A Stereo Object Segmentation Algorithm Based on Disparity and Temporal-Spatial Information

摘要

A novel stereo object segmentation algorithm based on disparity matching in rank domain and temporal spatial information is proposed in this paper. By doing a rank transform, stereo videos are mapped to the rank space from the gray-level space, which eliminate the interference of the environment variations and diverse cameras. The disparity is estimated by a region-based matching method with fast searching stratagem. A disparity compensation step is done by smooth area based algorithm to refine the matching result. Combine with temporal and spatial segmentation method in 2D, the proposed disparity segmentation enhance the performance of stereo segmentation. The experiment result indicates the performance and practicability of our algorithm.