Privacy-preserving Motion Detection In Videos

In the era of big data,the processing of video data is widely used.At the same time,the question of privacy leakage by the video data is becoming increasingly serious.As a result,the conflict between information use and privacy preserving rises.To preserve the privacy,it is easy to damage the structure of the video data when encrypting the video,so that the compression ratio of the video data decreases.The way to make use of the video data with high efficiency while preserving the privacy becomes a worth studying project.Among the various usages of video data,the moving object detection in video is a common one.The privacy-preserving moving object detection in video aims to preserve the privacy of the video,and at the same time,detects the moving objects in the video precisely.Moreover,the complexity of the algorithm is low.The existing privacy-preserving moving object detection methods have the following disadvantages: the hardware condition of the algorithm is hard to achieve,the computation complexity of the method is high,and the compression ratio is not high enough.The above questions need to be solved.The algorithm consists of two parts.One part is encryption and compression,and the other is detection part.In the encryption and compression part,the algorithm compresses the video data by the HEVC standard.And at the same time,encrypts the video data by the format compliant encryption.The aim of using format compliant encryption is to guarantee the safety of the video data and at the same time makes the algorithm be able to decode the video to do motion detection.While in the detection part,in order to detect the moving objects in each frame,the algorithm first detects the motion region.After that,the moving objects in different frames are connected by moving objects tracking step.And finally,the object shapes are refined by moving object trajectories.The experimental results show that the privacy-preserving moving object detection algorithm can detect the moving objects precisely in the video.The hardware condition is easy to achieve and the computation complexity of the algorithm is low.