Video event representation for abnormal event detection

In recent years surveillance video analysis has become an emerging area of research as video surveillance system is used in all places for monitoring the happenings to ensure safety. Automatic video surveillance system is useful for all applications like military surveillance, traffic monitoring, health monitoring of elder people at home, street surveillance. It is essential to represent video events for the further processing like video browsing, retrieval, summarization. Hence this paper presents a novel method for representing visual events in a video for event detection. For efficient visual event representation shape and motion can be used as features. Hence, pyramid of HOG (PHOG) feature is extracted for shape information and then it is combined with Histogram of magnitude, orientation and entropy of Optical Flow (HMOEOF) to have motion information. The extracted features are used to train the SVM classifier in order to detect and classify the events in the video as normal or abnormal. The proposed method results an equal error rate of 16.71% which shows that it outperforms well compared to other event detection approaches.