Detecting abnormal activities in video surveillance with multi-models

In this paper, an adaptive method for detecting abnormal activities in video surveillance is proposed. In this method, a multi-Gaussian distribution called activity model is used to model a moving object activities. The activity model parameters are updated to satisfy the object motion attributes in a real-time when every new frame comes, and at same time this moving object current activity can be recognized by means of its possibility in the activity model. The advantage of this method is that the proposed activity models can update itself adaptively to match the current motion style of the object. The models are robust to the light change in the style of the object activity, and they are sensitive to these activities that do not meet the models. Several experiments are given to show that the proposed method is efficient.