A computer vision-based system for recognition and classification of Urdu sign language dataset.

Human beings rely heavily on social communication as one of the major aspects of communication. Language is the most effective means of verbal and nonverbal communication and association. To bridge the communication gap between deaf people communities, and non-deaf people, sign language is widely used. According to the World Federation of the Deaf, there are about 70 million deaf people present around the globe and about 300 sign languages being used. Hence, the structural form of the hand gestures involving visual motions and signs is used as a communication system to help the deaf and speech-impaired community for daily interaction. The aim is to collect a dataset of Urdu sign language (USL) and test it through a machine learning classifier. The overview of the proposed system is divided into four main stages , data collection, data acquisition, training model ad testing model. The USL dataset which is comprised of 1,560 images was created by photographing various hand positions using a camera. This work provides a strategy for automated identification of USL numbers based on a bag-of-words (BoW) paradigm. For classification purposes, support vector machine (SVM), Random Forest, and K-nearest neighbor (K-NN) are used with the BoW histogram bin frequencies as characteristics. The proposed technique outperforms others in number classification, attaining the accuracies of 88%, 90%, and 84% for the random forest, SVM, and K-NN respectively.