Development of an Arabic Sign Language Recognition System Utilizing Deep Convolutional Neural Network

The ability to recognize sign language movements, especially Arabic Sign Language (ArSL), is necessary to increase the conversation accessibility for the those who are hard of hearing or deaf. However, there are a number of drawbacks to the current ASLR techniques, such as their dependency on manually created features, lack of datasets, and inefficiencies in real-time applications. This study suggests a novel method that utilize Deep Convolutional Neural Network (DCNN) model designed especially for ASLR applications to overcome these difficulties. The suggested DCNN overcomes the drawbacks of conventional handmade feature-based techniques by automatically extracting discriminative features from unprocessed sign language images using hierarchical feature learning. Preprocessing techniques such as normalization and scaling are applied to improve the 7,857 completely labeled images in the RGB Arabic Alphabet Sign Language (AASL) the data set, which is provided for ASLR model evaluation and training. The results of the study demonstrate how successful the recommended DCNN model is, as evidenced by its astounding 99% accuracy on the AASL dataset. Comparison with similar approaches like VGG-16, ResNet-18, and other Efficient Net based architecture, shows the accuracy to which the proposed method can classify Arabic sign language alphabets. The findings presented evidence for how the chosen architecture for DCNNs can significantly enhance ASLR improvement, open up opportunities for improving communication interaction with Deaf and hard-of-hearing people and their integration. In doing the proposed work, we use Python.