摘要
由于单目摄像头尺度缺失,导致鲁棒位姿估计结果不准确,为此提出单目主动摄像头真实三维重建方法。通过融合机器人的控制信息和尺寸信息,得到单目相机的位置信息,利用位置信息计算出等效基线,根据变基线等效双目策略理论获取精度更高的深度估计;采用GoDec-RANSAC算法排除样本点中的局外点,估计鲁棒位姿;通过获取的位姿,估计最后一帧图像与第一帧图像之间的相对位姿,得到全局位姿估计,完成单目主动摄像头真实三维重建。仿真实验结果表明,该方法能够准确的估计鲁棒位姿。
Due to the lack of monocular camera scale, the result of robust pose estimation is not accurate, so a real 3 D reconstruction method of monocular active camera is proposed. By fusing the robot’s control information and dimension information, the position information of monocular camera is obtained, the equivalent baseline is calculated by using the position information, and the depth estimation with higher accuracy is obtained according to the variable baseline equivalent binocular strategy theory;the robust pose is estimated by using godec-ransac algorithm to exclude the local points in the sample points;the last frame image and the first frame image are estimated by the acquired pose. The relative pose between the images is estimated and the real 3 D reconstruction of monocular active camera is completed. The simulation results show that this method can accurately estimate the robust pose.
作者
卫彦智
吴杰
WEI Yan-zhi;WU Jie(School of Mechanical&Automative Engineering,South China University of Technology,Guangzhou 510641 China)
出处
《自动化技术与应用》
2021年第8期1-5,10,共6页
Techniques of Automation and Applications
关键词
多模态融合
主动双目匹配
鲁棒位姿估计
主动闭环优化
单目摄像头
multimodal fusion
active binocular matching
Robust pose estimation
active closed-loop optimization
monocular camera
