Rubik’s Cube Pause vs. Moving Detection

This is my first machine learning project, where I built and trained a model to detect whether a Rubik’s Cube is paused or moving in a video.

The most important part of this project was creating my own dataset from scratch, which taught me a lot about the challenges of data preparation in ML.

---

Project Overview

• Built a dataset of 9500+ frames extracted from Rubik’s Cube videos
  
• Labeled frames into two categories: Paused and Moving
• Trained a MobileNetV2-based CNN model using TensorFlow/Keras
• Achieved:
  • ~86% accuracy
  • 0.95 ROC-AUC score
    
• Implemented real-time video detection with OpenCV

---

Features

• Custom dataset creation (frames extracted + labeled manually)
• Model training with class balancing and augmentation
• Real-time detection with borders + labels:
  • Red = Paused
  • Green = Moving
• Time analysis graph showing pause time, move time, and total solve time

---

Example Output

• Paused cube frame (red border + label)

• Moving cube frame (green border + label)

• Solve time breakdown graph

  

---

Installation

Clone the repository:

git clone https://github.com/yourusername/rubiks-cube-detection.git
cd rubiks-cube-detection