一种复杂场景下的视频流人脸隐私保护技术

查看视频监控的过程中,一些场景存在因为人脸面部信息暴露在监控视频中导致个人隐私信息泄露的风险,有必要对实时视频流中的行人隐私信息进行马赛克处理。目前市面上常见的基于人脸检测的打码方法在实时监控视频流上打码效果受行人姿态、光线影响较大,存在实时性差、漏检较多等问题。针对以上问题,提出了融合人脸检测算法、目标物体检测算法和前置帧关联检测方法的多检测模型,并与传统的人脸检测模型进行对比。实验结果表明,在人脸检测召回率上,所提模型相较于传统人脸检测算法提高了532%。

A novel approach for face privacy protection based on surveillance video in complex scene

During the process of monitoring video surveillance,potential risks of leaking personal privacy information may occur due to the exposure of facial information to the surveillance video.It is necessary to mosaic the pedestrian privacy information in the real-time video stream.At present,the commonly used coding method based on face detection is heavily affected by pedestrian appearances and illumination,which often leads to weak real-time performance and undetected errors.To solve the problems above,a multi-detection model was proposed,combining face detection algorithm,target object detection algorithm and pre-frame association detection method,and compared with the traditional face detection model.Experimental results show that the recall rate of the proposed model is 532%higher than that of the traditional face detection algorithm.