Fast Stereo Matching using Adaptive Window based Disparity Refinement

Aiming at the trade-off between accuracy and efficiency in current local stereo matching, a fast stereo matching method based on adaptive window is proposed. The census transform combined with absolute differences (AD) is used for matching cost initialization. Then an iterative cost aggregation based on exponential step adaptive weight (ESAW) is adopted to improve the parallelism and execution efficiency. In disparity refinement stage, an adaptive window based on pixel’s color similarity and Euclidean distance is built for each unreliable point. Then the unreliable points are further classified as ‘occlusion’ and ‘mismatch’, and different refinement strategies are taken for different classifications. Finally, the proposed method is optimized with compute unified device architecture (CUDA) and evaluated on graphic processor. The experiment results show that the proposed method is the most accurate among the real-time stereo matching methods listed on the Middlebury stereo benchmark.