🤙 American Sign Language (ASL) Detection & Classification

A Machine Learning pipeline for feature extraction and robust classification of ASL alphabet images.

📖 Overview

This repository contains a complete pipeline for recognizing and classifying the American Sign Language (ASL) alphabet. Originally built as an end-of-semester college project, it demonstrates the entire machine learning lifecycle—from data preprocessing and feature extraction to training, evaluation, and visualization.

By applying Histogram of Oriented Gradients (HOG) with traditional machine learning models (SVM, KNN, Random Forest), the project identifies distinct visual features in hand gestures to map them to specific alphabet letters.

📊 Dataset

The models are trained using the publicly available Kaggle ASL Alphabet Dataset.

• The dataset features varying lighting conditions and angles.
• We balance the classes to prevent skewed predictions.
• Classes: 29 (A-Z, space, del, nothing)
• Image dimensions: 64x64 pixels (Resized for performance optimization)

⚙️ Methodology

Instead of using resource-heavy Deep Learning (CNNs), we leverage classical computer vision techniques:

1. Data Preprocessing & Balancing: The images are loaded, resized to 64x64, and balanced to ensure equal representation for all signs.
2. Feature Extraction (HOG): The core mechanism is the Histogram of Oriented Gradients, which counts the occurrences of gradient orientation in localized portions of an image. This effectively highlights the edges and contours of hand shapes.
3. Classification Models Evaluated:
   • Support Vector Machine (SVM): Utilizing a linear kernel to find the optimal hyperplane between high-dimensional features.
   • K-Nearest Neighbors (KNN): Predicting classes based on proximity in feature space.
   • Random Forest (RF): An ensemble learning method reducing variance and overfitting.

📈 Results Highlights

Our evaluation revealed competitive accuracies across the models, showcasing that structural edge-based features (HOG) capture the essence of ASL hand signs very effectively:

• Support Vector Machine: High dimensional boundary identification yielded excellent results.
• Random Forest: Solid robustness against noise in the images.

---

💻 Tech Stack

• Languages: Python
• Vision & Image Processing: OpenCV (cv2), scikit-image
• Machine Learning: scikit-learn
• Data & Visualization: numpy, pandas, matplotlib, seaborn

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/RohanBabbar/ASL-Detection-and-Classification.git
cd ASL-Detection-and-Classification

2. Set up the Environment

It is highly recommended to use a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate
pip install -r requirements.txt

3. Usage & Execution

We've modularized the code into production-ready scripts. Ensure that you have downloaded the Kaggle ASL dataset and placed it in an accessible directory (e.g., data/asl_dataset).

• Run the pipeline: You can explore the newly refactored scripts located in the src/ directory to run feature extraction and execute training models natively.

---

📂 Repository Structure

├── README.md               <- Project documentation
├── requirements.txt        <- Python dependencies
├── assets/                 <- Auto-generated graphs and charts
├── src/                    <- Refactored modular python scripts
│   ├── dataset.py
│   ├── features.py
│   └── train.py