Forest feature LiDAR SLAM (F2-LSLAM) for backpack systems

Recent advances in sensor and algorithmic technologies have led to the exploration of remote/proximal sensing for automated forest inventory at various scales. For detailed below-canopy mapping to derive critical forest biometrics, such as diameter at breast height (DBH), merchantable height, and debris volume, under-canopy mobile LiDAR mapping is preferred. These mapping systems typically rely on an onboard integrated global navigation satellite system/inertial navigation system (GNSS/INS) unit for point cloud generation. The main challenge of such under-canopy mapping is the intermittent access to the GNSS signal, which is crucial to deriving accurately georeferenced mapping products. Advances in Simultaneous Localization and Mapping (SLAM) offer an alternative to GNSS/INS-assisted georeferencing in GNSS-denied/challenging scenarios. In this study, we propose a general and comprehensive Forest Feature LiDAR SLAM framework that encompasses odometry and mapping threads for a 3D LiDAR unit mounted on backpack systems to achieve accurate forest inventories. In the odometry thread, ground/tree trunk features are extracted from LiDAR scans (i.e., points from a full revolution of the laser beams) and used to estimate the transformation between successive scans. The mapping thread performs local/global least squares adjustment (LSA) using derived features to register LiDAR scans to a common reference frame. One advantage of the proposed framework is that the tree trunk features and ground information – which are critical to forest inventory applications – are directly derived in the SLAM process, making it superior to other geometric feature-based approaches. Additionally, relative trajectory information provided by onboard navigation units and/or reference point clouds from other sources can be incorporated into the process. In this study, three in-house developed backpack systems with varying specifications were used to collect data in complex forest areas. The proposed SLAM strategy was performed on these datasets and compared with point clouds acquired by uncrewed aerial vehicle (UAV) and a commercial backpack LiDAR system. Experimental results suggested that the proposed framework can produce point clouds with satisfactory intra-dataset alignment quality (in the range of 2–4 cm) and positional accuracy (around 10 cm) for all backpack systems.