Real-Time-Rock-Paper-Scissors-Recognition-Using-YOLOv5

Real-Time Rock, Paper, Scissors Recognition Using YOLOv5

This project is a real-time hand gesture recognition system built with YOLOv5, capable of detecting and classifying rock, paper, and scissors using a webcam feed.

🎓 This was my official school project, and it means a lot to me — it marks my hands-on dive into real-world AI development. I'm passionate about building smart systems, and this project reflects my interest in solving interactive problems using machine learning.

---

📌 Features

• ✅ Real-time detection of rock, paper, and scissors gestures
• 📸 Custom dataset creation with varied lighting and angles
• 🏷️ Manual image annotation using LabelImg and Label Studio
• 💻 Model trained locally on a personal laptop (YOLOv5, mAP@0.5 = 0.96)
• ⚡ Real-time webcam detection with OpenCV (20–30 FPS)
• 📄 Full documentation of setup, training, troubleshooting, and deployment

---

🛠️ Tools & Technologies

• Python 3.12
• PyTorch
• YOLOv5
• OpenCV
• LabelImg, Label Studio
• Anaconda (virtual environment)
• Windows (training environment)

---

📂 Project Structure

📂 rock-paper-scissors-yolov5/
├── 📂 data
├── 📂 models
├── 📂 scripts/
│   ├── split.py           # Optional: dataset splitter
│   └── gen-data-yaml.py   # Script to generate data.yaml
├── 📂 utils
│   ├── classes.txt        # Class names (e.g., rock, paper, scissors)
│   └── data.yaml          # Dataset config file for YOLOv5
├── detect.py              # Detection script
├── requirements.txt       # Requirements needed to be downloaded
├── train.py               # Training script
├── val.py                 # Validation script
├── export.py

---

🧠 How It Works

1. Data Collection
   Hand gesture images (rock, paper, scissors) were captured manually using a webcam.

2. Annotation
   Bounding boxes were drawn manually using LabelImg and exported in YOLO format.

3. Training

   • Model: YOLOv5s
   • Epochs: 25
   • Batch Size: 16
   • Input Size: 640x640
   • Achieved mAP@0.5: 0.96

4. Deployment
   The trained model was used for live webcam detection using OpenCV on Windows.

   ---

🖼️ Project Highlights

📘 Technical Report

You can read the full technical documentation and implementation guide below:

📄 Download the PDF Report Report.pdf

🚀 Run It Yourself

Follow the steps below to run the real-time Rock, Paper, Scissors recognition system on your own machine.

1. Clone this repo and install dependencies

git clone https://github.com/Prosperteni/Real-Time-Rock-Paper-Scissors-Recognition-Using-YOLOv5.git
cd Real-Time-Rock-Paper-Scissors-Recognition-Using-YOLOv5
pip install -r requirements.txt

2. Download the Trained Model

The trained YOLOv5 model weights (best.pt) are too large to be stored on GitHub directly.

📦 Download best.pt from Google Drive

After downloading, place the best.pt file into your project folder:

📂Real-Time-Rock-Paper-Scissors-Recognition-Using-YOLOv5/
└── best.pt

3. Run Real-Time Detection

Make sure your webcam is connected. Then run:

python detect.py --weights best.pt --source 0

--source 0 refers to the webcam.

You can also test an image or video by changing the source: --source path/to/image.jpg or --source path/to/video.mp4

4. Download the Dataset

Download the dataset from the following link and place it into the main project folder (e.g., dataset/):

📂 Real-Time-Rock-Paper-Scissors-Recognition-Using-YOLOv5/
└── 📂 dataset

📥 Download Dataset

Make sure the folder structure matches what your config file (data.yaml) expects.

---

5. Optional: Train the Model from Scratch

If you want to retrain the model using your own dataset:

python train.py --img 640 --batch 16 --epochs 50 --data datasets/data.yaml --weights best.pt

Note: Make sure your dataset is correctly formatted and data.yaml is set up properly.