AP-Vision Pro is a cutting-edge, privacy-centric AI surveillance system designed for real-time personnel tracking and safety auditing. It combines the speed of YOLOv11 with the precision of MediaPipe to provide actionable insights at the edge.
- Real-Time Tracking: Utilizes YOLOv11 with ByteTrack for persistent identity management of multiple subjects.
- Behavioral Intelligence: Integrates MediaPipe Tasks API to analyze 33 skeletal landmarks for action recognition.
- Cloud-Native Architecture: Implements WebRTC to enable secure, low-latency video streaming from the user's browser to the inference engine.
- Data Sovereignty: Processes video frames locally/in-memory to ensure zero data retention and total privacy.
- Audit Reporting: Built-in system to generate and download CSV safety reports for industrial compliance.
The system follows a modular, asynchronous design to prevent UI freezing during heavy AI computation:
- Frontend: Streamlit-based mobile-responsive UI.
- Streaming: WebRTC peer-to-peer connection for live frame transfer.
- Inference Engine: YOLOv11 (Detection) + MediaPipe (Pose) + Custom Logic.
- Data Layer: Session-state logging for exportable reports.
git clone [https://github.com/YOUR_USERNAME/AP-Vision-Pro.git](https://github.com/YOUR_USERNAME/AP-Vision-Pro.git)
cd AP-Vision-Pro
pip install -r requirements.txt