Research And Applicated Of Content-Based Surveilliance Video Key Frame Detection Algorithm

With the fast development of computer hardware and software as well as network technique, Video surveillance system which possess many qualities such as effective、long-distance、visual data etc. becomes popular, more and more Video surveillance system has been applied to military forbidden zone, bank, traffic etc..The amount of surveillance video data is very huge, mostly are redundancy. Current manual method which finds valuable information in surveillance video rely on surveillance workers has turn up several radical problems, such as health of surveillance workers, real-time of information retrieval, false judgement or loss judgement. The fact is that we can not get rid of these problems unless replace surveillance workers with machine. How to effectively discover important contents in surveillance video becomes a challenging task both in research and industry.In order to resolve these problems, this thesis focuses on key frame detection methods and proposes to design a practical system to facilitate important information retrieval from huge surveillance video data. At first, defines the key frame, as well as analysis the file format, data structure, video quality with some statistic program and tools; Afterward, enlightened by normal video retrieval techniques, base on background difference algorithm and adjacent frame subtraction algorithm, according to the diversity of characters between key frame and redundant frame, together with the influence from sunlight, car-light, street-light etc. this thesis proposes two kinds of improved key frame detection methods adapt to surveillance video. The first improved method is base on the standard background difference algorithm, bring forward a dynamic background update strategy, which can detect the movement of background frame,then guide the dynamic replacement of background frame in surveillance video depend on the threshold set before. This improved method can get a effect and real time results, and reduce redundant frame distinctively. Differing from the first improved method, the second improved method is more effective for the case that require to detect high speed object from low speed background. By introducing entropy calculation of adjacent frame subtraction result, the second improved method can increase the detection veracity more significantly. Simulation experiments indicate these improved methods can meet the surveillance video detection well, get a more effective and accurate result than the former method.Finally base on the above improved methods, accompany with a comprehensive requirement analysis, using Visual C++design and develop a key frame detection system which oriented to surveillance video. This system has a facilitated user-machine interface, can realize many video edit and video processing functions, such as data input,video browse, key frame detection etc. Besides a multi-thread mechanism which in order to meets the parallel requirement for capturing and processing of surveillance video.Practical experiments for the key frame detection system indicate that this system run steady and fluent., can detect key frame from surveillance video effectively in time., and satisfies the practical demand.