Barcode & QR Code Detector

A real-time barcode and QR code detection system built with YOLOv8 and ZXing-CPP. It detects codes from a live webcam feed, decodes them, and automatically looks up product information from multiple online databases — no specialized scanning hardware needed.

---

How it works

The system runs a multi-stage pipeline:

1. Detection — YOLOv8 (Small) finds and localizes barcodes and QR codes in the webcam frame with bounding boxes
2. Decoding — ZXing-CPP reads the content of the detected code
3. Product lookup — the decoded value is queried against 5 product databases to retrieve item information
4. Visual feedback — color-coded results on screen:
   • 🟢 Green — detected, decoded, and product found
   • 🟠 Orange — decoded but product not in database
   • 🔴 Red — detected but couldn't decode

---

Model

• Architecture: YOLOv8-Small (CSPDarknet backbone, PANet neck, 3 detection heads)
• Parameters: ~11.1M trainable
• Input size: 640×640
• Classes: 2 (barcode, qrcode)
• Pre-trained on: COCO, fine-tuned on a custom dataset
• Training: 30 epochs, AdamW optimizer, batch size 16, AMP enabled

Dataset

• 325 barcode images + 275 QR code images
• 80/20 train/validation split
• Augmentations: rotation (±5°), perspective transform, HSV color jitter

Detection scales

The model uses three detection heads to handle different distances:

• 80×80 feature map → small/distant codes
• 40×40 feature map → medium distance
• 20×20 feature map → close-range codes

---

Product databases queried

• Open Food Facts (World)
• Open Food Facts (Egypt)
• Open Beauty Facts
• Open Pet Food Facts
• Open Products Facts

---

Challenges we solved

• Lighting variability — handled with HSV augmentation during training
• Viewing angle — perspective transformation augmentation lets it handle codes up to ~±20° off-axis
• Multiprocessing on Windows — set DataLoader workers=0 to fix PicklingError in Jupyter
• Decode failures — three-tier color feedback so partial detections are still useful

---

Tech

• Python 3
• YOLOv8 (Ultralytics)
• ZXing-CPP for barcode/QR decoding
• OpenCV for webcam capture and visualization
• Open Food/Beauty/Pet Food/Products Facts APIs

---

Team

Mohammed Tariq, Mazen Abdallah, Malak Elhawary, Maryam Hatem, Ahmed Deaiy, Marvel Fady, Youssef Nashaat