摘要
结合人脸检测算法和跟踪学习检测算法(Tracking Learning Detection,TLD)完成多个人脸的检测跟踪,用来实现对汽车4S店顾客的实时进店检测。由于采集图像序列帧率低,导致帧间人脸姿态变化大,容易出现目标丢失现象。本文采用Kalman滤波和最邻近数据关联方法,提出一种改进的基于TLD的顾客进店实时检测算法,有效改善了目标短暂丢失现象,增强了算法的鲁棒性。实验证明,该算法具有抵抗光线变化、小范围形变和短暂遮挡的优点,能够解决复杂环境中的实际问题。
This paper is mainly to complete the multi face detection and tracking based on face detection algorithm and TLD （Tracking Learning Detection） algorithm, to realize the real time detection of customers into the cars 4S shop. Due to the low frame rate image acquisition se- quence, resulting in inter face pose changes, prone to the phenomenon of missing target,this paper uses the Kalman filter and the nearest neighbor data association method, proposes a real time detection algorithm for customers based on TLD. So the detection algorithm is efficiently improved, and the robustness of the algorithm is enhanced. Experiments show that the algorithm has the advantages of resistance to light chan- ges, small deformation and transient occlusion. It can solve the problems in complex environment.
出处
《微型机与应用》
2016年第14期42-45,共4页
Microcomputer & Its Applications
