Abstract
Activities held in cities often lead to crowd gathering, and activities involving large numbers of people may have hidden public safety risks. Therefore, detecting the aggregation of the population and discovering the anomalies of the human flows in a timely manner are of significance in urban public safety, traffic management, emergency control, and terrorism prevention. To detect abnormal crowds in urban areas, the paper presents an approach Pherkad/A and implements Pherkad/A in the corresponding tool (named Pherkad/T). Pherkad/A adapts a three-state MMPP (Markov-Modulated Poisson Process) model to take continuous flow data as input. It thus can detect two kinds of anomaly, i.e., flow increase and flow decrease in real time from incoming traffic flows. On the other hand, Pherkad/T permits users to specify their interested monitoring areas and time intervals and then report the anomaly results in a visualized way. Extensive experiments are conducted on the real data. The experimental results show that Pherkad/A has high accuracy in detecting traffic flow anomalies and Pherkad/T can respond various monitoring requirements in real time.