TRACKABILITY AS A CUE FOR POTENTIAL OBSTACLE IDENTIFICATION AND 3-D DESCRIPTION

In many man-made environments, obstacles in the path of a mobile robot can be characterized as shallow, that is, they have relatively small extent in depth compared to the distance from the camera. We present a framework for segmenting shallow structures from their background ove r a sequence of images. Shallowness is first quantified as affine desc ribability. This is embedded in a tracking system within which hypothe sized model structures undergo a cycle of prediction and model-matchin g. Structures emerge either as shallow or nonshallow based on their af fine trackability. Two major contributions of this work are (i) aggreg ate object tracking based on 3-D motion and structure constraints in c onstrast with traditional primitive feature tracking based on image mo tion heuristics, and (ii) use of temporal behavior for object segmenta tion and 3-D reconstruction.