Abstract:

Published in TR-2000, May 2000, 15 Pages abstract: The ability to perceive depth is common in animals, including humans. It is an ability we take for granted in our everyday lives. To perceive depth, our brains interpret the two slightly different images it receives from our eyes. Similarly, multiple images are needed for computer depth perception. Depth perception algorithms must determine where objects are in one picture in relation to the others. The disparity data is then used to calculate depth. Recognizing these objects is difficult. Using more than two images helps the detection process. We present a trinocular camera setup to take three images of a scene instead of two. The depth extractor works well alone, and it works better with the preprocessor and postprocessor. Taking three views of a scene is more helpful than two, because occlusions are reduced and the effects of shadows are reduced. The algorithm can use the standard two-image format for comparison. With standard test images the algorithm performs well.