Human Target Tracking, Intelligent Video Surveillance

With the development of science and technology, the intelligent video surveillance which can monitor all kinds of scenes automatically and real-time, has been used in a wide variety of professions. Such as the video surveillance of banks, housing estate, factories, traffic flow etc. Human tracking is the core technology of intelligent video surveillance and a hot topic in computer vision. It contains a lot of kernel technology in computer vision, digital image processing, pattern recognition, artificial intelligence, automatic control and so on and has been used in intelligent video surveillance, traffic flow surveillance, medical diagnosis, weather report etc.Now, the particle filter is the most widely used method in human tracking. It is one of the Bayes methods. Transforming the human tracking to Bayes estimating is its main thinking. That's to say, suppose we have know the prior distribution and the current observations, then we estimate the posterior distribution constantly. In practice, all the systems are non-Linear and non-Gaussian and the particle filter can deal with the non-Linear and non-Gaussian systems well.Generally, when transforming the three-dimensional to two-dimensional image, much information has been discarded and some noise has been bringing in. So it's very difficult to build the human model from images, especially used single-vision-feature. Hence, this paper uses multi-vision-feature, color and shape, to track human. In the course of building human model, it can adjust the weights of two features automatically. Therefore, when one of two features has become unreliable because of the change of environment, the other feature can have more weight, that will insure that the tracking result is still accurate and stable.After comparing our method with single-vision-feature with experiments, we find our method is better. In complicated environment, it can still have good performance.