+ Computer Vision · Flagship
Realtime Vision Analytics
Real-time people and vehicle counting, zone occupancy and dwell-time analytics on CPU with YOLO11n + ByteTrack, served by FastAPI with a live dashboard.
19.3 FPSONNX on a laptop CPU
Overview
A video analytics service that turns webcams, video files or RTSP streams into footfall and traffic metrics. YOLO11n detections are tracked with ByteTrack and fed into directional line counters, polygon zones with occupancy and dwell time, and a heatmap. FastAPI serves an MJPEG stream, REST and WebSocket stats, and a SQLite event log with CSV export, shown on a live dashboard. It runs on an ordinary CPU, with an ONNX Runtime path benchmarked against PyTorch.
Highlights
- Directional per-class line counting and zone dwell analytics; demo-clip counts matched a manual count
- ONNX Runtime at 19.3 detector FPS vs 12.7 for PyTorch (320 px) on a 2-core laptop CPU
- Live MJPEG dashboard at about 9 FPS end-to-end on CPU with WebSocket KPIs and an occupancy chart
- Every line crossing and zone enter/exit logged to SQLite with CSV export
- 48 tests including end-to-end API/WebSocket/MJPEG tests and real YOLO inference
Tech stack
Gallery
