SafeSite AI

SafeSite AI is a Flask-based construction safety monitoring app that uses YOLO models to detect workers, PPE compliance, and restricted-zone intrusions in real time.

Best Model

https://drive.google.com/file/d/1keShImRescQkreYiGrvxLL_G49aH4vRY/view?usp=sharing

Features

• Real-time video monitoring from a webcam or uploaded video file
• PPE violation detection for missing helmets and safety vests
• Restricted-zone creation and intrusion alerts
• Bilingual voice alerts
• Violation logging in a local SQLite database
• Dashboard analytics and PDF report generation

Project Structure

SafeSIteAI/
|-- app.py                  # Flask entry point
|-- detection/              # PPE and zone detection logic
|-- alerts/                 # Voice alert system
|-- database/               # SQLite access layer
|-- dashboard/              # Analytics and charts
|-- reports/                # PDF report generation
|-- templates/              # Flask HTML templates
|-- static/                 # CSS, JS, and generated screenshots
|-- zones/                  # Persisted restricted-zone definitions
|-- uploads/                # Uploaded video files
|-- runs/                   # YOLO training/inference outputs
|-- ppe detection.v3i.yolov8/  # Local dataset folder
|-- requirements.txt
|-- .gitignore

Requirements

• Python 3.11+ recommended
• A webcam for live monitoring, or a local video file
• YOLO model weights placed in the project root:
  • ppe_best.pt
  • yolov8n.pt

Setup

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Run

python app.py

Then open http://127.0.0.1:5000 in your browser.

How It Works

• yolov8n.pt detects people in the frame.
• ppe_best.pt detects PPE items such as helmets and vests.
• A worker is flagged when their bounding box does not sufficiently overlap with the required PPE detections.
• Restricted zones are stored in zones/zones.json.
• Violations are stored in database/safesiteai.db.
• Incident screenshots are saved under static/screenshots/.

Main Routes

• / live monitoring page
• /dashboard analytics dashboard
• /reports report generation page
• /video_feed MJPEG stream

API endpoints include:

• POST /api/start
• POST /api/stop
• POST /api/upload_video
• GET /api/videos
• POST /api/toggle_detection
• POST /api/toggle_voice
• GET/POST /api/zones
• DELETE /api/zones/<name>
• POST /api/zones/clear
• GET /api/violations
• POST /api/report/generate
• GET /api/stats
• GET /api/recent_violations

Git Ignore Notes

The repository is configured to ignore local/generated artifacts such as:

• virtual environments and Python cache
• uploaded videos
• generated violation screenshots
• local SQLite databases
• YOLO training outputs in runs/
• large model weights like *.pt
• bulky dataset contents inside ppe detection.v3i.yolov8/

Dataset metadata files such as data.yaml and the Roboflow README files are still kept trackable.

Notes

• uploads/, static/screenshots/, and the database file are created and updated locally at runtime.
• The dataset folder in this project appears to be intended for local experimentation/training, not for normal source control.
• If you want this project to be easier for other people to run, the next good step would be adding sample environment/setup notes for model downloads and OS-specific voice dependencies.