An advanced real-time object-tracking system built with Streamlit and YOLO11, specifically designed to detect and count birds with high precision.
This application provides a user-friendly dashboard to monitor live camera feeds and track individual birds. It leverages the state-of-the-art YOLO11 object detection model, specifically configured to target bird classes (class index 14), with custom-tuned tracking via the BoT-SORT algorithm.
- Real-time Streamlit Dashboard: Web-based interface for live monitoring and interactive controls.
- AI Sensitivity & Calibration: Adjustable confidence thresholds and dynamic tripwire alignment.
- Unique Bird Tracking: Maintains a global set of unique rack_ids with BoT-SORT for accurate cumulative counting.
- Custom Telemetry: Real-time performance tracking and inference metrics.
- Optimized Performance: Utilizes high_quality_tracker.yaml with optimized BoT-SORT parameters, including ReID and camera motion compensation.
Birds.counting.mp4
- Python 3.x
- ultralytics (YOLO11 support)
- opencv-python
- streamlit
- pyyaml
Ensure the yolo11n.pt model file is available (automatically downloaded by ultralytics).
Launch the application:
python -m streamlit run new.py
Once running, use the web interface to:
- Select your camera channel.
- Adjust the AI sensitivity (Confidence Threshold) and Tripwire position.
- View live detection metrics and processing speed.
- Architecture: Upgraded to YOLO11 for superior detection accuracy.
- Tracking Algorithm: Implements a custom BoT-SORT configuration (high_quality_tracker.yaml) featuring:
- Sparse Optical Flow for camera motion compensation.
- Appearance-based re-identification (ReID).
- Optimized thresholding for stable tracking.
- Pipeline: High-efficiency frame processing with live Streamlit UI integration.