Robust Scan Registration for Navigation in Forest Environment Using Low-Resolution LiDAR Sensors.

Area of Science:

• Robotics and Automation
• Forestry Technology
• Computer Vision

Background:

• Automated forest machines are crucial due to hazardous conditions and labor shortages.
• Existing SLAM methods struggle in complex forestry environments with limited sensor data.

Purpose of the Study:

• To develop a robust SLAM and tree mapping method for forestry using low-resolution LiDAR.
• To improve navigation accuracy and tree localization without relying on GPS or IMU.

Main Methods:

• Utilized tree detection for scan registration and pose correction with low-resolution LiDAR (16Ch, 32Ch) and Solid State LiDAR.
• Implemented adaptive pre-processing and heuristic tree detection.
• Evaluated on private and public datasets.

Main Results:

• Achieved robust scan registration, outperforming generalized algorithms with >3m RMSE reduction for 16Ch LiDAR and 3.7m RMSE for Solid-State LiDAR.
• Increased tree detection by 13% compared to fixed radius search.
• Automated tree trunk diameter estimation with 4.3cm MAE (6.5cm RMSE).

Conclusions:

• The proposed method offers robust SLAM and tree mapping in forestry conditions using only low-resolution LiDAR.
• This approach enhances the automation of forestry machines, improving efficiency and safety.