Automated Vehicle Classification for Intelligent Toll Gate Systems Using Convolutional Neural Networks
TL;DR
The experimental results indicate that the ConvNet architecture achieved the most stable classification performance, reaching a validation accuracy of up to 1.000 with lower validation loss fluctuations compared to AlexNet and ResNet under constrained dataset conditions.
Abstract
Vehicle classification at toll gates plays an important role in determining toll tariffs accurately and supporting efficient transportation management. However, manual vehicle classification remains prone to operational inefficiency, transaction delays, and human error, particularly under high traffic conditions. This study aims to comparatively evaluate the performance of AlexNet, ResNet, and a lightweight custom ConvNet architecture developed for this study using limited CCTV image datasets collected from a real-world toll gate environment in Indonesia. The dataset consists of five vehicle categories based on Indonesian toll road standards. During preprocessing, RGB images with an initial resolution of 159×127 pixels were resized to 224×224 pixels, normalized, and divided into training and validation sets using an 80:20 ratio. Model performance was evaluated using accuracy and loss metrics during training and validation. The experimental results indicate that the ConvNet architecture achieved the most stable classification performance, reaching a validation accuracy of up to 1.000 with lower validation loss fluctuations compared to AlexNet and ResNet under constrained dataset conditions. Nevertheless, the reported performance should be interpreted cautiously because the experiments were conducted using a limited dataset without external cross-location validation. These findings demonstrate the potential applicability of CNN-based vehicle classification for toll gate systems; however, broader deployment requires larger datasets and more diverse real-world validation scenarios.