Research Of Pedestrian Tracking In Intelligent Video Surveillance

To further improve the accuracy, robustness and real-time performance of pedestrian tracking in intelligent video surveillance, this paper mainly studied the pedestrian detection method based on the moving target detection and classification and the pedestrian tracking method based on a combination of the feature representation and motion model.In the part of pedestrian detection, this paper proposed a moving target detection method based on Bayesian modeling that can extract more complete moving objects from the relatively complex monitoring scene by using different features to describe different parts of the background. Experimental results show that the method can effectively deal with the movement interference in the background and has strong adaptability to illumination changes. In addition, this paper reduced the amount of computation for HOG feature extraction by using the integral image and PCA dimensionality reduction processing and realized rapid pedestrian detection based on the improved HOG feature and SVM after having extract moving object regions using the detection method of moving object based on Bayesian modeling. Experiments show that our pedestrian detection method has a higher accuracy and efficiency compared to the pedestrian detection method based on a global searching.In the part of pedestrian tracking, this paper proposed a new tracking method combining SIFT and Kalman filter with Mean shift pedestrian tracking method which can select the corresponding tracking strategy according to different situations to make best use of the scale invariance of SIFT features and the motion estimation characteristic of Kalman filter. Experiments show that this method has a higher robustness when the color of moving targets and the background is similar or there are a big scale change, occlusions and a fast moving for the targets.Finally, this paper constructs an intelligent pedestrian tracking system based on videos, so that users can choose the corresponding method for detecting and tracking to achieve pedestrian tracking according to different monitoring videos.