With the development of society and economy, the range of traffic monitoring is expanding, and the number of the cameras used for traffic surveillance is explosively growing. As a result, traditional manual monitoring approaches no longer meet the needs of the current traffic monitoring systems. Integrating video analysis methods, intelligent traffic surveillance technology employs computers to complete the various monitoring tasks, which can not only save the human resources but also improve the accuracy of the surveillance system. Roadside camera calibration is the foundation for completing the various computer vision tasks, and thus is foundational for intelligent traffic surveillance. In this dissertation, we focus on how to calibrate the roadside camera used for traffic monitoring, and accordingly we have studied the camera calibration problem under the different types of traffic scenes. The major contributions of the dissertation are presented as follows: 1.Considering that in the existing vanishing-point-based calibration methods the number of the parameters to be estimated is relatively small or the calibration condition used is too strict to satisfy, we propose a roadside camera calibration method based on vanishing points and vanishing line. As for the estimation of camera intrinsic parameters and rotation angles, we first present a minimum calibration condition that consists of two vanishing points and a horizon line, which can be easily obtained in most traffic scenes; next, we model camera parameters and the observation errors of the vanishing points and incorporate the constraint equations between the horizon line and camera parameters, and eventually the camera calibration problem is converted into a least squares optimization problem instead of closed-form computation; in order to obtain more accurate calibration results, we propose a dynamic calibration method that fully employs the multiple observations of the vanishing points. As for the estimation of camera translation vector, we present a more flexible method that exploits known lengths in the road or known heights above the road. The experimental results on synthetic data and real traffic images demonstrate that the proposed calibration method outperforms the existing closed-form calibration methods in terms of accuracy and robustness. 2.Considering that the existing vehicle registration methods based on three-dimensional(3D) model often poorly perform when the clutter or occlusio...
修改评论