Virtual Stereo Display Techniques for Three-Dimensional Geographic Data Abstract (Beaton et al., 1987; Moellering, 1989; Robinson, 1990; Techniques have been developed which allow display of Hodges, 1992). While these technological innovations are volumetric geographic data in three-dimensionalform using likely to contribute to broader use of stereo images, new con- concepts of parallactic displacement. Image stereo pairs and cepts of vision have established that this leftlright separation synthetic stereo of digitized photographs, SPOT, Landsat m, is not necessary to establish a virtual three-dimensional im- shaded relief, GIs layers, and DLG vector data are imple- age Uules, 1971; Marr, 1982; Jones et al., 1984). mented. The synthetic stereo pair is generated from a single Stereovision is not a reflex action but is in fact an intel- image through computation of a parallax shifl proportional lectual activity in which one depth map, for example, the to the volumetric values at each location in the image. A left image of a stereo pair, is received by the eye-brain com- volumetric dataset, such as elevation or population density, bination and stored, and later merged with a second depth provides the single requirement for generation of the stereo map, the right image of the stereo pair (Marr, 1982). This image from an original digital dataset. To view the resulting concept of vision opens new approaches to the display of stereo images, softcopy methods, including traditional ana- three-dimensional geographic information and, with the use glyph techniques requiring complementary colored lenses of current computer graphics technology, can be imple- and an autostereoscopic procedure requiring no special mented at low-cost. Simultaneously, the development and viewing equipment, are used. Hardcopy methods include an- distribution of large-format color hardcopy devices, such as aglyph displays of combined vectorlraster datasets on large electrostatic plotters, have provided opportunities to use tra- format electrostatic plotters. Stereo results for the Driftess ditional anaglyph image separation techniques to yield pano- area of Wisconsin, for northern Georgia, and for the Grand ramic views of extensive geographic areas. Canyon are presented. Current alternatives for achieving virtual three-dimen- sional displays include motion parallax systems based on Introduction single images, stereo and alternating pair systems using two Extracting geographic information for topographic mapping images, and varifocal mirror and holographic systems using from image sources has been traditionally accomplished more than two images (Kraak, 1988; Hodges, 1992). Not all of using stereoscopic methods. However, the application of these technologies are applicable to cartographic feature ex- stereoscopy to support display and analysis of geographic traction and display of geographic phenomena, but the stereo phenomena in geographic information systems (GIS) has not and alternating pair methods hold particular promise for en- been effective because of prohibitive costs, data processing hancing the productive capability of cartographic systems limitations, and requirements for specialized equipment, in- (Fornaro et al., 1985). cluding visual aids for the analyst. With current develop- It is the primary objective of this paper to present virtual ments in computer graphics technology and an improved three-dimensional images as a viable alternative for display, understanding of human vision, these impediments are being mensuration, and analysis of geographic phenomena. The removed. thesis is that increased technological capability through digi- Cartographic and image processing researchers have de- tal and electronic media, and a better theoretical understand- veloped synthetic stereo generation techniques for both con- ing of human visual processing of depth information, now tinuous and discrete data distributions (Batson et al., 1976; permit virtual three-dimensional images to be used economi- Jensen, 1980). The methods of display have required visual cally for simple display, mensuration, and analysis tasks. separation of the left and right images of a stereo pair to the The approach is to provide the theoretical background of the left and right eyes, respectively (Wolf, 1983; Kraak, 1988). methods and discuss implementation alternatives and result- Separation of the left and right images has traditionally been ing images. accomplished passively in a simultaneous view time-parallel To accomplish the objective, the application of angular method with stereoscopes, polarized lenses, or anaglyph and motion parallax from volumetric data to the display of methods. Recently, new computer-based technologies offer- both raster and vector geographic datasets is explored. Spe- ing alternating views for time-multiplexing stereo images cifically, Landsat Thematic Mapper (TM)and SPOT stereo im- have been introduced. These techniques provide separation ages and digitized photographic data in both real and on 120 hertz non-interlaced displays with liquid crystal synthetic stereo forms are used. Monoscopic raster images shutters and radially polarized lenses or, alternatively, by ac- and U.S. Geological Survey (uSGS) digital line graphs (DLGS) tive lenses which include the liquid crystal shutters and are are modified using digital elevation models (DEMS) to create synchronized with the display by an infrared transmitter synthetic stereomates. The concepts of angular parallax and motion parallax are analyzed in the context of these datasets using anaglyph and VISIDEPt" realizations (Jones et a]., 1984). - Photogrammetric Engineering & Remote Sensing, These concepts are implemented within current computer Vol. 59, No. 12, December 1993, pp. 1737-1744. 0099-111219315912-1737$03.00/0 E. Lynn Usery 01993 American Society for Photogrammetry Department of Geography, University of Wisconsin-Madison, and Remote Sensing 384 Science Hall, 550 North Park street, Madison, WI 53706. PEER-REVIEWED ARTICLE 1 I graphics capabilities embedded in GI~and image processing are used to explain depth perception in photos and paint- systems. ings. This group of cues include relative size, linear perspec- tive, height of objects above the line of sight, shadow, relative brightness, and atmospheric attenuation. Motion par- Background allax is the only cue which involves motion of either parts of Three-Dimensional Display of Geographic Data the visual scene or the observer that has been considered by ~erce~tualtheorists. The focus of this paper is display and measurement of geo- . A graphic data in true three-dimensional form rather than as Motion parallax is a monocular physiological cue which perspectives, contours, shaded relief, or other techniques can be easily observed by scanning the objects in view. As which allow interpretation of volumetric information from a the eye moves from one position to another, close objects two-dimensional presentation. Two-dimensional presentation move faster across the retina than far objects. This cue is of three-dimensional geographic data is well-represented in consistently in use through motion of the body as a whole as the cartographic literature in standard textbooks and will not in walking or through motion of the eyes themselves as they be examined here [Keates, 1982; Wolf, 1983; Robinson et al., scan a scene and interpret phenomena. 1984). Kraak (1988) examines some aspects of three-dimen- Recent research indicates that human three-dimensional sional displays but also includes two-dimensional presenta- perception is not a reflex reaction to stimuli from two tions. Jensen (1978) reserves the term three-dimensional for sources received by the two eyes independently (Marr, 1982; only those displays resulting in a virtual image such as ana- McLaurin et al., 1986). Studies of persons with exceptional glyph stereo, holograms, and plastic relief models. This ap- eidetic ability have demonstrated human capacity to fuse proach will be followed in this paper, such that any image random dot stereograms and interpret the three-dimensional termed three-dimensional will use three full dimensions for information when time separation of the images spans two or display, although the third dimension may be virtual as in more days (Jules, 1971). This ability indicates a retention of an inaglyph display. the depth map in memory and later fusing that map with a Any spatial dataset containing quantitative information new source also containing a depth map. Ogle (1963; 1967), over an area creates a surface and may be represented as a Jules (1971), and others have demonstrated that this ability is three-dimensional virtual image. Human visual perception not unique to eidetikers but is a part of normal human visual requires two separate views of these surfaces to generate the perception when the time delay is reduced to 100 millisec- virtual three-dimensional image, and methods to generate the onds. Marr (1982) hypothesized that a depth map is actually second view by introduction of parallax based on the volu- stored in short term memory to be fused with a second depth metric information have been termed synthetic stereo in the map received after the time delay. image processing community (Batson et al., 1976). Jensen Jones et al. (1984) have used this concept to develop a (1980)discusses methods of applying this concept to point, three-dimensional imaging technique based on visual depth line, and area cartographic objects as well as raster-based perception by parallax induction