Towards Better Perception of Urban Information: A Visualization Perspective

PhD Thesis Proposal Defence


Title: "Towards Better Perception of Urban Information: A Visualization
Perspective"

by

Mr. Qiaomu SHEN


Abstract:

Rapid urbanization has become one of the most important global trends in the 
last 50 years. Although half of the world’s population live in urban areas and 
contribute to 80 percent of the world’s GDP, the ever more crowded urban areas 
result in a series of problems, such as traffic congestion, pollution, 
insufficient resources, and unbalanced urban infrastructure.  Fortunately, the 
development of sensing technology has made data collection and processing 
easier and cheaper, thus providing an opportunity for people to understand the 
phenomenon or even determine the solutions to address these problems. However, 
due to the high dimensionality, heterogeneity of the dataset, and the complex 
analytical tasks, the pure automated techniques are insufficient in the 
exploration of urban information. On the other hand, the human analysts with 
sharp perception and domain expertise cannot deal with large volumes of data 
without powerful tools. Visualization bridges the gap between analysts and 
automated techniques, and it has been widely applied in the exploration of 
urban information.

In this proposal, we introduce several novel visual analytics techniques that 
cover the three domains in urban information exploration: place, people, and 
technology. In the first work, we propose StreetVizor, a visual analytics 
system that helps urban planners to explore fine-scale living environments. The 
system automatically extracts the features of human-scale urban form from 
street view images through machine learning techniques. Then, a comprehensive 
analysis framework and novel visual designs are proposed to support free 
exploration from multiple levels. In the second work, we target the 
visualization of massive human movement data. We propose route-aware edge 
bundling, which visualizes the overview of origin-destination trails. By 
introducing the additional graph structure as constraints, the trail bundles 
can follow the traffic network in the city. In the last work, we focus on the 
model interpretation in the application of air pollutant forecast. We propose 
MultiRNNExplorer, which visualizes the recurrent neuron network behaviors in 
multi-dimensional time-series forecast. To validate the effectiveness of the 
proposed techniques, we conduct several case studies based on real-world 
datasets and domain expert interviews.


Date:			Thursday, 31 October 2019

Time:                  	3:00pm - 5:00pm

Venue:                  Room CYTG001
                         lifts 35/36

Committee Members:	Prof. Huamin Qu (Supervisor)
 			Dr. Qiong Luo (Chairperson)
 			Dr. Xiaojuan Ma
 			Dr. Pedro Sander


**** ALL are Welcome ****