This is a fine-tuned hand gesture recognition model designed to classify numerical sign language gestures for barrier-free kiosks.

The model is highly effective, achieving an average confidence of 90%, an average inference time of 76.92 milliseconds, and a compact model size of 8.1 MB. It is specifically designed for resource-constrained environments, such as kiosks and other embedded systems, making it ideal for scenarios requiring high accuracy and fast inference speed within limited computing resources.

Using Google's MediaPipe hand gesture recognition model, the model was trained and fine-tuned to recognize Korean Sign Language numbers.

Clone this project:

git clone https://github.com/Godsihyeong/Hand-Gesture-Recognition-for-Barrier-Free-Kiosk.git

Using pip and requirements.txt:

pip install -r requirements.txt

Using conda and environment.yaml:

conda env craete -f environment.yaml

You can run the fine-tuned model and test it using your laptop's webcam by executing the webcam.py script. This script utilizes the fine-tuned hand gesture recognition model and displays the predictions in real time.

Instructions:

1. Ensure that your environment is set up and all dependencies are installed.
2. Run the following command in your terminal:

python webcam.py

3. The webcam feed will open, and the model's predictions for hand gestures will be displayed.

This project is licensed under the Apache License 2.0. See the LICENSE file for details.