在双目视觉系统立体匹配中, 相似度计算是不可或缺的步骤, 而相似性计算的过程存在着精度和速度上的矛盾关系。就此提出了两个加速策略: 用以减少匹配搜索像素点个数的分层筛选策略和用以减小匹配窗口相似性计算量的自适应间隔策略。以绝对差值和(SAD)算法和归一化互相关(NCC)算法为例阐述了加速策略的实现步骤。实验证明使用分层加速策略和自适应间隔策略的可行性, 使得原始算法在基本保持精度的情况下获得了速度的大幅度提升。

In matching of stereo vision system, similarity measurement is an essential step, however, there is always trade-off between accuracy and computational efficiency. Accelerating matching cost calculation in two aspects is proposed : combining efficient method by hierarchical screening and reducing pixels in window by self-adaptive sampling. Sum of Absolute Differences (SAD) method and Normalized Cross Correlation (NCC) method are taken as an example to illustrate implementation of accelerating strategy. Experimental results prove that the proposed strategies achieve a significant improvement in computational cost while maintaining the accuracy of effective method.