AI-powered person tracking system with auto-scan and manual control capabilities.
# Start the project
./start.sh
# Check status
./status.sh
# Stop the project
./stop.shVisit http://localhost:5173 to view the frontend.
sharkbytes2025/
+-- docs/ # Documentation
| +-- README.md # Main documentation
| +-- QUICK_START.md # Quick start guide
| +-- INTEGRATION_GUIDE.md
| +-- PERFORMANCE_OPTIMIZATIONS.md
| +-- TENSORRT_SETUP.md
| └-- ...
+-- models/ # AI models
| +-- yolo11n.pt # Base YOLO model
| +-- yolo11n.onnx # ONNX intermediate
| └-- yolo11n_160_fp16.engine # TensorRT optimized
+-- scripts/ # Setup and control scripts
| +-- setup_venv.sh # Virtual environment setup
| +-- start_project.sh # Start all services
| +-- stop_project.sh # Stop all services
| └-- status_project.sh # Check service status
+-- tests/ # Test and build scripts
| +-- test_analyze_frame.py
| +-- test_tensorrt.py
| └-- build_tensorrt_engine.py
+-- sentry/ # Sentry tracking service
+-- web/ # FastAPI backend
+-- frontend/ # React frontend
+-- mobile/ # React Native mobile app
+-- gemini/ # Gemini AI integration
└-- requirements.txt # Python dependencies
- Real-time person tracking using YOLOv11 + DeepSORT
- TensorRT optimization for 2x YOLO speedup
- Auto-scan mode - automatically scans for targets
- Manual control - override with pan/tilt controls
- Face detection for improved tracking accuracy
- Web interface - React frontend with live video feed
- Average FPS: 12-15 (peaks at 20-25)
- YOLO inference: 13-30ms (TensorRT optimized)
- Resolution: 320x320 camera input
See the docs/ folder for detailed guides:
- Jetson Orin Nano (or compatible NVIDIA device)
- Python 3.10+
- CUDA 12.6+
- TensorRT 10.7+
- USB camera
- PCA9685 servo controller
Copyright © 2025 SharkBytes