中文摘要：

立体匹配通过计算和标识匹配图像的视差图来获得图像的深度信息，一般计算量大，无法满足实时性要求。本文聚焦立体匹配的匹配代价聚集和视差计算环节，在动态规划方法的基础上，提出了一种实时的立体匹配算法。根据连续性约束，提出了基于自适应形状窗口的快速匹配代价聚集算法，加速了臂长和匹配代价聚集的计算效率；利用边缘检测技术获得图像边界信息，修改动态规划的转移方程，使得边界像素可以在整个视差空间中选择视差值，降低边界处匹配视差的误匹配率。实验结果表明：通过结合上述两个步骤的改进算法，可以获得满足实时性要求、高质量的匹配视差图，整体的匹配准确率较高。

英文摘要：

Depth of image information in stereo matching is typically calculated from the disparity map of image sets. This method suffers from expensive computation, which is unable to run in real time. This paper focuses on the problem of cost aggregation and disparity optimization calculation and proposes an improved real-time stereo matching algorithm based on the dynamic programming method. According to continuity constraints, an adaptive-shape-window-based rapid cost aggregation strategy is utilized to increase the computational efficiency of arm length and cost aggregation. By means of edge detection technology, the boundary information is obtained and the transfer equation of dynamic programming is modified such that the boundary pixels can select the disparity in the whole disparity space. Thus, the error-matching rate of disparity can be reduced without increasing computational complexity. The experiments show that the integration of the above two improvements can achieve a fairly good disparity map in real-time together with better matching accuracy.