Segmentation of optical flow and 3D data for the interpretation of mobile objects

The detection of mobile 3-D objects using both stereovision and motion information is considered. First, the authors present an optical flow estimation method based on a multiresolution technique for the matching of contour chain points. Then an algorithm of segmentation of the chains using the continuity of the gradient of the apparent velocity is described. A similar algorithm is developed for stereo vision: the segmentation of the chains is performed in function of their 2-D proximity and the continuity of their disparity. These two segmentations are then used to interpret the scene in terms of isolated rigid or deformable objects. Results are shown for two complex real road scenes.<<ETX>>