Computer vision based indian sign language recognition using deep learning

Abstract

Speech is a human default and unique modality for language development and communication which is essential for memory and overall cognitive development. Excellency in language permits a child to be extrovert enriching the development of cognitive and psychosocial skills; whereas, for auditory deprived children, the misalignment of the brain and ear makes them impotent to communicate with the society which creates a central dogma that hearing-loss is a disability which further ignores their psycho-social identity. To fill such gaps and make their community more freewheeling in India, Indian Sign Language (ISL) - a complete language with its own linguistic and verbal elements was framed. Though ISL is appropriate and absolute in every linguistic approach, lack of prerequisite and proficiency enforces dedicated teachers to teach the curriculum through contrived signs for the sake of convenience that not only diminishes the distinctiveness of ISL but also dislodges the idea of learning their mother tongue. This creates an imbalance in the analogous learning of communication and curriculum language. In order to balance the level in learning, effective vision-based days of the week ISL model is developed through Convolution Neural Network (CNN) architecture which boasts independent learning of ISL. The proposed model comprises of six stages: dataset creation, preprocessing, splitting dataset into train, validation and test, applying various types of image augmentation techniques according to split, constructing CNN model for feature extraction and classification and finally evaluating the result through evaluation measures. Initially, an image dataset is created as there is a scarcity of standard ISL datasets in internet sources. The images are created on vision-based technique to avoid of carrying additional superfluous hardware gadgets for human computer interaction.