Monocular visual localisation optimisation method based on improved ORB

In the initialisation process of monocular visual SLAM, environmental factors, lighting conditions, and moving objects can cause errors in feature matching. We propose an optimised method for monocular visual localisation based on an improved version of the oriented fast and rotated brief (ORB) algorithm. By replacing rBRIEF descriptors with box average difference (BAD) descriptors, we enhance the robustness and computational efficiency of the feature matching algorithm. The use of the PROSAC algorithm eliminates mismatches and further improves accuracy. Experimental results show that our method improves the accuracy of feature matching in complex environments while maintaining computational efficiency. Compared to ORB, our method reduces runtime by approximately 10 ms. In typical complex scenes, the reprojection error is close to the sub-pixel level, and the matching accuracy is improved by 1.31% to 15.32%. Our method enhances computational efficiency during the initialisation process of monocular visual SLAM, thereby indirectly improving localisation accuracy. It is applicable in fields such as autonomous driving, robot navigation, mining exploration, virtual reality, and more.