Real-Time American Sign Language (ASL) Interpreter

This project uses Python, OpenCV, and TensorFlow to implement and train an unsupervised Real-Time Object Detection Model that can identify and translate American Sign Language (ASL) signs in real-time with up to 91% precision.

So far, the model has been trained to detect 5 ASL signs: Hello, Thank You, Yes, No, and I Love You. I'm currently working on adding more ASL signs and increasing the number of images per sign to increase the accuracy and precision of the model.

Implementation

Image Collection and Labelling

This project uses OpenCV to capture images and detect signs in real time. The labelImg annotation tool (cloned from https://github.com/HumanSignal/labelImg has been used to label and annotate the captured images. To collect and label more images and/or signs, use Step 1 - Capture and Label Images using OpenCV and labelImg.

Training and Testing the Model

The collected and labelled images have been manually split into the train and test folders. The TensorFlow Model Zoo (cloned from https://github.com/tensorflow/models) has been used to train the model.

How to Use

1. Clone this repository (git clone https://github.com/ahorna-c/real-time-ASL-interpreter.git)
2. Activate the venv virtual environment (source venv/bin/activate)
3. Execute Step 2 - Training Model using TensorFlow.ipynb through Jupyter Notebook. This will enable you to interact with the model through your webcam.