Real-time Analytics on Urban Trajectory Data for Road Traffic Management

The School of EECS is hosting the following thesis review 3 seminar:

Real-time Analytics on Urban Trajectory Data for Road Traffic Management

Speaker: Zichun Zhu
Host: Dr Rocky Chen

Abstract: Map-matching is a fundamental process that aligns timestamped location records with the actual routes traversed by moving entities within a road network. This critical task not only enhances trajectory accuracy by correcting imperfect data but also underpins reliable spatial analysis, navigation, and transportation planning applications. Over recent decades, reconstructing travel routes from GPS data has been widely studied, employing approaches that range from geometry-based and topology-based methods to probability-based techniques and, more recently, neural-network-based models.

Unlike GPS trajectories, Bluetooth-based map-matching poses unique challenges. Specifically, neural network-based models often suffer from low efficiency on extensive road networks due to an enormous output space. Furthermore, Bluetooth data exhibit distinct uncertainties compared to GPS signals, and labelled Bluetooth datasets are scarce.

This thesis addresses these challenges by focusing on map-matching for Bluetooth data. The key contributions include (1) the development of a turn-based sequence-to-sequence model that significantly reduces the output space, (2) a map-matching model that leverages contrastive learning to capture the unique patterns inherent in Bluetooth data, and (3) an integrated framework in which an error detection module filters pseudo-labelled datasets to enhance data quality. Collectively, these contributions advance the state-of-the-art in Bluetooth map-matching and improve its applicability to real-world transport scenarios.

Biography: Zichun Zhu received his Master of Computer Science degree from the University of Melbourne, in 2020. He is currently pursuing a Ph.D. degree, focusing on applying deep learning techniques to transport challenges, specifically map-matching on Bluetooth data.