Detection and Recognition of Abnormal Events in Crowds Based on Visual Technologies

摘要

In recent years, public security all over the world is deteriorating day by day. Crowd gathering behavior and crowd unexpected riots behavior are two typical potential security perils to public security. In this paper, these two typical abnormal activities were studied using different visual technologies. To detect the crowd gathering behavior, the foreground images were obtained firstly. Then, the crowd spread parameter and the crowd density were calculated. Finally, the crowd gathering was estimated by the crowd gathering factor which fused the above two cues. Alarm was raised if the crowd gathering factor was above a certain threshold value. To detect the crowd unexpected riots behavior, firstly, the optical flow of the image was calculated. Then, two values named chaos parameter and the crowd kinetic energy were obtained using the optical flow. Lately the crowd riots factor was calculated by fusing the above two cues. If the crowd riots factor was above a threshold, the crowd unexpected riots could be determined. A large number of experimental results showed that the proposed method had good robustness.