LiDAR–RGB Fusion-Based Forestry Robot for Tree Stem Mapping and Robust Radius Estimation

Accurate and real-time mapping of tree stems is essential for modern forestry management and ecological monitoring. Traditional approaches such as manual surveys, terrestrial laser scanning, and UAV-based sensing are often constrained by occlusion, limited coverage, and high operational cost. To overcome these challenges, this work presents a LiDAR–RGB fusion pipeline developed and deployed on a custom mobile robotic platform for tree stem mapping and radius estimation. The system fuses semantic features from a YOLO-based detector with 3D geometric data, followed by clustering and projection into two-dimensional stem slices. Circle fitting is performed on range-filtered and semantically constrained point sets, enabling reliable diameter estimation under noise and partial visibility without introducing explicit distance-based weighting. Temporal filtering using mean, median, exponential moving average (EMA), and Kalman filters are then applied, forming the basis for a comparative analysis of temporal filtering strategies for stable global tree mapping. Experiments conducted in forest-like environments demonstrate accurate and robust stem localization and radius estimation. Mean and median filters achieved the lowest absolute errors (0.021 m) and highest determination coefficients (R² = 0.93), while the Kalman filter maintained strong robustness against noise and occlusion. The full LiDAR–RGB fusion pipeline operates in real time and enables scalable, high-precision mapping of forest stands. Overall, the developed framework with mobile robots establishes a reliable foundation for forest inventory, contributing to sustainable and data-driven forestry operations.