DEVELOPMENT OF A SIGN LANGUAGE RECOGNITION MODEL FOR UZBEK WORDS USING DEEP LEARNING METHODS

Abstract

The development of a sign language determination model for Uzbek words using deep learning methods aims to address the communication barriers faced by the hearing-impaired community in Uzbekistan. This study leverages advanced deep learning techniques to create an accurate and efficient model for recognizing Uzbek sign language, thereby facilitating better integration of hearing-impaired individuals into society. The model is designed to recognize and translate sign language gestures into textual format, enhancing accessibility and communication. The methodology involves several crucial stages: image preprocessing, feature extraction, feature learning, and classification. Initially, video data capturing various Uzbek sign language gestures is converted into image frames. These frames undergo preprocessing, including resizing to a standard dimension and noise removal, to ensure consistency and clarity in the dataset. Feature extraction is performed using techniques such as Gabor filtering, which effectively captures the essential characteristics of the gestures. This is followed by feature learning, where a convolutional neural network (CNN) is employed to learn and identify distinctive features of each sign language gesture.

The proposed model utilizes the ResNet-50 architecture, a deep residual network known for its high performance in image recognition tasks. The model is trained on a comprehensive dataset of images representing various Uzbek sign language gestures. The training process involves fine-tuning the network parameters to optimize the recognition accuracy. The final classification stage uses the learned features to categorize the gestures accurately, converting them into corresponding Uzbek words. Preliminary results demonstrate a high level of accuracy in recognizing and translating Uzbek sign language gestures. The model achieved a remarkable testing accuracy, indicating its potential effectiveness in real-world applications. However, the study also highlights the need for further exploration and improvement, particularly in handling real-time data with complex background variations and ensuring robustness across different signers and environmental conditions.