Temporal-to-Spatial Transmutation for Enhancing Arabic Sign Language Recognition via ResNet

Sign Language Recognition (SLR) aims to overcome communication barriers for individuals with hearing impairments, enabling smooth interaction, easy access to information, and better social integration. Most dynamic SLR systems are heavily relying on sequential models such as Long Short Term Memory (LSTM). However, these models suffer from high computational cost and heavy training time. In this paper, we propose a novel approach that transforms the temporal characteristics of sign language data into a spatial representation while preserving all essential features. Instead of using sequential models, we leverage a ResNet-based feature extraction method, which enhances spatiotemporal modeling and improves recognition accuracy while significantly reducing model complexity and training time. Experiments on the ArabSign dataset reveal that the proposed model achieves 91.3% of accuracy, outperforming our previous LSTM-based approach. This work highlights the advantages of convolutional feature extraction for dynamic Arabic Sign Language Recognition and presents a more effective and scalable alternative to recurrent models.