Depth estimation of frames in image sequences using motion occlusions

摘要

This paper proposes a system to depth order regions of a\udframe belonging to a monocular image sequence. For a given frame, re-\udgions are ordered according to their relative depth using the previous\udand following frames. The algorithm estimates occluded and disoccluded\udpixels belonging to the central frame. Afterwards, a Binary Partition\udTree (BPT) is constructed to obtain a hierarchical, region based repre-\udsentation of the image. The nal depth partition is obtained by means\udof energy minimization on the BPT. To achieve a global depth ordering\udfrom local occlusion cues, a depth order graph is constructed and used to\udeliminate contradictory local cues. Results of the system are evaluated\udand compared with state of the art gure/ground labeling systems on\udseveral datasets, showing promising results.