HCMCTrafficDataset: Enabling Smart Mobility Solutions for Motorcycle-Dense Cities with Limited Infrastructure

Abstract

This paper addresses the critical challenges of Intelligent Transportation Systems (ITS) in developing countries, with particular emphasis on motorcycle-dominated traffic environments prevalent across Southeast Asia. Existing traffic datasets inadequately represent the complex traffic dynamics in these regions, where low-resolution cameras and limited infrastructure constrain monitoring capabilities. We introduce HCMCTrafficDataset, a comprehensive multimodal dataset capturing urban traffic patterns in Ho Chi Minh City, Vietnam, with specialized focus on motorcycle detection and counting. Our contributions include: (1) a novel traffic dataset with graph-structured spatial relationships; (2) optimized baseline models for motorcycle detection using semi-supervised learning; (3) extensive benchmarking of spatial-temporal prediction methods; and (4) analysis of traffic monitoring under resource-constrained conditions. Experimental results demonstrate that graph-based models leveraging spatial dependencies significantly outperform traditional methods, with GNN-based approaches reducing prediction error by up to 23 % compared to conventional time-series models. This dataset enables development of ITS solutions tailored to developing regions, with direct applications in the ASEAN Smart Cities Network and similar initiatives