Research And Implementation On The Key Techniques Of Online Video Synopsis

With the development of technology and informational society,the prevention and management in the field of public security has been paid much attention.The amount of the surveillance videos has also been increased dramatically due to the promoted security work.Consequently,the efficient usage of these massive surveillance videos has imposed great challenges not only to the information search by human force,but also to the data storage and updating.Under this situation,video synopsis technology which can compress the original lengthy videos by simultaneously showing multiple moving objects even when they originally occurred at different times and places has been utilized and researched widely nowadays.To condense lengthy movies as much as possible,occlusions between moving objects can't be avoided at sometimes,not to mention the alteration of the showing sequence between them will affect the original content structure.In addition color aliasing and unnatural blending results widely exist in the image blending procedure between moving tubes and background.So it is very significant to research and improve the video synopsis algorithms.Due to the above mentioned problems in video synopsis framework,this article did some research that focus on the tube reallocation and seamless image fusion algorithms.Main works are listed as follows:Firstly,this article has designed the technology framework of the online video synopsis based on objects.Under the base of offline synopsis technology,the framework is proposed based on the requirement of the condensation result and processing speed during practical application.Consequently,original videos can been dealt with and the synopsis videos are generated at the same time under this online framework,which greatly satisfied the need of practical application.Secondly,this article has researched and improved the tube reallocation algorithm to reduce the object occlusion and temporal inconsistency phenomenon which affect the final condensation quality.Under the base of previous work,this paper improve the definition of occlusion and temporal consistency,moving structure between moving objects has been maintained as the same time.Improved tube reallocation algorithm can fix the above question great enough and enhance the efficiency a lot.Thirdly,this article has improved the seamless image blending algorithm.Considering the fact that classic Poisson Image Editing algorithm needs relatively large processing times and multiple image blending operation should be conducted in the video synopsis,mean-value coordinate technology has been applied to generate the final blending images.Considering the symmetry quality of the rectangular blending area,the mean-value coordinates' blending technology can be further improved so that the coordinates calculation only need conducted in the upper-left subarea,which greatly improve the efficiency while maintain similar image blending results.Fourthly,the content of synopsis video can be further filtered by restricting characteristics of moving objects.With the description,quantification and classification of moving features,size of shape and showing location,interested video content can be conserved and these irrelevant ones can be filtered out,which can reduce the content interference a lot and greatly improve the browsing and indexing efficiency.