| In recent years, with the continuous development of science and computer technology, and people’s living standards improving continuously, people pay more attention to video surveillance system and it has been widely used in many practical applications. The traditional video surveillance systems which based on the fixed camera have become unable to meet people’s demand for intelligence because of its narrow view and passive monitoring. Recently, due to its initiative monitoring, intelligent video surveillance system has become a hot spot of research.PTZ camera can rotate 360 degree and can zoom. Because of these advantages of PTZ camera, so that, it widely used in intelligent video surveillance. The most important thing in intelligent surveillance based on the PTZ camera is to turn the PTZ camera to track the moving objects. Due to the background and the size of the moving object continuously changing, it is difficult for the PTZ camera to track the moving objects. Although a lot of people have done many related works and solved a lot of problems, there are still many problems to be solved.This thesis studies the moving object tracking under the PTZ camera, and focuses on active tracking of the moving object, i.e. how to adjust the PTZ camera so that it can follow the moving object accurately. And this thesis proposes an algorithm of PTZ camera adjustment. This thesis mainly includes three parts, camera self-calibration, moving object tracking and the PTZ camera adjustment.The main contents of this thesis are as follows:(1) PTZ camera self-calibration. It is difficult for PTZ camera to realize self-calibration, because the computation of the self-calibration is very complex. So, few people use it. This thesis uses the PTZ camera self-calibration, and proposes an improved algorithm in the process of self-calibration which makes the camera self-calibration more accurate by deleting the incorrect matching point. The algorithm of camera self-calibration is as follows:first, take five pictures by using the same PTZ camera at five different locations, and these five images have an overlapped area. Second, extracting the SURF feature points of the images, and then choose one image to match with the other four images and get the matching points, then use the graph transformation matching algorithm to remove the incorrect matching points, get more correct matching points, and then according to the feature points to calculate four homograph matrix H. At last, according to the formulas to calculate the camera intrinsic matrix K, complete the camera self-calibration.(2) Object tracking. The object tracking algorithm under the PTZ camera mostly adopts the traditional mean shift or kalman filtering tracking algorithm. Based on compressive sensing theory, an object tracking algorithm is proposed in this thesis. This algorithm is simple, good robustness and faster, and can meet the demand of real-time also. The algorithm constructs a very sparse measurement matrix to efficiently extract the feature for the appearance model. Then compress sample images of the foreground target and the background using the same sparse measurement matrix. Then the tracking task is formulated as a binary classification via a naive Bayesian classifier with online update in the compressed domain. The maximum response value of the classifier is the location of the object. Then select a number of positive and negative samples near the target location to update the classifier parameters.(3) PTZ camera control. Based on the camera self-calibration and the position of the moving object a PTZ active tracking algorithm is proposed in this thesis, in which the camera can follow the moving object accurately, namely the moving object is always at the center of the field of the PTZ camera. By the camera intrinsic matrix K, the target location and geometric projection and computer vision theory to calculate the value of the P and T of the PTZ camera, so that the camera can be adjusted according these two values. |
PDF Full Text Request