Millimeter Wave Radar and Vision Fusion based Semantic Simultaneous Localization and Mapping

This work proposes an advanced Radar-Vision Fusion based Semantic Simultaneous Localization and Mapping (SLAM) algorithm that consists of radar-vision fusion based ego-motion estimation, semantic constraint-based point set registration and closed-loop detection. The pose of the vehicle is estimated by the Particle Filtering (PF) algorithm combined with the calculated motion model. The semantic-constrained Iterative Closest Point (ICP) algorithm is applied in point set registration and closed-loop detection. Radar-vision fusion semantic matching results are applied to generate semantic grid map. The algorithm is qualified in the launched indoor experiment, and the analysis of the results prove the effectiveness of the proposed algorithm.