Drone images and their annotations of goats/small ruminants (for computer vision purpose)

This dataset was developed within the framework of the European Horizon 2020 project ICAERUS, specifically for the livestock monitoring use case. The objective of this work is to explore the potential of drone-based computer vision methods for monitoring small ruminants in real farming environments.

More information about the project is available on the project website: https://icaerus.eu

Objective

Counting sheep and goats is a significant operational challenge for farmers managing flocks that may contain hundreds of animals. Traditional counting methods are time-consuming and prone to errors.

The objective of this work is to develop a computer vision–based methodology capable of automatically detecting, tracking, and counting sheep and goats when animals pass through a corridor, gate, or other naturally constrained passage.

The proposed approach relies on low-altitude aerial videos (<15 m) acquired using drones, providing a top-down perspective that facilitates the detection and counting of animals.

Progress and Enhancements

Our work includes the development of datasets and models dedicated to low-altitude aerial imagery of sheep (<15 m).

• Datasets contributions:

Multiple datasets either with or without annotations, have been produced and enriched as part of this work during the 2023-2026 period (see the summary table).

• Model Development:

We developed computer vision models for small ruminant detection (0.99 mAP50 in its version 4), tracking, and counting.
The models and associated code are available on GitHub:
https://github.com/ICAERUS-EU/UC3_Livestock_Monitoring

To improve the performance and robustness of detection models such as YOLO, the datasets were enriched to increase variability in:

• Environmental conditions (background types and lighting conditions)
• Animal appearance, including non-white sheep and goats, which are often underrepresented in existing datasets.

Data set description

This dataset is a subset of an original dataset of images and videos without annotations, now enhanced with annotations of goats. The annotation is labeled as “sheep” since no distinction is made between small ruminants.

Find more images and videos in the original dataset:

•  Lebreton, A., Depuille, L., NICOLAS, E., & Helary, L. (2025). Aerial videos and images of goats (for computer vision purpose) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.14591324

This dataset encompasses the following data:

• Pradel: a directory encompassing images and videos from a goat farm in France, Ardeche from 5 flights.
• Flight directory
• Images: images extracted from 5 videos (287 images), originally extracted at a frame of 5 frames/sec, when images variance was low, only some images remain.
• Fllight_directory_name.zip:  a .zip directory with the annotations of goats at the YOLO format (2790 “sheep” bounding boxes). The bounding boxes are labeled as “sheep” since no distinction is made between small ruminants.

Warning

In the directory "CUT_oblique_DJI_20240522173449_0002_V.mp4", a large number of goats are located in a shaded area. In standard vision, they are barely discernible, but by adjusting contrast and brightness, they become more visible. Users are free to decide whether they want to keep these annotations or not, to avoid introducing too much noise under typical conditions.

Future Work

Following extensive efforts in data collection and annotation, our next objective is to finalize and deploy the sheep counting pipeline on an edge computing solution, enabling real-time livestock monitoring in operational farm environments.

In parallel, additional projects are exploring other computer vision applications in sheep farming, expanding the potential use cases of this technology.

Collaboration and Contact

We welcome collaborations on this topic. For inquiries or further information, please contact:
Adrien Lebreton
Email: adrien.lebreton@idele.fr