基于人脸识别的客流分析技术研究与实现

Research and Implementation of Passenger Flow Analysis Technology Based on Face Recognition

在复杂的场景中,多目标跟踪算法会由于遮挡出现目标丢失和误报的情况,导致目标跟踪失败。针对以上问题,本文提出基于人脸识别的多目标跟踪算法。首先,采用基于深度学习的YOLOv3进行目标行人检测;其次,对目标行人脸部特征点进行人脸姿态估计,用欧拉角表示人脸朝向信息,再对正向人脸通过人脸特征比对实现数据关联,非正向人脸采用基于Deep SORT的多目标跟踪算法进行目标跟踪。实验证明,在一定程度上提高遮挡情况下多目标跟踪的准确性,降低了行人ID跳变次数,在实时性和鲁棒性上也有良好的效果。

In complex scenes,the multi-target tracking algorithm will cause target loss and false alarms due to occlusion,resulting in serious target tracking failure. In response to the above problems,this paper proposes a multi-target tracking algorithm based on face recognition. First,YOLOv3 based on deep learning is used for target pedestrian detection,and then face pose estimation is performed on the target pedestrian,Euler angle is used to represent face orientation information,and then the positive face is used to realize data association through face feature comparison. The forward face adopts the multi-target tracking algorithm based on DeepSORT for target tracking. This method proves through experiments that the accuracy of multi-target tracking under occlusion is improved to a certain extent,the number of pedestrian ID jumps is reduced,and it has good effects in real-time and robustness.