LAUSR

For some applications that do not require a dense disparity map, we propose a fast semi-dense stereo matching method. First, we use grading rank transformation to obtaining the texture points of the image. Then, we use downsampling to reduce the matching complexity of texture points between consecutive frames. The disparity search range of the current frame is constrained according to the matching of texture points between consecutive frames and the disparity map obtained from the preceding frame. Finally, we perform multi-constrained stereo matching and disparity propagation to obtain the disparity map of the current frame. The proposed method reduces the complexity of stereo matching by the temporal disparity constraint and disparity propagation. Experimental results show that the semi-dense disparity map obtained is accurate and fast. For some video sequences, the proposed method can achieve real-time running speed.