Helmet Detection — Safety Compliance System

Real-time helmet detection system using YOLOv8 that monitors live camera feeds, logs violations, captures snapshots, and dispatches automated email alerts to site managers.

---

How It Works

1. YOLOv8 processes every webcam frame to detect person, worker, helmet, and hardhat classes
2. For each detected person, it checks if a helmet bounding box overlaps the head region (top 40% of the person box)
3. If no helmet is found on a person → violation is triggered:
   • Snapshot is saved to images/violations/
   • OCR runs on the person crop to read a chest/suit ID
   • Email alert with the snapshot is sent to the manager (throttled to once every 20 seconds)

---

Tech Stack

Component	Tool
Object detection	YOLOv8 (Ultralytics) — model/best.pt
Vision processing	OpenCV
Web server	Flask
Chest number OCR	EasyOCR / custom (ocr_utils.py)
Email alerts	Python smtplib + Gmail SMTP
Frontend	Vanilla HTML / CSS / JS

---

Project Structure

Helmet-Detection/
├── app.py                        # Flask server + detection pipeline
├── email_utils.py                # Gmail SMTP alert sender
├── ocr_utils.py                  # Chest number OCR extractor
├── cleanup.py                    # Auto-delete old violation images
├── model/
│   └── best.pt                   # Fine-tuned YOLOv8 helmet model
├── images/violations/            # Auto-saved violation snapshots
├── logs/violations.csv           # Violation event log
├── templates/index.html          # Professional dashboard UI
├── static/style.css              # Custom styling
├── Dockerfile                    # Containerization config
├── render.yaml                   # Deployment specification
├── .env.example                  # Environment variables template
└── requirements.txt              # Project dependencies

---

Setup & Run

1. Install dependencies

pip install -r requirements.txt

2. Configure Environment (Optional)

Copy .env.example to .env and set your credentials:

cp .env.example .env

3. Run Application

python app.py

3. Open dashboard

http://localhost:5000

---

Dashboard Features

Feature	Details
Live detection feed	YOLOv8 bounding boxes with confidence %
Stats bar	Uptime · Helmets detected · Violations · Persons tracked
Event log	Timestamped, filterable (All / Violations / Safe)
Violation gallery	Last 6 snapshots, click to enlarge
Alert banner	Flashes on new violation

---

Email Alerts

Configure credentials in email_utils.py:

EMAIL_ADDRESS  = 'your@gmail.com'
EMAIL_PASSWORD = 'your_app_password'   # Gmail App Password (not your real password)
MANAGER_EMAIL  = 'manager@company.com'

Generate a Gmail App Password: Google Account → Security → 2-Step Verification → App Passwords

---

Model Details

• Base: yolov8n.pt (YOLOv8 Nano)
• Fine-tuned on: helmet / hardhat / person / worker dataset (data.yaml)
• Inference threshold: 0.5 confidence
• Helmet-on-head check: helmet box must be within the top 40% of the person box

---

Configuration (app.py)

Constant	Default	Description
MODEL_PATH	model/best.pt	Path to YOLO weights
MANAGER_EMAIL	gmail address	Alert recipient
VIOLATION_FOLDER	images/violations	Snapshot save path
Email throttle	20 seconds	Min gap between email alerts
Confidence threshold	0.5	Min detection confidence

---

Requirements

• Python 3.10+
• Webcam or RTSP camera
• Gmail account with App Password enabled (for email alerts)

---

Author

Harsh Ramesh Nerkar
B.Tech CSE 2026 | AI/ML Engineer
Email: harshrameshnerkar@gmail.com