Volume 30, Issue 3, 2021
DOI: 10.24205/03276716.2021.5020
Recognition of Hand Signs Based on Geometrical Features using Machine Learning and Deep Learning Approaches
Abstract
Hand gestures are used by specially challenged people for interacting with the outside world. The gesture vocabulary is created with action words and alphabets with a focus on using them as a sign language interpreter. This paper focuses on the usage of the American Sign Language (ASL) gestures and recognition of few action gestures using single hand. The skin region is identified by marking the skin-colour pixels in Hue Saturation and Value (HSV) colour space. The hand contours are obtained after performing background subtraction techniques and morphological operations. The pre-processed hand mask is given as input to the Support Vector Machine (SVM), a machine learning classifier model, and to the Convolution Neural Network (CNN), a deep learning model. The performance analysis in the predicted output for all the models are analyzed. The accuracy of 97% is witnessed for recognition of ASL gestures and 93% for action gestures using CNN approach that surpasses SVM approach yielding an accuracy of 84% and 80% for the same.
Keywords
American Sign Language; Centroid; Contours; Convolutional Neural Network; Gesture recognition; Hand detection; Support Vector Machine