Multimedia Sensor Networks, Target Detection And Tracking Algorithm To Achieve

Images take more than 70% account of multiple media information collected by Multi-media Sensor Network (MSN), so the moving target detection and tracking algorithm based on image sequence is the emphasis of the researches of MSN. MSN analyses and processes image sequence and detect the interested moving target in order to abstract useful information for tracking and anticipation.By the review and analysis of current methods in this field, this article has proposed some novel algorithm in the research of moving target detection and tracking. Besides these, this article has developed the software which can detect moving target and track its trace based on MSN. My research work and contribution of this thesis are summarized as follows:1. By taking advantage of the results of moving target detection and tracking, this thesis presents an improved black movement estimation method based on Kalman filter. By the moving target's position in next frame which predicted by Kalman filter, the image block for computing the motion vector can be selected. This algorithm well considers the movement information correlation within inter frame, and uses the new 3-step search in order to increase the correctness of the search rule. The experiment results show this algorithm can enhance computation of modal parameters precisely and rapidly.2. Getting rid of changes in background which generated by moving camera is crucial and complex in moving target detection. By research of current primary algorithms of global estimation, this thesis adopts the 6 points model of camera to compute the parameters of the model. This method can decrease the time only by calculating the part black. Then the iteration algorithm can eliminate the infection from these moving target blacks. So we can get the parameters of the 6 points model of camera. Using this result every pixel can be complemented by bilinear interpolation.3. Getting rid of the noise is another emphasis in target detection. High frequency coefficients of noise and image details can be distinguished by multi-scale wavelet decomposition after studying the theory of wavelet. The noise of image can be removed by setting feasible threshold. The moving target can be detected by the difference of 3 frames which can be gained by motion compensating and noise distinguishing.4. The position and size of moving target region in image can be obtained after moving target detection in current frame. Using Kalman filter, the about position of moving target in next frame image can be predicted. Global estimation of next frame image can be implemented by the result of Kalman filter.5. The software of moving target detection and tracking which was based on vedio sensor node was designed and developed. After giving the flow chart and critical modal design of the software, I have designed a method of target orientation by using multi-camera in MSN simulation.Summaries and prospects have also been put forward in the final of the thesis.