LIDAR SLAM-Based Dense Reconstructions of Natural Environments: Field Evaluations

Abstract

The generation of accurate, dense 3D maps of natural environments is an important problem relevant to geological survey and wilderness search and rescue. In the search and rescue context it is imperative for map generation to occur in a timely manner. LIDAR SLAM based mapping algorithms run in real time and are resilient to the varying lighting conditions that cause issues in visual mapping systems. We evaluate LeGO-LOAM and UPSLAM for their trajectory and reconstruction accuracy in the Mecca Hills as well as the efficacy of visual fiducial markers for drift correction in natural environments. We find visual landmarks to provide significant drift improvements even on the 100 m scale and the LIDAR reconstructions of UPSLAM to perform comparably to the Metashape baseline 3D reconstruction.