Trajectory Based Abnormal Event Detection from Surveillance Video

Event detection systems have been invented to detect the defined occurrences of interest by checking a large amount of recorded video sequences. However, in such sequences, we are being faced to some happenings which cannot be completely predefined to the systems and also these events are often unpredictable. All these anomalies possess a common property in common; they have a much lower occurrence probability rather than normal events. This property of distinction attracts the interest of video surveillance operators. Regarding the mentioned feature of distinction, we propose an approach, which has the ability of detection of abnormal traffics. The proposed clustering based un-supervised system finds the moving objects in surveillance circle of thecamera and after implementing some post processing steps, tracks the objects of interest to the point of extracting appropriate features such as trajectory features. We apply an efficient clustering method and the gained clustering consequences indicate the normal and abnormal trajectories. Experimental results show the high performance of the system which outperforms previous works in efficiency. Finally some conclusions are made.

کلیدواژه ها:

video surveillance ، abnormal event detection ، trajectory ، tracking ، clustering