Abstract: To improve the detection accuracy, an abnormal crowd event detection and localization method is proposed. This algorithm includes two phases: abnormal crowd event detection and event source localization. In the first phase,an algorithm combining with space and time is proposed. From the spatial perspective, the average kinetic energy distribution histogram is extracted to describe the crowd movement characteristics, and SVM classifier is used to classify the crowd movement characteristics; From the time perspective, the hidden Markov model is built to detect continuous crowd behaviors in the scene. In the second phase, under the framework of RANSAC, the event source location is realized by calculating the intersection point of the reverse extension line of abnormal behavior crowd movement, where the multiple event sources can be labeled simultaneously. Experimental results on UMN benchmarks show that this algorithm can effectively detect crowd anomaly behavior, and the value of AUC is 0.967, respectively increases by 0.127, 0.074 and 0.007 compared with traditional optical flow method,SIFT point detection method and social force method. Furthermore, this algorithm successfully localizes the event sources.

Key words: abnormal crowd event detection, event source localization, combination of space and time, RANSAC