Publication

Abstract : Effective communication with the deaf community necessitates the recognition of sign language; however, identifying skeleton-based sign language poses numerous challenges. These challenges include capturing temporal dynamics, sequential patterns, and other gesture nuances such as signer variability, gesture class imbalance, and semantic ambiguity. Therefore, developing robust methods for creating skeletal representations that accurately capture temporal dynamics, address semantic ambiguity, and remain connected to various communities is crucial for enhancing usability and inclusiveness. This paper presents a Skeleton-Based Sign Language Recognition (SB-SLR) framework for sign language recognition, addressing challenges by leveraging features derived from pivotal frames of sign videos. Our technique comprises three steps: Identification of Indicator Frame (IIF), Identification of Pivotal Frames (IPF), and Generation of Skeleton Sequence (GSS). We utilize a Convolutional Neural Network model for classification, which takes these sequences of skeleton images as input to recognize sign words. The SB-SLR algorithm demonstrates superior performance in labeling specific sign words, potentially enhancing communication accessibility for the hearing impaired.