Reserch And Implementation On Key Technologies Of Massive Data Video Synopsis And Retrival

With the development of science and technology and the growth of awareness of public safety, surveillance cameras are widely applied to daily life, to meet the social security demand of each place. A large number of surveillance cameras produce massive data monitor video, but no enough human resources and time resources can be found to deal with the massive data video analysis. How to browse the surveillance video of a long period of time quickly and completely, and extract information of interest, become the important research content in monitoring fields.In this theses, video synopsis and content based video retrival are studied, in order to solve the above massive data surveillance video browsing and information of interest extracting problem. The main work contents are as follows:(1) According to the principle s of object-based video synopsis and the actual characteristics of monitoring scenes, the core technologies in video synopsis framework are analyzed, including motion detection and tracking, trajectories extraction and reset, and video abstraction generating. An approach based on gaussian mixture model to simulate the moving detection is presented; and an approach based on particle filter and the blob information is presented to track moving objects. For each moving object, extract its trajectory.(2)The trajectory optimization reset and video abstraction generating technologies are studied. Firstly, trajectories are properly described, and then set the objective energy function for trajectory optimization reset solution; secondly, simulated annealing algorithm is studied, which is employed for the combinational optimization; Finally resetted trajectories are pasted in the background to generate video abstraction.(3)The content based video retrieval technology is studied, according to the characteristics of monitoring scenes, select the shape features of the targets as the basis of retrieval. Use LibSVM to train samples and get benchmark data template containing object shape characteristics, according to the benchmark data set criterion, do size filtering and area filtering, and realize video retrieval functions.In the Windows platform, build VS2010 with OpenCV development environment, design and develop the video synopsis and retrieval system, in order to verify the feasibility of the algorithm.In this theses, the video synopsis and retrieval system, realizes the video synopsis function, which generates the synopsis video for quick browsing; at the same time, based on size filtering and area filtering, the system a lso realizes two video retrieval functions.