基于动态规划的分层立体匹配算法研究

动态规划是双目立体匹配的经典算法,针对控制点动态规划算法易产生横向条纹以及立体匹配普遍的边缘性和弱纹理区域问题,提出一种基于金字塔分层双向动态规划的改进立体匹配算法。该算法将分层模型加入传统动态规划方法,以低像素层级为高像素层级提供控制点集,并在匹配代价计算中采用一种自适应相关性测度函数,加以匹配代价滤波,提高算法精确度及实时性并获取高精度视差图。以Middlebury标准库中的图片以及实拍图片作为实验对象,实验表明所提出的方法具有较好的性能。

Improved dynamic programming based stereo matching algorithm

Dynamic programming is a classical algorithm of stereo matching. In order to solve the well known streaking effect of the dynamic programming algorithm and widespread marginal and weak texture region problem of the stereo matching, a kind of improved pyramid and bidirectional dynamic programming based stereo matching algorithm is presented. The algorithm takes the image pyramid model into the traditional dynamic programming method. The lower pixel level provides control points for the high pixel level, and in the matching cost computation, an adaptive correlation measure function is used. The proposed method improves the accuracy and real time performance, and the high precision disparity map can be obtained. The picture in the Middlebury standard library is used for the experiment object, and the experiment result shows that the proposed method has better performance.