Abstract
The increasing reliance on autonomous navigation systems necessitates robust methods for detecting and recognizing textual information in natural scenes, especially in complex scripts like Arabic. This paper presents a novel attention-based unified architecture for Arabic text detection and recognition on traffic panels, addressing the unique challenges posed by Arabic's cursive nature, varying character shapes, and contextual dependencies. Leveraging the ASAYAR dataset, which includes diverse Arabic text samples with precise annotations, the proposed model integrates Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks with attention mechanisms to accurately localize and interpret text regions. The architecture demonstrates state-of-the-art performance, achieving a Mean Intersection over Union (IoU) of 0.9505, precision of 0.953, recall of 0.934, F1-score of 0.929, and an overall recognition accuracy of 97%. Visualizations of attention weights and SHAP analyses highlight the model's explainability and focus on relevant features, ensuring reliability in real-world applications. Furthermore, the system's computational efficiency and real-time applicability make it suitable for use in Advanced Driver Assistance Systems (ADAS) and autonomous vehicles, reducing driver distractions and enhancing traffic safety. This study not only advances Arabic text recognition research but also provides insights into developing scalable, multilingual text detection systems for complex real-world scenarios.