Monocular SLAM Based on Point-Line Affine Invariant

At present, the mainstream monocular SLAM algorithms are still based on point features to estimate trajectory of camera and build environment maps. However, it is usually difficult to find enough point features in low texture environments, which will lead to system performance degradation or even collapse. Therefore, this paper introduces line features on the basis of ORB-SLAM2 and a monocular SLAM based on point-line affine invariant is proposed. It can reduce the segmentation and redundancy in the extraction process according to the relationship of line features, and construct the relationship between points and lines through point-line affine invariant to improve the speed of line features matching. At the same time, the idea of weight is introduced into the calculation of point and line reprojection error to further enhance the accuracy of the system. Finally, we have verified the feasibility of our system through multiple experiments on different datasets. The results show that our system has better stability and robustness in low texture environments.