Fast Object Hypotheses Generation Using 3D Position and 3D Motion

This contribution proposes a method to generate object hypotheses from stereo obstacle detection and image motion. Our algorithm is a general approach since it does not require any a priori information about the shape of the observed objects but relies on the basic assumption that the objects are rigid. The algorithm has two processing stages: First, obstacles are detected using stereo vision. Second, each obstacle is segmented into clusters of consistent motion in 3D space. The clustering process explicitly accounts for measurement uncertainties of stereo disparity and 2d motion. Our system may serve as a general feature for higher-level object detection and classification.