DR. SIDNEY BERTRAM

The Bunker-Ramo Corp.Canoga Park, Calif.

The Universal AutomaticMap Compilation EquipmentThe initial cost, when amortized over a reasonable estimate ofthe production, should show a lower cost per modelthan can be achieved with conventional instruments.

(Abstract is on page 247)

INTRODUCTION

T HE Universal Automatic Map Compila­tion Equipment (UAMCE) nearing com­

pletion at The Bunker-Ramo Corporation isintended to produce high-quality orthophotosand altitude charts from a variety of photo­graphic inputs, including convergent frameor panoramic pairs of up to 9- by is-inchformat. It will be able to set up its own stereomodels from measurements made whenoperating as a precision stereo comparator.

The design of the UAM CE is based on theprinciples demonstrated in the very success­ful Automatic Map Compilation System de­veloped at The Bunker-Ramo Corporation.*

* S. Bertram. "The Automatic Map Compila­tion System," PHOTOGRAMMETRIC ENGINEERING(July, 1963).

S. Bertram. "Application of Hybrid Analog and

DR. SIDNEY BERTRAM

It includes a number of new features intendedto enhance its utility in a production environ­ment. Like its predecessor, the UAMCE isbeing developed under the sponsorship of theU. S. Army Engineer Geodesy, Intelligenceand Mapping Research and Developmen tAgency.

This paper reviews the principles of auto­matic map compilation, describes the imple­mentation for the UAMCE, and outlines theanticipated operating procedures for com­parator measurements and for compilationoperations.

PRINCIPLE OF A TOMATIC MAP COMPILATION

The principle of automatic map compila­tion from stereo photographs can be under­stood by examining the geometry associatedwith typical aerial photography. An elevationview normal to the line of flight of a cameravehicle, with photographs taken from camerastations C\ and C2, is presented in Figure 1. Apoint P(X, Y, Z) in the common field of viewis shown imaged at Pl and P2 on the resultingphotographs.

Suppose it is desired to determine thealtitude of point P from measurements on thephotographs. If a low estimate of the altitudeis made (as shown in the illustration), theimage of P would appear to the left of theimage of its estimated position in C1 and tothe right of the image of its estimated posi­tion in C2 . If the two photographs are syn-

Digital Techniques in the Automatic Map Com­pilation System," Proceedings-Spring Joint Com­puter Conference, 1963.

244

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 245

ORTHOPHOTO ALTITUDE CHART

Map Sheets Prepared by Automatic Map Compilation System

(1)

chronously scanned by fine points of lightmoving from left to right along SI and S2 andcentered on the estimated positions of thepoint, the signal corresponding to p, will ap­pear early in its scan while that from P2 willbe late in its scan. \iVhile particular imageryis not identified, the differential time betweenthe appearance of related imagery provides ameasure of error in the altitude estimate.

CORRELATION OF SIGNALS

The recognition of the similarity betweentwo signals, basic to the measurement of thedifferential time, is achieved through a corre­lation process which implements the mathe­matical expression

1 J'SeT) = - g(t)h(t - T)dtTo I-To

where get) and h(t) are the two signals to becompared, T is a delay, and SeT) provides thedesired measure of the similarity of the twosignals oyer a range of integration; SeT) is ameasure of the average correlation level overthe integration time, To. The delay parameter,

T allows the signals to be compared with dif­ferent time offsets; the delay that maximizesSeT) provides the desired measure of the dis­placement of similar or "correia table" ele­ments in the two signals.

Figure 2 illustrates the correlation processfor the hypothetical case of a sharply definedimage element giving rise to a signal in theform of a rectangular pulse of unit amplitudeand time duration T. The first signal, J(t) ,is shown in (a). The second signal, shown in(b), is assumed to be identical to the first, ex­cept for a delay T. The instantaneous productof the two signals, shown in (c), again is asquare pulse, but the pulse duration extendsonly over the overlap of the two contributingpulses. The correlation function for T <To = 1(the area com mon to the two pulses as afunction of the delay) is shown in (d). Themaximum value of the correlation function,T, occurs when the pulses are co-incident(T=O) with a linear rise and fall on eitherside going to zero where the pulses do not havea common region.

246 PHOTOGRAMMETRIC ENGINEERING

try to pull in to the top of the peak it is on­hence, to an erroneous value.

The possibility of an ambiguous height de­termination (Figure 3) can be reduced byusing a low-resolution scanning system thatcan resolve the "forest" but not the "trees."This approach would broaden the correlationfunction so that it has a single peak extendingbeyond the limits of the original functionwith only a trace of the original secondarypeaks showing (Figure 4). It will be observedthat the position of the maximum cannot bedetermined as easily on this new function, butthere is no longer any possible ambiguity.

FIG. 1. Development of Height-Error Signalfrom Stereo Photographs.

It is interesting to examine the correlationfunction for a set of regularly spaced elements-for example, a row of houses or trees in anorchard. Figure 3 presents this function for aset of three idealized pulse elements. The cor­relation function has a large peak when 7

equals zero; i.e., when the three pulses of f(t)overlap with the corresponding pulses ofj(t-7). In this case, the principal peak issurrounded by smaller secondary peakswhere two pulses overlap and still smallerpeaks where only one pulse overlaps. Theseare of interest because it is possible -for asystem to have a su fficien t error to take itto one of these peaks. I n such a case, an ytracking device based on correlation would

In an automatic map compilation system itis advisable to use dual correIa tors, with ahigh-acuity channel taking advantage of theavailable resolution in the photographs anda low-acuity channel operating through elec­trical filters to effectively work only on lowresolution imagery in the photography. Thehigh-acuity channel provides tight trackingmost of the time, with the low-acuity channelhelping over difficult areas (such as abruptaltitude changes).

PRACTICAL CONSIDERATIONS

Imagery is usually characterized by gradualtransitions, yielding electrical signals con-

1_

(a) REPEATED PULSE SIGNAL f(t)

1_

t_

t_

T__[lJ1,===-:::-:-:-::-.--

Ie) INSTANTANEOUS PROOUCT f(l) f(t-T)

Cd) CORRELATION fUNCTION ff(t) f(I-T) dT

FTj------(-a)'""'P""ULSE SIGNAL '-;f(;:;I)~------

T~ ETj----7o(b":")-=D7::IS-=P:-:LA::;C-=~::-'D PULSE SIGNA-:-L-::f(-:-j.T::")----

EJ-----;-(c:-)I"'N:;:ST:;CA:-;N~T-::AN:;;E~l?USPRODU;-;C~T-::f(:;;-I)-::f("I-T:;;-)----

As:l,-------<-- 0(d) CORRELATION FU"CTION ff(l) f(t-T)dt

FIG. 2. Development of Correlation Functionfor Sharply Defined Element.

FIG. 3. Development of Correlation Functionfor Repeated Pulse Signal.

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 247

sisting of rounded pulses that merge together.The corresponding correlation functions donot have sharp corners, as shown in Figures2(d) and 3(d), but curve around the top andtaper off towards zero.

To optimize altitude tracking, it is neces­sary to ensure a high level of useful signal intothe correlators with a minimum of noise orextraneous signal. The doc level, which doesnot contain any lIseful information, couldobscure small image detail (particularly inhigh-light areas); letting through the fine de-

is critical for optimum correlation. Only sig­nals which can represent height differentialsare useful. For example, this implies that theimage of a straight line (such as would be pro­duced by a road running parallel to the lineof camera separation) should produce a mini­mum signal because it does not contain anyheight information. The scan implied inFigure 1 (parallel to the line of flight), satisfiesthis requirement. For elements in the photog­raphy representing tilted ground areas, theappropriate scan is one representing the pro-

ABSTRACT: The Universal A utomatic Map Compilation Equipment nearingcompletion at The Bunker-Ramo Corporation is being developed to p·roduceaccurate orthophotos and altitude charts automatically in a map-productionenvironment. It will serve as a precision comparator and, using the equipmentcomputer, will perform orientation calculations to obtain the data required forthe compilation operation.

The equipment consists of four identical scanning tabtes, a cont'rol console,a Bunker-Ramo Model 133 Digital Computer with input/output equipment,and associated electronics. A ny one of the four tables can be used to carry a dia­positive during compilation or comparator operations or to carry a film sheetfor exposure of the orthophoto or altitude chart during compilation.

The equipment is designed to perform one hundred altitude measurementsper second; detaited 9- by 18-inch, 100% lap, 30-degree convergent stereo pairscan be compiled in less than three hours with an expected accuracy correspond­ing to a Cofactor of 5000. Smaller models and/or a reduced accuracy require­ment would require lower compilation times per model.

The equipment permits the operator to outline possible trouble areas before acompilation is started and to command appropriate action to be taken when theareas are examined during compilation.

When operated as a comparator, the equipment permits virtually simultane­ous stereo measurements on common areas on four diapositives-one on eachtable. A measurement error of less than 4 p. rms is expected for such operations.

tail would result in an unnecessarily high levelof photomultiplier (and other) circuit noiseentering the correlators. For this reason, thesignal processing begins with the removal ofboth the average, or d-c, level and high fre­quency components near the resolution limitsof the system. In the case of the low-acuitycorrelators, the high frequency componentsare further attenuated in order to minimizethe possibility of multiple peaks in the cor­relation function.

The four modified signals (two for the high­acuity and two for the low-acuity height errorsensor) are brough t to the desired energy levelin independent, automatically gain-controlledamplifiers. Thus, only the signal componentsgoing to the respective correiators are in­fluential in controlling the input level.

The method of scanni ng the photographs

jection of such a scanning line on the tiltedareas as seen in the two photographs.

CORRELATOR CIRCUIT

The correia tors used in all Bunker-Ramoautomatic mapping equipments resemble a"quarter-square multiplier" or "phase-sensi-.tive detector" as illustrated in Figure 5.To aid in the description, the circuit is dividedinto an upper (heavy line) and a lower (lightline) portion.

In the upper portion, suppose the nonlinearelements Dl and D2 have current-voltage re­lationships expressed by

and

PHOTOGRA MMETRIC ENGINEERING

DEVELOPMENT OF HEIGHT-ERROR SIGNALS

An error in estimating the height gives riseto a corresponding time offset in the appear­ance of corresponding imagery in the twoscanning signals. This, in turn, results in alevel of correlation which is as much de­pendent upon the image content as on the

current for one polarity while passing a neg­ligible current for the opposite polarity. Thelight (lower) portion of the circuit (Figure 5)supplies the currents of opposite polarity.Thus, diode Dl supplies current to the capaci­tor when ea+eb >0, and diode D3 supplies acorresponding current for ea+eh <0. Similarly,diode D2 supplies a negative current foren -eh <0 and diode D4 supplies a negativecurrent for ea -eh >0.

In practice, the balance of the correlator isfar more important than its linearity as amultiplier. The transformers must cover thefrequency range of interest, and the capacitor(with associated resistors) must have a desira­ble time constant. In the UAMCE the capaci­tor is replaced by an operational integrator;this integrates linearly until the capacitor isshort-circuited by signals from the heightcounter or computer. One integrator receivesthe combined output of two correlators con­nected differential1y, as described in the nextsection.

eb + eb R

D1

+e. R

-::

- ehR

D2

-= +e. R

OUTPUT

-ebR

D3 Ie-e. R

":"

+ebR

04

-ea R

FiG. 5. Correiator Circuit.

248

The elements used for Dl and D2 are notsquare-law devices, but diodes which pass

Fig. 4. Correlation Function for Repeated PulseSignal Using Low Resolution Scanning Signal.

where the sign reversal is obtained by re­versing the element. Suppose further that thetwo transformers have inputs ea and eh. Ele­ment Dl has an effective input voltage, sup­plied through equal summing resistances,R, of (ea +eb)/2, while element D2 has aninput of (ea -eh)/2. The current into the inte­grating capacitor, C, is the difference of thecurrents from the two nonlinear elements:

Because the voltage across a capacitor is theintegral of the current into it, the output,taken from across the capacitor, is the desiredintegral of the product of the two inputsignals.

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 249

SIGNALSFROM AGe

AMPLIFIERS

FIG. 6. Height-Error Sensor

error; the output of such a correia tor providesno clue to the altitude errOl-.

To obtain a useful height-error signal, two­correlators are used in conjunction with timedelay units connected (as shown in Figure 6)so that correia tor 1 has a peak output forerrors on the low-altitude side while corre­lator 2 has a peak for errors on the high­al ti tude side. These, along with the corre­sponding difference output, are shown inFigure 7. It will be observed that the differ­ence is zero for LlZ = 0 (coincident signals),with a positive output for LlZ >0 and negativeoutput for LlZ <0; thus, the difference outputis appropriate fOI' use as a height-error signal.

The UAMCE has a high-acuity and a low­acuity height-error sensor. The outputs ofthese are appropriately summed to provide anet height-error signal.

HEIGHT-ERROR MEASURING UNIT

The height-error signal must be convertedto a digital measure of the error. This is ac­complished in the height-error measuring unit(Figure 8). The signal from the height-errorsensor is passed to an integrator whose outputis monitored by a positive and negativethreshold detector unit. \iVhen the inte­grated error signal exceeds one of the thresholdlimits, the detector steps a reversible counterin a corresponding direction and resets theintegrator to zero. The counter, in turn,operates an elementary digital-to-analog con­verter that produces a voltage to deflect the

center of one of the diapositive scans in adirection to reduce the observed time differen­tial. If the al titude error has not been com­pletely compensated, the remaining errorsignal may again cause the threshold to be ex­ceeded, resulting in a second count. Operation

DISPLACED CORRELATION fUNCTIONS

oz

DIFFERENCE CORRELATION FUNCTION

FIG. 7. Output of Differential Correlators.

250 PHOTOGRAMMETRIC ENGINEERING

COMPUTER RESET SIGNAL .

MEASURED HEIGHT ERROR TO COMPUTER

DEFLECTION

FIG. 8. Displacement-Error Measuring Unit

and

I n this case,j is the focal-length of the camerathe coefficients UI, U2' .• Wa are functions ofthe orientation of the camera at the time agiven photograph was exposed, and X" Y"and Zr are measured with respect to the posi­tion of the camera station. Equation (3) isaugmented by simple equations providingcorrections in the fil m coordi nates, x, y, fordistortion in the camera lens or film and forcorrections in the geographical coordinatesfor the effects of the curvature of the earthand atmospheric refraction. For panoramicphotography the relationships take a morecomplex form in order to account for thesweep motion of the camera and the motionof the camera vehicle during the exposure.

continues in this manner until the height­error has been compensated and appears asa corresponding count in the reversiblecounter. In operation, the counter position isthen read by the associated computer and theintegrator and counter reset to zero so that anew measurement can be initiated. Note thatthe height-error measuring unit is, in effect,an analog-to-digital converter that provides adigital measure of any height error. Becauseof the closed-loop operation, the measure­ment is independent of the quality of theimagery in the field of view. However, thetime required to make the measurement doesdepend on the image content.

MATHEMATICS OF AUTOMATIC

MAP COMPILATION

Operation of automatic map compilationequipment is limited to photography forwhich the geometry is sufficiently well definedso that the functional relationships

x = G(X, Y, Z)

and

(3)

For ordinary photography the relationshipsin Equation (2) take the well known form

* For the presen t purposes a rectangu1ar co­ordinate system is presumed. Where the curvatureof the earth is significant, a suitable correction isused.

which relate coordinates (X, Y, Z)* on theground to coordinates (x, y) in the photo­graphs, can be obtained. The functions, G andN, must be in a suitable form for computationby a digital computer.

y = JJ(X, Y, Z), (2) In the UAMCE a digital computer cal­culates the photographic coordinates, (x, y),for each of the photographs corresponding toa ground horizontal position (X, Y) and analtitude Ze obtained by extrapolation frompreviously measured positions; the (x, y)values are used to position the photographsover the respective scanning units.

The computer also calculates and outputsthe partial derivatives of Equation (2) re­quired to develop the appropriate scanningsignals. These can be written in terms of thederivatives of the functions in the form

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 251

Here, dx and dy are displacements in thephotograph corresponding to effective scan­ning displacements dX and d Y along theground;

are the components of the slope at (X, Y, Z). Inpractice, dX is a fast sweep parallel to theflight line and d Y is a slow sweep in the per­pendicular direction; these combine to form aTV-like scan along the ground_ The sweepcomponents, dx and dy, in the plane of thephotographs produce a one-to-one timecorrespondence between scanning of groundand image points.

At each iteration the computer calculatesthe photo coordinates of each new point forthe desired X and Yand the best estimate ofZ available. The measurement (by the analogsystem) of the resulting altitude error re­quires a knowledge of the scale factor re­lating observed displacements in the photog­raphy to altitude errors. In the case of verti­cal frame photography, the scale factor isessentially constant for a given model; for allother types of photography the scale variesover the compilation field. The scale is calcu-

The components of slope are measured byanalog circuits that implement the scans de­scribed by Equa tion (4), wi th independen tcounters representing Zx and Zy. The signalsto change the state of the slope counters arederived by comparing the differential heighterror signals on the positive and negative Xhalves of the scanned area for Zx and on thepositive and negative Y halves for Zy. Be­cause the slope measuring elements formclosed-loop analog-to-digital measuring units(as in the height measurement), the measure­ments are independent of the quality of theimagery in the area. The measured digi talslope values are transferred to the computerfor use in estimating the altitude for succes­sive elementary compilation areas.

y = II(X, Y, Z)

y' = II'(X, Y, Z)

x = C(X, Y, Z),

x' = C'(X, Y, Z),

and

0= 11x'dX + Hy'dY + JIz' (6)

When Equation (6) is solved for dX anddY and the results substituted into Eq. (5),the differential shifts, dx and dy, correspond­ing to a unit altitude error are determined.These val ues are ou tpu tted to the analogequipment. Because only a very modest ac­curacy is required, simple approximations arepossible and only infrequent updating is re­quired. The calculation represents a trivialpart of the total computer problem.

and

dy = IlxdX + [hdY + Hz (5)

In Equation (5), dX and dY are found fromthe corresponding equations for the seconddiapositive by taking dx' = dy' = 0; that is,

0= Cx'dX + CI,'dY + Cz'

The computer also outputs a signal to con­trol the intensity of the scanner exposing thealtitude chart. For this purpose, the altitudeis divided by three times the desired contour

be the equations defining the coordinates of apoint P(X, Y, Z) on each of the photos of thestereo pair. Assume that the second (primed)diapositive is used to create the orthophotoand that its scan is not disturbed by the alti­tude correction signals. The problem is illus­trated in Figure 9, which shows an elevationview normal to the line between the twocamera stations. A high height-estimate isindicated, where the image in the center ofthe field of view on the second diapositive isshown as Po (instead of the correct position,P). The image of the estimated point, P., ondiapositive 1 is displaced GzdZ from theimage of P, and by GJ[dZ from the image ofPo. In practice it is expedien t to make the cor­rection using G/fdZ at the first diapositivewith the orthophoto signal taken from the sec­ond diapositive; the displacement of Po fromP is ordinarily so small as to produce neg­ligible error on the orthophoto. By assumingt.Z = 1 (i.e., equal to the increment of theheight counter), the required height cor­rections are readily found. Thus,

dx = CxdX + Cl-dY + Cz

lated by the computer as often as required tohave a sufficiently current value for use. Thecalcula tions are based on the following:Let

(4)

azand -- = Zv

aY

azaX = Zx

all affdy =-dX+-dY

ax aY

alI (az az)+- -dX+--dYaz ax aY

aC acdx = -dX +--dY

ax aY

aC (az az)+-- -dX+-dYaz aX aY

and

252 }>HOTOGRAMMETRIC ENGINEERING

//

V

CAMERAPOSITION 2

p~-------

FIG. 9. Development of Height Correction Signal, View Normal to Flight Line.

interval and the remainder, R, examined forassignment of the printout level. The level isassigned as follows:

o~ R < 1/3 scanner ofT

1/3 ~ R ~ 2/3 scanner medium2/3 < R scanner bright

This creates a chart (shown on the frontis­piece) showing well-defined altitude bands.The rotary sequence of densities permits anunambiguous assignment of altitudes, start­ing from a point of known altitude.

REDUCTION OF Y PARALLAX

Creating an adequate stereo model for com­pilation requires that both X and Y parallaxbe removed at a number of points in thestereo field so that coordinates of commonpoints in the field can be accurately ascer­tained. The height-error sensor removes Xdisplacements caused by height errors; there­fore, it is directly applicable to the au tomaticremoval of X parallax in the comparatormode. The removal of Y parallax displace­ments requires additional circuity.

Sensing of Y errors is achieved, in effect,by introducing an artificial Y parallax analo­gously to the X parallax provided by the delaylines in the height-error sensor (Figure 6).The Y parallax is introduced on a time­sequential basis by shifting the scan in Yonone diapositive, positively for a defined timeand then negatively for an equal time. Theoutput of the associated correia tor is switchedinto a positive input of an integrator duringthe positive shift, and to a negative input

during the negative shift. If there is no Yparallax at the point, the inputs during thetwo time periods are equal, and the integratedou tpu t is zero. If there is any Y parallax, thecorrelator will have a larger output when theartificial parallax partially cancels the imagedisplacement and a lower output for theother half cycle where the two componentsare additive; therefore, the integrated outputprovides a measure of the Y parallax.

In the UAMCE, a positive and negativethreshold detector (similar to that used withthe height-error sensor) provides digital errorsignals to the computer for reduction of Yparallax during comparator operations. Forautomatic compilation operations, the inte­grated output is used to deflect the center ofone of the diapositive scans in a direction toreduce Y parallax; this ensures high sensitiv­ity in the height-sensing circuitry, even witha poor model.

SCANNING PATTERN

It has been noted that the principalscanning for automatic height determinationshould be along a line parallel to the linejoining the two camera stations (i.e., on anX-directed line). This must be augmented bya slow scan perpendicular to the fast scan (ona Y-directed line) to produce an effectivearea scan providing a TV-like coverage ofeach small area. This ensures continuity ofthe image for both the operator and forautomatic tracking. The signal from onescanner is used to recreate the scene elementfor exposure of the orthophoto film sheet.

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 253

For automatic operations the time requiredto sample a given area must be kept as lowas possible. For this reason the Y scan ismade in an unusual pattern. Every line on thepositive side of the center of the scan is fol­lowed by a correspondingly positioned lineon the negative side. The scanned area iscovered very coarsely at first, then graduallyfilled in. A total of 128 lines is used with sub­stantially uniform sampling of the area in 2,4, 8, 16, 32, 64 or 128 lines. The total framescanning of 128 lines takes 20 milliseconds;because a lO-millisecond measurement periodis used during compilation, only every otherline is used in exposing a given scene elementfor the orthophoto.

PRODUCTION OF THE ORTHOPHOTO

AND ALTITUDE CHART

A television-like reproduction of the smallarea under observation at a given time iscreated by displaying the ground scan (dX,d V), appropriately scaled, on a flying-spotscanner with the instantaneous brightnesscontrolled by the video obtained from one ofthe diapositives. This is centered at the sim­ilarly scaled ground position (X, Y) to exposethe corresponding area on a sensitized filmsheet used to create the orthophoto. Theprofiling operation ensures a complete cover­age of the model area. A second flying-spotscanner, similarly positioned but with itsbrightness controlled by the three-level con­tour interval signal from the computer, isused to create the altitude chart.

CHARACTERISTICS OF THE UAMCE

Automatic map compilation devices requirescanners to examine the stereo photographs,to produce an orthophoto, and to print analtitude chart. A positioning unit must beassociated with each scanner.

In Bunker-Ramo's earlier device, theAutomatic Map Compilation System, the twodiapositives and single film sheet for theorthophoto and altitude chart were mountedon a common carriage, whose position cor­responded to the geographic area being ex­amined. Thus, while fixed scanners and imag­ing lenses sufficed for the two outputs, thediapositive scan lenses had to be moved byservos under computer command to accom­modate for parallax and camera-tilt dis­placements. This red uced the accuracy re­quirement of the large carriage, but limitedthe system's application because of the re­stricted range of the lens servos used to re­move parallax displacements.

In the UAMCE, four identical tables withassociated flying-spot scanners, under com­mand of the Bunker-Ramo Model 133 com­puter, are used; they provide virtually anunlimited amount of flexibility. The severeparallax displacements of strongly convergentphotography, the distortions of panoramicphotography, and a considerable range ofoutput scales for a given input scale can beaccommodated. Reducing the output scalemakes it possible to automatically preparemosaic orthophoto and altitude charts fromcontiguous stereo pairs.

An artist's concept of the UAMCE isshown in Figure 10. The four tables in theforeground can be used interchangeably tocarry diapositives or film sheets. The con­trol console includes a stereo-viewer formonitoring the automatic compilation opera­tion; it is arranged to permit stereo observa­tion of images from any two of the four tables,thus providing a capability for making vir­tual1y simultaneous stereo comparator meas­urements on up to four diapositives. To assistin the comparator operation, the images canbe rotated 90 degrees to facilitate observationof Y parallax, and the left and right imagesinterchanged to obtain a pseudostereo view.A reference viewer on the console is intendedto carry a print corresponding to a diapositiveused in one of the tables. The viewer has acarriage which may be slaved to anyoneof the four tables in accordance with theposition of a selector switch. A light movedby the carriage illuminates a small area onthe photo print to indicate the area under ob­servation at the corresponding table. A posi­tion control, located under the referenceviewer, permits the operator to change(through the computer) the x- and/or y­position of any of the tables or the heightduring compilation. A set of pushbuttons onthe console permits the operator to selectthe desired operational mode.

A Bunker-Ramo Model 133 Digital Com­puter, shown to the left of the console inFigure 10, calculates the table coordinatesand scan coefficients and controls the opera­tions. A typewriter at the left of the consoleis used for computer checkout and off-lineprogram and data preparation. A two-bayelectronic rack carries the timing signalgenerators, scan generators, correlators, andcomputer interface equipment. Table servo­and scan-drive amplifiers are located in smallracks adjacent to the associated tables.

254 PHOTOGRAMMETRIC ENGINEERING

FIG. 10. Universal Automatic Map Compilation Equipment.

SCANNING TABLES

The configuration of a scanning table isshown in Figure 11. The UAMCE requiresfour of these tables.

The requirements for the table are quiteunusual. It must be capable of moving a 9­by IS-inch glass plate at peak speeds of overtwo inches-per-second on orthogonal axes(with an average speed of one inch-per-secondduring compilation) and of positioning thescanning to within a 4-micron rms error on

the plate. The required two-coordinate mo­tion is provided by dual carriages, with onecarriage riding on top of the other.

The base of the table is a composite struc­ture of a highly rigid, stabilized bed castingwhich is nested in a fabricated steel cradle.The assembly rests on three leveling-typevibration isolators. The flying-spot scannerassembly is precision-mounted from the bedcasting, with provisions for access for repairor removal of the unit through the cradle.

LIMIT SWITCHESAND STOPS

FIG. 11. Scanning Table-Artist's Sketch.

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 255

LIMIT SWITCHAND STOP

BALL RETAINERAND SYNCHRONIZER

lASE COYER

FIG. 12. y-Carriage Ways.

The main supporting sections of the cradlefall directly under the sections of the bedwhich carry the longer (y-direction) ways.The bed, a stabilized Mehanite casting, isribbed to provide substantial rigidity and ismounted in place along supporting sections;special screws capable of fine adjustment areused to remove any small errors in the y-direc­tion way geometry.

The scanning table has two carriages (asindicated in Figure 11). Each moves inde­pendently of the other; their motions com­bine to position the transparency or film tothe proper location over the scanner. Theyare made of cast magnesium alloy which isselected and treated for maximum dimen­sional stability. It is interesting to note thatthe magnesium alloy chosen, when properlystabilized and stress-relieved, has a highinternal damping factor that operates to pre­vent "ringing" of the carriages when they arerapidly accelerated by the servo; such vibra­tions, though of small excursion, could ad­versely affect the operation of the FerrantiMoire fringe counting element used for theposition measurement.

The carriages have extensive ribbing, whichgives maximum structural rigidity. The lowercarriage rides on the y-ways and supports theupper carriage on the x-ways; the relation­ship of these two ways determines the orthog­onality of the table. Fortunately, the centeropening in the lower carriage has to be onlylarge enough to clear the lens tu be for onedirectional motion ; therefore, it is constructedas an almost completely closed box, makingit very rigid structurally.

The x-motion carriage has a smaller effecton the performance accuracy of the table be­cause its motion is completely determined bythe ways on the y-carriage. Because it carriesthe transparency or film adapter, it must have

a center opening equal to the largest diaposi­tive area to be examined.

The ways have a very important effect onthe performance of a precision machi ne. Ball­bearing, anti-friction ways are used in theUAMCE because of their low friction and be­cause the small number of parts involved inthe guiding action makes it easier to achievethe desired accuracy. The guiding way makesuse of "V" grooves on each mem bel'. Thefloating way makes use of a linear ball-bearingwheel that is free to move and rotate on itsaxis to minimize side loads and to relax thesetup tolerances. The ways, made of hardenedsteel, are floated in the related member toprevent strains caused by temperature varia­tions. The y-carriage \\·ith its ways, with thex-carriage superposed is shown in Figure 12.The construction of the x-carriage ways issimilar to that of the y-carriage ways.

The illustration also shows the lens holderand y-carriage light shield. The lens holder isrigidly attached to the table base and appearsin a slot in the y-carriage. Extensive lightshielding is used throughout to ensure againstfogging of the photosensitive film sheetduring the long time required for a compila­tion.

Each of the carriages is driven by a servomotor through a gear train, which, in turn,drives a recirculating ball-bearing screw. Two­micron Ferranti Moire fringe measuring ele­ments, independent of the drive, operate withreversible counters to provide a running countof the table position on each axis.

The tables are provided wi th adapters forholding either glass plate transparencies orfilm sheets. A vacuum hold-down and an ar­rangement similar to that used in a cut-filmcamera to facilitate handling is used for thefilm sheets.

256 PHOTOGRAMMETRIC ENGINEERING

During operation of the UAMCE the com­puter calculates the desired x and y positionsfor the carriages on each of the four tables.The digital outputs of the computer are con­verted to eight analog voltages, correspondingto the desired x and y posi tions of the fourtables. The pulse outputs from each opticalmeasuring uni t are accumulated in a reversiblecounter; the state of the counter representsthe instantaneous position of the correspond­ing table axis. To obtain the servo drive errorsignal, the counter state is converted to ananalog voltage, which is then subtracted fromthe analog voltage representing the corre­sponding computed table position. Theservos, acting in response to the error signal,drive the appropriate table axis ina directionto make the actual table position equal tothe computed table position.

Because analog servos are limited in speed,the accurately scaled servo error signal is alsoused to move the position of the scanningraster (or printout raster) in a direction tonullify the effects of any servo error. The re­sponse of the deflection system is about 1000times faster than the analog servos; the areaunder investigation on the diapositive, or theprintout area in the film plane, is thereforeessentially free from delays and identical tothe desired value. The measuring elementsprovide counting pulses at two-micron inter­vals, so that approximately 218 pulses are re­quired for the 20 inches (500,000 microns) oftable travel; hence, 18 bits are required torepresent the position in the computer. Be­cause 18-bit digital/analog converters areprohibitive to implement, only the 11 least­significant bits of the table positions aretransferred to the analog units; once thetable comes under control of the computer,the position error will be a very small part ofthe total range of control, and the larget- bitswould be redundant. The total range of thedigital/analog converters is (2048) (2 mi-

crons) = 4096 microns (or 0.16384 inch) of car­riage motion. The servo system must keep theposition error less than half this value to pre­vent ambiguity (as described later).

The difference between the computed andactual table position is found by taking thedifference between the d-c voltages appearingat the outputs of the two digital/analog con­verters. This is accomplished by a differentialamplifier, whose output is the servo error sig­nal. This output is later summed with the de­sired scan signal for the particular table to ob­tain the composite signal used to position thescanning spot.

The servo error signal is also used in com­pensating for a problem caused by droppingthe largest bits from the positioning unit. Thedivision of the table position into 11-bit units,in effect, sets up a periodic measuring systemin which the digital/analog converters changeover their full range in each period. This isillustrated by scale Sl in Figure 13, where thenumbers represent digital/analog outputs forthe given table position. Suppose that at timeA the command position is at 6 and the tableposition at 4; then there is a two-unit errorsignal forcing the table to the right. At a latertime B, the command position, Ce , will haveexceeded the full count of the digi tal to analogconverters. Then the output is, say, 0.3 in­stead of 10.3, while the actual position, Ct, is8.3; the difference signal would be minus 8instead of 2.

This situation is corrected by setting up asecond scale, S2, displaced from the first byhalf the periodic interval (i.e., an amountequivalent to the largest bit used). Conver­sion between the two scales is then obtainedby complementing the largest bits in bothdigital/analog units (ce and Ct). When the ex­cessive error is observed at B, the comple­menting would take place to shift the scale

3456789

FIG. 13. Table Position Measurement Scales.

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT

FIG. 14. Stereo Viewer.

257

used to 52, where the readings (with the cor­rect error difference of two units) would be5.3 and 3.3.

As the system moves to the right, the prob­lem recurs at C, where complementing wouldmove the scale used back to 51. I t is obviousthat as long as the real error remains belowhalf the interval (5 units in the diagram), anobserved error greater than half the intervalmust be corrected by the complementingtechnique. The periodic interval used in theequipment is 4096 microns; operationalspeeds and accelerations are limited to rangesyielding expected errors that are small withrespect to the interval.

AUXILIARY EQUIPMENT

The video signals generated by the photo­multipliers at each scanning table are usedto drive the correlation circuitry. The signalsare also used to recreate images of the scannedarea for exposing the orthophoto and to pro­vide a "window" to the operation in the formof an electronic stereoviewer. The viewer isshown in schematic form in Figure 14. It in­cludes two 4-inch, square-faced cathode-raytubes whose images are superposed using ahalf-silvered mirror; separation for stereoviewing is achieved through cross-polarizedfilters in front of the two tubes with corre­spondingly polarized glasses worn by theoperator. An electronically generated floatingmark (a bright cross or dot) provides a stereoreference.

The UAMCE incorporate5 a "referenceviewer," which is designed to illuminate apoint on a print corresponding to the areaunder observation on the related diapositive.The viewer, at the upper right of the controlconsole, is designed to take full-size prints.

Switches permit the operator to slave the op­eration to anyone of the four tables. Thereference viewer permits the operator to lo­cate the observation area in order that hecan interpret the image in the stereoviewer.It is also used in certain operations not re­quiring full equipment accuracy (such as ini­tial calibration and defining potentiallytrou blesome areas).

A control keyboard provides operator se­lection of computer programs or operatingparameters. Keys are provided to select thetable to be manually moved, as previously de­scribed, and to change the sensitivity of thecontrol. Other keys cause the computer to gothrough a defined calibration routine, to gointo the comparator or compilation operatingprograms, to take in measured coordinatedata, or to allow manually inserted heightchanges during compilation.

The tables are manually controlled with aposition control, which takes the form of aball whose upper surface protrudes above thetable top. Movement of the ball is communi­cated to the computer by properly coded sig­nals; the computer then commands corre­sponding motions of the tables. This may takethe form of combinations of x and y motionsof a selected table during comparator meas­urements or of Z changes during compilation.

A symbol generator provides video for thegeneration of symbols on the orthophoto andaltitude chart scanners. These are used, undercomputer control, for marking importantfeatures, such as pass-points and map grid­coordinates on the orthophoto and altitudechart. The symbols are generated by a flying­spot scanner, whose light is modulated in ac­cordance with a symbol on a mask. The sym-

258 PHOTOGRAMMETRIC ENGINEERING

bol is computer-selected, from a field of six,by moving the center of the scanning raster.The light modulated by the symbol is sensedby a photomultiplier and the resulting signalampli fied for application to the prin tou t tubes.

OPERATING CHARACTERISTICS OF THE UAMCE

The UAMCE permits an unparalleledoperating flexibility. This is illustrated in thefollowing description of the principal opera­tions performed with the equipment.

Stereocomparator operations can be per­formed simultaneously on four diapositives.Before the plates are inserted, a tape is pre­pared which lists the pertinent data for theplates. This includes the nominal orientationdata, camera station, and type of camera(frame or panoramic). In addition, rough posi­tions of desirable pass points are entered. Aprogram sheet that defines the task of theoperator is also prepared, listing the measure­ments to be performed.

With the tape entered into the computer,the diapositives inserted, and an appropri­ately marked print on the reference viewer,the operator presses the appropriate controlkey to start the operation. The computerthen commands all tables having diapositivesshowing the first pass-point to move to thedesignated coordinates. It also adjusts theshape of the scans at the diapositives in ac­cordance with the nominal orientation data,in order to present a virtually undistortedvertical view at the stereoviewer (even withpanoramic photos). Following the instruc­tions on the program sheet, the operatorswi tches the stereoviewer to a selected dia­positiYe and proceeds to cen ter the pass-poin ton the crosshair, using the position control.One image of the stereoviewer is then switchedto a second diapositive and the position con­trol used to center the pass-point in the fieldof view. Two alternatives are available forthe final centering:

(1) The operator may take advantage ofthe expanded picture, 90-degree rota­tion, and interchange of left and rightimages to manually center the secondimage to coincide with the first.

(2) The operator, through the keyboardcontrol, can command the correlationcircuitry to make the final centeringadj ustmen t.

The second alternative is obviously the moredesirable. However, it is not always possible;for example, if the area in the field of view

has an appreciable variation in altitude, itwill be necessary for the operator to adjustthe coincidence for the desired point. Thestereoviewer is then used with the third andfourth diapositives (or as many as includethe pass-point) and the process repeated.When the operator decides that the centeringon all diapositives is satisfactory, he depressesthe "store data" button, causing the com­puter to place the measured coordinates in anappropriate store. The operation is then rp­peated for the next pass-point.

\iVhen the measurements have been com­pleted, the data (together with geodetic con­trol) will be used to calculate the orientationparameters required for compilation.

Compilation operations require the orienta­tion data (including lens and film distortioncharacteristics) and measurement of the ma­chine coordinates of two reference points (e.g.,two fiducial points) to establish the interiororientation; of course, the measurement is notrequired if the diapositives have just beenmeasured in the comparator mode. The in­puts also include the coordinates of thecorners of the area to be compiled, the de­sired scale for the output, and the coordinatesof points to be marked. The geographic posi­tion corresponding to the center of the outputfilm sheets is also defined.

Using a print of one of the diapositives onthe reference viewer, the operator next movesthe point of light around areas he judges to bepotentially troublesome. After an area hasbeen outlined, the operator presses a buttonto select the program to be used when thearea is being compiled. The program includesthe following instructions:

(1) Hold altitude (as over a water area)(2) Take smaller profiling increments(3) Move across the area, then stop for the

operator to determine the altitude(4) Proceed slowly under operator control

of al ti tude.

\iVhen the instructions have been completed,the film sheets are inserted into the outputtables (if the compilation is to be one of amosaic set, the film might already be in thetables). The control and map grid-points arethen suitably marked under computer com­mand and, finally, the computer moves thetables to the compilation start position.

The operator then uses the position control

UNIVERSAL AUTOMATIC MAP COMPILATION EQUIPMENT 259

to find the altitude of the start point. Withthis accomplished he can depress the "start"button to begin the compilation mode, afterwhich his job is to monitor the operationthrough the stereoviewer until manual actionis required. Operator action will be requiredonly in very difficult terrains; the operatorcan mark these for attention before the com­pilation.

FUTURE 0" AUTOMATIC MAP COMPILATION

The success of the Automatic Map Com­pilation System demonstrated the feasibilityof such equipment and ensured the success ofthe UAMCE. The new equipment has manyfeatures, not present in the original, thatmake it useful as a precision comparator andenhance its potential as a compilation instru­ment. However, it cannot be claimed that theUAMCE is the ultimate instrument. Untilthe time that automatic equipment will beable to produce fully annotated, properlycolored maps without the attention of anoperator, improvements will be sought. It isuseful to examine the potential in variousareas of the operation.

The UAMCE allows 10 milliseconds foreach altitude measurement-about a third ofthat required by the earlier system. The im­provement is a result of (1) the higher inten­sity flying-spot scanners, which result in asignificant improvement in the signal-to­noise ratio and (2) a more effective scanningpattern that provides a good averaging overthe area in a few scan lines, permitting highergain in the height-error loop before localizedimage detail can cause an undesired count.The improvement would be greater, exceptthat only information in the central part ofthe scan is used for correlation, resulting in asignificant time loss. The larger area providesthe operator enough of a picture to be useful,while correlation is limited to a smaller areato obtain a more meaningful measurement ofthe altitude for the central point.

Because the effectiveness of the UAMCEoperational cycle remains to be verified, it isnot too useful to speculate on how muchshorter the time can become. If the equip­ment were improved to a point where operatorattention was not required, the scan could becon fi ned to the area used for correlation. Thismight permit a factor of two improvement.Faster scanning would require a higher band­width and, consequently, brighter scanners;some gain might be obtained in this way, but

it would be at the cost of poorer resolutionand perhaps less life from the scanner tubes.

It is anticipated that significant improve­ments in the UAMCE will result from pro­gramming modifications that will permit thedensity of the measurements to adapt to theterrain. To obtain altitude information in con­formance with usual map standards, it will benecessary to have at least two measurementsper contour interval. In flat areas the meas­urement interval can be large, allowing suchareas to be covered quickly; in areas of highrelief, the density of measurement must bevery high (unless it is not important to main­tain the accuracy). It has been estimated thatthe UAMCE will compile 9- by 18-inch, 100%lap diapositives in about 3 hours using (atnominal scale) 2S0-micron (O.Ol-inch) spac­ing along the profile, and SOO-micron (0.02­inch) spacing between profiles. For very flatareas this could be changed to, say, 1,000microns between profiles and, with modifica­tion of the computer programs, an equalspacing along the proftle; this would permit a9- by l8-inch, 100% lap pair to be compiledin less than half an hour. The program shouldbe made to adapt to the terrain to minimizethe time required for each model, while main­taining the required accuracy.

The UAMCE will be improved graduallyas other new computer programs are written.These will permit manual measurement ofspot heights with hard-copy printout, so thatthe data can be added to the map. It willprobably be desirable to write a program toallow the operator to contour manually inselected areas (as in a forest, where the opera­tor can use judgment to extend whateverstereo coverage of the ground is available);the program would print out the altitudechart on the basis of the manually mademeasurements, but print out the orthophotoon the basis of machine-made measurements,preserving the dominant imagery on the or­thophoto. Undoubtedly, still other programswill be found desirable to optimize the opera­tion and minimize the demands on the opera­tor.

Of course, it would be desirable to print outcontour lines, rather than the present rotarysequence of three gray tones. However, untilannotated contour lines (not requiring man­ual touchup) can be automatically drawn,the product would not be as useful as thepresent output. The three-tone sequence is

260 PHOTOGRAMMETRIC ENGINEERING

very easy to follow and easily smoothed toobtain the final product.

The UAMCE should print orthophotos atup to 40-line/mm. resolution without anynoticeable joint between the individual smallareas making up the composite output. Thehigh resolution permits outputs at reducedscale, so that high quality mosaics fromcontiguous stereo pairs can be made. Whileorthophotos may be adequate and desirablefor large-scale maps, they are not useful forsmall-scale maps. Continued efforts will berequired to automatically abstract the de­sired planimetric detail from the orthophotos.

Many groups are working on the problemof recognizing important details in photo­graphs. The compilation problem demandsattention to a new class of recognition prob­lems (such as, in an area under examination,can the structure of the area be discernedfrom the resulting electrical signals?). Thestereo signals should make recognition easier(assuming the scale of the photography issuch as to reveal height variation in thefeatures to be identified). It may be useful tobe able to place upper and lower limits for thealtitude variations within an area rather thanbeing constrained to use the weighted-aver­age altitude where the weighting favors thehigh information content signals.

In the series of equipment developed thusfar, no attempt has been made to store anysignificant amount of the altitude data, ex­cept through the altitude chart. It should berealized that the compilation process requiresthat the measured height of each elementaryarea be available in digital form at some time.It would be a trivial task to store these dataon magnetic tape, making them available for

later use for many applications. For example,the data might be used for cutting reliefmodels, for calculations required for cut-and­fill operations, or for calculating the expectedcoverage of a radar as a functiQn of possiblesites.

CONCLUSION

The development of the Automatic MapCompilation Equipment at Bunker-Ramohas been sponsored by the U. S. Army Engi­neer Geodesy, Intelligence and Mapping Re­search and Development Agency. Bunker­Ramo is proud to be associated with this or­ganization in bringing about a significantimprovement in the art of map compilation.

However, the application of automaticmap compilation need not be limited to mili­tary applications. Its use will be practicalwherever the mapping problem is sufficientto warrant the initial investment. The initialcost, when amortized over any reasonableestimate of the production, should show amuch lower cost per model than can beachieved with conventional instruments.

ACKNOWLEDGMENT

The mapping programs at The Bunker­Ramo Corporation have been carried on by aproject team encompassing many talents andskills. Mr. Melvin L. Baker, project engineeron the UAMCE, and Mr. George P. Miller,senior mechanical engineer, have contributedgreatly to the different programs. Mr. GlennS. Kimball, electrical engineer, and Mr.Floyd Smi th, research assistant, were par­ticularly helpful in the successful demonstra­tion of the Automatic Map CompilationSystem which made it practical to undertakethe UAMCE program.