System Calibration of a Field Phenotyping Robot with Multiple High-Precision Profile Laser Scanners

TL;DR

This work addresses the challenge of obtaining precise, high‑resolution georeferenced 3D crop point clouds from a field robot equipped with two close‑range, high‑precision laser scanners. It introduces an omnivariance‑based calibration cost and a factor‑graph pose estimation framework that fuses IMU, GNSS heading, and total‑station prism measurements to accurately register scanner frames to the robot pose. The method is validated in a dedicated calibration field against a TLS reference, showing sub‑centimeter accuracy ($\approx 0.8$ cm RMSE) when a reference cloud is used and good transferability across datasets, with improved inter‑scanner consistency. Remaining non‑static deformations of the enclosure introduce residual biases, highlighting the need for ongoing kinematic calibration along trajectories to further reduce errors. This approach enables high‑precision, field‑deployable 3D crop phenotyping with dual high‑precision scanners.

Abstract

The creation of precise and high-resolution crop point clouds in agricultural fields has become a key challenge for high-throughput phenotyping applications. This work implements a novel calibration method to calibrate the laser scanning system of an agricultural field robot consisting of two industrial-grade laser scanners used for high-precise 3D crop point cloud creation. The calibration method optimizes the transformation between the scanner origins and the robot pose by minimizing 3D point omnivariances within the point cloud. Moreover, we present a novel factor graph-based pose estimation method that fuses total station prism measurements with IMU and GNSS heading information for high-precise pose determination during calibration. The root-mean-square error of the distances to a georeferenced ground truth point cloud results in 0.8 cm after parameter optimization. Furthermore, our results show the importance of a reference point cloud in the calibration method needed to estimate the vertical translation of the calibration. Challenges arise due to non-static parameters while the robot moves, indicated by systematic deviations to a ground truth terrestrial laser scan.

Figures (7)

• Figure 1: a) Right laser triangulation scanner. b) For pose estimation: front and back GNSS antenna, IMU, and 360-degree prism. c) Left laser triangulation scanner.
• Figure 2: Factor graph of the pose estimation.
• Figure 3: Calibration test field used to evaluate our calibration approach. The prism is tracked with the Total Station (TS) for precise pose estimation. The Terrestrial Laser Scanner (TLS) is used to create a georeferenced point cloud for calibration and evaluation.
• Figure 4: Pipeline of our calibration method. 1) Point cloud creation with 2D laser profiles, initial calibration parameters, and robot poses. 2) Omnivariance computation with point cloud from 1 and reference point cloud (optional). 3) Evaluation of the cost function. 4) Iterative least-squares optimization. 5) Estimation on multiple scales.
• Figure 5: Poses of the robot estimated by our graph-based pose optimization approach using TS, IMU, and GNSS heading data as introduced in \ref{['subsec: pose_estimation']}. The brown rectangles represent the planes as shown in figure \ref{['fig:calib_field']}.