用于行人头部特征提取的目标区域匹配方法

An Target Region Matching Algorithm for the Head Feature Extraction of Pedestrian

为了准确地定位与跟踪序列图像中的运动行人以获取精确的客流量信息,提出了一种基于目标区域匹配的行人头部特征提取新方法。与常用的基于致密视差图的头部区域视差获取方法不同,该方法基于"先分割后匹配"的思想,即首先借助单目图像处理方法对基准图进行分割,获取候选头部区域;然后直接将这些候选头部区域作为目标区域,在匹配图中搜索其匹配对应区域以获取候选头部区域的视差;再借助候选头部区域的视差提取出候选头部区域的深度与透视特征,用于去除虚假头部区域以获取最终的头部检测结果。性能测试与实验结果表明,该方法不仅视差提取精度高、实时性好,并且借助该方法获取的头部特征具有较高的区分度,可以有效去除候选头部区域中的虚假头部区域,使客流量检测的准确率达到90%以上。

In order to acquire accurate passenger flow information by locating and tracking moving pedestrians accurately in image sequence, a novel approach of head feature extraction based on target region matching is presented. Deferent from the common methods based on dense disparity image to obtain the disparity of the head region, the method in this paper is based on the idea of ＂segmentation before matching＂, i.e. the reference image is segmented firstly by monocular image processing and the candidate head regions will be acquired. Then these candidate head regions will be taken as the target regions to be directly used in the correspondence regions searching and matching to obtain the head disparity. Finally the disparity of the candidate head regions is used to extract the depth and perspective feature of the candidate head regions to remove the false head regions. The performance test and experiment results show that the method proposed in this paper has the advantage of higher precision of disparity extraction and better real-time performance as well as the great importance to extract the pedestrian head＇s feature with high degree of distinction to effectively eliminate the false head regions so that the accuracy of the passenger flow detection can reach over 90%.