ROS2 Multi-Modal Windrow Detection Dataset (Single-Row Subset)

This dataset contains synchronized multi-modal sensor recordings from a Fendt Vario 724 tractor during real baling operations. The released subset (123 s, 35.9 GB) captures traversal along a single windrow under stable lighting and surface conditions. Full windrow dataset available on request due to size limitations.
It is part of a larger multimodal agricultural dataset effort for reproducible field-robotics research.

Sensors include:

• Ouster OS0-128 LiDAR (18.3 Hz, 131 k pts/frame)

• ZED 2i stereo camera (23.5 Hz, ~230 k pts raw / 66 k filtered)

• NovAtel OEM7 GNSS + IMU for trajectory reference

• Five auxiliary mono cameras (ArkCam)

All data were recorded with ROS 2 Humble and stored in MCAP format.
Topics include /ouster/points, /zed/zed_node/point_cloud/cloud_registered, /zed/zed_node/odom, /novatel/oem7/fix, and synchronized camera streams.

The dataset supports research on real-time windrow detection, sensor fusion, and GPS-free navigation in agricultural environments.
Code and reference implementation are available at: github.com/Gunreben/Windrow-Centerline-ROS2-Node