No abstract

of these for certain types of scene, so we hope to be able to label lines and nodes successfully without any type of coding. We investigate recent ideas in the design of irinocular active range-sensors. Such devices have the advantage of freedom from mechanical scanning, and rapid image capture. The main technical problem is overcoming the correspondence problem. This requires careful geometric design to take account of epipolar geometry and thorough modelling of image-measurement error. We present a novel design that, so far, seems to work well. Curiously it involves setting up the projectorcamera geometry to be degenerate -so that depth computation is ill-conditioned -and then backing off a little.Work on active rangefinders for 3-D inspection and robot vision has been in progress for about two decades. Overviews of the most widely-investigated approaches can be found in (Bastuscheck 1989) and (Jarvis 1983a). Some of these are Moire fringing, ratio image techniques (Bastuscheck & Schwarz 1984), or time-of-flight range systems (Jarvis 1983b). We are concerned with structured-light systems (Shirai 1972, Altschuler et al. 1981, Altschuler et al. 1987. Typical of systems that are quite well-developed for industrial applications in this area (e.g. profiling turbine blades) is the system of Mundy and Porter (1987). This device uses special hardware to give 60,000 range readings/sec, albeit over a small depth range. Like the less depth-limited but slower system of Case, Jalkio and Kim (1987), it involves mechanical scanning of a light pattern across a scene.There are still other systems which, like ours, attempt to get away from the expense and complexity of a mechanical scanner and move closer to simultaneous acquisition of all range data, essential for a moving scene. Such systems include that of Godin and Levine(1989) and that of Hu and Stockman(1989), the latter of which uses a grid of light to illuminate the scene, providing easilymatched artificial "surface features". However there is a correspondence problem. Regarding the camera and projector as a stereo pair (figure 1), and given an image point, it cannot be uniquely matched to a node point on the grid mask. This is the node labelling problem. Some solutions which have been proposed include colour (Boyer and Kak 1987) or thickness (Le Moigne and Waxman) coding, or space coding (Posdamer and Altschuler 1982, Altschuler et al. 1987). There are objections to all The solution of Stockman and Hu to the node labelling problem can be explained as follows. Given a projected grid-crossing on the image plane, it is required to determine which node in the projected pattern corresponds to that crossing. The position of the grid-crossing in the image plane is associated with an epipolar line in the projector plane (on the left of figure 1). Possible solutions are those nodes that lie on (or sufficiently near) the epipolar line. This generates a set of possible solutions for each grid crossing which can then be reduced somewhat by constraint propagation, between crossi...