利用人头信息辅助的摄像头人员跟踪

Human Tracking Under Camera with Head Information Assistance

为提升摄像头视角下的人员跟踪精度,本文在对加油站便利店场景进行人员数据采集后,基于CenterNet网络实现人头和人体目标框的联合检测与匹配,进一步提出自适应函数计算人体躯干目标框用于跟踪算法的预测和关联流程。该方法缓解了因拥挤和遮挡造成的跟踪失败和ID交换等问题。实验结果表明,利用人头信息辅助进行人员跟踪,可以有效提升跟踪算法的关联能力,为智能视频监控相关应用的分析算法提供更可靠的数据支撑。

In order to improve the personnel tracking accuracy from the perspective of the camera, after collecting the personnel data from the scene of the gas station convenience store, this paper realizes the joint detection and matching of the head and human body target frame based on the CenterNet network, and further proposes an adaptive function to calculate the human body target frame for the prediction and correlation process of the tracking algorithm. This method alleviates the problems of tracking failure and ID exchange caused by congestion and occlusion. The experimental results show that using head information to assist personnel tracking can effectively improve the correlation ability of tracking algorithm and provide more reliable data support for the analysis algorithm of intelligent video surveillance related applications.