Photo Cloud Modeling

download Photo Cloud Modeling

of 54

Transcript of Photo Cloud Modeling

  • 8/10/2019 Photo Cloud Modeling

    1/54

    Yoshinori Dobashi (Hokkaido University)

    Yusuke Shinzo (Hokkaido University)

    Tsuyoshi Yamamoto (Hokkaido University)

    Modeling of Clouds from

    a Single hotogra!h

  • 8/10/2019 Photo Cloud Modeling

    2/54

    Overview

    "ntrodu#tion$elated %orkro!osed Method$esultsCon#lusion

  • 8/10/2019 Photo Cloud Modeling

    3/54

  • 8/10/2019 Photo Cloud Modeling

    4/54

    Introduction

    ro#edural a!!roa#h

    Use of sim!le heuristi#ally'defined rules

    &o #om!utational #ost

    Diffi#ult to adust !arameters

    hysi#ally'based a!!roa#h

    Simulating !hysi#al !ro#ess of #loud formation

    $ealisti# #louds

    High #om!utational #ost

  • 8/10/2019 Photo Cloud Modeling

    5/54

    Our Approach

    "mage'based a!!roa#h Use of a single !hotogra!h

    *ot to re#onstru#t the same #louds

    Using the !hoto as a guide to synthesizesimilar #louds

    Three ty!es of #louds

    #irrus alto#umulus #umulus

  • 8/10/2019 Photo Cloud Modeling

    6/54

    Overview

    "ntrodu#tion$elated %orkro!osed Method$esultsCon#lusion

  • 8/10/2019 Photo Cloud Modeling

    7/54

    Related Work

    ro#edural modeling +ra#tals ,-os./0

    Te1tured elli!soids ,2ar.30

    Metaballs 4 noise fun#tion ,5be670

    S!e#tral synthesis ,Sak6/0

    $eal'time modeling8rendering system ,SS5H9/0

    ,2ar.30 ,5be670 ,SS5H9/0

    Diffi#ult to adust !arameters

  • 8/10/2019 Photo Cloud Modeling

    8/54

    Related Work

    hysi#ally'based modeling

    *umeri#al solution of atmos!heri# fluid dynami#s

    ,:H.;< MYD*9=< MYD*9>0

    Controlling #loud simulation ,D:*Y9.0

    ,:H.;0 ,MYD*9=0 ,MYD*9>0 ,D:*Y9.0

    High #om!utational #ost

  • 8/10/2019 Photo Cloud Modeling

    9/54

    Related Work

    "mage'based modeling Use of infrared satellite images ,D*Y?6.0

    *ot a!!li#able to !hoto taken from the ground

    *o #loud ty!es

    satellite image synthesized #louds

  • 8/10/2019 Photo Cloud Modeling

    10/54

    Overview

    "ntrodu#tion$elated %orkro!osed Method$esultsCon#lusion

  • 8/10/2019 Photo Cloud Modeling

    11/54

    Overview of Our Method

    #irrus

    alto'

    #umulus

    #umulus

  • 8/10/2019 Photo Cloud Modeling

    12/54

    Overview of Our Method

    #irrus

    alto'

    #umulus

    #umulus

    Cal#ulation of #loud image

    (#ommon !ro#ess)

    Cal#ulation of #loud image

    (#ommon !ro#ess)

  • 8/10/2019 Photo Cloud Modeling

    13/54

    Calculation of Cloud Image

    ?vervie

    #irrus

    alto#umulus

    #umulus

    intensity o!a#ity

    intensity o!a#ity

    intensity o!a#ity

    (in!ut !hotogra!hs) (#loud images)

  • 8/10/2019 Photo Cloud Modeling

    14/54

    Calculation of Cloud Image

    ro#esses

    in!ut image sky image

    intensity o!a#ity

    =@ $emoving #loud !i1els >@ "nter!olating sky #olor

    /@ #al#ulation of intensity8o!a#ity

  • 8/10/2019 Photo Cloud Modeling

    15/54

    =@ $emoving #loud !i1els

    Calculation of Cloud Image

    ro#esses

    in!ut image sky image

    intensity o!a#ity

    >@ "nter!olating sky #olor

    /@ #al#ulation of intensity8o!a#ity

  • 8/10/2019 Photo Cloud Modeling

    16/54

  • 8/10/2019 Photo Cloud Modeling

    17/54

    Calculation of Cloud Image

    ro#esses

    in!ut image sky image

    intensity o!a#ity

    =@ $emoving #loud !i1els >@ "nter!olating sky #olor

    /@ #al#ulation of intensity8o!a#ity

    = cloud

    image

  • 8/10/2019 Photo Cloud Modeling

    18/54

    Removing Cloud Pixels

    Use of #hroma to identify #loud !i1els Clouds are generally hite (or gray)

    $emove !i1el if #hroma A threshold

    "n!ut !hotogra!h removed #loud !i1els

  • 8/10/2019 Photo Cloud Modeling

    19/54

    Calculation of k! Image

    "nter!olation of sky #olors by solving

    oisson eBuation

    0),( = csky pI

    sky image

    #loud !i1els pc

    pc: cloud pixel

    = R, G, B

  • 8/10/2019 Photo Cloud Modeling

    20/54

    Calculation of Cloud Image

    "ntensity of #louds (single s#attering)

    viewpoint

    sun

    sk!

    cloud

  • 8/10/2019 Photo Cloud Modeling

    21/54

    Calculation of Cloud Image

    "ntensity of #louds (single s#attering)

    viewpoint

    sun

    sk!

    Isun

    Isuncloud

  • 8/10/2019 Photo Cloud Modeling

    22/54

    Calculation of Cloud Image

    "ntensity of #louds (single s#attering)

    viewpoint

    sun

    sk!

    Isky

    Isun

    Isky

    Isuncloud

  • 8/10/2019 Photo Cloud Modeling

    23/54

    Calculation of Cloud Image

    "ntensity of #louds (single s#attering)

    viewpoint

    sun

    sk!

    IskyIcld

    Isun

    Isky

    Isuncloud

    (p: pixel, = R, G, B)

    ),()()()(),( pIpIppI skysuncld +=

  • 8/10/2019 Photo Cloud Modeling

    24/54

  • 8/10/2019 Photo Cloud Modeling

    25/54

    Overview of Our Method

    #irrus

    alto'

    #umulus

    #umulus

    Modeling of #irrusModeling of #irrus

  • 8/10/2019 Photo Cloud Modeling

    26/54

    Modeling of Cirrus

    Cirrus #louds Thin and no self'shados

    To'dimensional te1ture

    in!ut !hotogra!h

  • 8/10/2019 Photo Cloud Modeling

    27/54

    Modeling of Cirrus

    Use of #loud image as >D te1ture $emoving effe#t of !ers!e#tive transformation by

    s!e#ifying #loud !lane

    in!ut image#loud image#loud !lane

    #irrus #loud te1ture

  • 8/10/2019 Photo Cloud Modeling

    28/54

    Overview of Our Method

    #irrus

    alto'

    #umulus

    #umulus

    Modeling of alto#umulusModeling of alto#umulus

  • 8/10/2019 Photo Cloud Modeling

    29/54

  • 8/10/2019 Photo Cloud Modeling

    30/54

    "enerating Meta#alls

    Converting #loud image into binary image

    binary image#loud image

  • 8/10/2019 Photo Cloud Modeling

    31/54

    "enerating Meta#alls

    Distan#e transform of binary image Distan#e from a hite !i1el to the nearest

    bla#k !i1el

    binary image distan#e image

  • 8/10/2019 Photo Cloud Modeling

    32/54

  • 8/10/2019 Photo Cloud Modeling

    33/54

    binary image

    "enerating Meta#alls

    2enerating >D metaballs at hite !i1els Use distan#e for radius of metaball

    distan#e image

  • 8/10/2019 Photo Cloud Modeling

    34/54

    Optimi$ing Meta#all %ensit!

    Minimizing differen#e beteen #umulativedensity and #loud intensity

    = =

    N

    p

    M

    l

    lpl rfqp1

    2

    1

    min)()(

    #loud image #umulative density at !i1el p

    #enter density field fun#tion ,%yvill690

  • 8/10/2019 Photo Cloud Modeling

    35/54

    Optimi$ing Meta#all %ensit!

    51am!le

    #loud image #umulative density image

  • 8/10/2019 Photo Cloud Modeling

    36/54

    pecif!ing Cloud Plane

    "ntera#tive s!e#ifi#ation ?rientation of #loud !lane and vieing angle

    #loud !lane

  • 8/10/2019 Photo Cloud Modeling

    37/54

  • 8/10/2019 Photo Cloud Modeling

    38/54

    Computing %ensit! %istri#ution

    Cal#ulating bounding bo1 of all metaballsSubdividing bounding bo1 into gridCom!uting density at ea#h grid !oint

    bounding bo1 density distribution

  • 8/10/2019 Photo Cloud Modeling

    39/54

    Overview of Our Method

    #irrus

    alto'

    #umulus

    #umulus

    #umulus#umulus

  • 8/10/2019 Photo Cloud Modeling

    40/54

  • 8/10/2019 Photo Cloud Modeling

    41/54

  • 8/10/2019 Photo Cloud Modeling

    42/54

    Computing urface hape

    Distan#e transform of binary image51tra#ting medial a1es

    i1els here distan#es are lo#al ma1ima

    binary image distan#e imagedistan#e image

    medial a1is

  • 8/10/2019 Photo Cloud Modeling

    43/54

    Computing urface hape

    Use distan#e at medial a1is as thi#knessof #louds

    ro!agate thi#kness by o!timization

    #loud image

    medial a1is

    thi#kness image

    #olorization by o!timization ,&evin 9;0

  • 8/10/2019 Photo Cloud Modeling

    44/54

    Computing urface hape

    Constru#ting surfa#e sha!eEssuming symmetri# sha!e ith res!e#t to

    image !lane

    front vie side vie

  • 8/10/2019 Photo Cloud Modeling

    45/54

  • 8/10/2019 Photo Cloud Modeling

    46/54

    Overview

    "ntrodu#tion$elated %orkro!osed Method$esultsCon#lusion

  • 8/10/2019 Photo Cloud Modeling

    47/54

    Results

    Com!uter CPU: Intel Corei7 (3.33 GH)

    !"in memor#: $G%

    GPU: &'II Ge*orce G+ 2-

    Com!utation time

    ithin =9 se#onds

  • 8/10/2019 Photo Cloud Modeling

    48/54

  • 8/10/2019 Photo Cloud Modeling

    49/54

    Results

    Elto#umulus

    in!ut synthesized

    in!ut

    synthesized

  • 8/10/2019 Photo Cloud Modeling

    50/54

    Results

    Cumulus

    in!ut

    synthesized

    in!ut

    synthesized

  • 8/10/2019 Photo Cloud Modeling

    51/54

  • 8/10/2019 Photo Cloud Modeling

    52/54

    Results

    E #loud s#ene

  • 8/10/2019 Photo Cloud Modeling

    53/54

    Conclusion

    Modeling of #louds from a !hotogra!h Methods for three ty!es of #louds

    (#irrus< alto#umulus< #umulus)

    2enerating realisti# /D #louds similar to thein!ut !hotogra!h

    +uture %ork "m!roving sha!es of #umulus #louds

  • 8/10/2019 Photo Cloud Modeling

    54/54