Moving Object Detection And Tracking For Intelligent Visual Surveillance

In intelligent visual surveillance system, computer can locate, identify and track the changes in the scene automatically and more importantly understanding the behavior of moving objects by analysis of the image sequence recording by the camera using computer vision & pattern recognition algorithm. The scenes monitoring by actual monitoring system are complex and changeable, the detecting and tracking of the moving objects accurately are challenge, these is also the basis in the following processing of intelligent surveillance such as integration of multi-camera, understanding and analysis of the behavior. In this thesis, the related model and algorithm about moving objects detection and tracking are studied based on the visual information of time and space and their relationships.In this thesis, a moving objects detection algorithm based on a dynamic background model is proposed. Our ideas come from three classic algorithms for moving objects detection. By analyzing their advantages and drawbacks in experiments, our algorithm absorbed mixture Gaussian model's basis thought effectively suggested by Stauffer etc. It constructs a mixture Gaussians model for each pixel. By sequence frames subtracting the model classify the pixels in each frame into background area, uncovered background area and moving objection area. In order to quick restore the background covered by stagnated objects when they move again, the model set the update rate in uncovered background area larger than which is in background area. Compare to the Stauffer's model, in our model moving objection area no longer create new Gaussian distribution, so it can avoid classify slow-moving objects to the background. Shadows detection and noise processing are implemented to improve the accuracy of detection results. In experiments, the algorithm is implemented to vehicle and human motion detection in image sequences which are obtained under differently natural conditions. The result shows that the detection algorithm can detect moving objects effectively and correctly.Another major work of this thesis is to track moving objects' trajectory in image sequences obtained from the proposed detection algorithm. An objects tracking model based on Kalman filters is presented. The motion estimation based on Kalman filters can reduce noise disturbance and shrink the search range of characteristic and improved tracking efficiency. In the process of objects matching, location and shape of objects are considered together, a matching function integrates the centroid and area of object is introduced to match objects between two consecutive image sequences. In addition, the emendation between the detection results and the tracking results improves stability of the whole tracking system better. Results of experiments show that the proposed tracking model based on Kalman filter relatively conforms to practical conditions and the tracking algorithm can reliably forecast and track the motion trajectories of objects.In this thesis, the prototypes for Intelligent Visual Surveillance System are implemented such as detection of left and remove detection of extraordinary density, detection of extraordinary speed, detection of illegal intrusion and crossing the fence and automatically tracking the moving targets by camera, based on the method of detection and tracking for moving target. In the experiments, the surveillance system is applied to the actual monitoring with many uncertainties. And it shows that the Intelligent Visual Surveillance System can help persons monitor the scenes effectively.