Global Object Representation of Scene Surveillance Video Based on Model and Feature Parameters

Abstract

Scene surveillance video is a kind of video which are captured by stationary camera for a long time in specific surveillance scene. Due to regular movement of vehicles with similarity structures, models and appearances, surveillance video produce amounts of redundancy and needs to be efficiently coded for transmission and storage. In this study, we investigated the video redundancy generation mechanism of scene surveillance, exploit and presents a new redundancy type-Global Object Redundancy (GOR), it is proven that the vehicles occupy the mostly proportion which caused by amounts of vehicles movement. Secondly, aiming at global vehicle objects representation and GOR elimination, a global object representation scheme of scene surveillance video based on model and feature parameters is introduced, by establish a global knowledge dictionary and feature parameter sets, low bitrate with high quality compression can be achieved due to only few vehicle objects individual semantic and feature parametric be transfer and coded. Finally, we carried out preliminary experiments in simulation environment and shows that the object representation scheme can effectively improve the compression of long-term archive surveillance video which with a certain of image quality assurance.