Smart_Scan

Smart_Scan is a Flask web app for AI-assisted shelf/product scanning.
It detects objects from uploaded images, estimates quality/freshness, extracts OCR text for product scans, and exposes summary/history/trend APIs.

Features

• YOLO-based object detection for product and fruit modes
• Freshness scoring and shelf-gap summary metrics
• OCR extraction for product scans (EasyOCR)
• Annotated output image and confidence graph rendering
• Scan history + trends via SQLite-backed APIs
• Live alerts endpoint for latest scan

Tech Stack

• Python + Flask
• PyTorch + Ultralytics YOLO
• EasyOCR + OpenCV + Pillow
• Matplotlib + NumPy
• SQLite

Project Structure

• app.py - Flask app, inference pipeline, routes
• scan.py - scan-related helper logic
• database.py - schema + history/trend access
• alerts.py - alert evaluation rules
• config.py - thresholds/model paths/shelf settings
• templates/ - HTML templates
• static/ - CSS/assets
• test_smoke.py - runtime smoke tests

Setup (Windows / PowerShell)

Recommended Python: 3.12.x (runtime.txt pins python-3.12.5).

cd C:\github-all\Smart_Scan
py -3.12 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip setuptools wheel
pip install -r requirements.txt

Run

python app.py

Then open: http://127.0.0.1:5000/

Quick Health Check

With the server running in one terminal, run in another terminal:

.\.venv\Scripts\Activate.ps1
python test_smoke.py

Expected output includes:

• PASS /api/history
• PASS /api/trends
• PASS /detect
• All smoke tests passed.

API Endpoints

• GET / - main UI
• POST /detect - run detection (image file + mode=product|fruit)
• GET /api/history - scan history
• GET /api/trends - trend/aggregate data
• GET /api/alerts/live - alerts for latest scan
• GET /image - last annotated image buffer
• GET /graph - last confidence graph buffer

Operational utility routes:

• GET /favicon.ico returns 204
• GET /.well-known/appspecific/com.chrome.devtools.json returns 204

These are intentionally handled to keep development logs clean.

Notes

• First install can be slow due to large ML dependencies (torch, ultralytics, easyocr).
• CPU mode is supported; CUDA is used automatically if available.
• Model files (*.pt, *.pkl) are loaded at startup from paths in config.py.