Video-based surveillance systems are in widespread use. However, it would be very time-consuming to browse the large amount of video data and trace abnormal events if no preprocessing is conducted. Hence, we propose a trajectory clustering-based surveillance video synopsis technique, which contains the following advantages: (1) the generated synopsis videos of similar event trajectories are concise for users to efficiently view and trace; (2) abnormal event trajectory detection can split out the possible suspicions; (3) the clustering results promote the efficiency of searching for a specific target. Under a normal surveillance environment, the accuracy of trajectory clustering can be higher than 90% with up to 70% data saving rate.
Fig. (a) Human trajectories extracted from the surveillance video. (b) Groups of starting and ending points of trajectories. (c) Original surveillance video frame. (d) Video frame of the synopsis video, wherein the people who appear in the monitored area will be displayed like in a queue, improving the efficiency of video browsing.