Praxis II — Automated Screw Sorting with Computer Vision & AI ScrewSorter is a computer-vision pipeline that automatically identifies screw head types and estimates screw lengths from images using Google Gemini AI and OpenCV. It was built for the Praxis II engineering design course and provides real-time screw analysis through a camera interface.

---

Table of Contents

• Features
• How It Works
• Supported Screw Head Types
• Prerequisites
• Getting Started
• Configuration
• Project Structure
• Limitations & Future Work
• License

---

Features

• 🔩 Three-class screw head classification — Flat Head, Oval Head, Round Washer Head using Google Gemini AI
• 🤖 AI-powered classification — Zero-shot screw head identification with Google's Gemini vision model
• 📏 Screw length estimator — Uses OpenCV contour analysis and QR code calibration to measure physical length in millimetres
• 📷 Real-time camera interface — Live video feed with capture functionality
• 🔒 Secure API key management — Environment variables with .env file support
• 💾 Result visualization — Saves annotated images with measurements and classifications

---

How It Works

Camera feed  ──►  QR detection (scale calibration)  ──►  Screw contour analysis
                        │                                       │
                        ▼                                       ▼
                 Physical measurements (mm)          Head shape classification
                        │                                       │
                        └───────────────────►  Gemini AI analysis  ──►  Results

1. QR Code Detection — Detects a QR code in the image to establish a scale reference (calibrated to known size)
2. Screw Detection — Uses OpenCV thresholding and contour analysis to identify screw shapes in the image
3. Length Measurement — Calculates screw shaft length using the QR scale and fitted bounding rectangles
4. Head Classification — Sends the image to Google Gemini AI with a prompt to identify the screw head type
5. Result Display — Shows measurements, classification, and saves an annotated image

---

Supported Screw Head Types

Class	Description
Flat_Head	Countersunk head that sits flush with the surface
Oval_Head	Partially countersunk with a rounded, decorative top
Round_Washer	Dome-shaped head with a built-in washer bearing surface

---

Prerequisites

Package	Purpose
google-generativeai	Gemini AI API client
opencv-python	Image processing and computer vision
Pillow	Image I/O and manipulation
numpy	Numerical operations
matplotlib	Result visualization
python-dotenv	Environment variable loading

Install all dependencies in a virtual environment: pip install google-generativeai opencv-python Pillow numpy matplotlib python-dotenv

---

Getting Started

1. Clone or download the repository

2. Set up a Python virtual environment:

   python -m venv .venv
   .venv\Scripts\activate  # On Windows
   pip install google-generativeai opencv-python Pillow numpy matplotlib python-dotenv

3. Get a Google Gemini API key:

   • Visit Google AI Studio
   • Create a new API key

4. Configure the API key:

   • Create a .env file in the project root
   • Add your API key: GEMINI_API_KEY=your_api_key_here

5. Run the application:

   python main.py

6. Usage:

   • Place a screw next to a QR code sheet for scale reference
   • Press SPACEBAR to capture and analyze
   • Press Q to quit
   • Results are displayed in the console and saved as result.png

---

Configuration

The following constants at the top of main.py can be tuned:

Constant	Default	Description
CAMERA_INDEX	1	Camera device index (0 for default webcam)
QR_SIZE_MM	21.9	Physical size of QR code in mm for calibration
SAVE_PATH	result.png	Path to save the annotated result image
MODEL_ID	gemini-2.5-flash	Gemini model to use for classification

---

Project Structure

ScrewSorter/
├── main.py                 # Main application script
├── .env                     # Environment variables (API key)
├── .gitignore              # Git ignore rules
├── README.md               # This file
├── LICENSE                 # License information
├── demo.py                 # Demo/test script
└── Older-Backup Files/     # Archived files

---

Limitations & Future Work

• Camera calibration — The QR size (QR_SIZE_MM) needs to be measured accurately for precise measurements
• Lighting conditions — Performance may vary with different lighting; optimal results with even, bright lighting
• Screw orientation — Best results when screws are positioned with heads clearly visible
• API rate limits — Gemini API has usage limits; includes retry logic for quota exceeded errors
• Real-time performance — Processing time depends on image complexity and API response time

---

Contributors

This project was developed by the Praxis II team:

• Team Members: Aditya Jain, Eva Deochakke, Jocelyn Li, Navneet Saxena, Peyton Smith

---

License

This project is licensed under the terms of the LICENSE file included in this repository.