基于区域快速生长的双目视觉立体匹配算法研究

Binocular Vision's Stereo Matching Algorithm Research Based on Regional Rapid Growth

立体匹配是立体视觉的重要研究内容之一,视差图的可靠性和计算复杂度直接影响了立体视觉的性能。针对目前双目视觉立体匹配算法计算量过大,本文提出了一种基于快速生长的区域立体匹配算法,应用差错能量作为匹配代价分别进行线性生长和跳跃式生长,遍历整幅图像后得到视差图,并采用均值滤波器滤除不可靠视差。实验表明该算法在保证较好可靠性的同时具有很好的实时性。

Image stereo matching is one of the key research topics in stereo vision. Reliability of disparity maps and computational cost of algorithm directly affect the performance of stereo vision. Because the binocular vision stereo matching algorithm is complex, in this paper, a region matching algorithm based on the rapid growing is proposed. In the algorithm, the error energy is used as the matching cost for the linear growing and leaping growth. After traversing the entire image the disparity map is obtained, and the mean filter is used to filter out unreliable disparity. Experiments show that the algorithm ensures reliability and has good real -time performance.