Accuracy enhancement for the front-end tracking algorithm of RGB-D SLAM

A robust and accurate simultaneous localization and mapping (SLAM) in working scenarios is an essential competence to perform mobile robotic tasks autonomously. Plenty of research indicates that the extraction of point features from RGB-D data that simultaneously take into account the images and the depth data increases the robustness and precision of the visual odometry method, used either as a self-reliant localization system, or as a front-end in pose-based SLAM. However, due to pure rotation, sudden movements, motion blur, noise and large depth variations, RGB-D SLAM systems often suffer from tracking loss in data association. The front-end tracking process of the ORB-SLAM system requires screening step by step, which is more likely to cause tracking loss. In order to solve the above problems, this work is intended to improve the ORB-SLAM front-end tracking algorithm based on the uniform speed model tracking effective frame and the matching of nearby frame algorithms. Then three datasets selected from TUM datasets with more motion blur are used to further verify the effect of the improved front-end algorithmic architecture. The experimental results suggested that the proposed improved scheme can not only effectively increase the number of tracked frames, but also reduce the amount of computation by about two times under the premise of guaranteeing the path accuracy.