🐄 BoviTrack: Real-Time Cattle Behavior Tracking & Monitoring

An end-to-end, edge-optimized computer vision pipeline designed to track and classify cattle behavior in real-time from farm surveillance feeds.

Instead of relying on a single heavy network, BoviTrack uses a decoupled dual-stage architecture to maintain high throughput and lightweight memory overhead suitable for edge deployment:

1. Target Localization & Tracking: A lightweight YOLO model detects cattle in the frame, paired with ByteTrack to maintain persistent IDs across frames despite occlusions and low-light transitions.
2. Behavior Classification: High-confidence detections are cropped and dynamically routed to a specialized YOLO classification head to identify target behaviors (eat, drink, stand, lie, ruminate).

---

📊 Dataset & Training

• Dataset Source: Beef Cattle Behavior Dataset (Kaggle)
• Training Pipeline: All data auditing, video frame extraction (with strict video-level splits to eliminate data leakage), and model training were conducted on Kaggle using NVIDIA T4 GPU acceleration.
• Pretrained Weights: Both optimized Nano-scale models are pre-packaged in the final_models/ directory for plug-and-play inference.

---

📁 Repository Structure

BoviTrack-RealTime-Behavior-Tracking/
├── data/
│   └── bovine_data.yaml              # Dataset configuration
├── final_models/
│   ├── detector_best.pt              # Trained YOLO cattle detector
│   └── classifier_best.pt            # Trained YOLO behavior classifier
├── notebooks/
│   └── notebook58828f625d.ipynb      # Kaggle training & validation pipeline
├── scripts/
│   └── main.py                       # Real-time inference & video processing script
├── videos/                           # Directory for input and output videos
└── README.md

---

🚀 Quick Start & Inference

1. Clone the Repository

git clone [https://github.com/Faissal-00/BoviTrack-RealTime-Behavior-Tracking.git](https://github.com/Faissal-00/BoviTrack-RealTime-Behavior-Tracking.git)
cd BoviTrack-RealTime-Behavior-Tracking

2. Install Dependencies

pip install ultralytics opencv-python numpy

3. Add Input Footage

Create a videos directory in the project root if it does not already exist, and place your target .mp4 video files inside it:

mkdir -p videos
# Copy your test video (e.g., day_video.mp4 / night_video.mp4) into the videos/ folder

4. Run the Pipeline

Execute the main deployment script:

python scripts/main.py

The script will automatically load the models from final_models/, process your video frame-by-frame with active ByteTrack ID persistence and behavior overlay badges, and export the final annotated video directly into the videos/ folder.