Multi-Sensor SLAM Assisted by 2D LiDAR Line Features

In the domain of indoor localization, visual SLAM has gained prominence as a popular approach. However, visual information is frequently limited by feature degradation, resulting in diminished accuracy or location completely lost. To address this challenge, IMU and expensive 3D LiDAR are conventionally employed as solutions. Considering the low cost, we propose a multi-sensor fusion system with a camera, IMU, and 2D LiDAR. Firstly, we extract straight-line features from 2D LiDAR by Random Sample Consensus (RANSAC). Secondly, in scenarios where visual features degrade, the real-time pose will be performed based on the 2D LiDAR and IMU. Thirdly, in instances where visual features remain adequate, we further enhance the accuracy of pose estimation by the 2D LiDAR straight-line features. The experimental results demonstrate that our method has reliable accuracy and robustness in situations where visual features are degraded. Furthermore, our proposed method outperforms both the visual method and the visual-inertial method in terms of accuracy, particularly within indoor environments such as in rooms and corridors.