To develop and validate a robust pipeline for the automated detection and tracking of individual sheep in aerial video data captured by drones, enabling enhanced livestock monitoring and management in open-field environments.
- Follow the Conda preparation instructions detailed at: Getting-Started
- Download the UoL Shepherding Dataset, and others from Aerial-Datasets
- Some materials on Geo-Referencing: DJI Aerial Geo-Referencing
Sheep Detection – Develop and optimise a vision-based pipeline for real-time sheep detection and individual tracking from UAV-captured video streams.
Geo-Referencing – Estimate the geo-referenced position of each tracked sheep in a global coordinate frame using drone telemetry and camera pose data, constructing a map of their positions, with respect to the field boundaries.
Synthetic Data – Work with Group 5 to connect a video stream from the simulation to incorporate synthetic image into the training pipeline.
This work package is expected to deliver a validated end-to-end system capable of detecting and persistently tracking sheep in aerial footage, with output locations accurately mapped to global coordinates. The dataset and methodology will be documented and shared with the research community, laying the groundwork for publication in leading robotics or precision agriculture venues.
Members of the group will be encouraged to engage with future work exploring multi-species detection, behavioural pattern recognition, and integration with animal health monitoring systems.
Added real-time sheep clustering and enhanced map visualization capabilities:
- Sheep Clustering: Automatic grouping of nearby sheep (15m radius) with cluster size tracking
- Interactive Map Overlays: Toggle controls for sheep trails, clusters, and camera FOV
- Trail History Management: Configurable history limit (120 points default) to prevent memory growth
- Enhanced Sidebar: Displays cluster count, locations, and sheep per cluster
- Multi-layer Maps: Satellite (default), street, and hybrid view options
Published:
/sheep_paths(nav_msgs/Path): Individual sheep GPS positions with tracking IDs/sheep_clusters(nav_msgs/Path): Clustered sheep centroids with cluster sizes/sheep_detections(sensor_msgs/Image): Annotated video with bounding boxes and IDs
Subscribed:
/drone/gps(sensor_msgs/NavSatFix): Drone position/drone/attitude(geometry_msgs/Vector3Stamped): Drone orientation/drone/gimbal(geometry_msgs/Vector3Stamped): Gimbal angles/drone/image(sensor_msgs/Image): Raw camera feed
Access the live map at http://localhost:8080 to view:
- Real-time sheep positions (orange markers)
- Movement trails (green polylines)
- Cluster formations (purple markers scaled by size)
- Camera field of view (blue polygon)
- Live annotated video feed
- Drone telemetry and sheep/cluster lists
See MAP_VISUALIZER_README.md for detailed documentation.