Unsupervised VOP Segmentation of Stereo-Captured Video Sequences

A modified active contour-based scheme is incorporated in this paper for unsupervised stereoscopic semantic segmentation. The scheme depends on the exploitation of depth information provided by stereo analysis methods. Assuming that a depth segments map is available, for each depth segment an active contour is unsupervisedly initialized on the depth contour. Then the active contour moves towards the VOP's boundary guided by an external force. This external force originates from a simplified edge map, constructed by a newly proposed scanning process, and it is used as directional attraction field. Finally a greedy approach is incorporated to lead the active contour from its initial position to the VOP's boundary. Experimental results indicate the reliable performance of the proposed scheme on real life stereoscopic video sequences.