三维监控系统中基于三维重构的交互式标定

发布时间：2018-04-27 21:54

本文选题：三维监控系统 + 摄像机标定　；参考：《电子学报》2017年03期

【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.
[Abstract]:Calibrating the position and pose of multiple surveillance cameras is the basis of intelligent monitoring system. The traditional calibration method needs to calibrate each target camera one by one, and it is difficult to deal with the non-overlapping monitoring field of view, camera motion and disturbance. In this paper, an interactive calibration framework based on 3D reconstruction is proposed. By introducing the 3D feature point cloud as the intermediate layer, the geometric transformation relationship between the scene and the reference background model is only established. The adaptive calibration can be realized by matching the target camera image with the 3D point cloud, and the workload can be significantly reduced. Because the matching is built between the monitored image and the 3D point cloud instead of between the monitored image, it can deal with the situation of monitoring field of view nonoverlapping. For camera motion and disturbance, an online relative attitude transfer method is proposed, which can overcome the problem of attitude change caused by camera disturbance and motion. Experiments show that the proposed method is effective.
【作者单位】：清华大学自动化系;清华信息科学与技术国家实验室;智能技术与系统国家重点实验室;
【基金】：国家自然科学基金(No.61225008,No.61373074,No.61373090) 国家重点基础研究发展计划(973)(No.2014CB349304) 教育部基金(No.20120002110033)
【分类号】：TP391.41

【相似文献】