THOMAS 'vV. MILLEH*

The Perkin-Elmer Corp.Norwalk, Conn.

Testing anImage-VelocityMeasuring DeviceThe device, which providesimage motion compensationfor aerial cameras, istested by means oflaboratory simulation.

(A bstract on page 848)

INTRODUCTION

T HE EVALUATION of an image-velocitymeasuring device and the theory of its

design requires the simulation of targets andmotion and the employment of precise mea­suring techniques. The laboratory techniquesllsed to gain confidence in the theory of thedevice are the subject of this article. Beforedescribing the test methods, it is consideredappropriate to describe briefly the devicebeing tested.

In its final form, the device is intended tobe a passive, airborne unit which will measureoptically the angular rate of a reconnaissanceaircraft as it travels over the ground. Theou tpu t is to be a vol tage Ii nearly proportionalto the angular rate (also termed V/ H) andwould be used to provide image motion com­pensation (IMC) for the camera and film.

A pape!- by Hufnagel, Hering, and Lands­man presented at the October 8, 1964 meet­ing of the Optical Society of America de­scribed the device in detail. Figures 1 and 2schematically illustrate its operation andFigure 3 is the sensor head itself. The elec­tronics were in breadboard form. The sensorhead was to be mounted on or near a camera,stabilized to the same degree as the camera,and aligned with ground track. The ground isimaged by the sensor head lens on a fixed gridconsisting of many uniformly spaced lineswhich are perpendicular to the ground track.

* Presented at the Annual Convention of theAmerican Society of Photogrammetry, \\lashing­lon, D. c., March 1965.

PHOTOCELL

DIRECTIONOF FLIGHT

SCENE OFGIVEN /

BRIGHTNESS

FIG. 1. VI II sensor opera lion.

Ligh t transmi tted by the grid is cond ucted toa light-sensitive detector (photodetector) viaa lucite pipe. As the image moves across thegrid, due to the relative motion of aircraft andground, the ligh t of the image is modulatedby the grid at a frequency proportional to themotion. The photodetector converts thismodulated image to an electrical signal which

.......---PHOTOCONOUCTIVEDETECTOR

FIG. 2. Optical schematic, VIH sensor head.

846

TESTING AN IMAGE-VELOCITY MEASURING DEVICE

vI H Preamplifier

V/H SensorLens Assembly

FIG, 3, VIH sensor head,

Folding Flat

847

is electrically processed to produce a final out­pu t as a dc vol tage.

LABORATOR Y SIMULATION

As fligh t testing is a long and costly process,it is more economical to simulate conditionsin the laboratory. To this end, test methodsand techniques were devised and are the chiefsubject of this article. Emphasis is placed onthe opto-mechanical tests with minimum dis­cussion of the electronics.

The geometric relationship between theground motion and the frequency output ofthe sensor head is given by the formula:

VF=Ja

J1(1)

where:

F=Olltplit frequency in cycles per second,at infinity focus

j= Focal length of sensor lens in milli­meters

a=Spatial frequency of grid in lines permillimeter

V=Ground track velocity of aircraft infeet per second.

H=Altitude in feet of aircraft abovegrou nd track

and

1/» f.

The general test technique is described asfoIlO\\'s. The sensor head \\'as focused (at finiteobject distance) on a positi\'e print of anaerial photograph. This print was mounted ona device \\'hich translated it linearly at rightangles to the sensor head optical axis. Thetest consisted of illuminating the aerialphotograph (or scene), moving it past thesensor head at a known velocity, and mea­suring the output frequency. The desiredoverall test accuracy was to be better than 1

THOMAS W. MILLER

848 PHOTOGRAMMETRIC ENGINEERING

per cent in the V/ Ii range of 20 to 50 milli­radians per second.

As most readers are aware, the step be­tween basic concept and successful imple­mentation is a big one. Let me take youthrough each step of the way and bypass theblind alleys we followed on occasion.

LINEAR TRANSLATION

The choice of linear rather than rotarytranslation was a necessi ty, as the sensorwas intended to work on a parallel-planeprinciple (i.e., the focal plane is parallel to theobject plane) ..At finite object distances only

By reversing the travel of the stage, we couldmake use of its motion in each direction forthe tests. The discontinuity at reversal wasovercome electrically by an external memorycircuit.

Specifically, we wanted to know:

(a) How accurately does the VIH devicework?

(b) Over what types of terrain featureswill it work?

(c) How lowIhigh must the scene con­tract be for specific accuracy?

(d) Over what ambient light levels will itwork?

ABSTRACT: A laboratory method and technique for testing an image-velocitymeasuring device yielded accurate and useful results. The device being tested wasa passive, airborne unit which detects the angular ground rate (VIH) of theaircraft and provided a voltage proportional to VIH for use in I MC. A movingtarget, consisting of various aerial photographs, was used to simulate groundrates. Scaling for finite object distances and scene contrast were important con­siderations. Simulated light levels over the range of 50 to 700 foot-Iambertswere used. Test results approaching instrumentation accuracy of ± 0.3 percent were obtained.

(e) How repeatable is it?(f) 'Wha t happens over cloud cover?(g) What are the effects of pointing In

pitch, roll and azimu th?(h) What is instrumentation error?

Many of these questions are interrelated sothat the choice of test method involved acompromise which would yield an optimumtest condition.

GEOMETRIC CONSIDERATIONS

The most straigh tforward approach was totackle the geometrical relationship expressedin Equation 1. Each of the quantities can bemeasured separately and its accuracy estab­lished. One addition to this equation is neces­sary for working at finite object distances:a change in the image distance. For finite ob­ject distances the change inf and the value ofH can be related by simple thin lens for­mulae. Making the appropriate substitutionsfor f and Ii in Equation 1 we get

JaVF j = -- (2)

k

linear translation satisfies this condition forextended fields. The field of view of the sensoris 30 degrees long in a direction parallel to theground track and 1 degree wide.

The 3D-degree length is intended to pro­vide averaging of the small altitude variationscreated by terrain features (hills, valleys,cliffs) and is basic to the sensor design con­cept. Thus the longer the duration of stagetravel the better the averaging and accuracyof VIH measurement. Continuous lineartranslation is possible wi th an endless bel tscheme. However the problems of uniformvelocity are more formidable than one wouldexpect, inasmuch as we were looking for anabsolute velocity accuracy of 0.1 percent. Wefinally settled for a linear translation of finitetravel which consisted of a carriage on rails(tIle carriage supporting the scene).

The moving carriage design followed goodmechanical engineering and model shoppractices. A synchronous motor drove acapstan wheel through a series of belts andpulleys. A fine wire, wrapped about the cap­stan and an idler provided linear translation.The moving stage rode on two parallel barsand was clamped to the wire. The entire sys­tem was essentially open-loop from a servoviewpoint, and depended on the accuracy ofthe power-line frequency for uniform velocity.

and

where

J(j + k)11=---

k(3)

TESTING AN IMAGE-VELUCITY MEASLJRING DEVICE 849

u = finite object distance, in milli­meters (equivalent to Ii)

k = increase in focal distance, I, forfinite object distances, in milli­meters

F, = frequency for finite object distance(u) in cycles pcr second

a, l' and I are as defined earlier, with \0 inmillimeters pcr second

By this technique, we could make k a pre­cisely known shim to shift the focal plancfrom infinity focus and also set the computedobject distance u. Xeedless to say, we did nothave a thin lens in practice and were facedwith having to establish the nodal planes ofthe lens if we werc to mechanically posi tionthe object distance u. A simpler, more precisetechnique, which was uscd successfully, wasto back-project the grid and adjust the objectdistance to get a known, computed angularfield at the object plane. This pro"ed simpleto do because: (1) the grid was made with thelens and matched with it (including radialorientation); and (2) the grid was accuratelycalibrated and the back-projected image (atthe object plane) can be measured with ascale to an accuracy of 0.25 per cent. Thusthe geometry and adjustment of the sensorhead was accurate to the combined, inde­pendent accuracy of the measurement of I(typically ±0.1 per cent), k (typically ±0.17per cent), and 1t (as ±0.25 per cent). Thisleaves only the stage velocity to measure(typically ±0.14 per cent). The measure­ment of these values was straightforwardusing routine optical inspection methods ofgood quality control.

The stage velocity was measured by inter­rupting a photocell with a series of accurate,uniformly-spaced holes in a plate attached tothe stage, measuring thc interruption periodwith an electronic counter, and taking anaverage of each of ten interruption periodsduring one pass of the stage. (O"er any singleperiod the stage rate was known to bc betterthan 0.1 per cent). The stage rate divided bythe object distance was the Vj Ii rate of thetest. (It is interesting to note that later testsfound us testing the stage motion with theV j Ii sensor rather than the other wayaround.) The usage of Equation 2 and carefulalignment should permit the prediction of F,and confirmation by test.

Having disposed of the geometric aspects,let us consider some other interesting param­eters.

TERRAIN SIMULATION

As there is an infinite variety of terrains

and thcre was no restriction to the selection oftest targets, \I'e let availability dictate ourchoice of aerial photos. Though such photosdid not give us a 3D model, they did give uslight modulation variations. Our search wasrestricted by the rcquirements that thephotos be of continuous ground cover, besufficient for the stagc length, and have ascale which would gi,'c us some scmblance ofaltitude. Since the sensor lens was nominallytwo inches in focal length, thc combination ofa 3D-degree I'jeld of view and stagc travellimited us to object distanccs on the order of650 millimetcrs (25.6 inches) which made thephoto problem a\l·k\'·ard. HO\l'e"er, as aerialcameras are our busi ness, we were able to ob­tain somc aerial scenes satisfactory for ourneeds.

The effective altitude at which the sensorsees the scenc can be computed from theknowlcdge of the al ti tude and focal length at\I'hich the original was taken. For a contactprint the geometric relationship is

Jz 1£It = - X - X II

11 v

"'here

h=apparent altitudc, In fcet, as scen bythe sensor

h = focal length of sensor lens, in milli­meters

I, = focal length of lens which took originalphoto, in millimeters

Ii =altitudc at which original photo \I'astaken, in feet

u = finite object distance when tcstingsensor, in millimeters

v =hllj(u-h).

By controlling the contrast in the prints, wearri ved at some objective, high-, medium-, andlow-contrast scenes. Thc scenes (or targets)as actually assembled consisted of two pieccsin order to increase the total scene length.This was done as follows: two enlargementswere carefully made of each scene, one amirror image of the other. They were joinedat this mirror fold line and mounted. Tominimize cross correlation of right- and left­hand scenes, the joining line (and entirescene) was til ted 3° abou t the vertical to thescene. 1aking the prints was an art ratherthan a science as it was necessary to controlenlargemen t, exposure and contrast. Persever­ance and repeated printings yielded the neces­sary prints. The final print was judged by eyefor con trast and exposure while the correctenlargement was proven by the ability tojoin left and right prints. Figure 4 is a re-

850 PHOTOGRAMMETRIC ENGINEERING

STAGE TRAVEL

STAGE TRAVEL

_ ~ LOW CONTRAST

. "-----......,.------ ~.----~- - ~

91"AGE TRAVEL ,.....I--.....,......,..------;":'----;-------'"~"""'-'~--- "..! i

I ~ .•

FIG. 4. Sensor output at various contrasts.

TESTING AN IMAGE-VELOCITY MEASURING DEVICE 851

cording of the photodetector output for high-,medium-, and low-contrast sections of ascene for the same illumination level. TheVIH information is contained in the fre­quency of the output. The scene was illumi­nated by two halide type, high-intensitylamps, specifically Sylvania DXN lamps.

Early in our usage of these lamps we dis­covered that they must be powered from a"'ell filtered and regulated (±0.01 per cent)DC source because AC power gave us a de­tectable light flicker (to say nothing of thevarious transients on a power line in a busylaboratory). As frequency was our chief in­formation, the usage of a dc source was anecessity. Most of our test work was run at a700 foot-lambert level which gave a comfort­able illumination level without creating toomuch heat and provided an adequate elec­trical signal-to-noise ratio (approximately50:1 SIN). The absolute value of the lightlevel was not critical because the sensitivityof the detector was fairly uniform over alarge brightness range. The spectral qualityof the ligh t and the spectral response of thephotodetector were compatible. Compu ta­tions showed that as long as the modulationlevel was greater than the equivalent narrow­band noise of the electronics and dc level wasunimportant. Thus, scene contrast was moreimportant to performance than actual brigh t­ness.

CONTRAST AND BRIGHTNESS

The worst test condition was a high VI1-[

rate at low light level with a low-contrastscene. The high VI H rate reduced the aver­aging time which is fixed by the electronictime constant and the field of view. The lowlight level placed the sensor photodetector atits lowest sensitivity region (plus increasingits time constant, which increases with lowerlight levels), while the low-con trast scenegave us low modulation levels. The results ofa typical "worst condition" test were:

Scene Low contrastBrigh tness. . . . . . . . . .100 foot-IambertsV/ H Rate. . . . . . . . . . . .44.27 mrad/sec.Sensor Measured Rate 44.67 mrad/sec.Error +0.4 mrad/sec.

For scenes with more contrast, the typicalerror was ± 0.2 mrad/sec.

For the low-contrast scene used (and ad­mittedly "low contrast" is a subjective term)we found that a light level of 20 to 40 foot­lamberts was the lowest brightness at whichwe could still obtain an accurate output­that is, no worse than ±0.5 milliradian per

second. The lowest detectable scene modula­tion was about 0.04 per cent. Thus, twothings were at work to set a lower limi t of per­formance:

(1) the decreased photodetector sensitivityand increased response time at low ligh tlevels; and

(2) the system noise with respect to scenemodulation.

SENSOR PERFORMANCE

The tests gave us some interesting insighton sensor performance. After many test runs,we observed a pattern of sensor error with theinstantaneous stage position. Investigationshowed that during its travel the stage wasslow to start, reached its calibrated value atmid-motion and decreased near the end of itstravel. To reduce the instrumentation errors,the output was read at the same stage posi­tion for each traverse of the stage. In thismanner the random error due to stage mo­tion was essentially eliminated. The accuracywith which the object distance was set con­tributed a bias which showed in the errorspread of many readings. Repetition taughtus the experimental techniques of how much(or little) care was needed in aligning thesensor head to the stage. Reasonable care wassufficient to set the object distance by theback-projection method. Proof of this lay invariations of error polarity. Once the skillshad been acquired, the error spread in out­put was randomly plus and minus for tenreadings or more.

The accuracy with which the input (VIHrate) was known was the rms error of each ofthe errors of the independent parameters(j, a, k, 1t and stage rate). This was ±0.16 percent, or 0.030 to 0.08 milliradian per secondover the range of VIH rates used.

Repeatability of the tests was excellent.The basic instruments used were an elec­tronic counter (good to ±0.05 per cent) and adigital voltmeter (good to +0.05 oer cent).For a range of 20 to 50 milliradians per sec­ond, from low to high contrast, and 100 to 700foot-lamberts the maximum error in the sen­sor, including instrumentation errors, was± 0.5 milliradian per second.

o attempt to simulate cloud cover wasmade either synthetically or within the scenephotograph itself. Intermittent cloud coverconstitutes a lack of contrast and thus a "no­signal" condi tion. Any protection against thiswill have to be solved in the electronics.

The question of the effect of local variationin terrain altitude was not tested because of

852 PHOTOGRAMMETRIC ENGINEERING

the problems of implementation. However,reasoning told us that if these altitude vari­ations are less than one per cen t of the meanaltitude, the error in sensed VjH, due to thiscause alone, must also be less than one percent. Further, terrain fluctuations are essen­tially averaged due to the 30-degree field (asdescribed earlier) and to filtering techniquespossible in the electronics.

One or two experiments were made on theeffect of pitch, roll and azimuth positions(static not dynamic). The magnitude ofpitch, roll and azimuth angles used waslimited. Pitch was essentially limited by theamoun t of defocusing that could be tolerated.Assuming a permissible focal plane variationof ±0.5 millimter at the finite focus used, thepitch angle on the test stage could not ex­ceed ±0.5 degree. At infinity focus, the pitchangle could be greater than this as shown bydesign calculations, and the system should re­main within a ± 0.5 milliradian per seconderror. Roll presen ted no problem except forthe physical problem of holding the sensorwith respect to the stage and compensatingfor the increased object distance. Roll

angles up to 15 degrees were tested. At in­finity focus, the effect of roll position shouldbe only a cosine variation. Azimuth angleswere the easiest to arrange and tests up to ± 7degrees azimuth showed sensor output re­duction by the cosine of the angle also.

SUMMARY

~'hat has been presented herein is a de­scription of the parameter tested, the natureof the test arrangement, and identification ofthe chief parameters contributing to testaccuracy of the sensor head.

The chief parameters were velocity (pro­vided by a moving stage), object distance,focal length, and grid spacing. The overall in­strumentation accuracy was ±0.16 per cent.Consideration was given to scene con trast,average brightness level and VjH rates.Brightness levels ranged from 50 to 700 foot­lamberts over a V j H of 20 to 50 milliradiansper second. Accuracy of the system beingtested approached the instrumentation ac­curacy. The effects of pitch, roll, and azimuthpointing on performance were briefly investi­gated.

ERRATAPage 639, July 1965: the caption for the photograph should read as indicated

below.

FIG. 1. Mrs. Clarice L. Norton receiving the Sherman Mills Fairchild Photogrammetric Award fromMr. John Carter representing the Fairchild Camera and Instrument Corp. The purpose of the award is tostimulate the development of the art of aerial photogrammetry.