Video Abstraction And Content Retrieval System Based On Spatial-Temporal Trajectory Analysis Algorithm

With the rapid development of society,the requirements for security in all kinds of business activities and people's daily life are increasing day by day,therefore surveillance camera systems are deployed everywhere around us.However,the video data captured by cameras is so huge that it's nearly impossible to be processed manually.For the reasons above,we need intelligent surveillance system,which provides video abstraction and content retrieval functions.Video abstraction process consists of foreground detection and trajectory estimation,while video content retrieval searches trajectories by specific features and removes irrelevant ones.In order to solve these problems,the major research work of this paper is as follows.Firstly,choose appropriate foreground detection algorithm based on the foreground image result and execution speed.Besides,smoothing and empty hole removal of foreground image,after which the outlines of moving objects in foreground image become more complete and helps the following trajectory analysis.Secondly,spatial-temporal neighborhood trajectory analysis algorithm is introduced in detail.By tracking the spatial-temporal neighboring moving objects,we can get their whole trajectories.Thirdly,this paper introduces and implements six video content retrieval algorithm,namely search by time,search by moving direction,search by area,search by color,search by event type,search by texture.Last but not least,this paper introduces the implementation of the video abstraction and content retrieval system.Video abstraction modular consists of event abstract and all-event compressed abstract.Event abstract puts labels on original video image sequences and forms a brief video clip which describes the motion trajectory of each moving object,while all-event compressed abstract fuses all motion trajectory occurred in the video with a background image and forms a compressed abstract which describes all motion trajectories at the same time.Video content retrieval supports multi-feature complex retrieval.