(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 10, No. 4, 2019

Using Space Syntax and Information Visualization for

Spatial Behavior Analysis and Simulation
Sheng-Ming Wang1
Chieh-Ju Huang2
Department of Interaction Design
Doctoral Program in Design
National Taipei University of Technology
National Taipei University of Technology
Taipei, Taiwan
Taipei, Taiwan

Abstract—This study used space syntax to discuss user route distribution to provide recommendations on transport and
movement dynamics and crowded hot spots in a commercial spatial planning as well as to outline a design method for future
area. Moreover, it developed personas according to its onsite integration between spaces and services in commercial areas.
observations, visualized user movement data, and performed a This study is to discuss the predictions of user spatial behavior
deep-learning simulation using the generative adversarial and pedestrian movement simulations within the established
network (GAN) to simulate user movement in an urban business area. For the movement of hanging around and the
commercial area as well as the influences such move might consumption movement except walking are not discussed in
engender. From a pedestrian perspective, this study examined the this study.
crowd behavior in a commercial area, conducted an onsite
observation of people’s spatial behaviors, and simulated user In the following of this paper, we extracted the diverse
movement through data-science-driven approaches. Through the elements of a city by space syntax, observed and collected
analysis process, we determined the spatial differences among pedestrians’ movement and behaviors from different groups in
various roads and districts in the commercial area, and according a commercial area, segmented personas, and then used a deep-
to the user movement simulation, we identified key factors that leaning GAN to integrate and analyze collected data and
influence pedestrian spatial behaviors and pedestrian perform simulations. Then we discussed how data-science-
accessibility. Moreover, we used the deformed wheel theory to driven concepts and data visualization could be used in the
investigate the spatial structure of the commercial area and the
establishment of user movement and environmental behaviors
synergetic relationship between the space and pedestrians;
in urban commercial areas. In the end of this paper, we tried to
deformed wheel theory presents the user flow differences in
various places and the complexity of road distribution, thereby make the recommendations for further study.
enabling relevant parties to develop design plans that integrate II. LITERATURE REVIEW
space and service provision in commercial areas. This research
contributes to the interdisciplinary study of spatial behavior A. Space Syntax and Deformed Wheel Theory
analysis and simulation with machine learning applications.
Urban space planning, which is focused on the relationship
Keywords—Spatial behavior; space syntax; information between spaces and people, may be analyzed using space
visualization; generative adversarial network (GAN); user syntax. Space syntax involves factors such as the main
movement entrances and exits of cities, means of transport for cross-
regional travel, and the structure of public spaces. Factors in
I. INTRODUCTION the discussion of urban space planning include land use, urban
The spatial movement of people has been presented in functions, and the spatial behaviors of people in an area [1].
various forms in different fields. Approaches to demonstrating Hillier and Han [2] introduced the space syntax theory in “The
and discussing movement results are determined by the Social Logic of Space”; it suggests that the movement scale of
relationships perceived by researchers between people and user activities and spatial distribution can be divided into the
spaces. In addition to the quantitative analysis of spatial data, following basic elements: axial, convex, and isovist spaces.
this study discussed from a pedestrian perspective the Space syntax emphasizes the patterns of pedestrian activities in
relationship between crowds and spaces in commercial areas as a city, as well as relevant spatial measurements and
well as observed the spatial behaviors of pedestrians onsite. relationships between spaces and social lives [3]. Space syntax
Underpinned by a data-science-driven approach, this study presents graphic spatial data that are then used to discuss
simulated pedestrian movement in different scenarios. Through environmental factors influencing user’ spatial behaviors and
a simulation, this study determined the spatial differences of experience. Such quantified data reveal the relationships
user movements in local and global environments in a among urban roads, blocks, and buildings [4]. Accordingly,
commercial area, identified the key factors affecting pedestrian space syntax provides attribute information that can be used in
spatial behaviors, and discerned pedestrian accessibility within designs for solving or alleviating problems that exist in urban
the space. Using a deformed wheel theory, we investigated spaces [5].
spatial structures and their synergistic relationship with Taiwanese studies using space syntax are predominantly
pedestrians. Moreover, we examined the differences of user related to urban planning and geographic environments; space
movements in various areas and the complexity of space and syntax can be used to provide comprehensive and quantitative

510 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

analyses and recommendations for urban and route planning

improvement in business activities. Therefore, this study used
space syntax as a tool to analyze spatial behaviors and user
movement, and then give service design suggestions in the
future business environment.
The operation of an urban system can be interpreted using
syntactic relation systems, such as symmetry–asymmetry,
integration–segregation, the determination of control values,
Axial space Convex space Isovist space
integration of features and control cores, and intelligibility and
predictability [6] [7] [8] [9]. The fundamental concept of space Fig 1. Models of Space Syntax’s Three Spatial Elements [20a].
syntax is that urban areas can all be represented using matrices
of connected spaces; such matrices can be simulated and Space syntax transforms physical spaces into topological
calculated using a computer and reveal space attributes using structures that comprise nodes and lines. Fig. 2 depicts the
data [10]. The nature of space. syntax is to show the interaction process of using quantitative indicators to assess the structural
among and presentation of spatial patterns, user behaviors, and characteristics of global or local spaces. These quantitative data
cultures [11]. Relationships between the spatial patterns of are generated using the location of each spatial element and the
human settlements and activities are presented using business connections between them in a system. Connected axial lines
models, social structures, and ideologies [12]. Space syntax can intersect at a node, which reveals the relationships among
predict the most crowded destinations, which should have the structural elements in a spatial system and enables
most highly integrated roads placed adjacent to them [13]. investigation of the relative depth from one structural element
to another.
Space syntax is a digital tool that facilitates data
quantification and provides spatial measurements of personal
space and overall layout. It measures the connectivity between
spaces as well as a space’s connectivity to the overall
environment, and furthermore, it calculates the depth and
spatial integration of a space. Depth is defined as the minimum
steps required to reach a space from another, and connectivity
refers to the number of spaces that a space is immediately
connected to [14]. (a)Road map (b)Axial map (c)Topological map
In quantitative studies, space syntax presents the spatial Fig 2. Spatial Transformation of Space Syntax [20b].
layout through topological distance and provides dynamic
locations and distribution of people. More crucially, it reveals TABLE I. USER SEGMENTS EQUATIONS IN SPACE SYNTAX [21]
the walkability as well as pedestrians’ movement and route M distance of
choices [15]. In space assessment, space syntax has three 1 𝑛−1 the i-th axis
𝑀𝐷𝑖 = ∑𝑗=1 𝑑𝑖𝑗 , 𝑖 ≠ 𝑗 from all the
𝑛−1
indicators of distance [16] [17] [18]: other n – 1.
Basic centrality measure: it accounts for the distance
 Topological distance: the route with the fewest number Mean Depth between each axis and all the others, with the shallowest
of turns. axis being the closest to all the others and the deepest
being the farthest one. In a city with a perfectly circular
 Geometric distance: the route with the smallest angle axial map, the shallowest axis would be close to the center
and the deepest would be on the perimeter.
change. Number of axes connected to the i-
𝐶𝑖 = 𝑘𝑖
 Metric distance: the route with the shortest length. Connectivity
th axis.
Measures how much an axis is directly connected to the
others. Axes with high Ci make it easier to pass through
According to the theory of topological structures, space the areas.
syntax contains three basic spatial elements for discussing the 𝑅𝐴𝑖 =
2(𝑀𝐷𝑖 −1)
Normalize
relationships among buildings, physical urban spaces, and their 𝑛−2
𝑛+2 d value of
functions for activities ([19] Fig. 1): 𝐷
2 ,𝑛 *log 2 (
3
) − 1+ + 1- MDi being
𝑅𝐴𝑁 = min(MDi)
(𝑛 − 1)(𝑛 − 2)
 Convex space: a space where any two of its points are Global 𝑅𝐴𝑁𝐷
=1 and
Max(MDi)
intervisible points. Integration
RN =
𝑅𝐴𝑖 = n/2.
RA expresses the centrality of an axis comparing its actual
 Axial space: a space formed by straight lines, such as Mean Depth with the theoretical highest and lowest values
roads. that Mean Depth could have in the given graph. Compared
to Mean Depth alone, Relativized Asymmetry is a
 Isovist space: the total space that can be viewed from normalization between 0 and 1.
𝑘𝑖 Sum of the inverse values
any point, in which the relationships between each 1 of Ci for the ki axes con-
connecting spatial element can also be viewed. 𝐶𝑇𝑅𝐿𝑖 = ∑
𝐶𝑗
nected to the i-th axis.
Control Value 𝑗=1

Sum of the inverse values of Ci for the ki axes connected

to the i-th axis.

511 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

Table I summarizes the Space Syntax measures analyzed in characteristics of a road network. By overlaying pedestrian
this research. walking distribution on simulated traffic, conflicts between
user and traffic can be identified and predicted on certain
 Mean Depth (MD): This is a global indicator that routes [22]. Accordingly, a quantitative analysis of space
measures the mean shortest route of each spatial syntax provides insight into users’ spatial behaviors as well as
element to other elements in a system through the reveals connectivity among spaces through topological
relative depth between axial lines. structures derived from axial maps, which also enables
 Number of connections (Cn): This is a local indicator discerning the structures of movement networks. Thus, space
showing the number of elements adjacent to a syntax can be used to investigate how the structure of a spatial
particular element; a high CN value (a high number of network affects people’s movement in an urban environment or
immediately connected elements) indicates high an indoor space.
accessibility and high publicness of that element. To simulate and visualize user movement in commercial
 Global Integration (Rn): This is a global indicator; a areas, this study employed a deep-learning generative
higher RN value of an element indicates higher global adversarial network (GAN) to simulate a large amount of data
integration of that element, namely it being more using user movement data collected during our onsite
integrated in a spatial system. Specifically, this means observations. The study compared visualizations of the
that pedestrians can reach other roads in fewer steps simulated data with space syntax results, discussed the
from such an element. similarities and differences between the two methods, and
predicted future environmental behaviors of users.
 Control Value (CV): This is a local indicator revealing
the extent to which adjacent elements control each Deep learning is a type of machine learning in artificial
other. A high number of elements adjacent to a intelligence. Machine learning involves training large amounts
particular result in high CVs of that element. of data through algorithms to establish mathematical models
that facilitate data prediction. Using artificial neural networks
Deformed wheel theory in space syntax reveals the original (ANNs), which simulate the mechanism of human neural
structures of spaces that show the synergic relationships among networks, deep learning increases the computational efficiency
the spaces. Deformed wheel is a semigrid or hub of lines near of central processing units and graphic processing units in
the central line that reveals a space’s relevant information and computers and has developed rapidly with the increased
characteristics. The spokes of a deformed wheel connect the prevalence of parallel computing. Data analysis of artificial
center to the edge and form an enclosed space. The formation intelligence is underpinned by data collection and algorithms
of a space comprises the following six steps: that simulate user analysis patterns. Specifically, artificial
intelligence entails a deep-learning mechanism fulfilled by
 Place interaction: From the perspective of nodes and
computing.
lines, a higher number and longer duration of user
activities in a node indicate a lower number and shorter The mechanism of a GAN comprises two deep-learning
duration of people staying in the lines. By contrast, a ANNs, namely a generator and a discriminator. The mutual
higher number and longer duration of people staying in learning between the two networks enables the GAN to
the lines indicate fewer and shorter durations of user simulate data that are comparable to the actual data [23]. Fig. 3
activities in the node. presents the mechanisms of a GAN. First, the generator
generates random data and the discriminator uses the actual
 Place identity: Place identity is related to people in a data to determine whether the generated data are flawed;
place. People coexist with places, and place identity subsequently, the generator corrects its network model
emphasizes user in a place. accordingly. Through mutual learning and optimization of the
 Place release: This refers to unexpected and accidental model, the generator will be able to generate data that are
events that elicit user enthusiasm and interests. highly comparable to the actual data. Therefore, when large
amounts of simulated data are required under circumstances
 Place realization: This refers to distinct events in a where few actual data are available, this algorithm may be
place that underlie its historical or cultural meanings. employed.
 Place creation: This reveals the real needs of a place, This study employed a GAN because such a network can,
such as its commercial or environmental using the existing actual data, simulate an unlimited amount of
characteristics. data that are similar to the actual data. Therefore, when large
amounts of data are required yet only limited sample data are
 Place intensification: This denotes a place’s features,
available, a GAN might be used to simulate logical data, as
which are formed through its design, policies, and city
opposed to randomly simulated data. Thus, a GAN is useful for
structure.
simulation analyses and prediction.
B. Simulation and Visualization of User Movement
Pedestrian movement is predominantly presented using
spatial configuration variables, which represent the

512 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

Fig 3. Mechanism of a GAN.

C. Research Site
This study focused on the Fengjia night market commercial
area (the basic cartographic data are shown in Fig. 4), which is
adjacent to Fengjia University and near downtown Taichung;
thus, it has become a major tourist destination in Taichung.
This commercial area plays a crucial role in the development
of Taichung city and is often considered a highlight of a trip to
central Taiwan. Fengjia night market, a combination of local
culture and geography, has established connections with
adjacent areas over more than 40 years of evolution; such an
interaction between business and the university has created a
unique marketing ecosystem in the region.
From the perspective of city-image design, the basic
Fig 4. Map of the Fengjia Commercial Area.
elements of a city image are edges, paths, nodes, districts, and
landmarks [24]. Moreover, urban user activities start and grow
This study compiled the locations of nodes and road
outward from these elements. Therefore, this study defined the
sections and finalized 67 paths in the commercial area; the
spatial structure and elements of the Fengjia commercial area
commercial area was separated into food, commercial, and
from a city-image perspective. We conducted onsite
residential districts, which were represented by different colors.
observations to outline the commercial area’s spatial structure.
The landmarks were determined as McDonald’s, Fengjia
According to the observation results (Fig. 4), the section of
University, and MOS Burger. To provide a clear demonstration
Wenhua Road located at the upper-left corner (near MOS
of this area, the map is oriented horizontally with a north
Burger) is divided into two parts by Fengjia University. We
reference provided.
observed that the customer flow split when customers reach the
dividing point, and incorporated this observation into our III. RESEARCH METHODOLOGY
analysis and discussion. In Fig. 4, the major districts of
commercial activities are distinguished from other regions; the A. Research Framework
borders represent the edges of Fengjia commercial area, According to our literature review, people are the
namely the study’s research site; and each line represents a path foundational elements in the formation of cities; therefore,
in this district. Visitors usually entered the commercial area research that improves the understanding of and simulates
through an entrance near McDonald’s (Fengjia Road) or one user-behavior patterns and interactive experiences in urban
near the food district on Wenhua Road, which is permits only spaces can contribute to the development and improvement of
pedestrian entry at night, whereas students of Fengjia cities and commercial areas as well as improve related
University usually entered the night market through the information services. Therefore, this study focused on these
university’s main entranceway (Fengjia Road). The central topics and proposed the research framework in Fig. 5. This
road of the commercial area is constituted by a section of study proposed an approach to extract the diverse elements of a
Fengjia Road between McDonald’s and the university’s main city, observed and collected the movement and behaviors of
entranceway, and the intersections between the routes are the user from different groups in a space, segmented personas, and
nodes where activities occur. The landmarks (represented by used a deep-leaning GAN to integrate and analyze collected
triangles) in this commercial area were the two ends of the data and perform simulations. Moreover, this study discussed
Fengjia Road section, namely McDonald’s and the university’s how data-science-driven concepts and data visualization could
main entranceway, and another university entranceway near be used in the establishment of user movement and
MOS Burger. The three landmarks served as the starting points environmental behaviors in urban commercial areas.
in our analyses.

513 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

1
∆𝜃𝐷 ∑𝑚
𝑖=1 *log (1 − 𝐷(𝐺(𝑧𝑖 )))+ (3)
𝑚

4) GAN algorithm:
Input: (Source, Destination), Training dataset
For k = 1 to I do
Draw m noise samples from pg(z)
Draw m real examples from pd(z)
Update the discriminator D based on (2)
Draw m noise samples from pg(z)
Update the discriminator D based on (3)
End for
This study visualized the large amount of data simulated
from our experiment and analyses to identify differences
between the simulated data and onsite observation results. Such
data visualization facilitates interactions for user and reveals
relevant information such as visitor backgrounds, purpose of
visit, route taken, duration of stay, crowded areas, and
distribution of business infrastructure. Moreover, we
incorporated scenario simulation into the data visualization and
Fig 5. Research Framework.
used personas to simulate the spatial experiences for user with
B. Research Methodology and Assessment various needs and develop a decision support system.
This study focused on how user movement and a deformed IV. RESEARCH RESULTS AND DISCUSSION
wheel interacted differently in the visualized presentations of
data yielded using space syntax and data simulation methods. A. Basic Analysis of Space Syntax
Therefore, this study first observed user movement and We labeled all the roads presented in Fig. 4 (see Fig. 6) and
behaviors in the research site, set research objectives, and used a topological map to analyze the relationships among
obtained an overview of the research site using space syntax. them (Fig. 7). According to the topological map, the Fengjia
Subsequently, this study used relevant software to analyze the commercial area is divided into two main subareas by sections
axial lines and predict user movement in the research site. We 1 and 2 of Fengjia Road. At deeper spatial layers are the food,
defined personas according to our onsite observations and used commercial, and residential districts to the north, whereas a
these personas in space syntax for axial and isovist analyses of food–commercial hybrid district to the south is at a shallower
the research site, the results of which were used to predict hot layer because it has fewer roads and connections. Table II
spots where people predominantly gathered. From the compiles locations of the commercial area’s city-image
deformed wheel perspective, the persona analysis results were elements, such as districts, landmarks, edges, and paths.
used to discuss the dynamics of crowd behavior in different
districts, which enabled us to identify crucial elements of
service design as well as assessments. Finally, movement data
of people with defined personas were inputted into simulation
software to produce a large amount of user movement data.
Subsequently, this study visualized these data and compared
the visualization results with those generated using the
deformed wheel theory; finally, it provided recommendations
for service design and assessment methods that could be
adopted in future research.
Mohammadi et al. [24] proposed the following GAN Fig 6. Road Labels in the Fengjia Commercial Area.
algorithm:
1) The discriminator (D) and generator (G) perform
minimization and maximization algorithms simultaneously:
𝑉(𝐷, 𝐺) = 𝔼𝑥~𝑝𝑑𝑎𝑡𝑎(𝑥)[log 𝐷(𝑥)] +
minＧ maxＤ (1)
𝔼𝑧~𝑝𝑧(𝑧)[𝑙𝑜𝑔(1 − 𝐷(𝐺(𝑧)))]
2) The weight of the discriminator is updated using the
following equation:
1
∆𝜃𝐷 ∑𝑚
𝑖=1 *log 𝐷(𝑥𝑖 ) + log (1 − 𝐷(𝐺(𝑧𝑖 )))+ (2)
𝑚

3) The weight of the generator is updated using the

Fig 7. Topological Map of Roads in the Fengjia Commercial Area.
following equation:

514 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

TABLE II. DISTRIBUTION OF CITY-IMAGE ELEMENTS IN THE FENGJIA The mean depth was 6.152, mean global integration was
COMMERCIAL AREA
0.723, and mean number of connections was 3.582. According
to Fig. 8, RN and CN show synergy with a weak positive
District Landmark Edge
correlation (R2 = 0.1322); a large R2 (maximum = 1) implies
that which user are more able to understand the spatial
Food Area A 33, 34, 35, 36,
(Nearby MOS 37, 38, 39, 40, 39, 40 distribution within an area. Furthermore, high global
Burger) 66 integration would indicate a high number of connections;
however, the positive correlation between them was weak and
Food Area B 5, 6, 32,
(Food Street)
5, 6, 32, 31, 30,
31, 30, minor differences between the number of connections and RN
was observed. The number of connections did not directly
Mcdonald’s：1, 3, 67 39, 5, 3, 7,
Fang Chia University 10, 11, influence RN. Instead, it was because the Fengjia commercial
Commercial 1~29, 38, area includes only short sections of roads and has a small area,
Area 41~53, 65~67 ：2, 5, 39 23~29, 41,
MOS Burger：48, 49, 42, 48, 65, thereby leading to a predominantly high RN.
54 67
54, 57, The global integration values of roads are visualized in
Living Area 49~62 Fig. 9, where a red line represents a high RN and a blue line
59~62,
represents a low RN. In addition, Fig. 9 depicts the distribution
TABLE III. SPACE SYNTAX MEASUREMENTS of road RN values in the commercial area; the highest RN was
observed at the middle section of Fuxing Road, with the RN
Items Paths Max. Value Min. Value Average decreasing toward the two ends. Overall, the RN decreased
gradually upward from McDonald’s, and roads with relatively
MD 67 9.694 4.439 6.152 high RN were located in the commercial district.
Rn 67 1.044 0.467 0.723
Cn 67 5 2 3.582

According to Table II, the top three landmarks are all

located within the commercial district, which may be because
the commercial district constitutes the largest area, and thus has
the most roads passing through it. Food districts are distributed
in two places; the one near MOS Burger has more roads
passing through it than does the one on Wenhua Road.
Using space syntax theory, we incorporated relevant
cartographic data into DepthmapX for calculation and
analyses, which revealed layers of roads in the Fengjia
commercial area; specifically, the mean depth, global
integration, and number of road connections were revealed
(Table III). Fig 9. RN of Roads in the Fengjia Commercial Area.

B. Persona Design
1) User segmentation per attribute: This study recorded
the stop points and observed purchasing behaviors of 65
people from 40 groups in the Fengjia commercial area during
the night market’s opening hours (6–8 PM). According to the
onsite observation results, this study divided the research
subjects into three segments and compiled these people’s stop
points as well as the routes they took. Subsequently, to analyze
the distributions of RNs and customers from different
segments in-depth, we divided the research subjects and
defined each segment’s purchase characteristics (Table IV).
Fig 8. Distribution of RNs and Numbers of Connections.

515 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

TABLE IV. USER SEGMENTS

Numbers of
Purchase
Segment groups and Spatial behaviors
characteristics
people

Are familiar with the

Six groups district distributions
Know what and
Locals (Seven and routes and entering
how to purchase
people) the commercial area
from various entrances

Have researched the

Fig 10. Space Syntax Analysis Map: Stop Points of Segment 1.
area, they know where Have rough
30 groups to locate services they purchasing plans
Backpackers
(46 people) require, and enter the and are open to
commercial area from changes
various entrances

Are unfamiliar with the

area’s spatial structure
Are uncertain of
and enter the
Tour group Four groups what to buy, and
commercial area from
travelers (12 people) thus explores the
the entrance near
commercial area
McDonald’s on Fengjia
Road

 Segment 1: Locals consisting of six groups (one group Fig 11. Space Syntax Analysis Map: Stop Points of Segment 2.
comprised two people, with seven people in total);
 Segment 2: Backpackers consisting of 30 groups (46 Backpackers (Segment 2; Fig. 11) started their journey
people); and mainly from road section with high RN (Fengjia Road) and
moved south toward the yellow district, namely the night
 Segment 3: Tour group travelers consisting of four market section on Wenhua Road. The intersections of Fuxing
groups (12 people). Road and Wenhua Road became hot spots because they were
adjacent to transport and parking spots.
Table IV shows that the people most familiar with the
area’s spatial structure were locals (Segment 1), whereas the Tour group travelers (Segment 3) predominantly stopped at
people least familiar with the spatial structure were tour group places with high RN (Fig. 12); however, their stop points were
travelers (Segment 3). Regarding purchase characteristics, scattered all throughout this area. They mainly spread along the
Segment 1 exhibited the most certain attitudes, whereas roads from two landmarks, namely the night market entrance at
Segment 3 exhibited the most uncertain attitudes. Segment 2’s the university’s main entranceway and that near the
spatial behaviors and purchase characteristics, namely McDonald’s.
familiarity with the area and certainty of purchases, were all
between those of Segments 1 and 3. Accordingly, the
characteristics of the three segments were distinct and their
spatial behaviors revealed their distributions in the area.
2) Persona and Spatial Behaviors: The spatial behaviors
(routes taken and stop points) of people from the three
segments were inputted into the QGIS software package to
identify hot spots. According to Fig. 10, locals (Segment 1)
started their journey predominantly from entrances at the
university’s main entranceway or McDonald’s and mainly
proceeded north toward the yellow district (food district) or
south toward the night market section on Wenhua Road.
Fig 12. Space Syntax Analysis Map: Stop Points of Segment 3.

516 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

C. Crowd Behavior Analysis Signage for Fengjia night market were provided at the
Crowd behavior analysis involves using a computer to intersection of Fuxing Road and Wenhua Road as well as near
simulate pedestrian movement and crowd behavior in a space. the parking lot located at the southern end of the commercial
It provides insights into pedestrian spatial behaviors, namely area. Fig. 15 illustrates the predicted crowd behavior of people
the relationship between space and pedestrian behavior and the entering the commercial area at the intersection of Fuxing
interaction between space and pedestrians. According to the Road and Wenhua Road; the most crowded area is on the
descriptions of user personas, pedestrians entered the Fengjia section of Wenhua Road between the intersections with Fengjia
commercial area from various entrances; therefore, the crowd Road and Fuxing Road. According to the night market’s layout,
behavior we observed at different entrances was analyzed this road section was one of two food districts in the
separately. Crowd density was presented using a color gradient commercial area. Food street vendors were located on both
from red (highest density) to blue (lowest density). sides of this road section, which was adjacent to the
commercial district; therefore, this road section exhibited the
Fig. 13 presents the predicted crowd behavior without any densest crowds. When we compared a panoramic photograph
focused entrance. The most crowded area (red section) is of the intersection of Fuxing Road and Wenhua Road with the
located at the center of Fengjia Road, with crowd density map in Fig. 15, the right end of the panoramic photo indicated
decreasing outward along the road. the entrance shown in Fig. 15. Because a signage for Fengjia
Fig. 14 illustrates the predicted distribution of crowds using night market was provided at this intersection, people driving
the entrance near McDonald’s; the results were consistent with there along Fuxing Road usually entered the commercial area
those shown in Fig. 14 (no focused entrance), which might be from this entrance.
because the landmarks in this commercial were located at both Fengjia University is a crucial landmark in this commercial
ends of this road section. Accordingly, future research could area as well as a commonly used entrance into the commercial
use McDonald’s as the main entrance to investigate crowd area for students. Fig. 16 presents the predicted crowd
behavior and hot spots in this commercial area. In addition to distribution in the commercial area of people entering from the
McDonald’s being a landmark, it was also a crucial indicator university’s main entranceway. Fengjia Road has the densest
because it was located at the top of a key road in this area. crowds, which is similar to the result when people enter
through the entrance near McDonald’s. This means that people
entering the commercial area from the two ends of Fengjia
Road mostly gathered along this section of road.

Fig 13. Predicted Crowd Behavior in the Fengjia Commercial Area without
any Focused Entrance.
Fig 15. Predicted Crowd Behavior in the Fengjia Commercial Area:
Intersection of Fuxing and Wenhua Roads as the Entrance.

Fig 14. Predicted Crowd Behavior in the Fengjia Commercial Area: Using
McDonald’s as the Entrance. Fig 16. Predicted crowd behavior in the Fengjia Commercial Area: Fengjia
University’s Main Entranceway as the Entrance.

517 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

of the three segments (Table V), which served as the

conditions and constraints used for subsequent simulations.

Fig 17. Predicted Crowd Behavior in the Fengjia Commercial Area:

University Entranceway Near MOS Burger as the Entrance.

Another entranceway to Fengjia University is near MOS

Burger located at the northern end of the commercial area
(indicated by the red arrow in Fig. 17). Therefore, we also
predicted the distribution of people entering the commercial
area from the university entranceway near MOS Burger.
According to Fig. 17, the most crowded area is on the Wenhua
Road section near MOS Burger, and this road section passes Fig 18. Road Edges and Centerlines in the Research Site.
through both food and residential districts. A reason might be
that students at the university usually entered the commercial
area through the university entranceway near MOS Burger.
These maps of predicted crowd behavior can be used to plan
and design services for students of the university.
D. User Movement Simulation and Visualization Analysis
1) Collection cartographic data of the research site: To
record the road locations of pedestrians in the Fengjia
commercial area, this study employed cartographic data
provided by the Taiwan MAP Service N
(https://maps.nlsc.gov.tw/) to pinpoint the locations of
buildings, outline edges, and centerlines of the roads in the
research site, as well as to define the site borders. Using
digital cartographic data, Google Maps street view, and 360° Fig 19. Zoomed View of the Road Map.
panoramic photos and videos captured onsite, we produced a
TABLE V. SPATIAL BEHAVIOR SETTINGS OF PERSONAS
map of the research site with the edges and centerlines of
roads indicated (Fig. 18 and 19). The blue-outlined road Max. walking Max.
sections are those within the research site and the red lines are Segments distance per stopping Spatial behaviors
minute time
the road centerlines. According to our onsite observations, They know what to purchase,
both vehicles and pedestrians are allowed on Fuxing Road and and thus tend to head to their
Xi’an Street; thus, the edges of the research site along the two destinations for purchases
30
Locals 50 m immediately after entering the
sections of road were defined as being 3–5 m from buildings minutes
commercial area and stop at
outward toward the roads. The remaining road sections were shopping spots for relatively
all only for pedestrian entry at night, and thus their road edges short durations.
They usually have planned
were defined as those shown in the cartographic data.
their routes in advance
2) Setting conditions for persona movement: Following according to shopping
relevant studies related to GANs, which have predominantly 120
categories they are interested
adopted graphic data in their training processes, the present Backpackers 40 m in. They orient themselves in
minutes
this area after entry and make
study collected graphic data of user movement, adapted its necessary changes to their plan.
source code from another study [25], and employed a Keras- They stop at shopping spots for
GAN. A sufficient amount of actual data is required before varying durations.
They are relatively open to
training a GAN; therefore, randomly simulated user what to purchase; bound to a
movement data were produced before sufficient data were Tour group 90
20 m tour group, they stop at limited
travelers minutes
collected. According to the persona definitions in Table IV, we places in the commercial area
and for limited durations.
defined their spatial behaviors according to the characteristics

518 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

According to the purchase characteristics of the three

customer segments and our onsite observations, the customers’
spatial behaviors were defined as being distinct from one
another. To generate pedestrian movement, we defined the
maximum walking distance per minute and stopping durations
in the commercial area of the three customer segments
according to their spatial behaviors.
The interval between each two consecutive simulated
positions of a pedestrian was determined at 1 minute in our
simulation experiment. Therefore, in the commercial area, we
recorded pedestrian positions, defined their maximum walking
distance per minute, and tracked their movement sequences
and directions to generate their movement tracks, which
consisted of the recorded points. The largest point of each track
indicated the position where the pedestrian spent the longest Fig 21. Simulated Results of Movement Tracks at the 1600th Training Cycle.
time during the recorded journey. For example, a recorded
duration of 30 minutes yielded 31 points (including the starting We continued the training and outputted a graphic result for
point) on a pedestrian’s movement track. each 100 training cycles. Fig. 22 presents the movement track
results of the 2000th training cycle; the variation among the
3) User movement visualization and analysis: According graphic results became smaller. Fig. 23 is the result of the
to the characteristics (Table IV) and spatial behaviors 2400th training cycle, after which the movement track results
(Table V) of the customer segments, this study proceeded with started to converge. If the overfitting problem is encountered
data collection in the commercial area and finalized the during ANN training, the ANN might produce pictures that are
sample size at 45 groups of people, with a total of 89 people; highly similar to the original inputs. Therefore, this study set
the sample comprised 32 groups of locals (55 people), 10 the GAN’s maximum training cycle to 2400.
groups of backpackers (22 people), and three groups of tour
group travelers (12 people). The main entrances of the
commercial area comprised those at the three landmarks and
one at the intersection of Wenhua Road and Fuxing Road.
These graphic data were inputted into the Keras-GAN to
simulate graphic user movement data; the process was as
follows. The collected movement data were inputted in the
Keras-GAN for training, and each training cycle involved 30
pictures. Fig. 20 presents the results at the 1000th training
cycle, during which a large amount of noise was observed.
Fig. 21 shows the results at the 1600th training cycle, during
which the road edges became clearer and the simulated
movement tracks were mostly located within the research site.

Fig 22. Simulated results of movement tracks at the 2000th training cycle.

Fig 20. Simulated Results of Movement Tracks at the 1000th Training. Fig 23. Simulated results of movement tracks at the 2400th training cycle.

519 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

not among the most commonly occurring movement tracks,

whereas the other two road sections from McDonald’s are
N (toward Fengjia University to the east and southward along
Fuxing Road). These results indicated that McDonald’s as a
landmark showed the highest connectivity with other roads
within the commercial area; pedestrians had to pass the
McDonald’s when they entered and exited the area.
2) Hot spot analysis of pedestrian segments and training
results of the movement track simulation: When we compared
Fig. 10–12, which show the hot spots of different pedestrian
segments, with Fig. 24, we observed the pedestrian hot spots
(1) (2) (3)
to all be located on the most commonly occurring movement
Fig 24. Pictures of the Three Most Common Movement Tracks Simulated
During Training. tracks in the simulation results. Therefore, the results of the
movement track simulation and those of the pedestrian hot
We extracted one simulated result every 100 training cycles spots were positively correlated; future researchers who wish
between the 1600th and 2400th training cycles; the nine to analyze the movement tracks of different pedestrian
simulated results provided a total of 270 pictures of movement segments more in-depth may also include predictions of
tracks, and the three pictures with the most occurrences were pedestrian hot spot distributions in their research.
combined with the cartographic data for subsequent analyses. 3) Deformed wheel and movement track simulation
Fig. 24(1) shows that most pedestrian movement was results: As previously described, deformed wheel theory is
concentrated at the south of the commercial area, namely from
used to discuss the synergetic relationships between two
Fengjia Road southwards to the night market section on
Wenhua Road and back to Fengjia Road through Fuxing Road. spaces. According to Fig. 13–17, the Fengjia Road section is
Fig. 24(2) presents a pedestrian movement track on Lane 20 of in the center of the commercial area, connecting roads on the
Fengjia Road (food district), in addition to that shown in (1). commercial area borders, and thus integrates places near the
Fig. 24(3) includes both pedestrian movement tracks shown in borders. This result is consistent with that in Fig. 24,
(1) and (2) and shows an additional pedestrian movement track according to which Fengjia Road is where pedestrian
starting from Lane 20 of Fengjia Road through to Lane 127 of movement is mostly concentrated. In addition, our onsite
Wenhua Road and finally back to Fengjia Road. observations verified that Fengjia Road was the central road of
The Keras-GAN-simulated movement results indicated that the commercial area; specifically, this road divided the
the pedestrian movement was predominantly concentrated at commercial area in the middle and connected the two food
the south of the commercial area, despite it having four districts with the commercial district. Wenhua Road and
entrances at the McDonald’s on Fengjia Road, the main Fuxing Road served as the commercial area borders, and their
entranceway of Fengjia University, the university entranceway intersections with Fengjia Road were two main entrances of
near MOS Burger, and the intersection of Wenhua Road and the commercial area. The section of Fengjia Road was short
Fuxing Road. Our onsite observations showed that the southern but served as a crucial hub of the commercial area, and thus
half of Wenhua Road constituted the main part of the night should be seen as the primary focus of the area’s service
market, and therefore, most pedestrian movement tracks were
observed in this part of the commercial area. Moreover, various design. According to our onsite observations and the length
food vendors operating along the northern part of Lane 20 of and integration values of roads located in the northern part of
Fengjia Road to Lane 127 of Wenhua Road attracted people to the commercial area, namely the area to the north of Fengjia
walk around this area. Road, only the food district had relatively dense crowds,
whereas the other places in this area did not exhibit high
V. RESEARCH CONCLUSIONS AND RECOMMENDATIONS spatial characteristics as a commercial area.
This study used the global integration of space syntax to
Overall, the global integration values and pedestrian hot
analyze hot spots where pedestrians most frequently stopped;
spots in the Fengjia commercial area were highly mutually
the Keras-GAN to simulate three movement track graphic
dependent; therefore, the global values of places in this
results; and the deformed wheel theory to investigate the
commercial area may be incorporated into the spatial design
relationship between the commercial area and pedestrians. The
for its service provision. The deep-learning simulation of
following conclusions were derived from the results:
pedestrian movement tracks and spatial characteristics revealed
1) Global integration and movement track simulation by space syntax could be used as a reference for each other.
results: Global integration refers to the connectivity and From the perspective of place creation in the deformed wheel
integration of a road in a global space. Fig. 9 shows that road theory, district layout greatly affected the affordances of roads
in terms of service design. The southern part of the Fengjia
intersection with the highest global integration value is
commercial area was officially planned to be the main food
McDonald’s. Similarly, the simulated movement results district, and thus, frequent place interaction was observed along
(Fig. 24) revealed that only one of the three road sections roads here. Regarding place identity and user behaviors in the
starting from McDonald’s (northward along Fuxing Road) is night market, place creation on the road that constitutes the

520 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 4, 2019

commercial area’s southern border facilitated the fulfillment of Tunnelling and Underground Space Technology, vol. 55, pp. 205-213,
user needs in this part of the commercial area; this approach to 2016.
user-need fulfillment is widely applied in urban planning. [6] Yesiltepe, D. & Kubat, A., “Impact of Bridges on Spatial Transformation
of Urban Pattern: The case of Golden Horn, Istanbul”, En 24th ISUF
Moreover, place realization was observed on Fengjia Road; International Conference. Book of Papers. Editorial Universitat
specifically, it is connected to Fengjia University’s main Politècnica de València. pp. 935-942. doi:10.4995/ISUF2017.2017.6059,
entranceway and plays an essential role in the place identity of 2018
both the commercial area and the university. [7] Hillier, B., Space is the Machine, A Configurational Theory of
Architecture. Cambridge University Press, 1986.
Following are our recommendations for future studies [8] Daneshpour, S. A., Abbaszadegan, M., & Elahi, M., “The Morphology
discussing the relationship between and elements of persona- of Qom; The Study on Spatial Configuration Changes of The City
based movement track simulation and space syntax: (1956-2021)”, Space Ontology International Journal, vol. 6, pp. 27-36,
2017
 Studies may consider examining whether tour group [9] Önder, D. E., & Gigi, Y., “Reading urban spaces by the space-syntax
and backpackers stay overnight in the Fengjia method: A proposal for the South Haliç Region”, Cities, vol. 27, pp.
commercial area. Those staying in hotels located in this 260-271, 2010.
commercial area will spend more time moving around [10] Choi, A.-S., Young-Ook Kim, Oh, E.-S., & Kim, Y.-S, “Application of
the area, and thus will stop at more places than will the space syntax theory to quantitative street lighting design”. Building
those who do not stay overnight. A longer duration of and Environment, vol. 41, pp. 355-366, 2006.
user movement data collection will facilitate more [11] Amorim, L. M. d. E., Filho, M. N. M. B., & Cruz, D, “Urban texture and
space configuration: An essay on integrating socio-spatial analytical
comprehensive possibilities of simulated graphic techniques.” Cities, vol. 39, pp. 58-67, 2014.
results. [12] Hanson, J., “Decoding Houses and Homes”, Cambridge: Cambridge
University Press, 1988.
 This study inputted 45 groups of people (89 people) for
GAN simulation, which included relatively limited [13] Koohsari, M. J., Kaczynski, A. T., Giles-Corti, B., & Karakiewicz, J. A.,
“Effects of access to public open spaces on walking: Is proximity
data from tour group travelers (three groups; 12 enough?”, Landscape and Urban Planning, vol. 107, pp. 92-99, 2013.
people). Thus, future research may expand the sample [14] Manzo, L. C., & Devine-Wright, P., Place Attachment: Advances in
size of tour group travelers. Moreover, future research Theory, Methods and Applications. New York: Routledge, 2014.
may perform Keras-GAN simulations for the three [15] Carpio-Pinedo, J., “Urban bus demand forecast at stop level: Space
personas separately to examine their movement Syntax and other built environment factors”, Evidence from Madrid.
simulation results individually. Procedia - Social and Behavioral Sciences, vol. 160, pp. 205-214, 2014.
[16] Li, Y., Xiao, L., Ye, Y., Xu, W., & Law, A., “Understanding tourist space
 In this study, we did space syntax analysis and at a historic site through space syntax analysis: The case of Gulangyu,
pedestrian movements simulation by GAN in a specific China”, Tourism Management, vol. 52, pp. 30-43, 2016.Alfonzo, M. A.,
commercial area. There is currently no relevant overall “To Walk or Not to Walk? The Hierarchy of Walking Needs.”
application and interdisciplinary research. In the future, Environment and Behavior, vol. 37, pp. 29, 2005.
the relevant evaluation methods can be used to [17] Turner, A., “From axial to road-centre lines: a new representation for
space syntax and a new model of route choice for transport network
understand the practicality of each function in the analysis” Environment and Planning B: Planning and Design, vol. 34,
cross-domain research from the professional pp. 539-555, 2007.
perspective in order to extend service design for this [18] Turner, A., Penn, A., & Hillier, B., “An algorithmic definition of the
research. axial map”, Environment and Planning B: Planning and Design, vol. 32,
pp. 425-444, 2005.
ACKNOWLEDGMENT [19] Klarqvist, B., “A Space Syntax Glossary”, NORDISK
ARKITEKTURFORSKNING, vol. 2, 1993.
The authors acknowledge the technical support on GAN
simulation application by Mr. Kai-Wan Cheng. [20] Jiang, B., Claramunt, C., & Klarqvist, B., “Integration of space syntax
into GIS for modelling urban spaces”, Journal of Applied Gerontology,
REFERENCES vol. 2, 2000.
[1] Soufiane, F., Said, M., & Atef, A., “Sustainable urban design of [21] Enrico di, B., Luca, P., & Matteo, C., “A Multivariate Analysis of The
historical city centers”, Energy Procedia, vol. 74, pp. 301-307, 2015. Space Syntax Output For The Definition Of Strata In Street Security
[2] Hillier, B., & Hanson, J., The Social Logic of Space. Cambridge. Surveys”, DEP - series of economic working papers, 2011.
Cambridge University Press, 1984. [22] Lerman, Y., Rofè, Y., & Omer, I., “Using Space Syntax to Model
[3] Ratti, C., “Urban texture and space syntax: some inconsistencies”, Pedestrian Movement in Urban Transportation Planning”, Geographical
Environment and Planning B: Planning and Design, vol. 31, pp. 487- Analysis, vol. 46, 2014.
499, 2004. [23] Arjovsky, M., Chintala, S., & Bottou, L, “Wasserstein GAN”, Retrieved
[4] Volchenkov, D., & Blanchard, P., “Scaling and Universality in City from arXiv:1701.07875, Retrieved Sep. 2018.
Space Syntax: between Zipf and Matthew”, Physica A, vol. 387, pp. [24] Lynch, K., The Image of the City. The Technology Press & Harvard
2353-2364, 2008. University Press, 1960.
[5] Zhao, J., & Künzli, O., “An introduction to connectivity concept and an [25] Mohammadi, M., Al-Fuqaha, A., & Oh, J.-S., “Path Planning in Support
example of physical connectivity evaluation for underground space”, of Smart Mobility Applications using Generative Adversarial
Networks”, Retrieved from arXiv:1804.083966v1, Retrieved Mar. 2019.

521 | P a g e
www.ijacsa.thesai.org