Synoptic video based human crowd behavior analysis for forensic video surveillance

Abstract

Modeling human blobs in crowd for analyzing the behavior is an important issue for video surveillance and is a challenging task due to the unpredictability. Huge video dataset is captured by using various resources like surveillance cameras in many places including the public environment like railway station, airport etc. It is very time consuming to watch the whole video manually for forensic applications of analysis. Most of the computer vision algorithms are concentrating for real time solutions. But, still after the fact also, so many issues could not be analyzed well. For example, if one wants to analyze a specific activity in a video, watching the video for so many hours is hectic. In this paper, Video synopsis is used to represent a short video while preserving the essential activities for a long video. In the existing methodology, active objects are shifted along the time axis and hence the video is compressed much, it causes collisions among the moving objects. Therefore, compact video synopsis is proposed by using a spatiotemporal optimization, which will shift the active object along the space as well as time and thus avoid collision among them. This synthesized compact background is introduced by using multilevel patch relocation (MPR) method to provide a larger virtual motion space for shifted objects. The synoptic video is proposed here to detect if any anomalies of human crowd(s) is present in the scene in a quicker time. Experimental results obtained by using extensive dataset show that the proposed algorithm is effective in detecting anomalous events for uncontrolled environment of video surveillance.