Real-Time Traffic Sign Detection For Autonomous Vehicles Using Improved YOLOv8

In recent years, fatalities in traffic accidents have risen to a concerning amount. Moreover, the main causes of road collisions are speeding and driver behavior. Detection of traffic signs algorithms that is applied to autonomous vehicles can solve the rise of road traffic accidents and highly improve the safety of road drivers. However, in real-life cases, many aspects often come unnoticed such as the distance, small size, and lighting conditions of traffic signs that is captured from the camera, leading to fatal mistakes that might jeopardize the overall detecting ability of traffic signs. Using the German Traffic Sign Recognition Benchmark Dataset, this paper aims for detecting traffic signs in real-time with the proposed RAPID YOLOv8 model that incorporates SPD-Conv, Contextual Attention Mechanism, and Canny Edge Detection to handle poor and low-resolution images while maintaining high accuracy and sustainable speed. Data augmentation is applied prior to training to further increase the effectivity of each model. The Rapid YOLOv8 has a 6.5%, 2.8%, and 15% increase in mAP, precision, and recall, respectively from the original base model of YOLOv8. Hence, indicating a superior and efficient model on detecting traffic signs both on accuracy, speed, and overall detecting small and poor images, resulting in a robust and flexible model.