| The detection and tracking of the moving target are the core of modern video surveillance systems, and has always been a hot and difficult research point in computer vision. In actual monitoring of the scene, multiple moving targets make the detection and tracking much complex. As a result, the dynamic changes of the scene, the nutation of the light, the detection and removal of the shadow of the moving targets and the crossing and separation between the moving targets have become the key problems and hot points.In this thesis, some practical solutions are proposed after studying the classic moving targets detection and tracking algorithms. Meanwhile, the function of the moving targets detection and tracking system is extended. Based on the binocular stereo vision, the depth information of the moving target can be extracted. When using Gaussian mixture modeling based background subtraction to detect the moving target in dynamic scene, the changes of the light can't be responded and the shadow of the targets can't be judged. To solve these problems, two methods are proposed in this thesis. For the tracking of the moving targets, we proposed a strategy based on the target blob tracking, target features tracking, target corner tracking and combined with Kalman predictor. This strategy can realize the stable tracking of multiple targets in complex environment. The extracting of the depth information based on the binocular stereo vision including the calibration, stereo calibration, stereo matching and depth calculation of the camera.The moving target detection and tracking algorithm in this thesis has been tested on multiple databases. The results show that this algorithm is effective when using it to adapt the mutations of the light, detect and remove the shadow of the target and stably track multiple targets. Meanwhile, the depth information extracted has high accuracy within a certain range, suggesting that the introduction of the binocular vision is feasible. |
PDF Full Text Request