A Surveillance Video Content Deduplication Technology Based On Motion Vectors

Currently,mankind is increasingly relying on a large number of High-Definition video surveillance to help improve the city's management and life.However,it is unworthy storing all the surveillance video data because of its large capacity.Note that many video surveillance points only record the same picture for a long time,or only part of the picture changes,such as video of a street at night.Therefore,how to identify the static picture or background correctly and efficiently to remove the redundant information,and then improving the storage efficiency is an important research topic.This paper aims at the scene of deduplication when transferring inventory surveillance video to cold data storage devices.According to the dynamic and static characteristics of the surveillance video itself,a surveillance video deduplication technology based on macroblock motion vectors is proposed.It avoids the traditional video decoding and image analysis processing to reducing time and space overhead.Specifically,the video is divided into two parts,dynamic part and static part,and the motion vectors are extracted from the video stream as a reference for video motion detection.After a series of filtering and other processing,the static part of the video,which is considered as the redundant information,will be removed.When the video is finally sliced and stored,the indices are established based on the time period of the slices.If there is no movement in a video slice,an image in the video will be stored instead.Eventually the goal of surveillance video data deduplication is achieved.A surveillance video content deduplication prototype is designed,implemented and tested.After several experiments on the method,it has been proved that the system of surveillance video deduplication based on motion vector can complete surveillance video deduplication mission.This system is 1.5 times faster in processing speed and the CPU occupancy is less than 60% compared with the typical implementation scheme based on OpenCV method.With a recall rate of 89%,the deduplication rate can reach 94% in the case of no movement in the video,and the total deduplication rate can reach 46%.However,there is still room for improvement in the accuracy rate of motion detection,which will be studied in the future.