面向视频监控场景的人脸检测与跟踪方法研究  

Research on Face Detection and Tracking for Video Surveillance Scenes

作　　者：严秋实 Yan Qiushi(Liangjiang New District Bureau of Chongqing Public Security Bureau,Chongqing401122,China)

机构地区：[1]重庆市公安局两江新区分局,重庆401122

出　　处：《长江信息通信》2021年第9期52-55,共4页Changjiang Information & Communications

摘　　要：针对传统的人脸检测与跟踪方法中,人脸遮挡和旋转而造成跟踪失败、重复抓拍以及抓拍质量不佳问题,提出了一种基于深度学习的人脸检测和跟踪方法。首先,采用改进的MTCNN网络对人脸进行检测,防止人脸检测过程中造成人脸的丢失;其次,采用了KCF跟踪方法,实现人脸的跟踪;最后,提出了一种人脸正脸判定方法,对检测到的高质量人脸进行输出。工程实践表明提出的人脸检测与跟踪方法,在视频监控场景下具有良好的鲁棒性,为人脸比对等算法提供了可靠的、信息含量高的人脸图像。Aiming at the problems of tracking failure,repeated capture and poor capture quality due to face occlusion and rotation in traditional face detection and tracking methods,this paper proposed a face detection and tracking method based on deep learning.First,the improved MTCNN network is used to detect the face to prevent the loss of the face during the face detection process;Second,the KCF tracking method is used to realize the tracking of the face;Finally,a positive face determination method is proposed to output the detected high-quality face.Engineering practice shows that the face detection and tracking method proposed in this paper has good robustness in video surveillance scenarios,and provides reliable and high-information face images for face matching algorithms.

关 键 词：MTCNN 人脸检测 目标跟踪 人脸正定判断 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]