A head-mounted three dimensional display*
`
`by IV AN E. SUTHERLAND**
`
`The University of Utah
`Sa.It Lake City, Utah
`
`~;-TRODUOTION
`
`-
`
`the three-dimensional
`idea behind
`_ e fundamental
`is to present the user \\-rith a perspective im;J.ge
`~play
`· ch changes as he moves. The retinal image of the real
`· ects which we s& is, after all, only two-dimensional.
`_ us if we can place suitau:2 +,\·n-dimensional images on
`observer's retinas, we can create the illusion that
`is seeing a three-dimension ... ">] object. Although stereo
`to the three-dimensional illu(cid:173)
`ntation is important
`than the ch::mge th~t takes
`-_ n, it is less important
`in the image when the observer moves his head.
`e image presented by the three-dimensional display
`change in exactly the way that the image of a real
`ject would change for similar motions of the user's
`. Psychologists have long known that moving per(cid:173)
`three-dimensional
`images appear strikingly
`tive
`the three-dimensional
`without stereo presentation;
`lay described in this paper depends heavily on thir,
`""\::inetic depth effect." 1
`In this project we are not making any effort to rnea(cid:173)
`rotation of the eyeball. Because it is very difficult
`measure eye rotation, we are fortunate that the per(cid:173)
`ive picture presented need not be ch:mged as the
`moves his eyes to concentrate on whatever :i:,art of
`picture he chooses. The perspective picture presented
`only be changed when he moves his head. In fact,
`measure only th(; position and orientation of the op(cid:173)
`system fastened to the user's head. Because the op(cid:173)
`system determines the virtual screen position and
`
`-
`
`in this pa.per was performed at Harvard
`"The work reported
`in part by the Advanced Research ProJ(cid:173)
`supported
`·,ersity,
`~ .-\geacy (ARP A) of the Department of Defense under con-
`SD 265, in part by the Office of :\"a.val Resean:h under con(cid:173)
`O.\'R 1866(16), and in part by a long standing agreement
`een Bell Telephone Laboratories and the Harvard Compma(cid:173)
`Laboratory. ThP early work at the ~[IT Linc{)ln Laboratory
`also supported by ARPA.
`
`--Formerly of Harvard Gniversity
`
`the user's point of view, the position and orientation of
`the optical system define which perspective view is
`ar,propriate.
`Our objective in this project has been to surround the
`user with <lisplayed three-dimensional infonnation. Be(cid:173)
`cause we use a homogeneous coordinate representa
`tion, 2 ,3 we can display objects which appe,i.r to be close
`to the user or whi<:h appe::.r to be infinitely far away. We
`can display objects beside the user or behind him which
`will become visible to him if he turns around. The user
`is able to move his head three feet off axis in any direc(cid:173)
`tion to get a better view of nearby objects. He can tum
`completely around and can tilt his head up or down
`thirty or forty degrees. The objects displayed appear
`to hang in the space all around the user .
`The desire to surround a user with information has
`forced us to solve the "windowing" problem. The "clip(cid:173)
`ping divider" hardware we have built eliminates those
`portions of lines behind the observer or outside of his
`field of view. It also performs the division necessary to
`obtain a true perspective view. The clipping divider can
`perform the clipping computations for any line in about
`10 microseconds, or about as fast as a modern high-per(cid:173)
`formance display can paint lines on a ORT. The clip(cid:173)
`piUJ divider is described in detail in a separate paper<
`issue. Because the clipping divider permits
`in this
`three-dimensional
`display of
`dynamic perspective
`dra~rings and arbitrary magnification of two-dimen(cid:173)
`sional drawings, we feel that it is the most significant
`result of this research to date.
`In order to make truly realistic pictures of solid
`three-dimensional objects, it is necessary to solve the
`"hidden line problem." Although it is easy to compute
`the perspective positions of all parts of a complex ob(cid:173)
`ject, it is difficult to compute which portions of one
`object are hidden by another object. Of the soft(cid:173)
`ware solutions now available, 2 -0- 10 only the :.\IAGI9
`and the Warnock 10 approaches seem to ha,·e poten(cid:173)
`tial as eventual rea1-time solutions for reasonably com-
`
`;.Epinted with permission from Proceedings of the AF/PS Fall Joint Computer Conference, Washington, D.C.:
`pson Books, 1968. 757-764.
`
`295
`
`Falkbuilt Ex. 1012, Page 001
`
`

`

`plex situations; the time reqllired by the other methods
`appears to grow with the square of situation complexity.
`The only existing real-time solution to the hidden line
`problem is a very expensive special-purpose computer
`at NASA Houston 11 which can display only relatively
`simple objects. We have concluded that showing
`"opaque" objects with hidden lines removed is beyond
`our present capability. The three-dimensional objects
`shown by our equipment are transparent "wire frame"
`line drawings.
`
`OperaJ,ion of the display system
`
`In order to present changing perspective images to
`the user as he moves his head, we have assembled a wide
`variety of equipment shown in the diagram of Figure 1.
`Special spectacles containing two miniature cathode ray
`tubes are attached to the user's head. A fast, two-dimen(cid:173)
`sional, analog line gener::1,tor provides deflection signals
`to the miniature cathode ray tubes through transis(cid:173)
`torized defledion amplifiers. Either of two head position
`sensors, one mechanical and the other ultrasonic, is used
`to measure the position of the user's head.
`As the observer moves his head, his point of view
`moves and rotates with respect to the room coordinate
`system. In order to convert from room coordinates to a
`coordinate system based on his point of view, a transla-,
`tion and a rotation are required. A computer uses the
`measured head position information to compute the ele(cid:173)
`ments of a rotation and translation matrix appropriate
`to each narti.cular viewing position. Rather than chang(cid:173)
`ing the rnfonnation in the computer memory as the user
`
`HEAD
`POSITION
`
`.·.~
`'
`
`IMEISURESHEAD
`POSITION
`I
`
`COMPUTER
`PROGRAMS
`
`EYEGLASS OISPLAI
`OPTICS
`
`3-0 LINE ~ ~
`ti~
`SPECIFICATION
`IN ROON
`:l; o
`COORDINATES ~ ~
`
`MINIATURE
`CATHODE RA! TUBE
`
`ANALOG
`DEFLECTION
`SIGNALS
`
`3-0 LINE
`CLIPPING
`MATRIX
`SPECIFICATION
`DIVIDER
`MULTIPLIER
`INEIE
`COORDINATES
`
`1-0 LINE
`ANALOG
`SPECIFICATION
`DISPLAY
`IN SCOPE
`DRIVER
`COORDINATES
`
`FIGURE 1-The parts of the three-dimellllional display system
`
`moves bs head, we transform information from roor::.
`coordinate::: to eye c;)()rdinates dynamically as it is dis(cid:173)
`played. A new rotation and translation matrix is loadec
`into the digital matrix multiplier once at the start o:
`each picture repetition. As a part of the display proces.(cid:173)
`the endpoints of lines in the room coordinate system arc
`fetched from memory and are individually transfonnec
`to the eye coordi.nate system by the matrix multiplier
`These translated and rotated endpoints are passed vi&
`an intermediate buffer to the digital clipping divider
`The clipping divider eliminates any information out(cid:173)
`side the user's field of view and computes the appropriat€
`perspective image for the remaining data. The final out(cid:173)
`puts of the clipping divider are endpoints of two-di(cid:173)
`mensional lines specified in scope coordinates. The two(cid:173)
`dimensional line specifications are passed to a bufferec
`display interface which drives the analog line-drawinf
`di:,play.
`We built the special-purpose digital matrix multiplie,
`and clipping divider to compute the appropriate per(cid:173)
`image dynamically because no availabl1c
`spective
`general-purpose computer is fast enough to provide "
`flicker-free dynamic picture. Our equipment can pro(cid:173)
`vide for display of 3000 lines at 30 frames per seconci.
`which amounts to a little over 10 microseconds per line
`Sequences of vectors which form "chains" in which thf
`start of one vector is the same as the end of the previous
`one can be processed somewhat more efficiently than
`isolated lines. Assuming, however, two endpoints for
`every line, the matrix multiplier must provide coordi(cid:173)
`n. . .,.:; tra..,~fonnation in about 5 microseconds per end(cid:173)
`point. Each matrix multiplication requires 16 accumu(cid:173)
`la.ting multiplications; and therefore a throughput o;
`about 3,000,000 multiplications per second. The clip(cid:173)
`ping divider, which is separate and asynchronou.5.
`operates at about the same speed, processing two end(cid:173)
`points in slightly over 10 microseconds. Unlike the fixeci
`time required for a matrix multiplication, however, the
`processing time required by the clipping divider de(cid:173)
`pends on the data being processed. The time required
`by the analog line generator depends on the length oi
`the line being drawn, the shortest requiring about :3
`microseconds, the longest requiring about 36 micro(cid:173)
`st:eonds and an average of about 10 microseconds.
`The matrix multiplier, tlipping divider, and line(cid:173)
`generator are connected in a "pipe-line" arrangement.
`Data "stream" through the system in a carefully inter(cid:173)
`loc~ed way. Each wit is an independently timed digita
`device which provides for its own input and output
`synchronization. ~a.ch unit examines an input flag
`which siguals the arrival of data for it. This data are
`held until the unit is ready to accept them. As the um,
`accepts ll datum, it also reads a "directive" which tells i·
`what to do with the datum. When the unit has acceptee
`
`296
`
`Falkbuilt Ex. 1012, Page 002
`
`

`

`datum, it clears its input flag. When it h~ completed
`operation, it presents the answer on output lines and
`sets an output flag to signal that dak. is ready. In some
`eases the unit will commence tne neX!, task before its
`tput datum has been taken. If so, it will pause in the
`if it would have to destroy its output
`ew computation
`tum in order to proceed. Orderly flow of information
`Uirough the system is ensured because the output flag of
`esch unit serves as the input flag of the next. The aver(cid:173)
`the average
`age rate of the full system is approximately
`te of the slowest unit. Which unit is slowest depends
`on the data. being processed. The design average rate is
`&bout 10 microseconds per line.
`The computer in this system is used only to process
`once per frame,
`lihe head-position sensor infonnation
`the three-dimensional
`&nd to contain and manipulate
`wing. No available general-purpose computer would
`be fas1 enough to become intimately involved in the per(cid:173)
`required for dynamic perspec(cid:173)
`spective computations
`to
`·ve display. A display channel processor serves
`·etch from memory the drawing data required to recom(cid:173)
`pute and refresh the CRT picture. The channel proces(cid:173)
`in many way~ so that it is also
`sor can be "configured"
`and clipping
`pos.5ible to use the matrix multiplier
`·vider independently. For example, the matrix multi(cid:173)
`plier can be used in a direct memory-to-memory mode
`hich adds appreciably to the arithmetic capability of
`the computer to which it is attached. For two-dimen(cid:173)
`it is also possible to bypass the ma(cid:173)
`sional presentations
`trix multiplier and provide direct input to the clipping
`dividei; and display. These facilities were essential for
`debugging the various uruts independently.
`
`Presenting images to the user
`The special headset which the user of the three-di(cid:173)
`mensional display wears is shown in Figure 2. The opti(cid:173)
`eal system in this headset magnifies the pictures on each
`of two tiny cathode ray tubes to present a virtual image
`&bout eighteen inches in front of each of the user's eyes.
`Ea.ch virtual image is roughly the size of a conventional
`CRT display. The user has a 40 degree field of view of
`information displayed on the miniature
`the synthetic
`cathode ray tubes. Half-silvered mirrors in the prisms
`through which the user looks allow him to see both the
`images from the cathode ray tubes and objects in the
`room simultaneously. Thus displayed materul can be
`made either to hang disembodied in space or to coincide
`with maps, desk tops, walls, or the keys of a typewriter.
`The miniature cathode ray tubes mounted on the
`optical system form a picture about one half of an inch
`F-qu.are. Because they have a nominal six tenths mil
`l!p()t size, the resolution of the virtual image seen by the
`user is a.bout equivalent to that avail.a.hie in standard
`
`FIGURE 2-The head-mounted display optics
`with miniature CRT's
`
`large-tube displays. Each cathode ray tube is mounted
`the
`in a metal can which is carefully grounded to protect
`user from shorts in the high voltage system. Additional
`is provided by enclosing the high voltage
`protection
`wiring in a grounded shield.
`The miniature cathode ray tubes have proven easy to
`drive. They us,., 1ectrostatic deflection and focussing.
`Because their deflection plate8 require signals on the
`order of only 300 volts, the transistorized deflection am(cid:173)
`plifiers are of a relatively straightforward d~ign. Com(cid:173)
`to
`followers are used
`emitter
`plementary-symmetry
`to
`drive four sma.11 coaxial cables from the amplifier
`each cathode ray tube. Deflection and intensification
`signals for the roiPiature cathode ray tubes are derived
`from a commercial analog line-drawing display which
`can draw long lines in 36 microseconds (nominal) and
`(nominal).
`lines as fast as three microseconds
`short
`The analog line generator accepts picture information
`in the coordinate system of the miniature cathode ray
`tubes. It is given two-dimensional scope coordinates for
`the endpoints of each line segment to be shown. It con(cid:173)
`nects these endpoints with smooth, straight lines on the
`two-dimensional scope face. Thus the analog line-draw(cid:173)
`ing display, transistorized deflection amplifiers, minia(cid:173)
`ture cathode ray tubes, and head-mounted optical sys(cid:173)
`tem together provide the ability to present the user with
`any two-dimensional line drawilljl.
`
`Head posi,tion sensor
`The job of the head position sensor is to measure
`and report to the computer the position and orientation
`of the user's head. The head position sensor should pro-
`
`297
`
`Falkbuilt Ex. 1012, Page 003
`
`

`

`--
`
`vide the user reasonable freedom of motion. Eventually
`we would like to allow the user to walk freely a.bout the
`room, but our initial equipment allows a working
`volume of head motion a.bout six feet in diameter and
`three feet high. The user may move freely within this
`volume, may turn himself completely about, and may
`tilt his head up or down approximately forty degrees.
`Beyond these limits, head position cannot be measured
`by the sensor. We suspect that it will be possible to ex(cid:173)
`tend the user's field of motion simply by transporting
`the upper part of the head position sensor on a ceiling
`trolley driven by servo or steppin.r motors. Since the
`position of the head with respect to the sensor is known,
`it would be fairly easy to keep the sensor approximately
`cent.ered over the head.
`The head position measurement should be made with
`good resolution. Our target is a resolution of 1/100 of
`an inch and one part in 10,000 of rotation. Resolution
`finer than that is not useful because the digital-tn-r>.na(cid:173)
`log conversion in the display system itself results in a
`digital "grain" of a.bout that size.
`The accuracy requirement of the head position sensor
`is harder to detennine. Because the miniature cathode
`ray tubes and the head-mounted optical system to(cid:173)
`gether have a pin-cushion di6tortion of abouL three per-
`
`FIGURE 4-The ultrasonic head position sensor in use
`
`cent, information displayed tc. the user may appear
`t,.., '"s much as three tenths of an inch out of place. Our
`head positioo sensor, then, should have an accuracy or:
`the order of ooe tenth of an inch, although useful per(cid:173)
`formance may be obtained even with less accurate head(cid:173)
`position information.
`We have tried two methods of sensing head position.
`The first of these involves a mechanical arm hanging
`from the ceiling as shown in Figure 3. This arm is free
`rotate about a vertical pivot iJ:, its ceiling mount. It ~
`two universal joints, one at the top and one at the bot(cid:173)
`tom, and a sliding center eection to provide the six
`motions required to measure both translation and ro(cid:173)
`tation. The position of each joint is measured and pre(cid:173)
`sented to the computer by a digital shaft position en(cid:173)
`coder.
`The mechanical head position sensor is rather heavy·
`and uncomfortable to use. The information derived
`fzom it, howevH, is easily converted into the fore
`needed to generate the perspective transforms.Lion. We
`built it to have a sµre method of measuring head posi(cid:173)
`tion.
`We have also constructed a continuous wave ultra(cid:173)
`sonic head position sensor shown in Figure 4. Three
`transmitters which troosmit ultrasound at 37, 38.6, and
`
`FIGURE 3-The mechanical head position sensor in use
`
`298
`
`Falkbuilt Ex. 1012, Page 004
`
`

`

`lliaking twelve measurements. We have gone to consid(cid:173)
`erable effort to write programs for the ultransonic hei>d
`position sensor. These programs embody several tech(cid:173)
`niL: ues to resolve the measurement ambiguities. Al(cid:173)
`though we have had some encouraging results, a full
`rep, rt on the ultrasonic head position sensor is not yet
`possible.
`
`The perspective tran~format,ion
`
`G~erating a perspective image of three dimensional
`information is relatively easy. Let us suppose that the
`information is represented in a coordinate system based
`on the observer's eye as shown in Figure 6 If the two(cid:173)
`dimensional scope coordinates, X, and Y,, are thought
`of as extending from -1 to + 1, simple geometric reason(cid:173)
`ing will show that the position at which a particular
`is related to its
`point should be displayed on the scr~n
`position in three-dimensional space by th;:, simple rela(cid:173)
`tio::s:
`
`x'
`a
`X. = - cotan-
`z'
`2
`
`y'
`a
`Y, = - ootan-
`z'
`2
`
`If an orthogonal projection is desired, it can be obtained
`by making the value of z' constant. Because the per(cid:173)
`spective (or orthogonaJ.) projection of a straight line in
`line, division by
`three-dimensional space is a straight
`the z' coordinate need be performed ovl) for the end(cid:173)
`points of the line. The two-dimensional analog line-
`
`eye
`
`x'
`
`FIGURE 6--The x' y', z' coordinates system based on the
`observer's eye position
`
`299
`
`FIGURE 5---The ultra.sonic head position sensor logic
`
`40.2 kHz are attached. to the head-mounted optical sys(cid:173)
`tem. Four receivers are mounted in a square array in the
`ceiling. Each receiver is connected to an amplifier and
`three filters as shown in Figure 5, so that phase changes
`in sound transmitted over twelve paths can be measured.
`The measured phase shift for each ultrasonic path can
`be read by the computer as a separate five-bit number.
`The computer counts major changes in phase to keep
`track of motions of more than one wavelength.
`Unlilre the Lincoln Wand 12 which is a pulsed ultra(cid:173)
`sonic system, our ultransonic head position sensor is a
`continuous wave system. We chose to use continuous
`wave ultrasound rather than pulses because inexpensive
`narrow-band transducers are available and to avoid con(cid:173)
`fusion from pulsed noise (such a~ typewriters produce)
`which hed caused difficulty for the Lincoln Wand. The
`choice of continuous wave ultrasound, however, intro(cid:173)
`duces ambiguity into the measurements. Although the
`ultrasonic head position sensor makes twelve measure(cid:173)
`ments from which head-position information can be de(cid:173)
`rived, there is a wave length ambiguity in each of the
`measurements. The measurements are made quite pre(cid:173)
`cisely within a wave, but do not tell which wave is being
`measured. Because the wavelength of sound at 40 kHz
`in air is about 1/3 of an inch, each of the twelve m~(cid:173)
`surements is ambiguous at 1/3 inch intervals. Because
`the computer keeps track of complete changes in phase,
`the ambiguity in the measurements shows up as a con(cid:173)
`at&nt error in the measured distance. This error can be
`thought of as the "initialization error" of the system.
`It is the difference between the computer's original
`guess of the initial path length and the true initial path
`length.
`We believe that the initialization errors can be re(cid:173)
`solved by using the geometric redundancy inherent in
`
`Falkbuilt Ex. 1012, Page 005
`
`

`

`generating equipment can fill in the center P?rtio~ of a
`line by drawing a two-dimens10nal
`three-dimensional
`line. The digital perspective generate, computes values
`only for the endpoint coordinates of a i: ne.
`infomul tion to be presented
`The three-dimensional
`by the three-dimensional display is stored in the corn(cid:173)
`pu ter in a fixed three-dimensional coordinate systeIL.
`Because tl-iis coordinate system is basf'.,d on the room
`around the user, we luwe chosen to call it the "room"
`coordinate system. The drawing data in the room coor(cid:173)
`dinate system is represented in homogeneous coordi(cid:173)
`nates. This means that each three-dimensional point
`or end of a three-dimensional line is stored as four·se(cid:173)
`the
`to
`first three correspond
`parate numbers. The
`ordinary X Y and Z coordinates of three-dimensional
`space. The fourth coordinate, usually called W, is a scale
`factor which tells how big a value of X Y or Z represents
`a unit distance. Far distant material may thus easily
`be represented by ma:~in~ the_ scale factor, W, small.
`Infinitely distant points are reµrt~?nted by setting the
`scale factor, W, to zero, in which case the first three co(cid:173)
`represent only the direction to the point.
`ordinates
`Nearby points are usually represented by setting the
`scale factor, W, to its largest possible value, in which
`case the other three coordinates are just the familiar
`fixed-point r':lpresentations of X Y and Z.
`
`The mat,rix multiplier
`
`We have designed and built a digital matrix multi(cid:173)
`plier to convert information dynamically from the fixoo
`"room" coordinate system to the moving "eye" coordi(cid:173)
`nate system. The matrix multiplier stores a four-by-four
`matrix of 18 bit fixed-point numbers. Becaurn the draw
`in homogeneous coordinates,
`ing data are represented
`the single four-by-four matrix multiplication provides
`for both translation and rotation.' The matrix multi(cid:173)
`plier accepts the four 18 bit numbers which represent an
`them as a four-component vector
`treating
`endpoint,
`which it multiplies by the four-by-four matrix. The
`result is a four-component vector, each component of
`which is truncated 1iO 20 bits. The matrix multiplier de(cid:173)
`livers this 80 bit answer to the clipping divider in ap(cid:173)
`proximately 5 microseoonds. It
`therefore performs
`about three million scalar multiplications per second.
`The matrix multiplier uses a separate multiplier
`module for each column. Each module contains an ac(cid:173)
`cwnulator, a partial product register, storage for the
`four matrix elements in that column, and the multipli(cid:173)
`cation logic. The entries of a row of the matrix serve
`simultaneously as four separate multiplicands. An in(cid:173)
`dividual component of the incoming vector serves as
`the common multiplier. The four multiplications for a
`
`single row are thus performed simultaneously. For
`additional speed, the bits of the multiplier are examined
`four at a time rather than individually to control multi(cid:173)
`ple-input adding arrays.
`
`The dipping or windowing task
`
`The job of the clipping divider is to accept three(cid:173)
`dimensional information in the eye coordinate system
`two-dimensional end(cid:173)
`and convert it to appropriate
`points for display. If both ends of the line are visible,
`the clipping divider needs merely to perform four divi(cid:173)
`sions one for each two-dimensional coordinate of each
`'
`end of the line. Enough equipment has been provided ir:
`the clipping divider to perform these four divisions
`simultaneously.
`If the endpoints of a line are not within the observer'~
`field of view, the clipping divider must dteide whethecr
`any portion of the line iE within the field of view. If sc.
`it must compute appropriate endpomts for that portic =
`as illustrated in Figure 7. Lines outside the field of ,ie-;.
`or behind the user must be eliminated. Operation of ·t::
`clipping divider is described in a separate pa.per' in th·(cid:173)
`issue.
`Like the matrix multiplier, the clipping divider is EC
`independently-timed digital device which provides fc~
`its own input and outpu~ sy!'c!i.ronization. It has an iL(cid:173)
`put and an output flag which provide for orderly flow _:
`information through the clipping divider. If a lina li~
`entirely outside the field of view, the clipping divide
`will accept a uew input without ever raising its ou . .:_
`flag. Thus only the visible portions of lines that are s._
`the clipping divide:-
`or partly visible get through
`
`CUPPING IN 3 DIMENSIONS
`
`p
`
`.. ---::
`
`-+---+---+--+-Xs
`
`i_REJECTF:~-
`085ERVE;;
`
`-
`
`SCOPE COORDINATES
`
`EYE COORDINATES
`
`FIGURE
`
`a.nd perspective projec: ..
`,-Clipping
`in three dimensions
`
`300
`
`Falkbuilt Ex. 1012, Page 006
`
`

`

`I did some preliminary three-dimensional display ex(cid:173)
`periments during late 1966 and early 1967 at the MIT
`Lincoln Laboratory. We had a relatively crude optical
`system which presented information to only one of the
`observer's eyes. TLe ultransonic head position sensor
`operated well enough to measure head position for l few
`minutes before cumulative errors were objectionable.
`The coordinate transfonnations
`and perspective com(cid:173)
`putations were perfonned by software in the TX-2. The
`clipoing operation was not provided: if any portion of a
`line was off the screen, the entire line disappeared.
`Even with this relatively crude system, the three
`dimensional illusion was real. Users naturally moved to
`positions appropriate
`for the particular views they
`desired. For instance, the "size" of a displayed cube
`could be measured by noting how far the observer must
`move to line himself up with the left face or the right
`face of the cube.
`Two peculiar and >\S yet unexplained phenomena oc(cid:173)
`curred in the preliminary experiment. First, because the
`displayed. information consisted of transparent
`"wire(cid:173)
`frnme" images, ambiguous
`interpretations were still
`possible. In one picture a small cube was placed above a
`larger one giving the appearance of a chinrney on a
`house. From viewpoints below the roof where the
`"chimney" was seen from inside, some concentration
`was required to remember that the chimney was in fact
`further away than the building. Experience with physi(cid:173)
`cal obJects insisted that if it was to be seen, the chimney
`must be in front.
`A second peculiar phenomenon occurred during the
`display of the bond struc1ure of cyclo-hexane as shown
`in Figure 8. Observers not familiar with the rippling
`hexagonal shape of this molecule misinterpreted
`its
`shape. Because their view of the object was limited to
`certain directions, they could not get the top view of the
`molecule, the view in which the hexagonal shape is most
`clearly presented. Observers familiar with molecular
`shapes, however, recognizedthe object as cyclo-hexane.
`In more recent experiments with the improved optical
`system and vastly improved computation capability,
`two kinds of objects have been displayed.. In one test, a
`"room" surrounding the user is displayed. The room i.s
`shown in Figure 9 as it would look from outside. The
`room has four walls marked K, S, E, and W, a ceiling
`marked C and a floor marked F. An observer fairly
`quickly accommodates to the idea of being inside the
`displayed room and can view whatever portion of the
`room ho wishes by turning his head. In another test a
`small cube was displayed. in the center of the user's
`operating area. The user can examine it from wlu tever
`side he desires.
`
`FIGURE
`
`8-A
`
`perspective view of the
`computer-displayed
`cyclo-hexane molecule
`
`FIG{JRE 9--A
`
`perspective view of the
`computer-displayed
`"room" as seen from outside
`
`The biggest surprise we have had to date is the favor(cid:173)
`able response of users to good stereo. The two-tube opti(cid:173)
`cal system presents independent images to each eye. A
`mechanical adjustment
`is available to accommodate to
`the different pupil separations of different users. Soft(cid:173)
`ware adjustments
`in our test programs also permit us to
`adjust
`the virtual eye separation used for the stereo
`computations. With these two adjustments
`it is quite
`easy to get very good stereo presentations. Observers
`capable of stereo vision unifonnly remark on the realism
`of the resulting image&.
`
`301
`
`Falkbuilt Ex. 1012, Page 007
`
`

`

`:\fEXT
`ACK~OWLEDG
`When I started work on the head-mounted display I
`had no idea how much effort would be invohed. The
`project would have died many times but ,or the spint
`of the many people who have become involved. The
`ultrasonic head-position sensor was designed and built
`at the :\HT Lincoln Laboratory by Charles Seitz and
`Stylianos Pezaris and is available for our continued use
`th.rough the cooperation of Lincoln Group 23. Seitz, as a
`the matrix multi(cid:173)
`Harvard employee, later designed
`olier. Robert Sproull, a most exceptionally capable
`Harvard Senior, simulated, designed most of, built
`parts of, and debugged the clipping divider. Two gradu(cid:173)
`ate students, Ted Lee and Dan Cohen have been an es(cid:173)
`sential part of the project throughout. Our many argu(cid:173)
`cl.ippi~, hid(cid:173)
`ments a.bout perspective presentation,
`den-line algorithms, and other ~ubjectE form one of the
`most exciting educational experiences I have had. Ted
`Lee's programs to display cury,:,.--1 surfaces in stereo have
`t_ ·.•i.1.en'~ programs
`boon the basis for rr::-ny cxperimen~s.
`to exercise the entire system form the basis of the dem(cid:173)
`onstrations we can make. I would also like to thank
`Quintin Foster who supervised construction and debug(cid:173)
`ging of the equipment. And finally, Stewart Ogden, so
`called "project engineer," actually chief administrator,
`who defended us all from the pressures of paperwork so
`·
`that something could be accomplished.
`
`REFERE~CES
`Ja
`1 BFGREEX
`Fiqure coherence in the kinetic depth effect
`Journal of Experimental Psychology Vol 62 X o 3 2i2-282 1961
`2 LG ROBERTS
`Machine perceptum of three-dirnensumal solid.,
`MIT Lincoln LaboNtory Technical Report ~o 315 yfay 22
`1963
`3 LG ROBERTS
`Homogeneous 1ruuri,x represen/,alum and manipulation
`
`of N-
`
`dim.ensional constructs
`The Computer Display Review Adams As.sociates \1ay 1965
`IE SUTHERLA~D
`4 RF SPROULL
`.{ clipping divider
`Proceedings of the Fall Joint Computer Conference 1968
`this issue
`5 'J COHE~
`A program for drawing bodie3 with the hidden lines removed
`f"' course 6.539 MIT Fall 1965
`A term-prcject
`0 '.-IT HA YXES
`A. computer method for perspective drawing
`Master's Thesis Texas A&M University Aug 1966
`i P LOUTREL
`to IM hidden-line problem for computer-drawn
`A solutum
`polyhedra
`~ew York Cniversity Tech.weal Report 400--167 (Thesis)
`Bronx Xew York September 1967
`8 A APPEL
`The notion of quantitative invisibility awl the machine rendering
`of solids
`Proceedings of 22nd ~ational Conference ACM
`ACM Publication p 6i Thompson Book Company Washington
`DC 1967
`Spplications Group Inc (\L-\GIJ
`9 \lathematical
`S-D simulated graphu:s
`Datamation February 1968
`!O J E W AR~OCK
`A hidden line algorilhmfor halftone pi.cture represenlal.um
`C:niversity of C'tah Technical Report 4-5 \fay 1968
`installed at the \tanned Space Craft Center at
`l I Equipment
`the direction of the
`is under
`Houston Texas. The project
`under
`General Electric Company Electronics Laboratory
`'.llASA Contract No NAS 9-3916
`12 LG ROBERTS
`TM Lincoln wand
`MIT Lincoln Laboratory Report June 1966
`13 AC TRAUB
`Stereoscopic display using rapid varifocal mirror o&eillaticru
`Applied Optics Vol 6 number 6 June 1967
`14 P VLAHOS
`The three-dimensional display: /1.$ cuu and techniques
`Journal of the Society for Information Display Vol 2 Number
`6 ~ov /Dec 1965
`1.5 R LA~D IE SUTHERLAND
`Real time color sl.ereo compuJer displays
`To be published in Applied Optics
`
`302
`
`Falkbuilt Ex. 1012, Page 008
`
`