SLAM过程中的机器人位姿估计优化算法研究

Research on pose estimation optimization algorithm in mobile robot SLAM process

为了在图像位姿估计过程中充分利用更多空间点信息及减少噪声的影响,本文提出了一种基于EPnP算法的相机位姿估计优化方法。该方法先基于EPnP算法估计相机位姿,再把得到的相机位姿和空间点位置都作为优化变量,与相机的观测数据一起构建关于相机位姿的最小二乘优化问题,进行BA优化,来提高位姿的精确度。同时在SLAM后端对不同时刻视觉里程计的位姿和回环检测的信息进行优化,再把优化后的相机位姿作为观测数据和激光雷达信息进行融合,从而进一步提高SLAM过程的可靠性。实验结果证明了本文所提出的算法在机器人室内自定位和建图方面的准确性和可靠性。

A camera pose estimation optimization algorithm based on EPnP（ efficient perspective-n-point） is proposed to make full use of spatial points and reduce the effect of noise on camera pose estimation process. Firstly,the camera pose is estimated based on EPnP. And then the least squares optimization problem for the camera pose is constructed by observing data,the camera pose and positions of spatial points which are seen as optimization variables,which can improve the accuracy of camera pose. Meanwhile camera poses and the data of loop-closure detection at different moments are optimized at SLAM back-end,and the camera pose is fused with laser radar data to improve the effectiveness in SLAM process. Experimental results show the effectiveness and accuracy of the proposed method on 3 D localization and mapping.