The Research Of Key Technology In Efficient Video Abstraction For Surveillance Videos

With the advancement of surveillance technology, surveillance cameras could be found nearly everywhere around. However, the amount of information and the space for storage of these surveillance videos bring about some challenges. Video abstraction or video summarization, which is capable of efficient information browsing and retrieval, turns out to be the solution to this difficulty.At the beginning, this thesis gives a brief introduction of surveillance videos, video abstraction and video synopsis. Then it focuses on two video abstraction methods. One uses key frames which are obtained by GMM and frame difference, and the other uses video synopsis which is generated by median frame and background subtraction. As for the former, we first construct the background model via GMM. Second, we select key frames from the surveillance video based on two dynamic thresholds. Finally, we extract interested target by training the liner SVM. As for the latter, we first choose the median frame over a period time as the background. Then we extract objects after the process of background subtraction and region filling. Finally we insert objects into the chosen background with Poisson image editing. These two methods are simple and efficient. They are able to maintain ideal results while avoid complex calculation. This paper also provides evaluations on the produced video abstractions based on current regulations.