Research on SLAM Drift Reduction Mechanism Based on Point Cloud Segmentation Semantic Information

Abstract

This paper combines the semantic segmentation of scenes with Simultaneous localization and Mapping (SLAM) technology to build a three-dimensional semantic map. The input sequence is selected by ORB-SLAM for key frame selection, and the scene’s semantic segmentation is performed in the corresponding point cloud data. We use a new 3D segmentation framework, which can effectively simulate the local structure of point cloud. A drift reduction mechanism based on semantic constraints and Bundle Adjustment (BA) constraints was proposed. This mechanism considers the three-dimensional objects, feature points and camera pose for semantic recognition in the scene, and integrates them into the back-end BA to optimize them. The final experimental results show that compared with the current popular ORB-SLAM, this mechanism can reduce the system’s translation drift error by 18.8%.