Mobile robot vision odometer based on point- line features and graph optimization

Abstract

In order to solve the problem of real-time and accuracy of autonomous positioning of mobile robot, a monocular visual distance calculation algorithm based on local and global optimization of point-line feature fusion is proposed. First of all, feature points and Canny edge features are extracted from the input image, and the point features are detected by ORB features with better robustness and real-time performance. After that, we use the similarity function of the feature block to find the optimal matching position and solve the resulting interior point for the incremental estimation of the robot pose. Finally, based on the idea of graph optimization in SLAM, this paper designs a pose optimization method based on nonlinear optimization, which can achieve the optimization of local pose and the global pose optimization to reduce the accumulation of position and attitude errors and trajectory drift. The experimental results show that the algorithm meets the real-time requirements, and can significantly reduce the pose error, with higher positioning accuracy.