The benefits and the costs of three-dimensional displays are first discussed. Then the different depth cues which are used to perceive depth are reviewed and the conclusions of psychological studies which have examined these cues in combination are reported. These conclusions generally support an additive linear model of depth perception with heavy weighting given to binocular disparity, motion parallax, and interposition. However, the presence of motion sometimes reduces the salience of disparity. Techniques for implementing both stereoscopic and perspective displays are then reviewed in detail.

This paper reports on two experiments in which subjects judged the relative depth ordering and subjective quality of depth of simple, geometric figures (planar circle, square, and triangle). The 3-D images were presented on a Tektronix SGS 620 field-sequential stereoscopic CRT. Four sources of depth information (cue types) were combined factorially to construct exemplary 3-D images: Relative Size (angular subtense decreased with increasing depth); Disparity (binocular disparity varied from crossed to uncrossed with increasing depth); Interposition (closer figures overlapped ones farther away in depth); and Luminance (luminance decreased with increasing depth). Inclusion of each of the three monocular cues produced significantly faster depth judgments. However, there was a lack of significant response time effects associated with binocular disparity. Conversely, stereo presentations strongly improved ratings of subjective image quality. These data indicate that stereoscopic images may provide subjectively more compelling depth information than images containing only monocular cues. However, they also provide evidence for at least one limitation of stereoscopic display utility.

Taking advantage of stereo hardware shouldn't require major software development, but choosing the correct projections and transformations isn't easy. Stereo done incorrectly will not only be ineffective, but probably genuinely uncomfortable for the viewer. This paper discusses the correct way to do stereo perspective projections, and how to implement these in software on a Silicon Graphics Iris Workstation.

Two experiments are described which examine the effect of distance in the depth plane, as defined operationally by binocular disparity, on focused and divided attention. In both experiments, disparity was manipulated with a Tektronix stereoscopic 3D display system. In Experiment 1 (focused attention), 10 subjects classified letter stimuli presented at zero disparity. The targets were surrounded, either vertically or horizontally, by two irrelevant stimuli. The distractors were presented at 7 different depth planes relative to the target (i.e., 3 in front, 3 behind, and 1 at the same depth). The distractors showed evidence of being processed when at the same depth plane as the central relevant stimulus, but diminished processing when presented at different depth planes. This pattern of data provides evidence for a fairly "narrow" attentional bandwidth in depth, for focused attention. In Experiment 2 (divided attention), 8 subjects performed a one-axis pursuit tracking task at 0 disparity, while concurrently responding to stimuli presented immediately below, and at 5different depth planes (1 at the same disparity as the tracking task, 2 in front of, and 2 behind the tracking display). While there was considerable interference between the two tasks, there was no evidence that this interference was modulated by similarity of depth when the depth plane of stimulus presentation was known in advance (i.e., was "blocked"). These results suggest that in divided attention, where stimulus location is known in advance, the attentional bandwidth in depth is sufficiently broad to accommodate the same range of depth planes over which filtering had been observed in focused attention. When however the depth plane was randomized and unpredictable, evidence for a depth gradient of divided attention was found. The results are discussed in terms of attentional theory and the usefulness of stereoscopic displays.

The cockpits of both military and civilian aircraft now contain color, multifunction displays. This capability, coupled with advances in graphics generators, has given the crew station designer the ability to create display formats which are very intuitive. However these display formats are limited to 2-D (flat -- such as a road map) or 2 l/2-D (perspective -- such as railroad tracks narrowing to a distant vanishing point) when trying to portray spatial concepts. The recent development of technology to allow for the incorporation of 3-D stereo into the display formats has the potential to dramatically decrease the pilot's information processing load through the creation of formats which can more realistically portray real world scenes that are inherently three dimensional. This paper describes a series of experiments designed to determine if there is payoff in 3-D stereo in combination with various monocular depth cues such as size coding and aerial perspective. Guidelines are presented discussing the types of display formats in which 3-D stereo will have the greatest contribution.

The application of stereopsis (true depth) cuing to advanced heads-down flight display concepts offers potential gains in pilot situation awareness and improved task performance, but little attention has been focused on a fundamental issue involving their use. The goal of this research was to determine whether or not the short-term use of heads-down stereoscopic displays in flight applications would degrade the real-world depth perception of pilots using such displays. Stereoacuity tests are traditionally used to measure the real-world depth perception of a subject. This study used such a test as part of the experimental protocol. Eight transport pilots flew repeated simulated landing approaches using both non-stereo and stereo 3-D heads-down pathway-in-the-sky displays. At the decision height of each approach, the pilots transitioned to a stereoacuity test using real objects rather than a two-dimensional target test apparatus. Statistical analysis of stereoacuity measures (averaged over pilots and replicates), comparing a control condition of no-exposure to any electronic flight display with the transition data from non-stereo and stereopsis displays, revealed no significant differences for any of the conditions. Clearly, transitioning from short-term exposure to a heads-down stereopsis display has no more effect on realworld depth perception (based on stereoacuity) than transitioning from a non-stereo display. However, depth perception effects based on size and distance judgements, and long-term exposure remain issues to be investigated.

Using perspective and stereo, computers present 3-D objects on 2-D display surfaces so that we see the objects as 3 dimensicnal. Extrapolations of perspective and stereo permit display of 4 dimensions. Nine types of images are produced by extrapolating perspective and stereo to 4 dimensions, namely images with: 1. Perspective from w, 2. Perspective from w and stereo from z, 3. Perspective from w and z, 4. Horizontal perspective from w and vertical perspective from z, 5. Stereo and perspective from w, 6. Stereo from w and z, 7. Horizontal stereo from w and vertical stereo from z, 8. Horizontal stereo and perspective from w, and vertical stereo and perspective from z, 9. Horizontal stereo from w and vertical perspective from z. Types6, 7, and 8 produce as many as 4 images. As few as 2 may be sufficient to preserve dimensional information; These images encode information from 4 dimensions in discernible visual cues. Though potentially confusing because they conflict with our conventional interpretation ofdepth cues, they present 4-dimensional objects in an unambiguous way. These images can help us understand 4-dimensional objects.

A new morphological algorithm for automated interpolation of height grids from contour images is developed. Explicit identification of contours within the image is not necessary since interpolation is done directly in image space using morphologic iransforms and labeling schemes. 3-D depth-shaded perspective images of height grids for specified viewing angles are generated using transformed height columns. These methods avoid the processing associated with intermediate contour and polygonal representations. Co-registered continuous tone (gray scale) image data is mapped onto the 3-D views. Advantages of this new recursive algorithm include the ability to interpolate and label whole families of points simultaneously and the basing of interpolated values on surrounding contour morphology. In addition, the methodology for decomposition and labeling of contour intervals is directly amenable to parallel implementation.

Three-dimensional pictorial displays, incorporating depth cues via stereopsis, offer a potential means of displaying information in a natural way to enhance situational awareness and provide increases in operator performance. Conventional computational techniques rely on asymptotic transformations and symmetric clipping to provide the stereo pair. New techniques that replace these conventional computations were developed to increase the control of the stereo-viewing space. Also, the effective region of stereopsis cuing was determined empirically by comparing perceived depth against computed depth. Conventional asymptotic transformations, used to map the visual scene to the stereo viewing volume, allow a single, specific scene distance to be fixed at the screen location. The new piece-wise linear approach allows creative partitioning of the depth viewing volume, with freedom to place the depth cuing emphasis where desired. Asymmetric clipping makes better use of the available display surface than symmetric clipping, and provides increased fields-of-view throughout the depth-viewing volume. The results of the experiment determining the effective region of stereopsis cuing indicate that a practical viewing volume falls between -25%and+60% of the viewer-to-screen distance. Also, the data revealed that increasing viewer-to-CRT distances provide increasing amounts of usable depth.

Existing, documented techniques for the presentation of higher dimensionaL information are characterized. Techniques include: Symbolic Star PLots, Chernoff Faces, Gtyphs, Boxes, Profile PLots, SymboLic Scatter PLots, KLeiner-Hartigan Tree SymboLs, GeneraLized Draftsman DispLays, Andrew's PLots, ParaLLeL Axes Graphics, and Cartesian Hyperspace Graphics. Each technique is evaLuated based on accuracy, simpLicity, clarity, appearance, well-designed structure, information leveL, dimansonaL capacity, flexibility, interpretability, visual impact, mastery time, and computational tractability. Strengths, weaknesses and applicabilities of each technique are determined. Techniques are categorized as symbolic and non-symbolic. Characteristics of each category are identified.

This paper considers the issue of total system lag in real-time interactive computer graphics environments. In these systems, such as virtual environments and simulators, system lag dramatically effects the usability of the system. There are two types of lag discussed in this paper: transmission lag time, the time difference between the moving of a sensing device (such as a position tracker) and the display of that device's motion on a graphic display; and position lag, the difference between the actual position of a tracker in motion compared and the displayed position of the tracker at the same time. Using the Virtual Interactive Environment Workstation being developed at NASA Ames Research Center as the system to be measured, a method of measuring these types of lag using a video technique is described. The relationship between the two types of lag is observed and modeled, as well as a relationship between system lag and graphic update rate. It is found that the position lag can be understood in terms of the transmission lag, so that optimizing a system for small transmission lag will also optimize for small position lag. Using the results described in this paper the lag in other systems can be estimated and the effect of graphics performance on system lag can be predicted.

A stereoscopic workstation is under development for visualizing transparent 3 dimensional (3D) volumes of data. The hardware and software are summarized, and initial results are presented.

The Virtual Environment Workstation Project (VIEW) at NASA's Ames Research Center has developed a remotely controlled stereoscopic camera system that can be used for telepresence research and as a tool to develop and evaluate configurations for head-coupled visual systems associated with space station telerobots and remote manipulation robotic arms. The prototype camera system consists of two lightweight CCD video cameras mounted on a computer controlled platform that provides real-time pan, tilt, and roll control of the camera system in coordination with head position transmitted from the user. This paper provides an overall system description focused on the design and implementation of the camera and platform hardware configuration and the development of control software. Results of preliminary performance evaluations are reported with emphasis on engineering and mechanical design issues and discussion of related psychophysiological effects and objectives.

Stereoscopic displays enhance the ability of the viewer to comprehend spatial relationships. With the commercial introduction of liquid crystal stereoscopic shutters (LCSS), it has become practical to incorporate stereoscopic 3D into moderately-priced graphics workstations, greatly extending the applicability of this technique. This paper reports on the results of an evaluation of stereoscopic 3D for Air Force applications. This work focused on the identification and development of optimized stereoscopic display techniques for the visualization of terrain data.

This paper describes the implementation and integration of the Ames counterbalanced CRT-based stereoscopic viewer (CCSV). The CCSV was developed as a supplementary viewing device for the Virtual Interface Environment Workstation project at NASA Ames in order to provide higher resolution than is currently possible with LCD based head-mounted viewers. The CCSV is currently used as the viewing device for a biomechanical CAD environment which we feel is typical of the applications for which the CCSV is appropriate. The CCSV also interfaces to a remote stereo camera platform. The CCSV hardware consists of a counterbalanced kinematic linkage, dual-CRT based stereoscopic viewer with wide angle optics, video electronics box, dedicated microprocessor system monitoring joint angles in the linkage, host computer interpreting the sensor values and running the application which renders right and left views for the viewer's CRTs. CCSV software includes code resident on the microprocessor system, host computer device drivers to communicate with the microprocessor, a kinematic module to compute viewer position and orientation from sensor values, graphics routines to change the viewing geometry to match viewer optics and movements, and an interface to the application. As a viewing device, the CCSV approach is particularly well suited to applications in which 1) the user moves back and forth between virtual environment viewing and desk work, 2) high resolution views of the virtual environment are required or 3) the viewing device is to be shared among collaborators in a group setting. To capitalize on these strengths, planned improvements for future CCSVs include: defining an appropriate motion envelope for desk top applications, improving the feel of the kinematics within that envelope, improving realism of the display by adding color and increasing the spatial resolution, reducing lag, and developing interaction metaphors within the 3D environment.

Cyberspace is a new computer-based medium that enables groups of people to play the roles of characters in simulations of three dimensional worlds. This paper describes a theatrical approach to design in cyberspace, in which the designer, called a spacemaker, plays the role of a magician in an evolving space. In most computer aided design systems, the designer is regarded as a third person observer of a static three dimensional model. In the approach described here, the spacemaker is considered to be a direct participant in a virtual reality.

Dimension Technologies has developed a black and white 640 X 480 pixel flat panel autostereoscopic computer monitor that is compatible with the IBM PC family of computers and most compatibles. The monitor employs a liquid crystal display as the image generating element. The LCD is used in combination with a patented illumination system to produce vivid stereoscopic Images that can be seen by several observers from several locations in front of the screen. This paper describes the monitor and how It works, the advantages of DTI's autostereoscopic technology, and the results achieved with the monitor as a display device for computer generated images and still television images.

We have adapted home videogame glasses from Sega as workstation stereo viewers. A small (4x7x9 cm.) box of electronics receives sync signals in parallel with the monitor (either separate ROB-Sync or composite video) and drives the glasses.The view is dimmer than with costlier shutters, there is more ghosting, and the user is feuered by the wires. But the glasses are so much cheaper than the full-screen shutters ($250 instead of about $10 000) that it is practical to provide the benefits of stereo to many more workstation users. We are using them with Sun TAAC-1 workstations; the interlaced video can also be recorded on ordinary NTSC videotape and played on television monitors.

Head mounted stereo displays for virtual environments and computer simulations have been made since 1969. Most of the recent displays have been based on monochrome (black and white) liquid crystal display technology. Color LCD displays have generally not been used due to their lower resolution and color triad structure. As the resolution of color LCDdisplays is increasing we have begun to use color displays in our Eyephone. In this paper we describe four methods for minimizing the effect of the color triads in the magnified images of LCD displays in the Eyephone stereo head mounted display. We have settled on the use of wavefront randomizer with a spatial frequency enhancement overlay in order to blur the triacis in the displays while keeping the perceived resolution of the display high.

Binocular parallax graphics and parallel axes graphics are described and contrasted. With appropriate hardware, binocular parallax graphics can make effective, realistic three-dimensional images. Parallel axes graphics uses simpler hardware to render its images, which, though not life-like, present data without loss of information even in more than three dimensions. Binocular parallax graphics is superior for real-world objects; parallel axes graphics is superior for presenting information with more than three simultaneous variables. Binocular parallax graphics benefits from but often requires a substantial repertoire of cues and scan conversion techniques; parallel axes graphics uses fewer cues and maps naturally to the display surface. Binocular parallax graphics offers the advantages and pitfalls of familiarity; parallel axes graphics, in its developmental stage, precludes both.

A new technique for the presentation of cyclopean stereograms is described, in which complete information for the two eyes is contained in a single, printed image. Such "autostereograms" may be generated to contain an unlimited range of 3-D depth forms within certain constraints. The depth image also occurs at multiple depth planes in front of and behind the physical plane, and may be designed to progress recursively through these multiple planes. Specific autostereograms may be generated so as to be used to continuously tile an indefinitely large surface. Further types can be devised to produce different depth images in the two orthogonal viewing orientations. Finally, the autostereogram principle is used to explore cyclopean perception based on the percept ofbinocular luster, which has surprising properties.

A volumetric computer graphics system for interactive three-dimensional object design has been designed and implemented with capabilities which are not avai1abk in traditional polygonal-based systems. These new capabilities provide the user with a sculpting-like environment. The key to the system is its ability to generate and manipulate objects of arbitrary shape and complexity in (volumetric) object space while displaying shaded projection images of the objects in (two-dimensional) image space at various orientations and in "real time" for motion parallax depth cueing. This ability is realized with a volumetric object definition and new techniques for volumetric display and interaction, such as integer-based ray tracing and multiple image buffering. System modules include: 1) the Volume Object Space Module, where the volume definition and object information is stored and accessed; 2) the Image Generation Module, which generates arbitrary two-dimensional views with depth cue information from the three-dimensional object space; 3) the Image Display Module, which handles storage of and access to the actual images displayed; 4) the Edit Module, which performs various object modification tasks; and 5) the User Interface Module, which handles user interaction. A system for interactive volumetric rendering and editing provides great flexibility and extensibility for interactive design and display of arbitrarily complex shapes.

The paper describes the development and evaluation of 3-D Television Systems. 3-D TV had been developed with a view to proving whether it can be a useful remote handling tool which is easy to use and comfortable to view. The paper summarizes the principles of operation, the initial development, the evaluation trials at UK facilities and it reviews the developments which have been found to give improved performance in terms of speed and accuracy of operations and to reduce the number of camera views required.

Recording disparate left and right eye views on separate, synchronised tape recorders has obvious advantages, but the cost, unwieldy post-production, and storage requinments may exclude small scale and industrial users. There is a need for a simple system of recording 3D TV images on a single tape. It may be possible to accommodate stereo pictures in little more than the bandwidth of conventional video, for example by coding as left channel plus left.right difference signal. Simpler techniques involve discarding infonnation, but in practise this can be of little visible consequence. Two methods of combining left and right views into a video signal which can be recorded on any format video recorder are described. The first involves multiplexing fields from the left and right channels. On playback the fields are separated and each displayed twice to avoid ificker. Vertical resolution is reduced, but this is subjectively acceptable. With the second technique portions of the left and right views are combined in a side-by-side half frame fonnat. Horizontal, vertical and temporal resolution are unaffected, but picture size is reduced and aspect-ratio changed. This format is useful when recorded data is input to an image processing system.

An improved assessment methodology has been implemented at NOSC and tested using an instrumented peg-in-hole (PiH) taskboard. Several aspects of the methodology are discussed in light of their implications for future studies of manipulator performance. Using a simple (but highfidelity) force-feedback manipulator, a group of 9 trained operators showed a consistent advantage for stereoscopic TV viewing over monoscopic TV viewing when performing the PiH task. To introduce a controlled element of spatial uncertainty into the testing procedure, taskboard orientation relative to the manipulator and remote video camera head was changed in a randomized order on a trial-by-trial basis. The stereoscopic advantage demonstrated by this study can reasonably be expected to be even more pronounced as the quality of the stereo TV interface is improved and force-feedback provided through the manipulator system is diminished and/or distorted.

The benefits of stereoscopic vs monoscopic TV were explored on remote manipulator simulations designed to minimize object familiarity and monocular cues. A visually complex three dimensional maze constructed of twisted wire defmed the task. In experiment 1, operators were timed while manipulating a rod through the maze to attach wire hangers at predesignated locations. Initial performance was vastly superior with stereo as compared to mono view even though uncontrolled motion cues were present. The occurrence of extreme scores with mono view suggested operators' use of strategies and tactics to compensate for information loss. In Experiment 2, a new maze and task which controlled motion cues replicated the superiority of stereo over mono TV. Changing the maze/camera position to require new motor positioning problems resulted in extreme time scores that decreased across trials and days. Subjective scales and interviews piovided insights into the reactions and strategies developed by operators to overcome performance problems during mono conditions. For wire maze tasks, stereo view provides immediate veridical visual information sufficient to easily and accurately guide motor performance. Results suggest that images provided by mono view are highly ambiguous requiring trial-and-error strategies that produce erratic, time consuming movements by the operator.

Display of a stereoscopic scene requires the generation of two differently projected views of a scene, referred to as the left-eye image and the right-eye image. Stereoscopic images exhibit spatial locality in that the relative displacement between the left-eye and right-eye viewing positions is small enough such that the left and right-eye images are similar in many respects. In this paper we present an algorithm that takes advantage of spatial locality to significantly reduce the computation time needed to ray trace the right-eye image of a stereo pair by using the information available in the left-eye image. The algorithm is derived by combining the repmjection algorithm presented described by Badt for fast generation of animation frames with the stereoscopic perspective projection technique presented in Hodges, et. 1,4

A laproscope capable of providing three-dimensional images in real time has been designed and constructed. Initial tests have been made with a prototype instrument that is 1 1 mm in diameter, a size that is typical of scopes currently in use. Images from two separate viewpoints are combined through special folded optics and brought out along a single light path. Electro-optic shutters are used to alternately switch between the two stereoscopically related beams so that the resulting image information can be delivered to a single video camera. The resulting three-dimensional image may then be viewed on a standard television monitor. This laproscope can be operated with the same ease as other scopes presently available.

This paper discusses (1) a new proofing methodofprinting Cibachrome from monochrome films, producing higher saturation, spatial resolution, and dimensional stability; (2) experiments with inks and materials for mass printing of barrier-strip and lenticular autostereograms; (3) photographic enlargement and barrier-strip scaling; (4) techniques for combining photographs of real objects with computer backgrounds; and (5) the mathematics of projection and interleaving cylindrical (non planar) autostereograms producing up to a 360° viewing angle.

Anplar fields of view filling at least the region directly viewable on the optical axis of the eyes -about 90 - are necessary to support the illusion of being immersed in another space, be it computer generated or the product of remote cameras. Any greater view, up to 270° laterally, enhances the illusion. Stereopsis, and completing the 360° visual sphere by head motion, round out the illusion; they require the wearing of a head-mounted-display (HMD). This paper describes the optical viewing system used in almost all existing HMD systems, and makes the case that, compared to video monitors, these HMD systems provide a qualitatively different kind of access to remote or computer generated reality -a difference that, even in the present crude state of the art, is striking enough to herald a new order of computer interfacing and real time telepresence. A caveat is presented: that if the spatial presentation is false - not "orthospace" - then the charm and power of the illusion, and the utility of the system, are sharply diminished.

The stereoscopic depth perception is known to be a very sensitive function of the human visual system. Against the background of a necessary data-compression in threedimensional television systems (3DTV) with 'look-around capability' (multi-viewpoint-systems) some experimental results about irrelevance reduction in respect of the depth infovmation will be given. The main emphasis is on the quanlization of disparities, which represent the binocular depth information. In the first of two experiments, 3D b/w stills of four natural scenes were used. The disparities of the stereo-image pair were calculated by a correlation method and subsequently used for reconstruction of test stereo pairs. Several stereo-image pairs, differing by degree of quantization errors, were assessed by non-expert observers. The second experiment was carried out with stereo-image sequences of simple moving patterns. Results show that the depth resolution of a TV-stereo-image pair could be considerably coarser than is predicted from visual depth acuity. Some aspects of spatial bandwidth iduction of the disparity data as a further step to irrelevance iduction will also be discussed.

Since the invention of stereoscopy (WHEATSTONE 1838) reasons for and against 3-dimensional images have occupied the literature, but there has never been much doubt about the preference of autostereoscopic systems showing a scene which is 3-dimensional and true to life from all sides (perfect 3-dimensional image, HESSE 1939), especially since most stereoscopic movies of the past show serious imperfections with respect to image quality and technical operation. Leave aside that no convincing perfect 3D-TV-system is in sight, there are properties f the stereoscopic movie which are advantageous to certain representations on TV and important for the 3-dimensional motion picture. In this paper we investigate the influence of apparent motions of 3-dimensional images and classify the different projection systems with respect to presence and absence of these spectacular illusions. Apparent motions bring dramatic effects into stereoscopic movies which cannot be created with perfect 3-dimensional systems. In this study we describe their applications and limits for television.

A variety of low cost devices for capturing, editing and displaying field sequential 60 cycle stereoscopic video have recently been marketed by 3D TV Corp. and others. When properly used, they give very high quality images with most consumer and professional equipment. Our stereoscopic multiplexers for creating and editing field sequential video in NTSC or component(SVHS, Betacain, RGB) and Home 3D Theater system employing LCD eyeglasses have made 3D movies and television available to a large audience.

A stereoscopic vjewing system increases the efficiency of the operator of a remotely controlled mobile platlorm equipped with a manipulator. Two production stereoscopic viewing systems will be described. The test data shows the efficient operation of the plafform in 3D compared with the plodding when doing the task in 2D.