Multi-Sensor Terrestrial SLAM for Real-Time, Large-Scale, and GNSS-Interrupted Forest Mapping

TL;DR

This paper introduces a real-time, large-scale forest mapping method using sensor fusion and SLAM that accurately estimates tree counts and diameters without relying on GNSS, suitable for challenging environments.

Contribution

A novel SLAM algorithm combining lidar and IMU data with hierarchical clustering for detailed forest inventory in GNSS-interrupted areas.

Findings

Achieved high accuracy in tree detection and DBH estimation.

Demonstrated real-time performance on handheld and robotic platforms.

Outperformed existing SLAM solutions in forest mapping accuracy.

Abstract

Forests, as critical components of our ecosystem, demand effective monitoring and management. However, conducting real-time forest inventory in large-scale and GNSS-interrupted forest environments has long been a formidable challenge. In this paper, we present a novel solution that leverages robotics and sensor-fusion technologies to overcome these challenges and enable real-time forest inventory with higher accuracy and efficiency. The proposed solution consists of a new SLAM algorithm to create an accurate 3D map of large-scale forest stands with detailed estimation about the number of trees and the corresponding DBH, solely with the consecutive scans of a 3D lidar and an imu. This method utilized a hierarchical unsupervised clustering algorithm to detect the trees and measure the DBH from the lidar point cloud. The algorithm can run simultaneously as the data is being recorded or afterwards on the recorded dataset. Furthermore, due to the proposed fast feature extraction and transform estimation modules, the recorded data can be fed to the SLAM with higher frequency than common SLAM algorithms. The performance of the proposed solution was tested through filed data collection with hand-held sensor platform as well as a mobile forestry robot. The accuracy of the results was also compared to the state-of-the-art SLAM solutions.