Design and Development of EfficientNet for Traffic Sign Recognition

Lee, Yu Li (2024) Design and Development of EfficientNet for Traffic Sign Recognition. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text LEE YU LI_Full Text.pdf Restricted to Registered users only Download (3MB)

Abstract

This dissertation investigates the adaptation and optimization of EfficientNet, a state-ofthe- art deep learning architecture, for Traffic Sign Recognition (TSR) systems embedded within Autonomous Driving or Advanced Driver Assistance Systems (ADAS). Given the critical role of traffic sign recognition in enhancing road safety and driving efficiency, this study explores the potential of EfficientNet due to its superior performance on image classification tasks. The research methodology involved redesigning the baseline EfficientNet-B0 model to enhance its feature extraction capabilities specifically for TSR. Experimentswere conducted using three diverse traffic sign datasets: GTSRB (German Traffic Sign Recognition Benchmark), CTS (China Traffic Sign dataset), andMTS(Malaysia Traffic Sign dataset). These datasets presented unique challenges in terms of variability in sign appearance and environmental conditions. The performance of the optimised EfficientNet was evaluated against traditional convolutional neural network (CNN) models through rigorous testing, focusing on accuracy, and computational efficiency. The outcomes demonstrated that EfficientNet, with specific enhancements, could outperform existing models in recognizing traffic signs accurately and swiftly, suggesting its high applicability in real-world scenarios