IoT Safety Helmet Detection System

AI on Edge with 3-Second Updates • December 2018

87% Accuracy

Edge AI

Real-time

IoT

🏭

Factory

📹

8MP • 1080p

👷

✓

👤

👷

✓

🖼️

2-3 FPS

🥧

Pi 3 Model B

Tiny YOLO • 6.2MB

1.2GHz Quad-Core

1GB RAM

~450ms/frame

⚠️ No Helmet!

🖼️

HTTP POST

Every 3sec

📡

WiFi

Devices

💻

📱

Access via Browser

HTTP GET

Detection Accuracy

Fine-tuned on 2500 factory images

Helmet

94.4%

No Helmet

~85%

Overall mAP50

87%

78% baseline→87% fine-tuned (+9%)

🎯

87%

Accuracy

⚡

1-2 FPS

Speed

💰

$95

Hardware

🔧

60%

Time Saved

📦 Hardware Cost

Pi 3 ($35) + Camera ($25) + SD Card ($10) + Power ($10) = $95 total

⚡ Processing Pipeline

1080p → 416×416 resize → Tiny YOLO inference → POST every 3sec

🏷️ Smart Labeling

Custom WinForms tool for model-assisted annotation • 60% faster than manual

Python

PyTorch

Tiny YOLO

Raspberry Pi 3

Flask

OpenCV