视觉客流检测中的动态轮廓匹配方法

Dynamic contour matching approach in vision-based passenger flow detection

为了利用深度信息提升视觉客流统计系统的准确率,提出了一种基于轮廓匹配的头顶部视差获取方法.该方法借助透视比例约束进一步缩小匹配搜索空间,利用头顶部轮廓相似测度获得最优匹配,并采用动态规划方法在多层路径匹配图中搜索最优匹配路径进而获取头顶部视差.左右视图的头顶部轮廓提取过程可以并行实现,并且视差提取过程避免了复杂的致密视差求取运算,因此该方法非常适用于对实时性要求很高的嵌入式场合.实验结果表明,将该方法应用于视觉客流统计系统,客流检测的准确率获得了较大幅度的提升.

A method based on contour matching to extract the head disparity was presented, in order to improve the accuracy rate of vision-based passenger flow estimation system by using the depth information. The perspective constraint was applied to reduce the search space, and the bead＇s contour similarity measure was employed to find the optimal matching, and then the head disparity was obtained in the optimal matching path which was searched by using the dynamic programming approach in the multistage graph. The method is very suitable to be applied to the real-time embedded system because it can extract the heads＇ contours in stereo pairs in parallel and avoid the complex dense disparity computation. The experimental results show that the accuracy rate achieves a great improvement when applying the method to the vision-based passenger flow estimation system.