Real-time windrow detection from onboard tractor sensors for automated following
Pith reviewed 2026-05-08 03:00 UTC · model grok-4.3
The pith
Low-cost stereo cameras match LiDAR performance for real-time windrow detection from tractors.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
We present a multi-modal dataset combining stereo vision and LiDAR from tractor-mounted sensors during real baling operations, with synchronized data and GNSS trajectories. Using this dataset, we implement a real-time (>20 Hz) centroid-based windrow-following method on an NVIDIA Jetson AGX Orin. Across the critical 4-10 m guidance range, stereo and LiDAR depth measurements show strong agreement (0.965 +/- 0.021), indicating that low-cost stereo sensors can approach LiDAR performance. Our open-source ROS 2 pipeline provides a reproducible benchmark for GPS-free windrow detection and supports development of practical autonomous forage-harvesting systems.
What carries the argument
The centroid-based windrow-following method applied to the synchronized multi-modal stereo and LiDAR dataset.
Load-bearing premise
The data from specific real baling operations is representative enough of varied field conditions, windrow shapes, and lighting for the centroid method and depth agreement to generalize reliably.
What would settle it
Collecting new stereo and LiDAR data from a different field or under changed conditions and finding a correlation below 0.9 or inaccurate following would disprove the claim that stereo approaches LiDAR performance reliably.
read the original abstract
Proprietary design in commercial windrow-detection systems restricts transparency and limits progress in open autonomous forage-harvesting research. We present a multi-modal dataset combining stereo vision and LiDAR from tractor-mounted sensors during real baling operations. The dataset includes synchronized sensor data with GNSS trajectories, partly released as ROS2 Humble bags on Zenodo, with additional data available on request. Using this dataset, we implement a real-time (>20 Hz) centroid-based windrow-following method on an NVIDIA Jetson AGX Orin. Across the critical 4-10 m guidance range, stereo and LiDAR depth measurements show strong agreement (0.965 +/- 0.021), indicating that low-cost stereo sensors can approach LiDAR performance. Our open-source ROS 2 pipeline provides a reproducible benchmark for GPS-free windrow detection and supports development of practical autonomous forage-harvesting systems. Dataset: https://zenodo.org/records/17486318
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper presents a multi-modal dataset of synchronized stereo vision and LiDAR data collected from tractor-mounted sensors during real baling operations, including GNSS trajectories. It releases part of the data as ROS 2 Humble bags and implements an open-source real-time (>20 Hz) centroid-based windrow detection and following pipeline on NVIDIA Jetson AGX Orin hardware. The key empirical result is strong agreement (0.965 ± 0.021) between stereo and LiDAR depth measurements across the 4-10 m guidance range, supporting the claim that low-cost stereo sensors can approach LiDAR performance for GPS-free windrow detection in autonomous forage harvesting.
Significance. If the depth agreement is shown to translate into equivalent lateral localization and following performance, the work supplies a valuable public benchmark, dataset, and reproducible ROS 2 pipeline that addresses the opacity of commercial windrow-detection systems. The hardware-timed implementation and empirical sensor comparison during actual operations are concrete strengths that could accelerate open research in agricultural robotics.
major comments (1)
- Abstract and Results: The claim that stereo approaches LiDAR performance for automated following rests on the reported depth correlation (0.965 ± 0.021) in the 4-10 m range. Because the implemented method is explicitly centroid-based, performance equivalence also requires comparable lateral (x, y) localization of the windrow in each modality. No comparison of centroid positions, no following-error metrics, and no detection-success rates between the stereo and LiDAR pipelines are supplied; depth agreement at detected points can coexist with systematic lateral offsets or differing robustness to lighting and windrow shape. This gap is load-bearing for the central claim of practical equivalence.
minor comments (2)
- Abstract: Details on centroid computation (e.g., how points are clustered or averaged), data exclusion criteria, and the full error analysis beyond the single correlation statistic are not provided, making it difficult to assess the robustness of the 0.965 value.
- The manuscript should explicitly state the number of samples, field conditions, and any filtering used to compute the depth correlation, and discuss whether the specific baling runs are representative of varied windrow shapes, lighting, and terrain.
Simulated Author's Rebuttal
We thank the referee for the constructive review and for recognizing the value of the released dataset, ROS 2 pipeline, and hardware-timed implementation. We address the single major comment below.
read point-by-point responses
-
Referee: Abstract and Results: The claim that stereo approaches LiDAR performance for automated following rests on the reported depth correlation (0.965 ± 0.021) in the 4-10 m range. Because the implemented method is explicitly centroid-based, performance equivalence also requires comparable lateral (x, y) localization of the windrow in each modality. No comparison of centroid positions, no following-error metrics, and no detection-success rates between the stereo and LiDAR pipelines are supplied; depth agreement at detected points can coexist with systematic lateral offsets or differing robustness to lighting and windrow shape. This gap is load-bearing for the central claim of practical equivalence.
Authors: We agree that depth correlation alone does not fully establish equivalence for a centroid-based follower and that explicit lateral (x, y) centroid comparisons, detection-success rates, and following-error metrics (leveraging the available GNSS trajectories) would strengthen the central claim. The current manuscript emphasizes depth agreement because it is the dominant factor in the 4–10 m guidance range and because the same centroid extraction logic is applied to both modalities; however, we acknowledge the referee’s point that this leaves open the possibility of systematic lateral offsets. In the revised version we will add (i) direct comparison of the computed windrow centroids in the tractor frame, (ii) per-modality detection-success rates across the collected sequences, and (iii) lateral following-error statistics relative to GNSS where windrow ground truth can be derived. These additions will be placed in a new subsection of the Results and will be summarized in the Abstract. revision: yes
Circularity Check
No circularity: empirical depth agreement computed directly from new paired sensor data
full rationale
The paper collects a fresh multi-modal dataset from tractor-mounted stereo and LiDAR sensors during actual baling runs, then applies a centroid-based detection pipeline whose output metrics (including the reported 0.965 +/- 0.021 depth correlation over 4-10 m) are calculated as straightforward statistical summaries of the synchronized measurements. No equations, fitted parameters, or self-citations are used to derive the central claim; the agreement figure is an independent empirical result on external data rather than a quantity forced by construction or prior author work. The open-source ROS 2 implementation further supports external reproduction, confirming the derivation chain is self-contained.
Axiom & Free-Parameter Ledger
axioms (1)
- standard math Standard assumptions of computer vision and point-cloud processing for centroid extraction
Reference graph
Works this paper leans on
-
[1]
Engineering in Agriculture, Environment and Food 7/2014, pp
[Ch14] Choi, J .; Yin, X .; Yang, L .; Noguchi, N .: Development of a laser scanner -based navigation system for a combine harve ster. Engineering in Agriculture, Environment and Food 7/2014, pp. 7-13,
2014
-
[2]
T.; Soitinaho, R .; Oksanen, T .: Ploughing furrow recognition for onland ploughing using a 3D -LiDAR sensor
[GSO23] Goetz, K . T.; Soitinaho, R .; Oksanen, T .: Ploughing furrow recognition for onland ploughing using a 3D -LiDAR sensor. Computers and Electronics in Agriculture 210/2023, pp. 107941,
2023
-
[3]
Robotics and Autonomous Systems 123/2020, pp
[KFB20] Kneip, J .; Fleischmann, P .; Berns, K .: Crop edge detection based on stereo vision. Robotics and Autonomous Systems 123/2020, pp. 103323,
2020
-
[4]
S.; Han, J
[Yu21] Yun, C.; Kim, H.-J.; Jeon, C.-W.; Gang, M.; Lee, W. S.; Han, J. G.: Stereovision-based ridge-furrow detection and tracking for auto -guided cultivator. Computers and Electronics in Agriculture 191/2021, p. 106490,
2021
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.