A Comparative Analysis of CNN and YOLOv8 for Real-Time Traffic Sign Detection

In recent years, real-time object detection has become a critical requirement in fields such as autonomous driving, surveillance, and robotics. Traditional Convolutional Neural Networks (CNNs) often deliver strong accuracy but can struggle with inference speed and the detection of small or overlapping objects. This paper addresses the need for a more robust and efficient solution by comparing a CNN-based model and YOLOv8 on a representative dataset, focusing on metrics such as accuracy, precision, recall, and inference time. Our objective is to determine which architecture best meets the demands of time-sensitive applications without compromising detection performance. We train both models under uniform conditions, employing consistent data augmentation and hyper parameters, and evaluate them in the same testing environment. The results reveal that YOLOv8 not only outperforms the CNN in accuracy and F1-score but also delivers real-time inference at approximately 90 frames per second. This significant improvement in speed and detection quality underscores the suitability of YOLOv8 for applications that require instantaneous object recognition. By highlighting these comparative insights, we offer a clear solution pathway for researchers and industry practitioners seeking an optimal balance between performance and computational cost in real-world detection tasks.