RoleSolo developer, end to end
TypeAI desktop application
ModelYOLO11 · 99.49% mAP@50 · 72.1 FPS
Overview
Smart Eye replaces manual vehicle counting and timing at a car wash with an automated pipeline. A YOLO11 detector, trained exclusively on the client's own CCTV footage, finds every car; ByteTrack handles frame-to-frame association; and a custom Re-ID module reassigns stable identities to vehicles that return after occlusion, using cosine similarity on HOG features. Built with PyQt5 over a WAL-mode SQLite database, it logs each wash session with entry and exit timestamps and snapshots, raises overtime and vacuum-capacity alerts, and surfaces historical analytics, all from a single desktop window.
Key features
- Custom YOLO11 detector trained on client footage: 99.49% mAP@50 at 72.1 FPS, with a YOLO11n ↔ YOLO11l hot-swap at runtime.
- Multi-vehicle tracking via ByteTrack plus a HOG and cosine-similarity Re-ID module for stable IDs through occlusion.
- Automated entry and exit detection with hysteresis point-in-polygon testing, plus JPEG snapshots of every session.
- Configurable overtime and vacuum-full alerts via on-screen notifications and Windows system-tray messages.
- Historical analytics: cars-per-day bar chart, average-duration line chart, and a 24-hour peak-hours heatmap.
- RTSP IP-camera support with auto-reconnect, visual area configuration, and CSV export of session history.
