MobileViTv2-MLP: Arabic Traffic Sign Recognition with Enhanced Lightweight Vision Transformers

Traffic Sign Recognition (TSR) plays a pivotal role in advanced driver-assistance systems (ADAS) and autonomous vehicles, serving as a fundamental element for safe and efficient road navigation. Due to a diverse array of factors such as variations in sign appearance, fluctuating lighting conditions, occlusions, and environmental influences, the task of TSR poses significant challenges. These challenges have been addressed with advanced deep learning techniques, particularly with Convolutional Neural Networks (CNNs) and other architectures proficient in handling visual data. While such approaches have shown promise, they are often limited in their ability to effectively manage the aforementioned challenges. To overcome these obstacles, we propose a novel approach called the MobileViTv2-MLP method. This approach aims to develop a resilient model specifically tailored to the task of identifying and categorizing traffic signs encountered during driving. Our methodology involves training the model on an extensive dataset comprising diverse representations of traffic signs under varying conditions, ensuring its adaptability and reliability in real-world scenarios. Through experimentation and evaluation, our refined approach has demonstrated exceptional accuracy, achieving a remarkable 99.75%. This outcome underscores the robustness and effectiveness of our proposed method in addressing the challenges inherent in TSR, thereby contributing to the advancement of intelligent transportation systems.