Research And Implementation On The Key Techniques Of Video Synopsis Based On Objects

With the development of technology and the increasing security level of our country, the amount the surveillance videos is increasing dramatically. In this situation, the video synopsis technology which can compress the original videos and realize the fact that moving objects which appeared in the original video at different time and different places can be shown in a short synopsis video simultaneously is being used widely nowadays. The video synopsis technology has two features, the first feature is that its computation data is giant, the objects are easy to be overlapped by each other and hard for viewers to get their temporal causal relationship after synopsized. The second feature is that there always have pixel color confusion between moving object tubes and synopsis video background. So the research and improvement in video synopsis algorithms are very significant.Since the video synopsis technology has the features mentioned above, this article did research on the key technology of the video synopsis technology: energy minimization synopsis video based on objects and seamless image fusion between object tubes and synopsis video background. The main work are listed as follows:1. Designed data structure for video synopsis technology based on objects and proposed the technology framework of the video synopsis based on objects. The article designed the data structure which can ensure the data consistency during the data translation between different models and the reusability of the data. Meanwhile the article also proposed the technology framework for video synopsis based on objects which keep the API interface for various potential functions in the future and it is significant for the development of the video synopsis technology in various fields.2. Researched the energy minimization algorithm for object tubes rearrangement and improved the tube collision energy and temporal consistency energy algorithm. For the problems that it is easy for moving objects to get overlapped by each other and hard to get their temporal relationship in synopsis video, the article added an object tube matrix similarity algorithm to justice the correlation and moving direction of each objects.What’s more, the article changed the temporal consistency energy definitions which solve the problem in video synopsis technology and improve the efficiency comparing with the classic algorithm.3. Research and improve the image fusion algorithm for moving object tubes and synopsis video background image. For the unique condition that the content in some part of object tubes and synopsis video are similar but the pixel value are different because of luminosity, the classic Poisson Editing algorithm always changed the both pixel value in different images to realize the image confusion. The article improved the algorithm and find the optimal boundary of the tubes so that can decrease the negative effect of retained background pixel in object tubes on synopsis video background and improve the qualification of the video.4. The article realized the intelligent clustering for object tubes by k-means clustering algorithm and delete the content redundancy of the long object tubes and keep the other object tubes unchanged. This function increase the compress ratio of the synopsis video while not decrease the visual effect of users.