in  

triangulation  blurry  images  are  very  tricky  for  pinpointing  the  position  of  features  and  

other  elements  that  are  out  of  focus.  Furthermore  the  resolution  is  a  very  important  

parameter  much  due  to  the  same  reason  mentioned  earlier  meaning  more  pixels  equal  

more  accurate  calculations.    Consistent  lightning  is  also  a  factor  that  helps  

photogrammetry  in  identifying  various  elements  in  the  photograph.  Varying  light  casts  

different  shadows  from  picture  to  picture  this  can  have  a  destructive  effect,  as  the  same  

feature  can  have  varying  intensity  and  color  depending  on  the  image  analyzed  at  the  

moment  (Greve  1996;  Schenk  2005).  

 

   

Page 15: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  8  

3 Method  3.1 Literature  study  To  see  what  has  been  done  in  the  field  of  3D  reconstruction  with  the  use  of  

photogrammetry  a  literature  study  has  been  done.    Google  Scholar  and  KTH  Royal  

Institute  of  Technology’s  “Primo”  service  have  been  used  as  primary  knowledge  wells  

for  academic  research  within  the  area.  Furthermore  high  detail  3D  reconstruction  from  

photogrammetry  is  a  relatively  new  concept  relevant  information  has  been  found  on  

blogs  and  other  technical  news  themed  websites.  Google  search  and  Wikipedia  have  

been  used  as  a  compliment  or  in  conjunction  to  the  named  sources  above.  

 

3.2 Quantitative  evaluation  Quantitative  methods  are  applied  in  this  report  to  measure  the  difference  in  distance  

between  points  on  several  reconstructed  3D  surfaces.  This  is  done  with  the  help  of  

software  that  generates  points  on  the  reference  surface  and  then  tries  to  match  them  

with  the  test  surface.  Quantitative  distance  data  is  received  for  each  of  these  points.  

Unfortunately  only  the  disparity  between  two  reconstructed  objects  can  be  compared  

quantitatively  and  not  between  the  reconstructed  objects  and  the  original.  This  is  due  to  

the  fact  that  the  original  is  not  in  the  digital  domain  and  any  “conversion”  from  physical  

to  digital  is  simply  another  form  of  reconstruction  method.  This  leaves  the  “quality”  

parameter,  from  original  to  reconstruction,  unquantifiable  and  a  subjective  matter.  What  

can  be  compared  are  the  disparities  of  the  surfaces  from  the  reconstructed  models,  

which  is  also  the  main  goal  in  this  thesis,  to  find  out  if  the  methods  are  comparable  and  

yield  similar  results  in  a  non  professional  environment.  

 

3.3 Qualitative  observations  Furthermore  the  resulting  reconstructed  models  will  be  assessed  from  a  qualitative  

approach  with  the  aim  to  give  some  way  of  categorizing  the  perceived  quality  and  

similarity  of  the  digital  reconstructed  object  compared  to  the  physical  original.  This  

assessment  is  purely  subjective  and  may  vary  depending  on  the  individual.  This  

observation  is  carried  out  with  the  purpose  of  another  input  into  the  main  research  

question,  whether  or  not  the  tested  methods  yield  comparable  results.  A  method  might  

Page 16: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  9  

give  relatively  small  quantifiable  differences  but  the  discrepancy  in  perceived  quality  

between  the  original  object  and  the  reconstructed  model  may  be  very  large.  

 

4 Evaluation  setup  The  photogrammetric  evaluation  process  consists  of  two  major  parts,  the  first  is  the  

reconstruction  of  the  physical  object  in  3D  domain  –  this  will  be  called  data  generation.  

The  second  part  consists  of  analyzing  the  reconstructed  objects  and  comparing  them  to  

each  other.  This  is  where  the  quantified  data  is  extracted  from  the  objects  –  this  will  be  

called  data  comparison.    

 

As  the  software  applications  used  in  this  thesis  are  commercial,  the  specific  algorithms  

and  techniques  used  are  not  available  to  the  public  domain.  Therefore  a  general  

approach  to  reconstruction  with  photogrammetry  and  laser  scanning  has  been  

described  and  for  the  specific  software  that  follows  all  information  that  is  available  for  

each  application  will  be  presented  and  assumptions  based  on  what  is  known.  

 

4.1 Data  generation  -­‐  Photogrammetric  approach  4.1.1 Photo  environment  setup  and  camera  parameters  The  data  generation  environment  was  set  up  in  a  home  setting  with  varying  lightning  

conditions  as  to  stay  in  line  with  the  thesis  main  goal  of  analyzing  the  possibility  for  

mainstream  public  to  apply  this  technique  in  a  non  professional  context.  A  tripod  was  

used  to  aid  the  stability  and  positioning  of  the  camera  and  the  lightning  consisted  of  

mixed  indoor  light.  The  camera  was  placed  as  close  as  possible  to  the  object  for  the  

majority  of  the  picture  to  cover  most  of  the  image  area  to  preserve  as  much  detail  as  

possible.    

 

The  camera  used  was  a  FUJIFILM  X100S.  The  pictures  taken  had  the  aperture  value  of  16  

as  to  maximize  the  depth  of  field  for  maximum  amount  of  the  element  to  be  in  focus.  

Depending  on  the  lightning  registered  by  the  camera  different  shutter  speeds  were  

applied  although  most  of  the  taken  pictures  had  a  shutter  speed  of  ¼  seconds.  The  

pictures  were  saved  digitally  in  .jpg  format  with  a  resolution  of  4896  x  3264  pixels.

Page 17: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  10  

 

As  mentioned  before  resolution  and  focus  are  two  very  important  parameters  for  the  

reconstruction  process  and  these  have  been  taken  into  consideration  in  this  setup.  

However  consistent  lightning  is  also  an  important  factor  for  best  results  from  

photogrammetric  reconstruction  but  there  has  been  no  effort  in  this  setup  to  minimize  

this.  This  is  to  mimic  home  conditions  of  mainstream  users  to  stay  in  line  with  the  thesis  

goal  and  as  in  a  home  environment  there  is  usually  no  way  to  achieve  evenly  lit  objects  

such  as  with  studio  lightning  conditions.  However  it  is  assumed  that  a  good  depth  of  

field  and  focus  can  be  achieved  in  a  home  environment.  

 

4.1.2 Photogrammetric  processing:  Agisoft’s  Photoscan  

The  procedure  of  photograph  processing  and  3D  model  construction  is  described  with  

the  following  four  stages  according  to  the  PhotoScan  manual  (Agisoft  LLC  2011).  

4.1.2.1 Camera  alignment    The  software  searches  for  common  points  in  the  collection  of  photographs.  To  be  able  to  

match  them,  the  software  also  calculates  the  position  and  orientation  of  the  camera  for  

each  picture.  It  is  very  probable  that  this  is  carried  out  with  some  form  of  triangulation  

described  earlier.  The  result  of  this  process  is  a  sparse  point  cloud  where  several  points  

have  been  identified  and  matched  over  the  different  camera  positions.  It  is  important  to  

have  in  mind  that  several  points  can  be  calculated  and  matched  per  camera  location  (a  

single  photo).  However  this  sparse  point  cloud  is  not  what  the  reconstruction  of  the  3D  

model  is  based  on  unless  explicitly  specified  by  the  user.  The  information  that  is  used  

further  in  the  process  of  reconstruction  is  mainly  the  set  of  camera  positions  gathered  in  

this  stage,  presumably  with  the  intention  of  using  them  as  a  starting  point  for  further  

triangulation  of  points.      

4.1.2.2 Dense  point  cloud  

The  generation  of  a  dense  point  cloud  is  done  with  the  help  of  the  estimated  camera  

positions  and  the  respective  matching  photographs.  Unfortunately  no  more  information  

is  available  on  this  stage.    It  is  likely  that  a  more  complex  triangulation  process  is  used  in  

this  stage  to  generate  a  dense  point  cloud  with  the  help  of  the  initially  calculated  camera  

positions.  

Page 18: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  11  

4.1.2.3 Mesh  construction  The  third  stage  consists  of  constructing  a  3D  polygonal  mesh  representing  the  object  

surface  based  on  the  dense  point  cloud.  

 

In  most  cases  two  algorithmic  methods  are  available  that  the  software  can  apply  for  3D  

mesh  generation:  

 

• Height  field  –  Optimized  for  modeling  of  planar  surfaces,  such  as  terrains  or  bas-­‐

reliefs.  For  aerial  photography  processing,  it  requires  a  lower  amount  of  memory  

and  allows  for  larger  data  sets  processing.  

 

• Arbitrary  -­‐  For  closed  objects,  such  as  statues,  buildings,  etc.  It  doesn't  make  any  

assumptions  on  the  type  of  the  object  modeled,  which  comes  at  a  cost  of  higher  

memory  consumption.  

 

In  this  thesis  the  arbitrary  algorithm  method  was  used  for  all  cases  of  photogrammetric  

reconstruction  processing.  

 

Once  the  mesh  is  constructed,  the  user  has  the  ability  to  somewhat  edit  it.  Non-­‐complex  

corrections  such  as  mesh  decimation  (simplification),  removal  of  detached  components  

and  the  closing  of  holes  can  automatically  be  performed  by  the  software  in  the  mesh  

generation  process.  

 

Furthermore  the  final  stage  of  the  model  construction  is  the  application  of  textures,  as  

this  is  not  relevant  in  this  thesis  it  will  be  ignored.    

 

4.1.3 Photogrammetric  processing:  Autodesk’s  123D  Catch  Unfortunately  very  little  information  can  be  found  about  123D  Catch’s  reconstruction  

process.  Presumably  the  process  is  somewhat  similar  to  that  of  PhotoScan.  It  can  also  be  

assumed  that  again  triangulation  is  a  key  algorithm  and  that  it  is  based  on  characteristic  

points  identified  in  the  set  of  pictures.  

 

Page 19: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  12  

The  practical  process  is  very  similar  to  PhotoScan’s.  The  user  takes  a  set  of  overlapping  

photos  of  the  object  from  different  angles.  The  difference  is  the  processing  is  done  via  

the  software  applications  server.  The  user  uploads  the  photoset  to  the  server  and  after  a  

while,  depending  on  image  size  and  quantity  the  server  returns  a  reconstructed  3D  

model.  

This  is  both  an  advantage  and  a  flaw  comparing  it  to  PhotoScan.  By  automating  the  

process  only  the  photographs  have  to  be  provided,  this  can  be  helpful  in  the  aspect  of  a  

mainstream  user  and  it  can  be  assumed  that  the  server  has  better  computational  power  

than  if  the  reconstruction  would  be  performed  locally.  However  this  limits  the  amount  of  

influence  the  user  has  on  the  reconstruction  such  as  key  parameters  for  reconstruction  

and  insight  into  the  process  and  therefore  also  the  end  result.  

 

4.2 Data  generation  -­‐  laser  scanning  approach  4.2.1 Environment  and  setup  The  scans  were  carried  out  in  Gliwice,  Poland  at  the  Institute  of  Theoretical  and  Applied  

Informatics  (part  of  the  Polish  Academy  of  Sciences)  in  one  of  their  offices.  The  laser  

scanner  was  a  Konica  Minolta  Vivid  9i  and  it  was  connected  to  a  mechanically  rotatable  

platform,  which  together  with  the  scanner  were  operated  through  the  complimentary  

native  software  for  the  scanner.    

For  the  scanner  to  encompass  the  whole  object  several  scans  had  to  be  carried  out  from  

different  angles,  this  was  controlled  with  the  rotatable  platform.  The  angle  of  rotation  

per  scan  was  30°  up  to  a  complete  circle  of  360°  resulting  in  12  overlapping  scans.

 

4.2.2 Konica  Minolta  Vivid  9i  The  scanner  used  for  the  conducted  tests  is  a  Konica  Minolta  Vivid  9i.  The  Vivid  9i  is  a  

scanner  created  for  small  to  medium  sized  objects  which  it  can  capture  with  high  detail.  

This  is  a  triangulating  laser  scanner  with  a  charged  coupled  device  (CCD)  receiver  like  

the  ones  described  in  chapter  2.3.1.  

 

4.2.3 Model  reconstruction  from  point  cloud  

When  the  laser  scanner  has  completed  its  registration  of  points  on  the  object,  each  of  the  

point  clouds  are  interpolated  individually  to  create  a  part  of  the  objects  surface.  Theses  

Page 20: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  13  

pieces  have  to  be  aggregated  into  a  single  surface  to  complete  the  process.  This  was  

done  by  aligning  each  individual  piece  in  respect  to  the  others  and  merging  them  into  a  

single  shape  to  construct  a  3D  model.  The  software  application  used  for  this  was  

Geomagic  Qualify  which  is  used  to  measure,  align  and  compare  fabricated  parts  to  their  

digital  blueprint  in  the  production  industry.    

 

     

4.3 Data  comparison  4.3.1 Alignment  

Alignment  is  a  very  important  part  of  the  measurement  process,  as  the  objects  need  to  

be  identically  aligned  for  the  distance  results  to  be  accurate.  This  is  in  turn  a  problem  as  

the  objects  vary  in  shape  and  form,  it  becomes  a  subjective  judgement  if  the  objects  are  

aligned  properly.  First  n-­‐point  alignment  is  used  for  a  rough  alignment  of  the  objects.  

This  is  based  on  a  physical  person  identifying  common  points  on  both  objects  and  

marking  these.  The  minimum  for  this  alignment  is  3  points  but  can  be  as  many  as  one  

would  like,  therefore  the  name  n-­‐point  alignment.  When  the  points  are  identified  the  

objects  are  matched  up  so  the  points  on  both  objects  align  with  eachother.  

After  this  rough  alignment  a  global  registration  algorithm  is  used.  Randomly  selected  

points  on  the  reference  surface  are  used  to  reposition  the  test  object  to  minimize  the  

overall  distance  measured  from  the  points.  This  somewhat  eliminates  the  subjective  

factor  from  the  alignment  process  but  the  first  alignment  step  is  still  based  on  human  

perception.  

 

4.3.2 Measurement  The  laser-­‐scanned  surface  is  the  reference  surface  for  our  comparison.  The  other  

surfaces  reconstructed  with  the  help  of  the  photogrammetric  applications  are  the  test  

surfaces.  For  each  comparison  an  amount  of  randomly  selected  points  on  the  reference  

surface,  presumably  depending  on  the  number  of  vertices,  are  used  to  measure  the  

distance  to  the  corresponding  point  on  the  test  surface  for  each  individual  point  with  the  

help  of  Geomagic  Qualify.  This  in  turn  generates  a  data  set  of  point  pairs  with  their  

measured  distance  that  is  used  as  the  data  for  the  quantifiable  comparison  of  the  

methods.  

Page 21: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  14  

 

Due  to  the  surfaces  being  different  in  form  and  quality  it  is  not  always  possible  to  map  a  

point  on  the  reference  surface  to  a  corresponding  location  on  the  test  surface.  In  these  

cases  the  difference  can’t  be  measured  between  the  two  points.  However  it  is  fair  to  

assume  that  the  deviation  in  the  current  point  in  this  case  is  larger  or  similar  to  the  

maximum  deviation  of  the  measured  points.  

   

Page 22: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  15  

5 Results  In  this  section  there  are  four  cases  presented  with  varying  objects  in  complexity  and  size  

to  represent  a  wide  spectrum  of  smaller  objects  that  may  be  found  in  a  home  

environment  for  amateur  3D  reconstruction.  For  each  case  a  view  of  the  original  and  all  

reconstructions  is  shown  with  a  description  of  each  surface.    Furthermore  color  maps  

are  presented  from  four  different  perspectives  of  the  deviation  between  the  compared  

surfaces.  The  deviation  is  always  measured  in  millimeters  and  the  legend  of  the  color  

map  shows  that  green  areas  are  the  ones  closest  to  zero  in  deviation.  Further  towards  

red  are  larger  positive  deviations  and  towards  blue  larger  negative  deviations.  

 

5.1 Case  1  –  Quadric  object  This  first  case  is  a  type  of  calibration  case  different  to  the  other  test  cases  that  follow  in  

this  section.  The  original  object  here  is  a  digital  3D  model  of  a  simple  quadric  surface  

(A).  The  digital  original  was  then  reproduced  into  the  physical  domain  (scale  1:1)  with  

the  help  of  a  3D  printer  (B)  to  then  again  get  reconstructed  to  digital  domain  (C&D)  with  

the  help  of  the  photogrammetric  methods  presented  earlier.  The  physical  object  is  small  

and  measures  100x95x45  mm  (height  x  width  x  depth).  

 The  interesting  thing  here  is  that  we  have  a  digital  original  giving  us  some  way  of  

comparing  the  deviation  between  the  original  and  the  reconstructed  models.  This  is  not  

possible  in  the  other  cases,  as  the  original  object  is  physical.    

 

 Figure  3  -­‐  A)  Digital  original  3D  model.  B)  Physical  3D-­‐print  based  on  A.  C)  Reconstructed  

digital  model  with  PhotoScan  based  on  B.  D)  Reconstructed  digital  model  with  123D  Catch  based  

on  B.  

 

Page 23: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  16  

 

5.1.1 3D  print  reconstructed  with  PhotoScan  

 

What  is  presented  here  are  four  different  perspectives  (top,  bottom,  isometric  and  side  

view),  of  the  deviation  between  the  original  (A)  and  the  reconstruction  with  the  help  of  

PhotoScan  (C).    

 

 Figure  4  –  Case  1,  3D  scan  compared  to  reconstruction  with  PhotoScan  

 

The  numerical  difference  between  the  digital  original  and  the  reconstructed  model  is  

very  small.  As  shown  in  the  full  table  of  points  in  appendix  chapter  9.1.1,  about  96%  of  

the  points  are  placed  in  the  interval  of  -­‐0,1386  to  0,304  mm.  So  the  numerical  difference  

over  96%  of  the  surface  is  less  than  half  of  a  millimeter.  

 

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

1.1312 -1.0498 0.1594 / -0.0785 0.1

Page 24: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  17  

 

5.1.2 3D  print  reconstructed  with  123D  Catch  

 

The  same  perspectives  as  in  the  previous  comparison  are  also  displayed  here.  This  is  the  

comparison  of  the  digital  original  (A)  and  the  reconstructed  model  with  the  help  of  123D  

Catch  (D).  

 Figure  5  –  Case  1,  3D  scan  compared  to  reconstruction  with  123D  Catch  

 

The  results  are  similar  to  the  reconstruction  with  PhotoScan  (C)  and  again  the  numerical  

difference  here  is  small,  although  a  bit  larger  than  the  results  from  PhotoScan.  What  we  

can  see  here  is  also  that  the  differences  gravitate  a  bit  more  towards  a  negative  distance  

difference  whereas  when  reconstructing  with  PhotoScan  we  see  the  differences  

gravitating  towards  a  positive  difference.  

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

0.845 -2.395 0.1825 / -0.2531 0.2282

Page 25: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  18  

5.2 Case  2  –  Angel  figure  In  this  case  the  original  object  is  a  physical  figure  of  an  angel,  which  consists  of  a  mix  of  

small  soft  and  hard  features  with  an  overall  complex  surface.  The  object  is  small  and  

measures  80x50x40  mm  (height  x  width  x  depth).  The  comparison  is  conducted  

between  the  photogrammetricly  reconstructed  models  (C&D)  from  the  two  different  

methods  and  compared  to  that  of  a  reconstructed  digital  model  from  a  3D  scanner  (B).  

 

 Figure  6  -­‐  A)  Physical  original.  B)  Digital  3D-­‐model  reconstructed  with  the  Vivid  9i  based  on  A.  C)  Reconstructed  digital  model  with  PhotoScan  based  on  A.  D)  Reconstructed  digital  model  with  123D  Catch  based  on  A  

 

 

Page 26: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  19  

5.2.1 3D  model  reconstructed  with  PhotoScan  

Here  we  see  four  different  perspectives  (front,  left,  back  and  right),  of  the  deviation  

between  the  3D-­‐scanned  reconstruction  (B)  and  the  reconstruction  with  the  help  of  

PhotoScan  (C).    

The  color  map  does  not  classify  the  grey  areas  in  this  comparison.  This  is  due  to  a  too  

large  discrepancy  between  the  two  compared  surfaces  in  the  specific  area  and  the  

deviation  measurement  algorithm  cannot  identify  the  corresponding  point  on  the  other  

surface.  Simply  put  a  point  on  surface  C  does  not  match  any  point  on  surface  B  and  

therefore  the  difference  cannot  be  measured.  

 Figure  7  -­‐  Case  2,  3D  scan  compared  to  reconstructed  model  with  PhotoScan  

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

8.5393 -9.6250 0.6051 / -0.6442 1.0395

Page 27: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  20  

The  reconstructed  model  is  very  noisy  and  distorted,  even  so  we  see  that  the  numerical  

difference  over  most  areas  of  the  surface  is  still  small,  only  0.5  mm.  With  that  said  it  is  

important  to  acknowledge  the  grey  areas  where  points  were  unable  to  match  we  can  

assume  the  difference  is  greater  than  +/-­‐5  mm.  

 

5.2.2 3D  model  reconstructed  with  123D  Catch  The  same  perspectives  (front,  left,  back  and  right),  of  the  deviation  between  the  3D-­‐

scanned  reconstruction  (B)  and  the  reconstruction  with  the  help  of  123D  Catch  (D).  As  

previously  the  color  map  does  not  classify  the  grey  areas  due  to  the  same  reasons  in  

5.2.1.  

 Figure  8  -­‐  Case  2,  3D  scan  compared  to  reconstructed  model  with  123D  Catch  

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

3.7093 -3.6926 0.6848 / -0.4766 0.8400

Page 28: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  21  

 

Although  both  photogrammetric  reconstructions  contain  artifacts  and  some  noise  the  

key  thing  to  notice  here  is  that  there  are  larger  areas  on  the  test  surface  (D)  which  can’t  

be  matched  to  the  reference  surface  (B)  than  in  the  previous  comparison,  C  vs  B.    This  

has  a  significant  impact  on  the  key  values  like  average  deviation  as  pieces  of  the  dataset  

are  missing.  

 

5.3 Case  3  –  Monkey  figure  As  in  the  previous  case  the  original  object  here  is  a  physical  figure  of  three  monkeys,  

which  consist  of  mostly  small  sharp  features.  The  overall  surface  of  the  object  is  very  

complex  with  a  high  amount  of  detail.    The  object  measures  70x105x50  mm  (height  x  

width  x  depth).  A  comparison  is  performed  between  the  photogrammetricly  

reconstructed  models  from  the  two  different  methods  (C&D)  and  compared  to  that  of  a  

reconstructed  digital  model  from  a  3D  scanner  (B).  

 

 Figure  9  -­‐  A)  Physical  original  object.  B)  Digital  3D-­‐model  reconstructed  with  the  Vivid  9i  based  on  A,  C)  Reconstructed  digital  model  with  PhotoScan  based  on  A.  D)  Reconstructed  digital  model  with  123D  Catch  based  on  A.  

Page 29: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  22  

 

 

5.3.1 3D  model  reconstructed  with  PhotoScan  Four  perspectives  (front,  left,  back  and  right),  of  the  deviation  between  the  3D-­‐scanned  

reconstruction  (B)  and  the  reconstruction  with  the  help  of  PhotoScan  (C)  are  shown  

here.  

 

 Figure  10  –  Case  3,  3D  scan  compared  to  reconstructed  model  with  PhotoScan  

 

This  statue  has  a  similar  size  comparable  to  Case  2  yet  the  discrepancies  here  are  

significantly  smaller.  The  deviation  is  also  very  evenly  spread  over  the  whole  object  

suggesting  there  is  little  to  no  noise.  

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

3.6367 -3.0904 0.1769 / -0.1312 0.1854

Page 30: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  23  

5.3.2 3D  model  reconstructed  with  123D  Catch  

 

The  same  perspectives  (front,  left,  back  and  right).  The  deviation  here  is  larger  than  in  

the  comparison  of  B  vs  C  and  the  deviation  is  not  as  evenly  spread  out  as  in  5.3.1.  

 

 Figure  11  –  Case  3,  3D  scan  compared  to  reconstructed  model  with  123D  Catch  

 

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

6.1359 -7.7362 0.8444 / -1.3935 1.4147

Page 31: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  24  

5.4 Case  4  –  Wooden  cat  This  case  is  unique  in  the  sense  that  the  object  here  is  significantly  larger  than  the  

previous  cases.  The  object  measures  410x150x50  mm  (height  x  width  x  depth).  As  

before  in  the  previous  cases  the  original  object  here  is  a  physical  figure  of  cat  carved  in  

wood  that  consists  mostly  of  large  soft  features  and  is  quite  simplistic  in  shape.  The  

comparison  is  conducted  between  the  photogrammetricly  reconstructed  models  from  

the  two  different  methods  (C&D)  and  compared  to  that  of  a  reconstructed  digital  model  

from  a  3D  scanner  (B).  

 

 Figure  12  -­‐  A)  Physical  original  3D  model.  B)  Digital  3D-­‐model  reconstructed  with  the  Vivid  9i  

based  on  A.  C)  Reconstructed  digital  model  with  PhotoScan  based  on  A.  D)  Reconstructed  digital  

model  with  123D  Catch  based  on  A.  

 

 

5.4.1 3D  model  reconstructed  with  PhotoScan  

We  can  see  from  the  color  map  that  a  lot  of  noise  has  distorted  the  model  quite  heavily  

in  some  areas  yet  still  most  of  the  surface  only  has  a  small  deviation  of  about  1mm  which  

may  seem  like  a  lot  compared  to  the  earlier  cases  but  taking  into  account  the  size  of  the  

object  it  is  actually  a  very  small  deviation.    

Page 32: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  25  

 Figure  13  –  Case  4,  3D  scan  compared  to  reconstructed  model  with  PhotoScan  

 

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

20.3581 -20.3864 2.0123 / -1.7814 3.2267

Page 33: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  26  

5.4.2 3D  model  reconstructed  with  123D  Catch  

This  reconstructed  model  is  not  as  noisy  as  the  previous  case  and  the  discrepancies  are  

much  more  evenly  grouped.  Again  even  though  some  areas  are  quite  deviant  most  of  the  

surface  has  a  very  small  deviation  of  0.5  mm.  

 

 Figure  14  –  Case  4,  3D  scan  compared  to  reconstructed  model  with  123D  Catch  

 

 

   

Deviation (mm) Max. Upper Deviation

Max. Lower Deviation

Average Deviation

Standard Deviation

10.6311 -7.8892 0.6068 / -0.9564 1.1290

Page 34: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  27  

6 Analysis  and  discussion  These  comparisons  are  carried  out  between  the  3D  laser  scan  of  the  object  and  the  

photogrammetric  reconstruction  so  small  numerical  differences  do  not  always  

correspond  to  an  ideal  reconstruction  as  errors  might  very  well  exist  in  the  laser  scan  of  

the  object  (Spytkowska  2008).  This  is  not  a  point  that  has  or  will  be  explored  further  in  

this  thesis,  as  the  goal  is  only  to  compare  the  methods  with  each  other  and  not  with  the  

physical  original  on  a  quantifiable  level.  As  explained  earlier  this  is  not  possible  in  most  

cases  and  neither  the  goal  of  this  thesis.  Nevertheless  it  is  important  to  keep  that  in  mind  

when  analyzing  these  results.  We  cannot  assume  the  digital  laser  scan  of  the  object  is  

ideal  but  it  is  the  reference  for  comparison  of  the  photogrammetric  reconstruction  

methods.  

 

6.1 Comparing  the  results  of  the  photogrammetric  

reconstructions  and  the  laser  scanned  reconstructions  Although  the  numerical  differences  in  all  these  cases  are  by  comparison  relatively  small  

excluding  some  minor  areas  it  is  important  to  note  the  difference  in  perceived  quality  of  

the  objects.  Looking  at  the  results,  the  conclusion  that  lower  average  deviation  also  

correlates  to  a  better-­‐perceived  quality  can  be  said  to  be  true.  Furthermore  we  can  see  

that  the  more  evenly  the  deviation  is  spread  out  over  the  surface  the  better  the  

perceived  quality  of  the  reconstruction.  Even  if  large  areas  are  quite  off,  the  key  factor  

affecting  the  perceived  quality  is  noise,  and  large  areas  with  similar  deviation  often  

suggest  small  amounts  of  noise  or  at  least  a  uniform  distortion.  This  of  course  assumes  

that  the  3D  laser  scanner  has  reconstructed  the  physical  original  in  a  satisfying  manner,  

which  is  assumed  here  as  discussed  earlier,  since  it  is  these  two  surfaces  we  are  

comparing  and  not  the  original.  These  conclusions  were  expected  but  cannot  be  said  to  

always  be  true  since  this  is  a  subjective  judgment  and  the  perception  of  quality  might  

vary  for  each  individual.  

 

Page 35: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  28  

6.2 Strengths  and  weaknesses  of  the  photogrammetric  

reconstruction  software  applications  It  is  also  interesting  to  try  and  analyze  the  strengths  and  weaknesses  of  the  algorithms  

of  the  two  photogrammetric  software  applications  in  their  ability  to  reconstruct  the  

surface  of  the  object.  In  general  we  can  see  that  the  algorithms  struggle  when  faced  with  

areas  and  whole  surfaces  that  lack  large  or  multiple  changes,  in  other  words  smooth  soft  

features  and  same  colored  surfaces  as  seen  in  case  2  and  4.  The  type  of  change  whether  

it  is  surface  based  or  color  based  seems  to  be  of  less  importance.  As  we  can  see  in  case  1,  

the  simple  and  smooth  surface  of  the  quadric  is  compensated  by  the  complex  pattern  

painted  on  the  surface.  Furthermore  surface  differences  like  the  fur  of  the  monkeys  in  

case  3  also  result  in  the  same  type  of  change  such  as  color  changes  due  to  lightning  

giving  the  features  a  shadow  and  thus  a  different  intensity.  Therefore  changes  in  either  

surface  or  color  result  in  a  similar  difference  on  a  photograph  and  it  can  be  assumed  that  

this  is  what  the  algorithms  use  to  identify  points  of  measurement  for  the  triangulation  as  

mentioned  in  chapter  2.  This  property  will  be  named  Δ  (Delta);  objects  with  a  high  

amount  of  Δ  produce  better  results  when  reconstructed.  

 

The  cat,  case  4,  is  a  good  example  where  we  see  quite  small  Δ,  as  changes  in  color  and  

surface  over  large  areas  contribute  to  confusion  for  the  photogrammetric  applications  as  

they  most  likely  have  no  way  of  identifying  the  points  on  the  surface  and  therefore  

struggle  aggregating  the  orientation  of  the  respective  images.    

 

We  can  also  come  to  the  conclusion  here  that  Δ  is  not  the  only  factor  for  good  

photogrammetric  reconstructions  but  also  the  density  of  the  change.  If  we  look  at  case  2,  

the  angel  figure  has  some  areas  with  soft  round  features  (low  Δ)  and  some  areas  like  the  

hair  with  high  Δ.  The  areas  with  high  Δ  have  been  reconstructed  quite  well  however  the  

lack  of  surface  nearby  with  high  amounts  of  Δ  create  a  challenge  for  the  reconstruction  

algorithm  and  the  result  is  quite  poor.  The  conclusion  that  can  be  drawn  from  this  is  that  

even  if  an  object  has  a  high  amount  of  Δ  it  has  to  be  spread  out  over  the  whole  surface  

for  a  good  reconstruction,  concentrated  amounts  will  only  give  a  good  result  in  that  

specific  area.  The  density  of  Δ  will  be  called  ρ  (rho).  

 

Page 36: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  29  

It  is  also  clear  that  there  is  a  difference  in  the  algorithms  between  the  two  applications.  

PhotoScan’s  reconstruction  generates  a  substantial  amount  of  noise  but  still  manages  a  

quite  small  overall  numerical  deviation  over  the  whole  surface,  suggesting  a  more  

intense  frequency  of  points  triangulated  or  a  large  difference  in  the  interpolation  

between  points.  In  the  case  of  123D’s  reconstruction  the  points  calculated  are  probably  

fewer,  this  contributes  to,  in  most  cases,  that  the  perceived  quality  is  much  higher  due  to  

the  lower  amount  of  noise  and  it  still  captures  the  overall  features.  However  it’s  

important  to  note  that  in  the  best  reconstruction,  case  3,  PhotoScans  point  frequency  

generates  a  more  accurate  reconstruction  by  capturing  far  more  detail  than  123D  Catch  

thanks  to  the  larger  amount  of  points  calculated,  although  as  mentioned  above,  this  is  

not  always  an  advantage  as  errors  can  become  plentiful.    

 

Another  interesting  difference  between  the  algorithms  is  that  PhotoScans  deviation  is  

usually  positive,  differences  between  the  reconstructed  surface  point  outwards  from  the  

object  however  123Ds  deviation  is  often  negative,  pointing  inwards  from  the  surface.  

This  results  in  PhotoScans  reconstructions  generally  ending  up  larger  in  size  than  their  

equivalent  laser  scan  and  123D  reconstruction  ending  up  smaller  than  their  counterpart.  

This  is  presumably  a  side  effect  of  each  applications  calculation  algorithm  and  it  is  hard  

to  know  if  it  is  related  in  any  way  to  the  perceived  quality  of  the  object  although  it  can  

have  some  impact  when  trying  to  recreate  objects  made  to  scale.  

 

6.3 The  photogrammetric  reconstruction  process  as  a  whole  Photogrammetry,  as  can  probably  derived  from  just  the  name,  is  highly  dependent  on  

the  input  data,  in  other  words  the  photographs  quality.  This  is  not  the  same  subjective  

qualities  of  the  human  perception  of  a  “nice”  picture,  but  rather  good  quality  as  in  

maximum  preservation  of  unaltered  visual  information  as  described  in  chapter  2.4.1  and  

4.4.1.  This  is  the  main  bottleneck  for  this  technology  and  as  photographs  with  altered  or  

low  amount  of  visual  information  also  give  bad  results  when  processed  by  the  

reconstruction  algorithm.  Bad  input  can  be  somewhat  mitigated  with  the  help  of  high  

amounts  of  Δ  and  ρ  on  the  object  but  only  to  some  extent.  This  thesis  focus  was  more  on  

the  results  of  photogrammetric  reconstruction  and  comparing  those  to  that  of  

reconstruction  by  3D  laser  scanning.    Even  though  the  results  are  fairly  impressing  in  

some  cases,  not  only  numerically  but  especially  in  perceived  quality,  such  as  case  3,  it  is  

Page 37: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  30  

fair  to  say  that  more  work  and  time  should  have  been  focused  on  the  process  of  

generating  the  input  data,  as  this  generates  the  biggest  differences  in  quality  in  the  

reconstruction  process.  Unfortunately  this  is  very  time  consuming  and  somewhat  

beyond  the  goal  of  this  thesis  as  we  focus  on  the  whole  process  rather  than  the  details  of  

each  part,  but  nevertheless  this  is  the  most  important  and  the  most  obvious  conclusion  

that  can  be  drawn  from  these  cases.    

 

The  dependency  on  input  data  goes  so  far  that  even  lack  of  data  is  better  than  low  

quality  data.  What  this  means  is  that  photographs  with  low  quality  visual  information  

such  as  a  blurry  or  heavily  shadowed  photographs  do  more  harm  than  good.  This  is  due  

to  the  need  to  aggregate  information  from  several  photographs  during  the  

reconstruction  algorithm.    Visual  information  of  poor  quality  can  confuse  the  algorithm  

during  the  aggregation  process.  In  the  attempt  to  identify  expected  features  or  rather  

points,  based  on  assumptions  from  earlier  presumably  proper  visual  information,  they  

cant  be  recognized  in  the  following  picture  as  they  differ  in  some  form,  even  though  it  is  

to  the  human  eye  the  same  point.  This  in  the  worst-­‐case  scenario  leads  to  earlier  proper  

visual  information  getting  discarded  and  also  becoming  useless.  

 

Photogrammetry  is  all  about  good  input  data,  and  as  mentioned  before  –  if  a  photograph  

was  an  ideal  medium  for  information  and  no  information  was  lost  in  the  transformation  

from  three  dimensions  into  two,  the  reconstruction  process  for  photogrammetry  

wouldn’t  need  more  than  a  single  photograph  –  unfortunately  this  is  not  the  case.  

 

6.4 Can  photogrammetry  yield  similar  results  to  3D  laser  

scanning  when  used  in  an  amateur  home  setting  for  

smaller  objects?  The  short  and  simple  answer  to  this  question  is  yes.  In  case  three  with  the  monkey  

statues  we  see  a  near  ideal  reconstruction,  not  only  in  numerical  terms  but  also  in  

perceived  visual  quality  of  the  object.  However  there  are  three  main  factors  along  the  

way  that  have  a  significant  impact  on  the  results  of  the  reconstruction.  The  quality  of  the  

visual  information  preserved  in  the  photograph,  focus  and  lightning  are  two  big  

components  in  maximizing  captured  visual  data.  Secondly,  the  features  of  the  object  

Page 38: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  31  

either  form,  color  and  intensity  –  Δ  and  the  density  of  these  changes  –  ρ,  enabling  the  

reconstruction  algorithm  to  establish  a  clear  and  robust  set  of  points  to  use  for  

triangulation  calculations.    The  third  and  final  piece  is  the  algorithm  used  for  

reconstruction  and  calculation  of  the  points,  the  transformation  from  two-­‐dimensional  

coordinates  into  three-­‐dimensional  ones.  This  is  also  where  some  questions  are  still  

unanswered  as  the  algorithms  of  these  commercial  software  applications  are  not  public  

domain  and  have  not  been  investigated  in  depth  in  this  thesis.  These  components  need  

to  be,  not  just  average,  but  of  good  quality  for  the  reconstruction  to  yield  similar  results  

to  that  of  a  good  3D  laser  scanner.  

 

7 Conclusion  This  thesis  has  described  the  background  and  general  technology  behind  both  methods,  

3D  scanning  with  laser  and  photogrammetry,  as  means  of  reproducing  physical  objects  

in  digital  three-­‐dimensional  space.  The  focus  has  been  to  compare  if,  with  the  help  of  

photogrammetry  software  in  a  non-­‐professional  setting,  it  is  possible  to  get  results  of  

similar  quality  to  that  of  a  laser  scanning  approach  which  has  been  the  preferred  

method  for  some  time  when  reconstructing  smaller  objects.  This  has  been  tested  with  

four  cases  of  objects  of  varying  surface  features  and  size,  these  four  cases  are  not  enough  

to  draw  generalized  conclusions  but  still  give  a  good  spread  of  data  as  to  see  where  

these  methods  strengths  and  weaknesses  lie  and  what  parameters  affect  this  type  of  

reconstruction.    

The  results  from  the  cases  indicate  that  it  is  indeed  possible  with  the  help  of  

photogrammetry  to  reconstruct  a  model  of  very  similar  quality  to  that  of  one  

reconstructed  with  a  laser  scanner.  However  it  is  also  obvious  that  these  good  results  

are  not  consistent  and  need  not  only  good  external  conditions  but  also  rely  on  the  shape  

and  features  of  the  objects  surface.  Objects  with  unique  non-­‐repeatable  variations  are  

the  ones  that  produce  the  best  results  when  reconstructed.  It  does  not  seem  to  matter  

whether  these  variations  are  physical,  color  or  intensity  based,  the  key  result  is  

differentiated  points  or  pixels  on  the  resulting  photograph  that  provide  the  algorithm  

with  highly  varying  points  used  to  identify  and  match  these  in  the  following  

photographs.    

 

Page 39: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  32  

It  is  fair  to  say  that  amateur  photogrammetry  can  yield  very  good  results  when  used  for  

digital  reconstruction  of  3D  objects  although  as  mentioned  before  these  results  of  good  

quality  are  not  very  consistent  and  require  a  certain  amount  of  variables.  If  these  

prerequisites  are  not  matched  the  results  in  most  cases  are  very  underwhelming  

compared  to  those  of  a  laser  scanner.  However  strictly  numerically  the  results  are  not  

very  far  off  and  it  is  debatable  if  the  methods  produce  comparable  results  but  when  we  

weigh  in  the  perceived  quality  of  the  object  the  results  are  usually  underwhelming  in  an  

amateur  setting.  

 

7.1 Future  research  Three-­‐dimensional  reconstruction  from  photogrammetry  is  still  relatively  new  

compared  to  other  reconstruction  techniques  and  it  is  interesting  to  see  what  can  be  

done  to  improve  these  methods  in  the  future,  as  there  is  certainly  potential  for  these  

methods.  For  further  research  on  this  topic  it  would  be  interesting  to  look  at  

experiments  with  more  focus  on  optimizing  the  process  of  photography,  as  this  is  the  

core  of  photogrammetric  reconstruction  –  the  input  data.  

 

Another  interesting  angle  of  approach  would  be  to  compare  amateur  laser  scanning  with  

the  results  of  amateur  photogrammetry.  As  in  this  thesis  the  focus  has  been  on  amateur  

photogrammetry  results  compared  to  professional  laser  scanning  it  might  be  an  unfair  

comparison  of  the  two  techniques.    Furthermore  there  seems  to  be  a  drive  to  produce  

more  and  more  equipment  for  3D  reconstruction  on  an  amateur  level  as  demand  for  

both  3D  printers  and  3D  scanners  rise.  

   

Page 40: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  33  

8 References  

Agisoft  LLC,  2011.  Agisoft  PhotoScan  User  Manual.  ,  p.37.  Available  at:  www.agisoft.ru.  

Baltsavias,  E.,  1999.  A  comparision  between  photogrammetry  and  laser  scanning.  ISPRS  Journal  of  Photogrammetry  and  Remote  Sensing,  54,  pp.83–94.  

Benard,  M.,  1984.  Automatic  stereophotogrammetry:  a  method  based  on  feature  detection  and  dynamic  programming.  Photogrammetria,  39,  pp.169–181.  Available  at:  http://www.sciencedirect.com/science/article/pii/0031866384900164  [Accessed  December  5,  2014].  

Blizard,  B.,  2014.  The  Art  of  Photogrammetry:  Introduction  to  Software  and  Hardware.  Available  at:  http://www.tested.com/art/makers/460057-­‐tested-­‐dark-­‐art-­‐photogrammetry/  [Accessed  March  2,  2015].  

Chen,  F.,  Brown,  G.  &  Song,  M.,  2000.  Overview  of  three-­‐dimensional  shape  measurement  using  optical  methods.  Optical  ….  Available  at:  http://opticalengineering.spiedigitallibrary.org/article.aspx?articleid=1075937  [Accessed  December  5,  2014].  

Crum,  R.,  2014.  Consumers  spurring  3D  printing  growth.  Dec  9,  2014  3:08  p.m.  ET.  Available  at:  http://www.marketwatch.com/story/consumers-­‐spurring-­‐3d-­‐printing-­‐growth-­‐2014-­‐12-­‐09.  

Dold,  C.  &  Brenner,  C.,  2006.  Registration  of  terrestrial  laser  scanning  data  using  planar  patches  and  image  data.  International  Archives  of  Photogrammetry,  …,  Vol.  XXXVI,  pp.78–83.  Available  at:  http://www.ikg.uni-­‐hannover.de/fileadmin/ikg/staff/publications/Konferenzbeitraege_full_review/DOLD_IAPRS06.pdf.  

Fassi,  F.  &  Fregonese,  L.,  2013.  BETWEEN  LASER  SCANNING  AND  AUTOMATED  3D  MODELLING  TECHNIQUES  TO  RECONSTRUCT  COMPLEX  AND  EXTENSIVE  CULTURAL  HERITAGE.  ISPRS  3DArch,  Trento,  …,  XL(February),  pp.25–26.  Available  at:  http://www.researchgate.net/publication/235742814_COMPARISON_BETWEEN_LASER_SCANNING_AND_AUTOMATED_3D_MODELLING_TECHNIQUES_TO_RECONSTRUCT_COMPLEX_AND_EXTENSIVE_CULTURAL_HERITAGE_AREAS/file/d912f51307c8faac65.pdf  [Accessed  March  19,  2014].  

Greve,  C.,  1996.  Digital  photogrammetry :  an  addendum  to  the  manual  of  photogrammetry,  Bethesda :  American  Society  for  Photogrammetry  and  Remote  Sensing.  

Matter  and  Form,  2014.  The  Matter  and  Form  3D  scanner.  Available  at:  https://matterandform.net/scanner  [Accessed  March  2,  2015].  

Page 41: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  34  

Mayer,  R.,  1999.  Scientific  Canadian:  Invention  and  Innovation  From  Canada’s  National  Research  Council,  Vancouver:  Raincoast  Books.  

Poznanski,  A.,  2014.  VISUAL  REVOLUTION  OF  THE  VANISHING  OF  ETHAN  CARTER.  Available  at:  http://www.theastronauts.com/2014/03/visual-­‐revolution-­‐vanishing-­‐ethan-­‐carter/  [Accessed  March  2,  2015].  

Schenk,  T.,  2005.  Introduction  to  Photogrammetry.  Department  of  Civil  and  Environmental  Engineering  and  Geodetic  Science,  The  Ohio  State  University,  pp.79–95.  Available  at:  http://gscphoto.ceegs.ohio-­‐state.edu/courses/GeodSci410/docs/GS410_02.pdf.  

Seitz,  S.  &  Curless,  B.,  2006.  A  comparison  and  evaluation  of  multi-­‐view  stereo  reconstruction  algorithms.  Computer  vision  and  ….  Available  at:  http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1640800  [Accessed  May  12,  2014].  

Slama,  C.C.,  Theurer,  C.  &  Henriksen,  S.W.,  1980.  Manual  of  photogrammetry,  

Spytkowska,  A.,  2008.  The  exploitation  of  Konica  Minolta  Vivid  9I  scanner  for  creating  a  virtual  copy  of  small  museum  objects.  Akademia  Górniczo-­‐Hutnicza.  

Wolff,  E.,  2004.  “Spider-­‐Man  2”:  A  Conversation  with  Visual  Effects  Guru  John  Dykstra.  Available  at:  http://www.awn.com/vfxworld/spider-­‐man-­‐2-­‐conversation-­‐visual-­‐effects-­‐guru-­‐john-­‐dykstra.  

   

Page 42: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  35  

9 Appendix  9.1 Case  1  –  deviation  results  9.1.1 Photoscan  model  compared  to  3D  scanned  model  

 

9.1.2 123D  Catch  model  compared  to  3D  scanned  model  

Deviation  interval  (mm)   #  Points   %  amount  -­‐0.8782  to  -­‐0.4990    1616    1.4827    -­‐0.4990  to  -­‐0.1198    66074    60.6217    -­‐0.1198  to  0.1198    23582    21.6361    0.1198  to  0.4990    16476    15.1164    0.4990  to  0.8782    1245    1.1423    

   

Deviation  interval  (mm)   #  Points   %  amount  -­‐1.1312  to  -­‐0.9658   7   0.0027  -­‐0.9658  to  -­‐0.8003   8   0.0031  -­‐0.8003  to  -­‐0.6349   6   0.0024  -­‐0.6349  to  -­‐0.4695   45   0.0176  -­‐0.4695  to  -­‐0.304   526   0.2061  -­‐0.304  to  -­‐0.1386   3268   1.2804  -­‐0.1386  to  0.1386   100728   39.4664  0.1386  to  0.304   144364   56.5634  0.304  to  0.4695   5774   2.2623  0.4695  to  0.6349   448   0.1755  0.6349  to  0.8003   47   0.0184  0.8003  to  0.9658   1   0.0004  0.9658  to  1.1312   2   0.0008  

Page 43: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  36  

 

9.2 Case  2  –  deviation  results  9.2.1 Photoscan  model  compared  to  3D  scanned  model  

 Deviation  interval  (mm)    #  Points    %  amount    -­‐5.0000  to  -­‐4.2469    175    0.2408    -­‐4.2469  to  -­‐3.4937    231    0.3179    -­‐3.4937  to  -­‐2.7406    232    0.3193    -­‐2.7406  to  -­‐1.9875    772    1.0624    -­‐1.9875  to  -­‐1.2344    3677    5.0600    -­‐1.2344  to  -­‐0.4813    7630    10.4998    -­‐0.4813  to  0.4813    45926    63.1998    0.4813  to  1.2344    8499    11.6957    1.2344  to  1.9875    2549    3.5077    1.9875  to  2.7406    1352    1.8605    2.7406  to  3.4938    721    0.9922    3.4938  to  4.2469    408    0.5615    4.2469  to  5.0000    156    0.2147    

9.2.2 123D  Catch  model  compared  to  3D  scanned  model  

 Deviation  interval  (mm)    #  Points    %  amount    -­‐4.2042  to  -­‐3.4084    25    0.0602    -­‐3.4084  to    -­‐2.6127    41    0.0987    -­‐2.6127  to  -­‐1.8169    89    0.2143    -­‐1.8169  to  -­‐1.0211    1310    3.1544    -­‐1.0211  to  -­‐0.2253    9811    23.6245    -­‐0.2253  to  0.2253    11712    28.2020    0.2253  to  1.0211    13912    33.4995    1.0211  to  1.8169    2411    5.8056    1.8169  to  2.6127    1205    2.9016    2.6127  to  3.4084    769    1.8517    3.4084  to  4.2042    244    0.5875    

   

Page 44: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  37  

9.3 Case  3  –  deviation  results  9.3.1 Photoscan  model  compared  to  3D  scanned  model    Deviation  interval  (mm)    #  Points    %  amount    -­‐3.6367  to  -­‐3.0609    1    0.0006    -­‐3.0609  to  -­‐2.4851    2    0.0012    -­‐2.4851  to  -­‐1.9093    0    0.0000    -­‐1.9093  to-­‐1.3334    1    0.0006    -­‐1.3334  to  -­‐0.7576    3    0.0018    -­‐0.7576  to  -­‐0.1818    16122    9.4107    -­‐0.1818  to  0.1818    107687    62.8591    0.1818  to  0.7576    47352    27.6403    0.7576  to  1.3334    143    0.0835    1.3334  to  1.9093    2    0.0012    1.9093  to  3.0609    0    0.0000    3.0609  to  3.6367    2    0.0012    

9.3.2 123D  Catch  model  compared  to  3D  scanned  model  

Deviation  interval  (mm)    #  Points    %  amount    -­‐7.7362  to  -­‐6.5113    2    0.0023    -­‐6.5113  to-­‐5.2864    0    0.0000    -­‐5.2864  to  -­‐4.0615    2653    2.9958    -­‐4.0615  to  -­‐2.8366    6492    7.3308    -­‐2.8366  to  -­‐1.6117    12105    13.6690    -­‐1.6117  to  -­‐0.3868    37165    41.9668    -­‐0.3868  to  0.3868    15998    18.0650    0.3868  to  1.6117    10853    12.2552    1.6117  to  2.8366    2943    3.3232    2.8366  to  4.0615    294    0.3320    4.0615  to  5.2864    51    0.0576    5.2864  to  6.5113    2    0.0023    6.5113  to  7.7362    0    0.0000    

   

Page 45: Photogrammetric software as an alternative to 3D laser ...826106/FULLTEXT01.pdf · DEGREE PROJECT, IN MEDIA TECHNOLOGY , SECOND LEVEL STOCKHOLM, SWEDEN 2015 Photogrammetric software

 

  38  

9.4 Case  4  –  deviation  results  9.4.1 Photoscan  model  compared  to  3D  scanned  model  

 Deviation  interval  (mm)    #  Points    %  amount    -­‐20.3864  to  -­‐17.1585    280    0.1521    -­‐17.1585  to    -­‐13.9307    1247    0.6774    -­‐13.9307  to  -­‐10.7028    1687    0.9164    -­‐10.7028  to  -­‐7.4750    2759    1.4988    -­‐7.4750  to  -­‐4.2472    5773    3.1361    -­‐4.2472  to    -­‐1.0193    46904    25.4802    -­‐1.0193  to  1.0193    100389    54.5355    1.0193  to  4.2472    16875    9.1672    4.2472  to  7.4750    4868    2.6445    7.4750  to  10.7028    1761    0.9566    10.7028  to  13.9307    826    0.4487    13.9307  to  17.1585    528    0.2868    17.1585  to  20.3864    183    0.0994    

9.4.2 123D  Catch  model  compared  to  3D  scanned  model  

Deviation  interval  (mm)    #  Points    %  amount    -­‐8.9478  to  -­‐7.2646    5    0.0074    -­‐7.2646  to  -­‐5.5813    35    0.0518    -­‐5.5813  to  -­‐3.8981    233    0.3450    -­‐3.8981  to  -­‐2.2148    4101    6.0715    -­‐2.2148  to  -­‐0.5316    19616    29.0414    -­‐0.5316  to  0.5316    33827    50.0807    0.5316  to  2.2148    8675    12.8433    2.2148  to  3.8981    872    1.2910    3.8981  to  5.5813    137    0.2028    5.5813  to  7.2646    29    0.0429    7.2646  to  8.9478    10    0.0148    8.9478  to  10.6311    4    0.0059