PLP-SAM: A Point-Line-Plane Adaptive Fusion SLAM Method

Simultaneous Localization and Mapping (SLAM) plays a crucial role in the field of mobile robotics, particularly with visual and laser fusion SLAM systems being widely adopted due to their robust performance and superior scene recognition capabilities. However, traditional SLAM systems often rely on feature points, which poses challenges in low-texture and sparse environments. In response to this issue, this paper presents a Point-Line-Plane fusion SLAM system (PLP-SAM) based on a multi-sensor SLAM framework. The proposed system effectively extracts reliable point, line, and plane features from the scene through the fusion of visual and laser data. Additionally, it utilizes a "Feature Sparsity Grid" to characterize the distribution of point features across various scenes. This grid informs the regions and target quantities for line and plane feature extraction, as well as the number of features engaged in matching and optimization processes. Consequently, the system achieves accurate pose estimation and maintains good real-time performance across a range of environments, including low-texture scenarios. The effectiveness of the system was evaluated on two datasets in M2DGR and TUM, with performance comparisons made against other visual-laser SLAM systems and multi-feature fusion SLAM approaches.