A keyframe selection for summarization of informative activities using clustering in surveillance videos

Video informative activities summarization in surveillance video has become an important approach that leads to object detection, object classification, and multi-event detection, etc. Several approaches such as dictionary learning, representation learning, statistical approach, etc. have been used for identifying important events. However, the efficiency of these methods lack in identifying the important activities in a video effectively. To overcome this challenge, this paper presents a keyframe selection algorithm by adopting multi-level clustering using surveillance video. To efficiently identify the activities, orientation computation, Markov chain based clustering, and adjacent matrix based clustering are used. The Markov chain based clustering is used to analyze and group the adjacent activities efficiently. Adjacent matrix based clustering ensures inter-relationship among the activities and then integrates the positive activities. It leads to an efficient and informative activities summary. The experimental results are tested on the PETS 2009, VIRAT and UCLA datasets. Also, various existing video summarization algorithms are compared with the proposed method for evaluating the performance of the proposed method.