Research Of Moving Target Positioning And Tracking Techniques In Intelligent Video Surveillance System

With the surveillant scene becoming more complex, traditional analysis of surveillance video of 2D-image has been unable to meet the demand of applications. It is necessary to obtain more information from the scene added as the analysis factors. Today, the depth information is of particularly importance. Therefore, to transter the 2D image coordinates of moving targets to their GPS coordinates has become a hot issue in intelligent surveillance technology. This thesis has made investigation on the positioning, trajectory prediction and fusion of moving targets. The main points of this work are as follows:1、The research of positioning technology based on error compensation curve. Firstly,analysis of the radial error and tangential errors caused by the lens distortion based on linear camera model. Then according to the isotropic characteristics of the camera lens technology, namely the positioning error with the distance with the lens center, we can get the GPS coordinate of the moving target using the principle of the concentric rings. Then use the error compensation curve way to eliminate error introduced by the use of concentric ring. Then we can achieve the high precision GPS coordinate of moving target in real time.2、The optimization by using sub-pixel positioning and data confidence. Based on the study on point one, this point carries on the further perfection. Firstly, use Vibe algorithm to get the moving target pixel level edge coordinates. Then use a variety of sub pixel method such as Zernike moments, analysis of the spindle and Gauss fitting to obtain more accurate sub-pixel coordinates, so as to obtain moved higher accuracy GPS coordinates. On the other hand, regard the confidence level of reliability as criterion, and use confidence level as a weighting factor to do confidence level weighted arithmetic, in order to obtain more reliability GPS coordinate.3、trajectory prediction and fusion of the moving target. Based on the previous two research points, the GPS coordinates can be obtained in intelligent surveillance scene in real time accurately. So the GPS coordinates used as the prerequisite of the known conditions to judge the motion mode of mobile targets in the scene. When multiple moving targets in cross movement trajectory, we can distinguish multiple moving objects according to the previous state, and when moving target in obstacle blocked, its trajectory can be predicted by Kalman filter. The motion trajectory of a camera in the group because of obstacles occlusion and its formation of the motion trajectory is interrupted, the more complete and reliable motion trajectory can be obtained by the trajectory prediction and fusion.Proved by the experiment, the GPS coordinates of the moving target can be obtained with high accuracy positioning by error compensation. And sub pixel and data confidence level also do further promotion on positioning accuracy. According to the positioning results, trajectory prediction and the fusion curve can match with the real motion state well. All these show that this algorithm has a good effect in the real time surveillance system.Finally, the thesis made the summary of this work, and also analyzed possible directions for further study.