REAL-TIME LIDAR-INERTIAL POSITIONING AND MAPPING FOR FORESTRY AUTOMATION

Abstract. Accurate real-time positioning in forests is challenging due to GNSS signal degradation and an unstructured spatial environment that is difficult to conceptualize through visual sensing. Positioning is vital in any forestry automation application, such as collecting inventory, harvesting, or search and rescue missions. Lidar and inertial based solutions are popular, however they often obtain real-time computation by effectively compressing the number of measurement points to track utilizing regular geometric shapes that do not adapt well to forest. Other solutions sacrifice the high-frequency of positioning estimates or they rely on post-processing. We propose a real-time lidar-inertial SLAM-based approach that utilizes NDT scan registration, factor graphs and loop closure corrections to produce accurate and high-frequency pose estimates. To test our method, data was captured with a lidar and imu sensor mounted on a stick surveying forest sites. Ground truth trajectory for accuracy evaluation was computed by fine registering individual laser scans onto a high-quality reference point cloud recorded from the same forest area using terrestial and airborn laser scanning methods. Experiments shows, that our method can produce real-time position estimates up to 200 Hz within 2–15 cm error.