A Method for Automatic Detection of Crimes for Public Security by Using Motion Analysis
Koichiro Goya, Xiaoxue Zhang, Kouki Kitayama, Itaru Nagayama · 2009
In this paper, an automated video surveillance for crime scene detection using statistical characteristics is presented. The system is named Public Safety System(PSS). If the scene shows some peculiar situation such as purse-snatching, kid napping and fighting on the street, the PSS recognize the situation and automatically report to agency. Localization of moving targets in the scene and human behavior estimation are key processes of the proposed method. Three motion characteristics are determined from video stream: distance between objects, moving velocity of objects and area of objects. These characteristic are used to determine human behavior. Using these three metrics as a feature vector, the system classify video streams into criminal and non-criminal scenes. We use these two kinds of action sequences for the training data set. After constructing the classifier, we use test sequences those are continuous video stream of human behavior consists several actions in succession. The experimental results show the method is useful to detect criminal scene by the discrimination of human behavior.