SLAM-Based Forest Plot Mapping by Integrating IMU and Self-Calibrated Dual 3-D Laser Scanners

Highlights

• Disassembly-Free Dual-Scanner BLS System: Designed a backpack laser scanning system integrating IMU with dual laser scanners (horizontal + inclined), where scanners can be folded into the backpack for protection without requiring disassembly.
• High-Precision Automatic Self-Calibration: Proposed a robust self-calibration method based on angle compensation and fixed rotation angles, achieving rotation errors less than 1° and translation errors around 1 cm even with limited overlapping FOV.
• Forest-Specific SLAM Framework: Developed a tightly-coupled SLAM framework that fuses IMU measurements with forest natural features (tree skeleton nodes and plane features), utilizing an iterated error-state Kalman filter for accurate motion estimation.
• Centimeter-Level Accuracy: Achieved mean errors and RMSE less than 3.0 cm in both horizontal and vertical directions across three challenging forest plots (0.1 ha each) with varying crown densities (0.25–0.96).
• Superior Performance: Outperformed state-of-the-art SLAM methods (LIO-SAM, FAST-LIO2, R-LINS, MULLS) with 60% lower planimetric errors and more regular trunk cross-sections.

Methodology

The proposed system consists of three main components:

1. Hardware System Design:

   • Dual Scanner Configuration: One horizontal scanner (L1) for below-canopy information and SLAM processing, one inclined scanner (L2) for vertical forest information acquisition
   • IMU Integration: Rigidly connected IMU (200Hz) providing high-frequency motion estimation and point cloud de-skewing
   • Wearable Platform: Telescopic rod-based backpack system allowing scanners to fold for protection

2. Dual-Scanner Self-Calibration:

   • Ground Extraction: Automatic ground plane detection using RANSAC in both scanner coordinate systems
   • Angle Compensation: Cost function minimization using roll and pitch angles to align ground planes
   • Rotation Parameters: Euler angle computation with fixed yaw angles provided by manufacturer mounting base
   • Translation Parameters: Geometric calculation based on known installation distances and rotation angle θ (typically ~80°)

3. SLAM Framework (Iterated Error-State Kalman Filter):

   • State Propagation: IMU-based error state propagation with 200Hz frequency
   • Motion Compensation: Backpropagation for point cloud deskewing using IMU measurements
   • Measurement Model:
     • Tree Skeleton Constraints: Trunk skeleton nodes to centerline distance for strong horizontal constraints
     • Plane Constraints: Point-to-plane distance using ground and branch features for vertical constraints
   • Iterative Update: Maximum a posteriori (MAP) estimation combining prior IMU prediction and feature measurements

Experimental Setup

Validation conducted in three forest plots in Zhuhai, Guangdong Province, China:

• Sensors: Velodyne VLP-16 LiDAR ×2, LPMS-IG1 IMU, Leica BLK360 (reference TLS)
• Trajectory: “S”-shape scanning path at 0.6 m/s speed
• Reference: Multi-scan TLS with artificial reflectors for accuracy assessment

Experimental Results

Mapping Accuracy Comparison:

Planimetric (Horizontal) Accuracy:

Vertical (Z) Accuracy:

Forest Parameter Extraction Accuracy:

• DBH (Diameter at Breast Height, 1.3m): Mean error ~0.015–0.022 m across three plots
• Stem Curve: Mean error ~0.015–0.020 m for diameter measurements at various heights

Key Findings

1. Calibration Robustness: The proposed self-calibration method maintains consistency (rotation angle difference < 1°, translation < 1 cm) across diverse scenarios (room, road, grass, forest) without requiring additional calibration devices or overlapping FOV.

2. Feature Superiority: Tree skeleton nodes provide invariant geometric constraints compared to traditional edge/plane features, eliminating matching errors caused by cylindrical trunk symmetry and scanner motion.

3. IMU-Fusion Benefits: Tightly-coupled IMU integration prevents SLAM degradation in feature-sparse forest regions and enables effective point cloud deskewing, critical for mobile scanning in dense forests.

4. Scalability: Processing speed of ~0.03s per frame on standard laptop (Intel i5-12500H, 16GB RAM), enabling real-time forest mapping capabilities.

5. Limitations: Dense canopy environments (>0.9 crown density) still pose challenges for complete above-canopy data acquisition; future work suggests integrating UAV laser scanning (ULS) for full crown coverage.