UnPark AI-Powered Smart Parking & Automated Violation Detection System
UnPark is an intelligent parking surveillance system that uses Computer Vision and AI to detect parking violations in real time, assign priority scores, and generate structured reports for enforcement authorities.It transforms passive CCTV monitoring into an automated, intelligent enforcement system.
Problem
Illegal roadside and sidewalk parking is a common issue, especially in dense urban areas. It: Blocks pedestrian walkways Causes traffic bottlenecks Requires inefficient manual monitoring
Features
- Real-time vehicle detection (YOLO)
- Multi-object tracking (DeepSORT / ByteTrack)
- Violation detection logic (No-parking / Overstay)
- Urgency scoring system
- OCR-based number plate extraction (ANPR-ready)
- Evidence clip generation
- MongoDB-based violation storage
- FastAPI backend
- Conceptual AR-assisted enforcement interface
Tech Stack
Language : Python AI & Computer Vision: YOLO (Ultralytics) , DeepSORT / ByteTrack , OpenCV , OCR (for ANPR pipeline) Backend : FastAPI , REST APIs Database: MongoDB (Atlas-ready)
How It Works
1️⃣ Vehicle Detection Vehicles are detected using YOLO with a confidence threshold: Confidence≥τ Only detections above threshold τ are processed further.
2️⃣ Violation Urgency Scoring Each violation is assigned a priority score: Urgency=αT+βL+γV
Where: 𝑇 = Duration of violation L = Location severity weight V = Vehicle type weight α,β,γ = Tunable parameters
3️⃣ OCR & Reporting Number plates are extracted using OCR Violation metadata is stored in MongoDB Evidence clips are saved Structured reports can be generated for authorities
Project Structure unpark/ │ ├── app.py # FastAPI server ├── inference.py # Detection + tracking pipeline ├── database.py # MongoDB connection ├── uploads/ # Input videos ├── outputs/ # Generated violation clips └── models/ # YOLO model files
Running the Project 1️⃣ Install Dependencies pip install -r requirements.txt 2️⃣ Start FastAPI Server uvicorn app:app --reload 3️⃣ Access API
Open: http://127.0.0.1:8000/docs Upload a video and let UnPark process violations automatically.
What Makes UnPark Unique End-to-end AI pipeline (Detection → Tracking → OCR → Backend → DB) Urgency-based violation prioritization AR-assisted enforcement concept Scalable architecture for multi-camera environments
Future Improvements Integrate full Automatic Number Plate Recognition (ANPR) with higher OCR accuracy. Integrate payment or fine management systems for automated penalty workflows. Collaborate with municipalities and commercial complexes. Integrate AR-powered enforcement tools for real-time officer assistance.
Author Developed by Raama Bhatia