Real-Time Generation and Automatic Update of 3D Point Cloud Maps in Featureless Environments Based on Multi-Sensor Fusion

This paper proposes a novel approach to generate high-precision maps and rebuild the area that the vehicle passes again in real time, ensuring that the map remains consistent with the real environment. To address the limitations of LiDAR odometry, our framework employs a multi-sensor fusion approach, which includes an onboard RTK and an IMU to determine the initial position of the vehicle. The height information is calibrated using the Normal Distributions Transform (NDT) of local maps. Old points are replaced by new ones in the same area to ensure a real-time map. We evaluated our system through extensive experiments, including real excavator working scenes and parking lots with a large number of moving vehicles. The results demonstrate robust and accurate mapping performance.