Real-time Vehicle Recognition System Using You Only Look Once Model with Uniform Experimental Design

Abstract

As urban areas develop and technology advances, artificial intelligence technologies offer numerous applications to alleviate a pressing issue: traffic congestion. The importance of traffic management and safety monitoring underscores the crucial role of integrating vehicle recognition technology with the Internet of Things within intelligent transportation systems. In this study, You Only Look Once with Uniform Experimental Design (U-YOLOv4) is proposed to enhance the performance of vehicle recognition. The approach aims to optimize hyperparameters within YOLOv4, resulting in a high recognition rate in the model. Furthermore, two datasets were utilized: Vehicle from Beijing Institute of Technology (BIT) and Computational Intelligence Application Laboratory from National Chin-Yi University of Technology (CIA-NCUT). The experimental results revealed significant improvements when comparing U-YOLOv4 to YOLOv4. In the BIT-Vehicle dataset, U-YOLOv4 achieved a mean average precision of 97.84%, whereas in the CIA-NCUT dataset, it reached 89.19%, highlighting its superior performance over YOLOv4. The U-YOLOv4 model has overall demonstrated significant improvement in vehicle recognition, revealing its adaptability across different datasets and various scenarios. Its application is expected to play a crucial role in intelligent transportation systems, enhancing traffic management efficiency and road safety.