ASL Hand Sign Detection 🤟 | YOLOv5s & YOLOv8n

A computer vision project that uses deep learning to detect ASL (American Sign Language) hand signs from images and video streams. Built with YOLOv5s and YOLOv8n, this model helps identify individual letters using a custom-trained dataset.

📌 Overview

This project aims to bridge the gap between gesture-based communication and real-time AI-driven recognition systems. It uses object detection models to recognize ASL letters from hand images in various scenarios.

• ✅ Real-time letter detection
• ✅ Supports YOLOv5 and YOLOv8
• ✅ Trained on universal hand-sign dataset
• ✅ Model files and runs included for easy reproducibility

---

🧠 Models Used

Model	Framework	Notes
YOLOv5s	PyTorch	Lightweight, fast, suitable for low-resource environments
YOLOv8n	Ultralytics	Newer architecture, faster convergence, and better precision in certain scenarios

---

📁 Dataset

• Downloaded from Roboflow
• Dataset contains plain background + hand signs for each letter (A-Z)
• ⚠️ Real-world testing may result in misclassifications due to complex backgrounds and lighting conditions.

---

⚙️ Installation

1. Clone the repository:

   git clone https://github.com/Harshit-Kandoi/ASL-letter-detection.git
   cd ASL-letter-detection
   

2. (Optional) Create a virtual environment:

    conda create -n yolov-env python=3.10
    conda activate yolov-env
   

3. Install requirements:

   pip install -r requirements.txt
   

4. (Optional) If you want to run Docker:

• Make sure Docker is installed, then build and run:

  docker build -t asl-detection .
  docker run -it --rm asl-detection
  

---

📈 Results & Observations

• ✅ Letter detection was smooth for most of the dataset.
• ⚠️ Challenges faced:
  • Out of memory errors with larger models on limited GPUs
  • Real-world data often includes background objects and noise
  • Accuracy dropped slightly when tested outside controlled environments
• 💡 Suggestion: Augment the dataset with diverse lighting, backgrounds, and skin tones to improve generalization.

---

🙌 Credits

• Ultralytics YOLO
• Roboflow
• Contributors to open datasets and training utilities

---

🚀 Future Improvements

• ✨ Add full-sentence detection support
• 📷 Live webcam/mobile integration
• 🧠 Train on cluttered background real-world datasets
• 🤖 Create a user interface using Streamlit or Flask

---

📬 Contact

Built by Harshit Kandoi
If you found this helpful, feel free to ⭐ the repo and share!