Three-Dimensional Point Cloud Reconstruction of Unstructured Terrain for Autonomous Robots

In scenarios such as field exploration, disaster relief, and agricultural automation, LIDAR-based reconstructed terrain models can largely contribute to robot activities such as passable area identification and path planning optimization. However, unstructured terrain environments are typically absent and poorly characterized by artificially labeled features, which makes it difficult to find reliable feature correspondences in the point cloud between two consecutive LiDAR scans. Meanwhile, the persistence of noise accompanying unstructured terrain environments also causes certain difficulties in finding reliable feature correspondences between two consecutively scanned point clouds, which in turn leads to lower matching accuracy and larger offsets between neighboring frames. Therefore, this paper proposes an unstructured terrain construction algorithm combined with graph optimization theory based on LOAM algorithm further introducing the robots motion information provided by IMU. Experimental results show that the method proposed in this paper can achieve accurate and effective reconstruction in unstructured terrain environments.