American Sign Language Alphabet Recognition Using Machine Learning

Abstract

In this paper, we propose a machine learning model for recognizing all 26 letters in the American Sign Language (ASL) alphabet. The model is trained using a dataset obtained by recording a 30-frame video of hand movements. MediaPipe is used to detect hand positions in each frame and extract their coordinates, resulting in an array of 63 values. These sequences of arrays are then passed down to our Sequential model that uses LSTM as the input layer and Dense as the output layer. We evaluated two models, with Model 1 and Model 2 both achieving similar accuracy. Our study demonstrates that the proposed machine learning model consisting of MediaPipe's hand detector and a neural network can effectively recognize all letters of the ASL alphabet.