Author: Lan, Tian
Title: Automated generation of labelled schematic network maps
Advisors: Li, Zhilin (LSGI)
Degree: Ph.D.
Year: 2020
Subject: Cartography -- Data processing
Digital mapping
Hong Kong Polytechnic University -- Dissertations
Department: Department of Land Surveying and Geo-Informatics
Pages: xiv, 127 pages : color illustrations
Language: English
Abstract: Schematic network maps (often simply called schematic maps) are simplified representations of line networks. On such maps, the true geometry of line networks is distorted while topological relationships are preserved. One famous example of schematic maps is the London Underground map designed by Harry Charles Beck in the 1930s, where congested areas are enlarged, and lines are re-orientated along octilinear (i.e. horizontal, vertical, and diagonal) directions with the preservation of topological relationships. Such a design improves the clarity and readability of the map, allowing people to quickly and accurately perform route planning and orientation tasks. Schematic maps today have been widely used to represent various spatial networks, as well as some non-spatial networks. Schematic maps are currently generated by a computer-aided approach that not only requires skilled map designers but also is time-intensive. In the past two decades, researchers have been working toward the automated generation of schematic maps. However, two main problems still exist. Firstly, schematized network layouts are not as pleasing as those created by the computer-aided approach, even when various sets of constraints are used. This is because most of these constraints govern the geometric properties of individual features; only a few constraints address the relationships between these features; and in particular, none explicitly deal with the main structure of an entire network. However, it has been more recently noted that the preservation of the main structures is most important, while the preservation of relative relations is helpful. The two together have been used to form a set of "general principles for schematization". Secondly, automated labelling (i.e. the placement of name labels) has not been well considered. Although some studies have tried to solve this problem, the labelling rules used in them are too simple and unsuitable for schematic maps. This project aims to tackle both of these problems. To solve the first problem, we propose the following strategy: modelling the "general principles for schematization" into constraints, and then integrating these modelled constraints into optimization models. More concretely, this set of general principles are first modelled as detailed constraints (e.g. "the preservation of the main structures" modelled as "making major lines as straight as possible") before they are integrated into mixed-integer programming (MIP) to optimize network layouts. Experimental evaluation for this new method (here called the GP-MIP-based method) is conducted. Benchmarks are Beijing metro and high-speed railway network layouts generated using an existing MIP-based method. From these experimental results, it is found that, in term of "fractal dimension", the complexity of network layouts has been simplified from 1.25 to 1.14 for the high-speed railway network, and from 1.38 to 1.29 for the Beijing metro. The average usability scores in terms of "clarity", "recognition of major lines", "visual simplicity" and "satisfaction" have also been improved by 16.2%, 17.4%, 14.7% and 37.5% respectively. These results indicate that the GP-MIP-based method can generate network layouts with improved clarity and aesthetics.
We have also proposed a strategy to deal with the second problem, namely, acquiring the rules from some existing maps and then integrating them into a greedy optimization algorithm for labelling. Schematic maps of the top 20 longest metro systems in the world are employed to acquire labelling rules, including the potential positions of name labels and the preferences of these positions. The former is determined by the consideration of touching edges, and the latter by "descriptive statistics" (DS). In order to optimize labelling with the acquired rules, an algorithm is developed based on a greedy optimization strategy. Experimental evaluation for this new method (here called the DS-based method) has been conducted with the official metro maps of Hong Kong and Tianjin as its benchmarks. Results show that, in terms of both "uncrowded level among name labels" and "ease level of finding associated name labels of stations", the average scores obtained by this method are consistently similar to those obtained from official maps (2.96 versus 2.89 at Tianjin, and 3.56 versus 3.56 at Hong Kong for the former, and 2.78 versus 2.74 at Tianjin and 3.30 versus 3.34 at Hong Kong for the latter). These results indicate that the name labels placed by DS-based method are as pleasing as those of official schematic maps. A limitation of this DS-based labelling method is its lack of "learn" ability. That is, if certain situations do not exist on the sample maps used for rule acquisition, the DS-based method may not place the name labels under them properly. To solve this problem, a strategy based on artificial neural networks (ANNs) is proposed. In this strategy, samples are first prepared, and then used to train and test the ANN models. In each sample, four types of attributes (connection relations, label length, line directions, and point coordinates) are used as inputs, and two types of attributes (positions and directions of labels) are used as outputs. To evaluate this ANN-based method, two experiments have been conducted. Experiment 1 found that, when compared with the DS-based method, the ANN-based method yields slightly larger average scores for the Tianjin metro (i.e. 3.64 versus 3.51, 3.56 versus 3.36 and 3.73 versus 3.41), but slightly smaller ones for the Hong Kong metro (i.e. 3.80 versus 4.05, 3.53 versus 4.03 and 3.76 versus 3.85), in terms of the ease level, uncrowded level, and satisfaction level. Experiment 2 found that, among the unlearned stations of the Nanjing metro, the ANN-based method placed all the name labels into positions without overlaps, while the DS-based method did overlap some of the name labels. These results imply that the ANN-based labelling method can not only effectively place name labels (especially for complex network layouts) but also has ability of "learn" to infer relative positions of name labels for unlearned stations. In summary, this project has developed a GP-MIP-based method for automated schematization and two methods (DS-based and ANN-based methods) for automated labelling. These methods are proven to be both effective and satisfactory. To achieve the complete automation, automated generation of other components should also be considered, and further work will be conducted in this area.
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
5520.pdfFor All Users8.42 MBAdobe PDFView/Open


Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: https://theses.lib.polyu.edu.hk/handle/200/11061