Study On Surveillance Video Synopsis In Geographical Scene

As a new GIS data,surveillance video could express real-time changes in geographic scene,showing the interactions between men and geographic scene,and attracts more and more researchers' attention.Comparing to the other GIS data,surveillance video has two special properties:huge amount of data volume,and high-value information sparsely distributed.This situation constitutes a huge contradiction on the lack of artificial video retrieval and browsing,which restricts quickly understanding and effectively using of video information.It is necessary to building automatic and intelligent video processing methods to assist men on video retrieval and browsing efficiently.In order to extract high-value information from surveillance video for quick analyzing,the researchers had developed the approach called "video synopsis".The objective on building video synopsis is to construct a concentrated video filled with information of interests(usually the movements on pedestrians and vehicles)which is extracted from the original video,to facilitate the efficient monitoring of human on understanding surveillance video content.However,the temporal and spatial fusion analysis and efficient expression of video synopsis methods need to be further improved,especially in the process of video synopsis.Therefore,we need to develop a new video synopsis method in which geographical environment and video analysis are carried out together and the video data are efficiently and accurately data extracted and analyzed.Therefore,this paper collaborated analyzing the video data and GIS data as a means to explore the theory of video synopsis based on video spatial ization and trajectory clustering,developing geographic scene based video scene synopsis method,and the following achievements were obtained:(1)Propose video moving object trajectories' clustering and modeling method under geographical scene.To improving the efficiency of video synopsis,it is necessary to do clustering and modeling on moving objects' trajectories from video.Existing clustering methods have poor adaptability in geographic space;trajectory clustering and modeling results cannot express geospatial information.This charter presents video moving object trajectories' clustering and modeling method under geographical scene.The method is carried out based on the results of video scene spatialization.Trajectories' entrances and exits within sight range of the camera were clustered by QT algorithm.And the clustering results are combined to achieve the trajectory clustering.The method achieves cluster modeling by extracting trajectory cluster boundary lines and polynomial fitting.Experimental result shows that comparing to the methods operated in image space,the trajectory clustering results obtained by this method has better geospatial adaptability,which could effectively distinguish moving objects' trajectories;the cluster modeling result can effectively express cluster's geospatial features.(2)Propose the video moving objects query method in the geographical scene.The existing video moving objects query method is inefficient and lacks the ability to express the geometric features.In view of the above problems,this charter summarizes the existing video moving objects feature analysis methods,proposing the matching query method of video moving objects' trajectory in geographic scene on the basis of trajectory clustering.In this method,the video moving objects' query condition is matched with the video moving objects' trajectory cluster,and the trajectory class is used as the feedback result of the query.The experimental result shows that this method can improve the query efficiency of video moving objects and makes the query result more accurate.(3)Propose surveillance video synopsis expression method under geographic scene.Video synopsis in the form of short video has inherent defects:overlapping on the presence of moving objects,confusion on moving objects timing structure,video background's constantly updating etc.For easing these problems,this charter proposes surveillance video synopsis expression method under geographic scene.The method is carried out by building background in virtual geographic scene based on the result of video scene spatialization,creating video objects' representation model based on the result of video trajectory clustering and modeling,and selecting the pattern of actual-virtual fusion model.The experimental results show that the method can effectively preserve the temporal structure of moving objects,and realize the fusion expression of geospatial information and image information effectively.(4)Performance evaluation on geographic scene based surveillance video synopsis.As geographic scene based surveillance video synopsis is a brand new method on video synopsis which uses virtual scene model as the representation platform,it is necessary to build an independent evaluation system for making performance analysis on it.The concentration for evaluation in this charter includes:video synopsis data compression performance,video synopsis expression timing reduce performance,video synopsis expression positioning error.For each evaluation contents,this charter selects the impact factor,and constructs the corresponding evaluation model.Finally,the performance evaluation is made by analyzing the experimental results.