Performance Analysis of Machine Learning Models for Kannada Sign Language Recognition

Kannada Sign Language is the communication method used by individuals with hearing/speech impairments in Karnataka. A person uses hand gestures, arm posture, hand orientation, and facial expressions while communicating with sign language. The general public's limited knowledge of sign language is a challenge for differently-abled people. Therefore, sign language has to be converted into a form everyone can understand. In this study, we test various models to detect and recognize sign gestures, analyze their performance, and develop a mobile application that uses the best model. A custom dataset of 20 Kannada character sign gestures was created, key points were identified using MediaPipe, and these key points were used by the Machine Learning models to recognize the sign gesture. The Random Forest model achieved the highest accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 8. 8 5 \%}$</tex>. The LSTM model was utilized in the mobile application for its higher recognition rate in real-time compared to Random Forest.