Near-real-time stereo matching method using temporal and spatial propagation of reliable disparity

A stereo approach to resolve the occlusion problem in stereo video sequence is introduced. We define a measure to evaluate the reliability of an initial disparity in combination with a left-right consistency check. An initial matching cost volume is computed with an absolute difference-census measure. In the spatial propagation stage, the outlier with a low reliability value is replaced/updated with the reliable disparity information in the support region. Because previous methods establish correspondence on a per-frame basis, they cannot obtain temporally coherent disparity results over a stereo sequence. In order to overcome the occlusion problem in a dynamic situation, we employ the modified codebook with color, disparity, reliability, array of the matching cost, and final access time in a temporal propagation procedure. Experimental results show that the proposed algorithm with general-purpose computing on graphics processing units (GPGPU) provides better performance when applied to disparity maps of real-time indoor/outdoor scenes.