基于窗口化匹配估计的ORB-SLAM算法研究

Research on ORB-SLAM Algorithm Based on Windowed Matching Estimation

针对ORB-SLAM系统中求解相机位姿存在随机性而引起定位精度不稳定的问题,提出一种基于特征点窗口化匹配和解析ICP的改进位姿求解算法,构建移动机器人ORB-SLAM系统。通过对提取的特征点进行窗口化,在保证选取良好特征点匹配的同时提高匹配效率;利用具有解析解的ICP算法求解相机位姿避免迭代,选取误差最小的窗口位姿求解结果进行光束法平差优化,以减小局部信息丢失或误匹配引起的位姿误差。结果表明:改进算法能够提高相机轨迹求解精度,轨迹误差相比ORB-SLAM2和ORB-SLAM3平均减少30%以上,并可以降低相机跟踪失败的概率。

To address unstability of location accuracy of ORB-SLAM system caused by randomness of camera pose solution method,an improved pose solution method based on feature point windowed matching and analytical ICP is proposed,and the mobile robot ORB-SLAM system is constructed.The extracted feature points are windowed to improve matching efficiency while ensuring good feature point matching,the analytical ICP algorithm is used to solve the camera pose for avoiding iteration,and the windowed pose solution with the smallest error is selected for bundle adjustment to reduce the pose errors caused by local information loss or mismatching.The results show that the proposed method could reduce the trajectory error by more than 30%on average compared with ORB-SLAM2 and ORBSLAM3,and can decrease the probability of camera tracking failure.