60 / 2025-05-14 11:50:09

Traversability Analysis for Vehicle in Unstructured Road Based on Semantic Segmentation and Environment Geometry

Multi-sensor Data Fusion-, Traversability Analysis Method, Elevation Map, semantic segmentation

In unstructured road, the unmanned vehicles should accurately perceive and comprehensively analyze the surrounding terrain and ground attributes for guarantee the safety. Currently, common traversability analysis methods generally rely on single sensor data, which cannot fully integrate the ground attributes and geometric features of the terrain in off-road environments. This limitation results in weak anti-interference ability and poor robustness in environment perception, thereby affecting the traversability analysis performance. Hence, this paper proposes an innovative approach for solving above challenges. Firstly, it improves the robustness of the environment sensing process by fusing data from multiple LiDAR sensors. Then, it achieves sensing of the road-ahead ground attributes via road-surface semantic segmentation. Finally, it constructs a localized, dense traversability map to visually analyze the road-ahead. The experimental results show that the elevation maps and traversable layers generated by the proposed method are improved in terms of detail and accuracy. This helps unmanned vehicles achieve more effective and stable perception ability and navigation basis in off-road or complex environments. This method demonstrates outstanding traversability analysis performance and can accurately analyze the road status.