Research On The Generation Of Tubes And Key-Frames Based On Video Synopsis

Intelligent surveillance video system is important and necessary for constructing to the safe city.Video synopsis is one of the core which is difficult in intelligent monitoring video technology.We can find and lock the suspicious target and full monitoring network under the moving target association tracking quickly throughtout the surveillance video system,when a case occurs.However,due to the surveillance video scene,the moving target will receive a variety of environmental interference,moving objects will exist between the mutual occlusion,etc.So the generation of video synopsis is a challenge.In this paper,the following aspects are studied for video abstract.In the aspect of video generation.Firstly,we compared and analyzed two method of object change detection and depth learning,which support the change detection is more avalieve.Then,we use the breadth search to produce the event tube,which build up by parts of based event tubes that generate throughtout the maximum overlap area method.Finally,we propose a method to delete the false event tubes to eliminate misuse of targets in complex scenes.Firstly,the HOG feature is extracted after the moving objects in the event tube are normalized for selecting key frames.Then,the most separable criteria of the target based on the feature of HOG is chosen to be the largest distance in the support vector machine.With the size of the target area given,we support the value fuction for key-frame images to evaluate the best key-frames.Video synopsis technology in the large monitor network areas can efficiently manage and view the surveillance video,through the videos under the surveillance network video summary can get the moving target set under the network.Then collecting features through the image processing and machine learning methods can calculate the correlation of moving objects based on target similarity under the whole network.We had made a large number of experiment by our test videos,as well as the experimental results of multi-camera network tracking,verify the effectiveness of the algorithm.