Research On Human Abnormal Behavior Detection And Recognition In Intelligent Video Surveillance

In recent years, intelligent video surveillance has become a new application filed of computer vision. Compare with the traditional video surveillance, the intelligent video surveillance has its difference in intelligence. That is, the intelligent video surveillance not only uses cameras instead of human eyes, but also uses computers instead of people to complete the monitor tasks. Therefore, a lot of manpower, materials and financial resources are saved. What's more, the intelligent video surveillance can detect the abnormalities of the current scene to avoid the unusual events happening. Thus, the intelligent video surveillance is beginning to be attracted by domestic and foreign scholars and research institutions for its wide application prospect and great potential economic value.The intelligent video surveillance system is an integrated application which includes image processing, pattern recognition, artificial intelligence and many other technologies. This paper mainly focuses on two points: wandering trajectory detection and human behavior recognition. The main research contents and results are as follows:1. The development status of the intelligent video surveillance system is outlined. Also, the current methods of wandering trajectory detection and human behavior recognition, which include some judgment methods, Hu moments, Zernike moments, R transformation and so on, are learned and researched. Then, the shortcomings of these methods are analyzed.2. Against the shortcomings of the current method of wandering trajectory detection, a method for wandering trajectory detection based on angle is proposed in this paper. The experimental results prove that several common kinds of wandering trajectory can be judged by this method with a general algorithm, and the wandering trajectory can be detected accurately. The most important thing is that none of training samples is required in our method. Moreover, the method improves the real time performance of monitor system by minimizing the time and space complexity.3. A method based on Nonnegative Matrix Factorization (NMF) and Hidden Markov Model (HMM) is proposed in this paper for human behavior recognition. The method of NMF is used for extracting the human behavior feature. The feature matrix can be got through determining the basis matrix and the number of basis vectors of each video sequence. The HMM is used for identifying and classifying the human behavior. The optimal HMM parameters can be estimated by Baum-Welch algorithm and then the identification process can be completed by comparing the likelihood value of each type of behavior. The experiment compares with the human behavior feature extraction methods of Hu moments and R transformation. The results prove that this method is good for human behavior recognition, and the recognition rate is obviously higher than the other two methods. Therefore, this method is significant for improving the automated analysis of human behavior of intelligent video surveillance system.4. A small intelligent video surveillance system is designed and implemented in this paper. This small system integrates two new algorithms which are proposed in this paper with some other basic functions to deal with the data transports from the monitoring scene. At the same time, the algorithms which are proposed in this paper are proved their feasibility in practical application by this small intelligent video surveillance system.