Vehicle Classification for Traffic Signal Optimization via YOINS Transfer Learning Model

Abstract

Intelligent transportation systems are increasingly being used in urban environments to enhance road safety, reduce congestion, and optimize vehicular movement. Traditional vehicle classification methods often fail to address dynamic traffic conditions, reducing real-time efficiency. The proposed hybrid deep learning-based approach, titled YOINS (YOLOv7 + Inception V3) Deep Learning Model, integrates YOLOv7 for real time vehicle detection and Inception V3 for fine-grained classification into ten classes such as trucks, buses, SUVs, Family sedans, Fire engines, Heavy trucks, Jeeps, Minibuses, Racing cars and Taxis. Experimental results validate the effectiveness of this approach, with Inception V3 achieving an accuracy of 85% and YOLOv7 achieving 81%. When combined, the YOINS hybrid model improves classification accuracy to 87%, demonstrating a slight enhancement over the individual models. Additionally, the dynamic adjustment of green signal timing based on real-time vehicle density has significantly improved traffic flow and congestion management, contributing to smarter urban mobility. The choice of YOLOv7 and Inception V3 over alternatives like EfficientDet and MobileNet is based on their superior balance of speed, accuracy, and scalability in ITS applications. This approach offers a scalable and efficient solution for intelligent transportation management.