基于单目视觉的GPS辅助相机外参数标定

针对机器人视觉系统外参数标定的问题,提出了基于单目视觉ORB-SLAM的差分GPS辅助相机外参数标定方法。分析了单目视觉ORB-SLAM和GPS(Global Position System)定位数据之间的相似关系,建立了相机外参数标定的非线性最小二乘模型。基于随机采样一致性(RANSAC),通过三点法求得模型的初始解。设计了Levenberg-Marquardt(LM)迭代算法求解出最优解,从而得到了最优的相机相对位置和姿态参数。最后,对提出的方法进行仿真和跑车试验验证。结果表明:在试验半径为50m时,所设计标定方法的姿态标定精度可达0.1°,位置标定精度可达0.2%。该方法标定过程简单实用,不需要外界环境的先验信息和人工干预,具有很高的精度和显著的应用价值。

Calibration of camera extrinsic parameters based on monocular visual with GPS assistant

An extrinsic parameter calibration method with differential GPS(Global Position System) assistant based on monocular visual ORB-SLAM(ORB-Simullaneous Location and Mapping) was proposed aimed at extrinsic parameter calibration problem of a robot vision system.Nonlinear least square models of extrinsic parameter calibration were established based on analyzing the similarity relationship between monocular visual ORB-SLAM and GPS positioning data.The initial solution of model was obtained by three-point method based on Random Sample Consensus (RANSAC), and then an optimal solution was obtained by designing Levenberg-Marquardt (LM) iterative algorithm.Thus optimal relative position and pose parameters of a camera were obtained.Simulation and traffic-running experimental verification was performed for proposed methods.The result indicates that when experimental radius was 50 m,the pose calibration precision of designed calibration method could reach 0.1° and position calibration precision could reach 0.2%.It concludes that the calibration process of the method is simple and practical.It does not need prior information of external environment and manual intervention, and has high precision and a significant application value.