Application Of Object Tracking Based On Mean Shift Algorithm In Security Protection System

Video surveillance system is an important component of the security and protection system(SPS). Intelligence Video Surveillance System(IVSS) can greatly reduce human intervention, relieve operator's burden of, and improve the efficiency for monitoring. The detection and tracking technique for moving targets is the key to the IVSS. Therefore, object tracking techniques in intelligent video surveillance have a broad prospect and a high value in practical use.The background, the current situation and the significance of moving target tracking techniques are firstly investigated, in the meantime, researches of moving object detection and tracking are also reviewed in this thesis.Mean Shift algorithm(MSA) and it's applications in object tracking are discussed secondly. Some conclusions are gained by simulation experiments as follows:(1) when a tracked target has following properties:obvious characteristics, distinct difference between the target and its background, there exists an overlap area between two adjacent frames, and no interference in the target area, the MSA can reach an perfect tracking. (2) Since the MSA ideally deals with the targets matching problem between two successive frames, it decreases the blindness of target searching, and improves the efficiency for target tracking. (3) But the MSA has some defects yet. because the MSA has no ability of template updating in target tracking, the search box size does not change with the target size. Moreover, when the targets move rapaidly, there is little overlap area between two adjacent frames, as a result, the searching range is reduced, and the MSA tracking performance becomes worse simultaneously. Thirdly, when the targets are occluded by other objects, the target template changes, and the correlation of adjacent frames is dropped, the result is that the target template has incorrect information and a worse tracking performance is obtained.Another tracking algorithm using Kalman filter is also studied in this thesis. The algorithm predicts the future trajectory for the targets with its motion parameters. In tracking the target trajectory is modified continuously to improve the estimation accuracy of Kalman filter. Experimental results show that Kalman filter can adaptively update the size of the search box, even if the targets are occluded in tracking, the algorithm can also give a better performance.An improved MSA is addressed to overcome the MSA defects in worse tracking performance for the rapaid moving target and target missing for occluded problems. By introducing Kalman filter advantage into the MSA, the improved algorithm uses Kalman filter to predict the target location, and applies Mean Shift lgorithm to search in a neighborhood of the prediction location. It greatly promotes the adaptation ability of the classical MSA and provides a good stability and robustness.In order to modify the disadvantages of the classic MSA, such as, worse tracking performance for targets shape change, targets occluded and lightness variation, a new tracking algorithm CamShift is introduced. The algorithm takes the color histogram of H component in HSV space as a target model, and extracts geometric moments as image features, it reveals strong robustness in the case of target rotated, occluded and lightness changing.