English

Abstract

Sign language is a need for deaf pupils to communicate with one another. People who are not deaf often do not learn sign language to interact with deaf people. It is also necessary to have an interpreter to explain the sign's meaning to others who are unfamiliar with it. Several unresolved issues, such as uncontrolled signing situations, various types of light, and varying degrees of partial occlusion, have adversely impacted hand gesture recognition efficacy. The suggested technique is unusual because it employs integrated features created by combining features obtained using conventional handcrafted feature extraction methods with deep learning models. Understandably the combined characteristics will include some repetitive and unnecessary characteristics, increasing computation time and wasting resources. We prevent this by using feature selection (FS) before providing the classifier with the merged features. We present the improved version of the newly developed Battle Royale Optimisation, IBROA, for feature selection. The characteristics are fed into a classifier for classification. Experiments were carried out, and the findings show that the proposed IBROA, which utilises integrated features and feature selection, outperforms classifiers and shows novel and efficient techniques for feature selection in Sign language classification.