🅿️ Real-Time Parking Counter

Project Overview

A real-time parking occupancy monitoring system built as a Computer Vision final project at Vietnam National University. Combines two detection backends (Classical CV at >150 FPS and YOLOv8 vehicle detection), a YOLO11n-cls classifier trained on 711,856 PKLot images achieving 98.48% accuracy, 8-directional A* path planning to guide drivers to the nearest free slot, and a FastAPI live web dashboard with MJPEG stream. Role: Project lead — dataset pipeline, Classical CV & YOLO detection, model training.

Key Features

• Dual detection backends: Classical CV (>150 FPS, CPU) ↔ YOLOv8n vehicle detection (GPU) — switchable live
• YOLO11n-cls classifier trained on 711,856 PKLot cells — 98.48% test accuracy, 3.2 MB model
• 8-directional A* path planning guides drivers from entrance to the nearest free slot
• FastAPI web app: MJPEG live stream, /status JSON, /switch toggle endpoint
• Interactive slot marker tool: draw polygon/rectangle slots, obstacles, entry point via mouse
• Homography correction for bird's-eye view on angled cameras
• Multi-source support: file / RTSP / webcam via sources.json
• HUD overlay: free slot count, nearest slot ID, FPS, color-coded slot status