Parallel Dense Binary Stereo Matching Using CUDA

This paper addresses the problem of dense stereo matching from two rectified images without prior information about the structure of the scene. CUDA is proposed to exploit the data independence inherently present in local stereo matching to handle the accuracy-time trade-off. To address edge fattening, the parallel implementation utilizes an outstanding aggregation technique that is followed by a hybrid matching metric which is proved to enhance the matching accuracy. By porting the whole pipeline to the GPU, a speedup ranging from 34x to 108x is achieved without compromising the accuracy of the resulting disparity map. The Middlebury benchmark and the latest dataset are used to evaluate the results.