NABAHAH (نباهة) – AI-Powered Laboratory Safety System

Team Members

• Musab Alabdullatif
• Mawda Alguraafi
• Norah Bindaham
• Turki Akbar

1-Overview

NABAHAH (Arabic: نباهة, meaning alertness) is an AI-driven laboratory safety monitoring system that combines Computer Vision, Database Management, and Real-Time Analytics to enhance safety and compliance.

It automatically detects:

• PPE compliance (lab coat, gloves, mask, goggles)
• Unauthorized entry into restricted red zones
• Chemical and liquid spills

All detections are logged in a Supabase PostgreSQL database and displayed through a FastAPI-based web dashboard for safety officers.

---

2-Key Features

✅ Real-time detection of PPE violations, spills, and red-zone breaches
✅ Live monitoring dashboard using FastAPI + Ngrok
✅ Voice alerts with Edge-TTS (Arabic/English)
✅ Centralized Supabase database for detections, alerts, and clips
✅ RAG (Retrieval-Augmented Generation) chatbot that answers safety queries
✅ Continuous retraining and model updates using new data
✅ Secure admin login with bcrypt-hashed passwords

---

3-System Architecture

The system follows a three-layer architecture:

1. Data Acquisition Layer – Captures live video streams via OpenCV
2. AI Detection Layer – Uses YOLOv8 models to analyze PPE, spills, and red-zone activity
3. Monitoring & Alert Layer – Displays detections, triggers voice alerts, and logs data

---

4-Core Technologies

Category	Tools / Frameworks
Programming	Python, FastAPI, Jupyter Notebook
AI Models	YOLOv8, PyTorch, OpenCV, Norfair
Backend	Supabase (PostgreSQL), PostgREST
Audio	Edge-TTS, Pydub, MoviePy
RAG Chatbot	Databricks + Vector Database
Deployment	Ngrok, Google Colab
Visualization	Dashboard with analytics (EDA, compliance rate, violations)

---

5-Database Schema

Table	Description
users	Admin login and authentication
videos	Uploaded video metadata
persons	Tracked individuals and PPE status
detections	Frame-level detection logs
alerts	PPE or red-zone alerts
clips	Saved annotated video clips
spills	Chemical/liquid spill records

---

6-Model Performance

Metric	Result	Description
mAP	94 %	Mean Average Precision for detection
Precision	93 %	Correct detections among alerts
Recall	90 %	True positive rate
Inference Speed	0.8 s / frame	Real-time detection capability

---

7-Installation

Clone the Repository

git clone https://github.com/MUSAB10000/NABAHAH.git
cd NABAHAH

All Models

You can find all trained models here:
Models