The stereo matching algorithm for binocular stereo vision is a critical and challenge issue in the field of computer vision, which may has lower matching quality or higher time complexity. To improve that, a boundary extension is pre-processed to improve matching accuracy of the image boundary area, offered a 5% matching accuracy increase; and then a hierarchical multi-resolution block matching algorithm for high-resolution image pair is proposed in the paper; based on the hierarchical multi-resolution block matching, a novel fast method is proposed, whose complexity is only one-ninth of original algorithm; and a method of variable window is adopted to obtain more satisfactory result on the untextured regions, offered a further 2% matching accuracy increase.Stereo matching algorithms are computationally intensive algorithm, with great computational cost and available parallelism. With multi-core technology developing, more and more researchers pay attention to using multi-core processors and many-core processor. In this paper, based on the research of NVIDIA GPGPU, the GPGPU-based parallel stereo matching algorithm is proposed, which can be 190 times faster than the traditional CPU implementation. |