Study On Moving Object Detection And Tracking In Video Image

Moving object detection and tracking is one of the important research topics in the field of computer vision. It blends many related areas of knowledge such as computer vision, video image processing, artificial intelligence, pattern recognition, automatic control and so on. It has been extensive applied in traffic detection, security monitoring, navigation, industrial detection, visual warning, TV guided. But because of the influence from many factors such as light change, noise, shelter, background interference and so on, the existing algorithms is facing great challenges in practical application. Therefore, the research of this topic has important theoretical significance and practical application value.Based on the analysis of present development of the moving target detection and tracking technology, this paper mainly studies several common methods of target detection such as frame differential method, optical flow method, background difference method and Mean Shift tracking algorithm which is with better comprehensive properties. Through experiment and analysis, we acquire the advantages and disadvantages of these methods and their applicable occasions. Moreover due to the real problem as noise, background interference, shelter, this paper proposed the improved algorithm on target detection and tracking methods.In order to achieve the detection of the moving target, the advanced Kalman filtering algorithm is proposed. Considering the results of the target detection obtained in the traditional Kalman filter containing a lot of noise, the target images need to be conducted a second detection utilizing the airspace characteristics statistics information of each pixel in the target area, and then the obtained results of foreground detection are updated to the predicted background image. As a result, this method can get more accurate targets with stronger robustness and higher adaptability to the environment.This paper has carried on the thorough research of Mean Shift tracking algorithm, and found that this algorithm is sensitive to background pixels and can not handle the problem of shelter well. On the basis of the traditional algorithm, Kalman predictor was added into the prediction of target location to solve the problem of shelter; In the process of target matching it adopts the method of background weighted and block color histogram to establish target model and the candidate target model uses the nuclear weighted and block color histogram method to improve the background sensitive issues. The experiment verifies that improved method can better deal with shelter and background sensitive issues, and effectively improves the accuracy and reliability of the algorithm.