Edge-Optimized Multi-Object Surveillance

flowchart LR
  A[Video input] --> B[YOLOv8 detection] --> C[Tracking] --> D[Anomaly rules]
  D --> E[Annotated output and alerts]
  B -. optimize .-> F[ONNX] -. optional NVIDIA .-> G[TensorRT]

Loading

How to run

1. pip install -r requirements.txt
2. Use the default COCO model through python -m src.pipeline --source 0 --save-output.
3. Optional custom training: configure data/data.yaml, then call src.train.fine_tune().
4. Export ONNX with src.optimize.export_to_onnx, benchmark supported variants with benchmark_inference, and see docs/edge_deployment.md for hardware setup.

Results

Run the benchmark on target hardware to generate reports/optimization_benchmark.csv. It records actual FPS and detection counts for PyTorch, ONNX, and (when available) TensorRT; do not claim performance numbers before running it.