Skip to Main Content
The detection of human abnormal action is very important in video surveillance. In order to detecting human abnormal action in different scenarios, we can use a method of double-layer Bag-of-Words model. In this method, the video information is firstly processed using Bag-of-Words, and the information of motion text words and scenarios are included in another bag. A video sequence is represented as a codebook of spatial-temporal words by extracting space-time interest points. An action characteristic is determined by behavior text words in special scenarios. A model of Latent Dirichlet Allocation (LDA) is applied to automatically learn the distribution of spatial-temporal words and the topics correspond to human action categories. And it also learns the probability distributions of the motion text words in a scenario with supervisor and the intermediate topics corresponding to anomalous or normal action. All of these are achieved by using the Latent Dirichlet Allocation model. Through this method, we can categorize the human action normal or abnormal in a special occasion to a new video. Thus we can advance corresponding security measures.