多特征协同的双层组合结构行人监控识别方法

摘要

In the process of human traffic statistics in public places,in order to effectively handle the shortcoming of low detection rate, high false alarm rate and poor real-time performance in pedestrian recognition caused by pedestrian occlusion and adhesion,we perform the features extraction of the aggregated B-Haar and the Edgelet feature coordination,design a pedestrian recognition model with double-layer composite structure.The upper layer of the model is a Haar feature (referred to as aggregated B-Haar feature)improved in combination with local binary pattern in complete binary tree structure,in charge of the extraction of candidate pedestrian targets,and makes sure the higher detecting recognition rate.The tree structure in lower layer contains a four-branch tree structure in series,it uses Edgelet feature and combining Bayesian principle to construct tree decision-making structure,carries out multi-position detection on candidate target and then determines whether the target is a pedestrian or not,thus achieves the aim of lowering false alarm rate and ensuring real-time performance.It is revealed by experimental analysis that compare with traditional tree structure and serial-parallel structure,the multi-feature collaboration pedestrian recognition algorithm with double-layer composite structure designed in this paper has obvious overall advantages in real-time performance,detection rate and false alarm rate.%在对公共场所人流量统计的过程中，为了有效解决因行人遮挡、粘连所引发的在行人识别上的低检测率、高虚警率、实时性不足的缺点，对聚集型B-Haar特征和Edgelet特征协调进行特征提取，设计了双层组合结构行人识别模型。该模型的上层是在完全二叉树架构下结合局部二元模式改进的Haar特征（称作聚集型B-Haar特征），主管提取候选行人目标，确保较高的检测识别率；下层树状结构使用四分支串联树状结构，利用Edgelet特征并结合贝叶斯原理构建树状决策结构，对候选行人多部位检测然后判断候选目标是否为行人，实现降低虚警概率，保证实时性的目标。经过实验分析表明，所设计的多特征协同双层组合结构行人识别方法与传统的树状结构、串并联结构相比，在实时性、检测率和虚警率上具有明显的整体优势。