UGV 3D Scene Mapping and Real-Time Target Detection in Underground Space

This paper focuses on 3D scene mapping and real-time target detection techniques for underground space based on small UGV. Both techniques are essential for small UGV to accomplish tasks in underground spaces. Complex environment and insufficiently light make some techniques inapplicable in underground space. The limited calculation power of small UGV leads to complex technique that cannot satisfy real-time requirements. In this paper, LiDAR is chosen as the sensor for mapping and target detection to enable small UGV accomplish real-time tasks in underground space. Considering factors such as calculation power and accuracy, this paper applies the LeGO-LOAM algorithm for localization and mapping, and the PointPillars algorithm for target detection. The experimental environments in this paper are chosen in an underground garage and NCUT campus to verify the algorithm deployed in small UGV. The experimental results show that the deployed algorithm has good robustness to build accurate maps and target detection in real time. In summary, the work in this paper provides important technical support for underground space mapping and target detection based on small UGV.