TOWARDS A BETTER UNDERSTANDING OF URBAN ENVIRONMENT: A VISUAL ANALYTICS PERSPECTIVE

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "TOWARDS A BETTER UNDERSTANDING OF URBAN ENVIRONMENT: A VISUAL 
ANALYTICS PERSPECTIVE"

By

Mr. Zezheng FENG


Abstract

Thanks to the open data campaign, urban data are becoming increasingly 
ubiquitous, which provides vast opportunities for addressing challenging 
issues through big data analysis. Furthermore, most of the urban data 
exhibits spatial and temporal features that are widely used in solving 
severe urban problems, including urban planning and traffic management. 
However, when urban analysts make decisions on solving such issues, they 
may be limited by an unintuitive representation for understanding the 
urban environment, needing more interpretation of the data-driven model. 
Therefore, understanding the urban data so as to understand the urban 
environment and then making proper decisions is one of the goals of urban 
analysts. The visual analytics approach has recently been widely used to 
assist analysts in solving urban problems with its intuitiveness, 
interactivity, and interpretability. Aiming to help analysts better 
understand the urban environment, this thesis proposed visual analytics 
approaches from three aspects respectively:

1. Representing and analyzing the urban movement data: The moving objects 
in the urban environment are usually limited by the complex urban 
infrastructure (i.e., the road network); that is, the two geographically 
distant locations may have a short commute time due to road connectivity. 
Therefore, visualizing and analyzing the urban movement data cannot be 
limited to measuring the distance between two locations by Euclidean 
distance. To this end, a new method named Topology Density Map is 
proposed, targeting accurate and intuitive density maps in the context of 
the urban environment.

2. Providing transparency to the urban traffic prediction models: The 
spatial and temporal features of urban movement data usually play an 
important role in traffic flow prediction. However, most prediction models 
are indeed “black boxes" which lack transparency. To convince the users 
when predicting the traffic and revealing the impact on the traffic from 
surroundings, a visual analytics approach named TrafPS is proposed, which 
includes two measurements named region SHAP and trajectory SHAP. Moreover, 
a visual analytics interface is developed to support the users in 
multi-level analysis for understanding traffic prediction and making 
decisions.

3. Modeling and exploring the urban higher-order movement: Higher-order 
patterns in the urban environment reveal sequential multi-step state 
transitions, usually superior to origin-destination (OD) analysis, which 
depicts only first-order geospatial movement patterns. However, 
conventional DAG-based movement modeling usually exhibits sparseness and 
ignores considering the temporal variants that are critical for movements 
in urban environments. Therefore, HoLens is firstly proposed for modeling 
and visualizing higher-order movement patterns in the context of an urban 
environment.

This thesis contributes to the VIS community with novel urban movement 
data processing algorithms, visualization techniques, and interactive 
visual analytics tools. Case studies with real-world dataset and 
interviews with domain experts are conducted to demonstrate the 
feasibility, usability, and effectiveness of the proposed three studies in 
this thesis.


Date:			Friday, 13 January 2023

Time:			10:00am - 12:00noon

Venue:			Room 3494
 			lifts 25/26

Chairperson:		Prof. Masaru YARIME (PPOL)

Committee Members:	Prof. Huamin QU (Supervisor)
 			Prof. Siu Wing CHENG
 			Prof. Tristan BRAUD
 			Prof. Fan ZHANG (CIVL)
 			Prof. Yun JANG (Sejong University)


**** ALL are Welcome ****