SafeCityAI

Intelligent Traffic Violation Detection using YOLOv5.

Designed & Developed by Mohammed Zaid Nayaz
Project Owner: Mohammed Zaid Nayaz
Copyright © 2026 Mohammed Zaid Nayaz. All Rights Reserved.

Overview

SafeCityAI is a production-ready AI object detection project for detecting:

• Helmet
• No_Helmet
• License_Plate

It includes a FastAPI backend, a premium React + Vite dashboard, YOLOv5 training scripts, dataset preparation guides, deployment files, tests, diagrams, and a Windows launcher.

Project Structure

SafeCityAI/
  frontend/       React, Vite, TypeScript, TailwindCSS
  backend/        FastAPI, OpenCV, PyTorch, YOLOv5 inference API
  dataset/        YOLO dataset folders and data.yaml
  training/       train, detect, validate, export, evaluate scripts
  weights/        yolov5s.pt and trained best.pt location
  docs/           documentation and diagrams
  scripts/        setup and utility scripts
  screenshots/    demo screenshot outputs
  demo/           demo guide and presentation assets

Quick Start

cd SafeCityAI
.\scripts\setup_windows.ps1
.\RUN_SAFE_CITY_AI.bat

Manual backend:

cd backend
..\.venv\Scripts\uvicorn.exe app.main:app --reload

Manual frontend:

cd frontend
npm install
npm run dev

API

Backend runs on http://127.0.0.1:8000.

• GET /
• GET /health
• POST /predict
• POST /predict-video
• POST /webcam
• GET /metrics
• GET /docs

Example detection:

{
  "class": "No_Helmet",
  "confidence": 0.91,
  "bbox": [120, 220, 80, 110]
}

Training

Prepare images and labels inside dataset/, then run:

python training\train.py --epochs 80 --batch 16 --img 640
python training\validate.py
python training\export.py

YOLOv5 automatically writes loss graphs, precision, recall, mAP, confusion matrix, and TensorBoard logs to training/runs/.

Deployment

• Frontend: Vercel using frontend/vercel.json
• Backend: Render using render.yaml
• Optional Dockerfiles are included for containerized demos.

See DEPLOYMENT.md.