【摘要】 标定多个监控摄像机的位姿是智能监控系统的基础.传统的标定方法需人工逐一标定每个目标摄像机,且难以处理非重叠监控视场以及摄像机运动和扰动的情况.对此本文提出一种基于三维重构的交互式标定框架,通过引入场景三维特征点云作为中间层,仅需一次性建立其与参考背景模型间的几何变换关系,即可通过目标摄像机图像与三维点云的匹配实现自适应标定,可显著降低工作量.由于匹配是建立在监控图像与三维点云之间而非监控图像之间,因此可以处理监控视场非重叠的情况.对于摄像机运动和扰动,提出了一种在线相对姿态传递方法,能够克服摄像机扰动和运动带来的姿态变化问题.实验证明了本文方法的有效性.更多还原

【Abstract】 Calibrating camera is essential for intelligent surveillance systems. Conventional calibrating methods usually calibrate the target cameras one by one and can not handle non-overlapping cases or camera motion/disturbance. In this work,we present an interactive calibration framework based on 3D reconstruction. The reconstructed 3D feature point cloud is treated as the interface between the 3D background model and the target camera. Through 2D-3D matching,a target camera could be automatically calibrated against the 3D feature point cloud. Due to the matching is performed between the target image and the point cloud,non-overlapping cases can be well handled. Also,an online relative pose transfer scheme is proposed to deal with the problem of camera disturbance or motion efficiently. Experiments demonstrate the effectiveness of the proposed framework.更多还原