IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO.

11, NOVEMBER 2016

3159

An Artificial Intelligence-Based Approach for

Simulating Pedestrian Movement
Yi Ma, Eric Wai Ming Lee, and Richard Kwok Kit Yuen

AbstractThis paper proposes a novel approach for simulating pedestrian movement behavior based on artificial intelligence
technology. Within this approach, a large volume of microscopic
pedestrian movement behavior types were collected and encapsulated into an artificial neural network via network training. The
trained network was then fed back into a simulation environment
to predict the pedestrian movement. Two simulation experiments
were conducted to evaluate the performance of the approach.
First, a pedestrian-collision-avoidance test was conducted, and the
results showed that virtual pedestrians with learned pedestrian
behavior can move reasonably to avoid potential collisions with
other pedestrians. In addition, a critical parameter, i.e., defined
as reacting distance and determined to be 2.5 m, represented
the boundary of the collision buffer zone. Second, a pedestrian
counterflow in a road-crossing situation was simulated, and the
results were compared with the real-life scenario. The comparison
revealed that the pedestrian distributions, erratic trajectories,
and densityspeed fundamental diagram in the simulation are
reasonably consistent with the real-life scenario. Furthermore, a
quantitative indicator, i.e., the relative distance error, was calculated to evaluate the simulation error of pedestrians trajectories
between the simulation and the real-life scenario, the mean of
which was calculated to be 0.322. This revealed that the simulation
results were acceptable from an engineering perspective, and they
also showed that the approach could reproduce the lane-formation
phenomenon. We considered the proposed approach to be capable
of simulating human-like microscopic pedestrian flow.
Index TermsPedestrian flow, artificial intelligence, neural networks, lane formation.

I. I NTRODUCTION

HE development of a pedestrian flow model and approach

that can simulate pedestrian movement from a microscopic perspective is crucial for the design and management of
public transportation facilities.
Currently, numerous pedestrian flow models with different
features have been developed [1], [2] and can be categorized as
either macroscopic or microscopic. The macroscopic models

Manuscript received June 7, 2015; revised December 24, 2015 and March 6,
2016; accepted March 11, 2016. Date of publication May 4, 2016; date of
current version October 28, 2016. The work described in this paper was fully
supported through a grant from the Research Grant Council of the Hong Kong
Special Administration Region, China (Project CityU 11206714). The Associate Editor for this paper was X. Cheng.
The authors are with the Department of Architecture and Civil Engineering,
City University of Hong Kong, Kowloon, Hong Kong (e-mail: yima23-c@my.
cityu.edu.hk; ericlee@cityu.edu.hk; bckkyuen@cityu.edu.hk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TITS.2016.2542843

convert the scenario into a node-link presentation and focus

on determining movement time [3], [4]. These models are
rapid tools that provide an overview of the movement but are
unable to describe the movement patterns in different areas
at different time steps. The microscopic models provide more
detail by imposing the rules of movement onto each agent
and letting them interact to simulate pedestrian movement.
Thus, bottleneck areas in crowded public traffic spots can be
predicted and facilities optimized to alleviate congestion. Currently, popular microscopic models are categorized into force-,
grid- and agent-based models based on the ways in which the
pedestrians are represented and propelled [5]. One representative force-based model is the Social Force (SF) model [6],
in which Newtons Law of Motion is introduced to represent
the interaction between pedestrians and obstacles. Pedestrian
movement in continuous space is determined by the resultant
force experienced by the pedestrian. This approach treats each
pedestrian as a mass body in Newtonian mechanics. For gridbased models, a discrete spatial representation is used to represent the simulated environment. The simulated environment is
divided into uniform distributed grids, within which pedestrian
movement is subject to the pre-determined moving rules. Typical representatives of this type of model include the Cellular
Automata (CA) [7] and Lattice Gas (LG) [8] models. The agentbased model is another type of rapid development model used
in recent years. It has been applied in various situations due to
its ability to simulate the unique behavior of heterogeneous humans [1]. In the agent-based models, each pedestrian is usually
represented as an independent and autonomous entity capable
of adaptation. The utility maximization approach is used to
navigate the pedestrians moving direction. The pedestrians
adjust their moving speed in the moving direction in terms of
the perceived environmental information [9]. The abovementioned three types of main-stream pedestrian simulation models
have been used extensively to simulate various scenarios [8],
[10], [11] (e.g., unidirectional, counter, crossing and bottleneck
flows) and to interpret various pedestrian collective behavior
and self-organized phenomena such as arching in front of a
congested doorway, faster-is-slower evacuation [12], [13], laneformation in counterflow conditions [14], strip formation in
crossing-flow conditions [15] and oscillation in counterflow
through a bottleneck passage [16].
Although many significant results have been achieved by the
aforementioned models, the modeling of pedestrian movement
remains challenging due to the limitations of the models. For
example, grid-based models make it difficult to represent the
variety of pedestrian movement direction and speed because

1524-9050 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

3160

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

the movement is restricted by the limited directions and speed

range. This can cause the pedestrian movement to become
over-mechanized. The force-based models may generate unrealistic backward movement due to the excessive superposition
of repulsive forces from other pedestrians in a long range [17].
The excessive artificial parameters of force-based models can
leave the model accused of artifacts. Furthermore, calibrating
for these parameters may lead to extra computational burden. A
similar problem exists with agent-based models.
To overcome these limitations, in recent years, scientists
and researchers have attempted to introduce modern artificial/
statistical intelligence technologies into pedestrian traffic problems. Zarita et al. [18] introduced the probabilistic neural
network (PNN) into a CA model to reinforce pedestrians
decision-making about movement direction. The PNN was
trained by a set of rules determined by the authors and the
intelligence model served as an intelligent navigator by simply
reproducing the rules given by the user. However, the model
remained restricted to the CA framework. Castro et al. [19]
developed an expert system based on fuzzy logic to detect
future collisions with other moving objects on the street and
avoid pedestrian accidents. Szemes et al. [20] developed a
fuzzy-neural network (FNN) model to mimic the obstacle
avoidance behavior of pedestrians. The trained FNN can identify boundaries, turnings and other fixed obstacles in passage.
Toth et al. [21] developed an artificial neural network (ANN)
and fuzzy logic (FL)-based human locomotion model to support the personal navigation during GPS gaps. The navigation
accuracy of the model in the test was up to 35 m CEP (circular
error probable) 50% level. Zheng et al. [22] applied Gaussian
Bayesian networks to predict pedestrians route choice behavior
during evacuation. Shao [23] developed a pedestrian simulation
model based on CA model. In the model, the neural network
was embedded to improve the intelligence of the simulation.
In addition, artificial/statistical intelligence technologies were
also used to investigate the other pedestrian-related traffic problems, such as pedestrian recognition [24], pedestrianvehicle
interaction system [25] and vehiclevehicle interaction system
[26], [27].
Enlightened by these works, this paper proposes a novel
approach based on a modern artificial intelligence technology to
simulate pedestrian movement behavior. Within the approach,
an artificial neural network (ANN) with a multi-layered perceptron (MLP) is used to learn microscopic pedestrian movement
behavior via a learning process based on real-life data. The
learned ANN is then used to predict the general microscopic
pedestrian movement behavior. The approach differs significantly from traditional pedestrian flow simulation models in
three aspects. First, it is nonlinear because it is based on
an ANN that originates from the human nervous system. It
is a powerful nonlinear simulation tool that can reproduce
various nonlinear system behavior types. In other words, the
nonlinearity between the various environmental stimuli and
behavioral responses in pedestrian movement is easily captured
using the ANN. Second, it is adaptive, unlike force- and
agent-based models in which all of the unknown parameters
must be artificially calibrated and the parameters may have
to be adjusted under different pedestrian movement scenarios.

Fig. 1. Typical architecture of the MLP-based ANN.

Instead, each parameter value in the ANN model is determined

automatically via an efficient learning process that draws from
a large volume of historical data. This automatic and efficient
parameter-tuning process can guarantee the generalization and
objectivity of the parameter values. Third, it is open because
the developed model is neither grid-based, with the pedestrian
subject to a pre-determined set of rules, nor force- or agentbased, with the pedestrians moving direction and speed determined by fixed formulas. Instead, the pedestrians movement
state is determined by a trained network, an open system that
can learn new knowledge and experiences with regard to pedestrian movement behavior from the outer environment as human
beings and encapsulate them into the network. Given these
features, the approach is promising to serve as a simulation
tool to analyze the useful statistical quantities for pedestrian
traffic, and to provide auxiliary information for designing a
safe, comfortable and efficient pedestrian traffic environment
for transportation systems.
II. D EVELOPMENT OF THE B EHAVIORAL
L EARNING A PPROACH
In the real world, knowledge of and experiences with how to
stand, walk and move in crowds are continually accumulated
and encapsulated in the human brain or consciousness via
observation and learning. In turn, the stored knowledge and
experiences are fed back to the human to direct their rational
movement behavior in crowds. In this paper, we attempt to
mimic this learningfeedback process using an ANN. In the
machine learning field, the ANN with MLP is a widely used
tool for solving decision-making, intelligent learning and intelligent forecasting problems [28][32]. Theoretically, it has
been proven to be a universal function approximate, as long
as a sufficient number of hidden neurons and hidden layers
are provided [33]. The architecture of the MLP-based ANN
consists of the input layer, the hidden layer(s) and the output
layer. Each layer comprises several neurons, and each neuron
on a layer is connected with the neurons of the adjacent layers,
as shown in Fig. 1. A three-layered network architecture (i.e.,
only one hidden layer between the input and output layers) was
used in this paper.

MA et al.: ARTIFICIAL INTELLIGENCE-BASED APPROACH FOR SIMULATING PEDESTRIAN MOVEMENT

A. Input Layer
The input layer is used to accommodate various system
stimuli, the architecture of which is shaped by the system
itself. For the pedestrian traffic system, empirical study results
indicate that the movement behavior of pedestrians in crowds
reflects the perceived environmental stimuli and the properties
of the pedestrians [5], [9], [34][36]. Thus, the elements in
the input layer corresponding to the pedestrian traffic system
should consist of the following parts.
1) Part IInput Parameters Used to Represent the Pedestrians Current Movement Features: Pedestrians usually do not
change their current direction and speed abruptly or frequently
due to the inertia effect [37]. Thus, a pedestrians subsequent
moving direction and speed tend to be consistent with his/her
current moving direction and speed if there is no perturbance.
Likewise, a pedestrians current velocity should be selected as
an input parameter.
2) Part IIInput Parameters Used to Represent Interactions
With Other Pedestrians: As an element of group movement,
pedestrians direction and speed of movement are influenced
byand influencethe moving states of other individuals.
Such interactions between pedestrians are very complex and
difficult to quantify. Fortunately, empirical research results and
numerous simulation cases have indicated that the degree of
interaction is based on the relative positions of and velocity
between individual pedestrians. Some obvious facts in real life
also verify these conclusions. For example, the smaller the
distance between two pedestrians, the stronger the reciprocal
influence on the moving state. These facts and conclusions
indicate that it is reasonable to select the relative positions of
and velocity between pedestrians as input parameters.
3) Part IIIInput Parameters Used to Represent Interactions With Obstacles: Obstacles, such as boundaries and the
fixed obstacles placed in an environment may constrain or
obstruct the pedestrians movement. In fact, the interactions
between pedestrians and obstacles are similar to those between
pedestrians, particularly if we view an obstacle as being similar
to a pedestrian standing still. Thus, the interactions between
pedestrians and obstacles can be represented in a similar way
to the interactions between pedestrians. However, it must be
noted that obstacles are typically still, and the relative velocity
between pedestrians and obstacles equals that of the pedestrians, as introduced in Part I. Thus, to avoid repetition, we only
select the relative position as an input parameter to represent
interactions between pedestrians and obstacles in this part.
4) Part IVInput Parameters Used to Represent Interactions With Desired Targets: Target positions determine the
pedestrians desired moving direction; that is, the pedestrians
desired moving direction always orients on the target if there is
no perturbance. Here, the relative position between the pedestrian and the desired target is selected as an input parameter to
represent the interaction between them.
5) Part VProperties of the Pedestrian: For the heterogeneous pedestrian flow, some of the individuals subjective
properties also influence his/her movement behavior. These
properties include physiological properties such as health level,
energy level and sensing range and emotional properties such

3161

as attraction intensity, panic intensity and social group property [38]. Pedestrians with different individual properties may
engage in different moving behavior under the same environmental stimuli.
Note that although the architecture of the input layer consists
of the abovementioned five input parameter types, it does not
mean that the number of input parameters is fixed. The number
of input parameters can be added or reduced according to
specific situations. For example, for the homogenous pedestrian
flow, the input parameters in Part V can be ruled out. Such
a flexible architecture can guarantee the adaptability of the
network model.
B. Hidden Layer
The number of neurons in the input and output layers is
fixed, determined by the system and equal to the number of
input and output parameters, respectively. However, there is no
unified approach to determining the number of hidden neurons.
Fortunately, many rules have been proposed to assist users in
estimating the number of hidden neurons. In this paper, the rule
developed by Ward System [39] is used to estimate the number
of hidden neurons. (See (1).)

(1)
Nh = (Nin + Nout )/2 + Ns
where Nin , Nh , and Nout represent the number of neurons
on the input, hidden and output layers, respectively, and Ns
represents the number of training samples.
C. Output Layer
Pedestrians microscopic movement behavior consists of a
series of continuous stimulus-response processes. Pedestrians
respond to subsequent moving direction and speed in terms of
the perceived environmental stimuli and to their own properties
in each moment. Thus, the subsequent response to velocity is
regarded as an output parameter in the output layer, which is
unique to this paper.
Thus far, input and output parameters have been specified,
but unlike most of the existing pedestrian flow simulation
models in which the functional relationships between input and
output parameters are provided explicitly in terms of empirical
formulas, we do not define any parameters or give any formulas
to express these functional relationships. Instead, we obtain
these functional relationships via training/learning from the
real-life data about pedestrian movement. This feature ensures
that the captured functional relationships between input and
output parameters are more objective and persuasive. In the
following sections, we show how we captured these functional
relationships and used the developed approach to examine case
studies.
III. S IMULATION OF P EDESTRIAN M OVEMENT B EHAVIOR
Based on the aforementioned introduction for the developed
simulation approach, we conducted the pedestrian movement
simulation on a crosswalk scenario to test it. The flowchart of
the simulation process is presented in Fig. 2.

3162

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

Fig. 3. Typical interaction patterns between subject pedestrian and various

environmental stimuli in pedestrian movement on a crosswalk.

Fig. 2. Flowchart for simulating pedestrian movement behavior.

A. Input Parameters Selection

For the pedestrian flow on a crosswalk, Fig. 3 shows the
typical interaction patterns between the subject pedestrian and
various environmental stimuli. First, the pedestrians current
velocity (represented as vs for the subject pedestrian and vn
for neighbor j) should be introduced into the input layer to
represent the individuals inertia effect.
Second, the relative positions of and velocity between pedestrians (represented as dns and vns in Fig. 3, respectively)
should be taken into account. Although it is certainly possible
to choose the relative positions of and velocity between the
subject pedestrian and all others in the simulation environment
as input parameters, it may lead to a dimension explosion in the
input layer. Fortunately, recent field studies in animal crowds
have indicated that each individual actually interacts with a
fixed number of its nearest neighbors to determine its motion,
rather than all of the neighbors in a given interaction distance
[40][42]. This, referred to as a k-Nearest-Neighbour (kNN)
interaction pattern, has been successfully applied in pedestrian
flow [9], [43].
A pedestrian is only sensitive to what happens inside a
180-degree angle in front of him/her, as such an area roughly
corresponds to the visually perceived area in pedestrian move-

ment [38]. Thus, to balance the computational complexity and

reliability of our model, we only select the interactions between
the subject pedestrian and his/her five nearest neighboring
pedestrians ahead as input parameters.
For the input parameters in Part III, we only take the
interaction with boundaries into account because, in general,
the extra obstacles will not be placed in the middle of the
crosswalk. Furthermore, given that a vector is used in the
representation, the relative position is always perpendicular to
the boundaries; that is, its horizontal projection is a constant and
equals zero. Thus, the effective interaction between pedestrian
and boundaries only happens in the vertical direction and can
be represented as the perpendicular distance from the left and
right boundaries (i.e., dlbs and drbs in Fig. 3, respectively).
A similar analysis is also suitable for the input parameters
in Part IV, as the desired target is a piece of boundary line.
The desired moving direction is always horizontally directed
towards the boundary line. Thus, the interaction between the
pedestrian and the desired target can be represented as the
horizontal distance from the boundary line at terminal (i.e., dts
in Fig. 3).
It is extraordinarily difficult to measure the physiological
and psychological properties of each individual in real life.
Additionally, these differences in physiological and psychological aspects do not influence pedestrian flow characteristics
from a statistical perspective. Hence, we view the pedestrian
as homogenous flow and do not take his/her properties into
account.
All of the input parameters suitable to general pedestrian
counter flow on a crosswalk were specified in the above analysis, but it must be noted that each neuron in the input layer
can only accept scalar input. The relative velocity and positions
should be decomposed as two scalars, thus 25 input parameters
are introduced into the input layer to represent the current
moving state and interactions with various environmental information. The corresponding architecture of the network is shown
in Fig. 4 and Table I.

MA et al.: ARTIFICIAL INTELLIGENCE-BASED APPROACH FOR SIMULATING PEDESTRIAN MOVEMENT

3163

Fig. 5. Front-end of the developed data extracting program.

Fig. 4. Architecture of the developed ANN approach.

TABLE I
I NPUT AND O UTPUT OF THE D EVELOPED ANN A PPROACH

B. Data Collection and Processing for Network Training

Data collection and processing are key in simulations, so we
chose not to obtain pedestrian movement data from laboratory
experiments or drills. Instead, real-life pedestrian movement
in pedestrian counterflow scenarios was captured. Pedestrian

movement was recorded from a top-down view using a camcorder and then processed into video clips. The recording site
chosen was a crosswalk on a street in Hong Kong. The highrise buildings and dense pedestrian flow in Hong Kong made it
possible to capture sufficient pedestrian movement.
The video clips were used to extract the positions of pedestrians by computer and record their coordinates at every time step.
It is extremely time-consuming to extract such data manually,
but using a fully automatic pattern recognition approach does
not guarantee 100% accuracy. Thus, we used the semiautomatic
approach developed by Teknomo [45]. A simple computer
program was developed to complete this work. The front-end of
the program is as shown in Fig. 5. It played the video clip on a
screen in slow motion while we used a mouse pointer to indicate
the head of a pedestrian on-screen as he/she walked. In this
way, the time stamp and the 2D coordinates of the pedestrian
were recorded on the interface. The video clip was replayed
until the movements of all of the pedestrians in the clip were
captured. However, as the top view of a video is not 100%
horizontally projected, the collected data were geometrically
corrected, converting the 2D coordinates of the pedestrian on
the interface into 2D coordinates in realistic scenarios. Upon the
completion of this semi-automatic data collection process, the
positions of all of the pedestrians at every time step under realistic scenarios were recorded.
The collected data must be processed in advance to cater to
the necessity of network training; that is, the collected data
must be converted to the training samples. To complete this
work, the time step was first determined to calculate the velocity
parameters of each pedestrian at each time step. The rational
time step is crucial to network training because a small time
step may generate a large amount of noise that may reduce the
performance of network training, and a large time step may
be disadvantageous in describing the variety of velocity, as
the velocity direction may become increasingly narrow as the
time step increases. As such, we chose a time step of 0.48 s to
balance the performance and variety of moving velocity. Based
on the chosen time step, 4496 training samples were generated.
The input end of each sample (25 input parameters) represented

3164

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

a type of environmental stimuli faced by pedestrians, and the

output end of each sample (2 output parameters) represented
the corresponding velocity response. The 4496 training samples
generated represented 4496 types of movement behavior in
crowds that were then encapsulated on network via training and
fed back to subsequent simulation experiments.

C. Network Training
The collected samples were used to train the developed
approach. To complete this work, the Backpropagation [46]
(BP) training algorithm was used for its powerful adaptivity.
It calculated the errors between the training and target outputs
based on pre-set weights and activation functions, then adjusted
the weights of the links between the neurons to minimize the
errors. Next, a new round of training was conducted to calculate
the new training errors based on updated weights and adjust the
weights again. This procedure was repeated numerous times
until the training errors were reduced to an acceptable scope.
To prevent overfitting, the early-stop validation approach was
used to supervise the BP training. It splits off parts of the
samples as the validation sets from the samples, which are
trained using an intermediate-state trained network to obtain
validation errors. Such a validation error is used to detect
the overfitting. Specifically, the network training is terminated
when no reduction in validation errors is detected over a certain
number of epochs (15 in this paper). Eventually, the network
state with the minimum number of validation errors was taken
as the ultimate trained network.
The trained network was also applied to testing sets split
from the sample database to evaluate the training performance.
R-squared, as defined in (2) between target output T and
training output P was selected as the performance evaluation
indicator
R2 = 1

n

i=1

(Ti Pi )2 /

n


(Ti T )2

(2)

i=1

where n represents the sum of the number of the target output

and T represents the mean of the target output.
It should be noted that the samples were randomly divided
into training, validation and testing sets in certain proportions.
It is thus possible for a training set to comprise some fortuitous
samples to generate a good performance value. To prevent
such a prejudice, the training should be repeated many times
to minimize the influence of randomization. Thus, for each
training episode, we randomly picked 3597 samples (80%) and
used them for network training. Another 674 samples (15%)
were picked as validation samples to supervise the network
training in terms of the early-stop validation approach. The
remaining 225 samples (5%) were taken as independent testing
sample sets to evaluate the performance of the trained model.
Before conducting the network training, the number of
neurons in the hidden layer should be determined. Based on the
network architecture and the number of samples, the number
of neurons in the hidden layer was estimated to be 81 in terms
of (1).

Fig. 6. Distribution of the performance of the network approximated by the

beta distribution.

The network training under the same network structure and

different sample divisions was run 1000 times, and the distribution of the performance values is presented in histogram form
in Fig. 6. Meanwhile, a beta distribution function, as defined
in (3), was used to approximate the generated distribution of
performance over 1000 times through independent training
1
xa1 (1 x)1 /u1 (1 u)1 du.

f (x|, ) =

(3)

The parameters of the distribution were estimated to be =

389.0088, = 45.0984.
D. Fundamental Test for the Trained Network
Before applying the network to complete pedestrian flow
simulation experiments, a fundamental test was conducted to
reveal whether the network could handle the basic pedestrian
collision avoidance behavior.
In the test process, the mean of each input parameter was
calculated in terms of the samples data. Then, a traditional
control variate method [47] was used to analyze the varying of
velocity response against the specific input parameters. Fig. 7
shows the mechanism of the control variate method.
In the first test, the moving state (velocity and position) of
five neighbors were fixed at the mean conditions. As the subject
pedestrian approached a group of five neighbors horizontally,
the formers moving velocity and the vertical relative distance
to each neighbor were kept at mean conditions artificially. The
exception was the horizontal relative distance to the neighboring group, which ranged from 3.50.5 m. Meanwhile, the
horizontal relative distance to each neighbor changed with
the horizontal relative distance to each neighboring group,
respectively. Then, the trained model was used to predict the
subsequent velocity in terms of the accepted input parameter
values accepted by the subject pedestrian.

MA et al.: ARTIFICIAL INTELLIGENCE-BASED APPROACH FOR SIMULATING PEDESTRIAN MOVEMENT

3165

Fig. 9. Varying of predicted speeds against horizontal relative speed.

Fig. 7. Mechanism of the control variate method.

Fig. 8. Varying of predicted speeds against horizontal relative distance.

The purpose of this test was to examine how the subject

pedestrian reacted to potential collisions at different positions.
Fig. 8 shows the detailed varying of velocity response against
the current horizontal relative distance. For ease of analysis,
the output velocity was decomposed into forward and sideways
speeds.
The velocitys response tendency varied with the horizontal
relative distance. When the subject pedestrian approached from
3.52.5 m, he/she did not decelerate to avoid the potential
collision with an oncoming pedestrian group. The subject
pedestrian considered the 2.5-m buffer zone large enough to

cope with the collision. Subsequently, as the subject pedestrian

approached the pedestrian group from 2.50.5 m, the subject
pedestrian decelerated forward and accelerated sideways to slip
past the pedestrians ahead. Based on these results, the critical
or reacting distance was determined to be 2.5 m. The value
of the reacting distance was 2.5 m when the value of the fixed
input parameter was kept at the mean of the sample data, but
may vary under different input conditions.
To understand how the velocity response varies against the
horizontal relative speed, another test was conducted in which
the moving states of five neighbors were fixed. The vertical
relative speed and distance to each neighbor were kept at mean
conditions, except the horizontal relative distance and speed.
The horizontal relative distance to the neighboring group was
set at 2.5 m, the reacting distance. The horizontal relative
speed between the subject pedestrian and the nearest neighbor
was changed from 0 m/s to 2 m/s, as the movement state
between the subject pedestrian and the nearest neighbor varies
from following to counter-moving state. The horizontal relative
speed to the rest of the neighbors was changed correspondingly.
Then, the trained model was used to predict the subsequent
velocity responses of the subject pedestrian for the accepted
input parameter values.
The aim of this test was to examine how a subject pedestrian
situated at a fixed position reacts to a potential collision under
different horizontal relative speeds. Fig. 9 shows the detailed
response variations to the forward speed against the current horizontal relative speed. For comparison, we also investigated the
varying of forward speed when the horizontal relative distance
to the neighboring group was set at 1.5 m. As Fig. 9 shows,
we can observe obvious deceleration whether the horizontal
relative distance to the neighboring group was 2.5 or 1.5 m. This
was a reasonable reaction to avoid the collision. By comparing
the two varying curves, we found that the nearer the horizontal
relative distance, the sharper the deceleration under the same
horizontal relative speed.
The abovementioned test results indicate that the virtual
pedestrian with learned knowledge and experience can take

3166

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

model, the original position of pedestrian i is assumed to be Poi

i
and the modified position Pm
is determined by:
 i

i
Pm
= arg min Pm
Poi  .

(4)

i
Pm
is subject to:


 i
Pm P j  Rmin for j(j = i)

 i
P Li  Dmax
m

Fig. 10. Video screenshot of pedestrian counter flow scenario on crosswalk.

rational action to avoid a collision in pedestrian movement.

They also suggest that the trained results of the network were
satisfactory.
E. Simulation and Verification for the Trained Network
The trained networks were used to mimic the pedestrian flow
on a crosswalk. The width of the crosswalk was W = 6.73 m
and the length was L = 8.12 m. The pedestrian movement on
the crosswalk was recorded by camcorder, and the recorded
time lasted 20 s (44 time steps). Fig. 10 shows the video screenshot for the first frame of video. The recorded pedestrian movement on the crosswalk in the video clip involved 94 pedestrians,
57 of whom crossed the crosswalk from left to right while the
rest (37) crossed from right to left. The trajectories with the
time stamps of each pedestrian were extracted using the abovementioned semi-automatic data processing approach.
Next, in the simulation experiments, the experimental setup
was kept consistent with the real-life scenario. The width and
length of the simulation region were set to 6.73 m and 8.12 m,
respectively, and the virtual pedestrians in the simulations were
represented by 0.4-m circles, which is the typical size of a person. Each pedestrian entered the simulation region at the same
time, location and velocity as the real-life scenario. However,
it must be noted that the number of neighboring pedestrians
encountered by the subject pedestrian may have been less than
five at the beginning of the simulation. To address this, we
added an extra virtual pedestrian some distance from the subject
pedestrian until the number of neighbors reached five. These
added virtual pedestrians made up for the shortage of neighbors
without influencing the subject pedestrians movement due to
the far distance.
Another noticeable problem in the simulation experiments
was that the movement determined by the developed model
did not ensure that all of the pedestrians could fully eliminate
position overlaps due to the existence of network training
error, although such error was very small. To overcome this
drawback, a position-modifying model was introduced. In the

(5)
(6)

where, for (5), P j represents the position of other pedestrians,

and Rmin represents the minimum separation distance between
two pedestrians and equals 0.4 m. For (6), Li represents the
position of pedestrian i at the last time step and Dmax represents
the maximum moving distance during one time step. Here,
Dmax is set to 0.86 m. The position of each pedestrian is
modified in terms of (4) in a random sequence.
The position-modifying model effectively avoids position
overlaps, but the final pedestrian dynamics in the simulation
experiment may have been unacceptable due to the accumulation of errors resulting from modifications to every time step.
However, in the simulation experiments, we were not concerned
with the in-between state of pedestrian movement, only with the
pedestrian dynamics at specific time steps in most situations.
We neglected the slight position overlap at the in-between
state of pedestrian movement, but only executed the position
modifying at specific time steps to guarantee the accuracy of
the ANN approachs prediction results. Once the pedestrians
entered the simulation region, the shuffled sequential update
[48] was used to update their coordinates. At each iteration
of the position update, the pedestrians were updated in a
random sequence. The simulation experiment was run for the
same time span as the real-life scenario recorded in the video
clip. Fig. 11(a) shows the pedestrian distribution at the 11th,
21th, 31th time step in the simulation, respectively, Fig. 11(b)
shows the corresponding pedestrian distributions in the real-life
scenario. Besides, Fig. 12(a) gives the pedestrian trajectories in
the simulation, and Fig. 12(b) gives the pedestrian trajectories
in the real-life scenario. The comparison indicates that the
pedestrian distributions and trajectories in the simulation were
basically consistent with the real-life scenario.
Apart from the pedestrian distributions and trajectories, we
also made a comparison on density-speed fundamental diagram
between the simulation, the real-life scenario and the empirical
results made by Weidmann [49]. The density and speed data
both in the simulation and real-life scenario were extracted in
a 2 2 squared area bounded by x = 5.2, x = 7.2, y = 0.4,
y = 2.4 (x, y represents the width and length direction in the
scenario respectively) considering the pedestrians were very
dense in this area. The results are shown in Fig. 13. The overall
distribution of the data points of the simulation is basically in
line with the real-life scenario, although some fluctuations in
speed can be found. Besides, many speed values both in the
simulation and the real-life scenario are higher in comparison
to Weidmanns results in the same density. This is due to the
especial conditions of urgency in the pedestrians march given
that the scenario is a crosswalk. Generally, the pedestrian would

MA et al.: ARTIFICIAL INTELLIGENCE-BASED APPROACH FOR SIMULATING PEDESTRIAN MOVEMENT

3167

Fig. 11. Comparison of pedestrian distributions at the 11th, 21th and 31th time step, respectively. The black and white circles represent pedestrians moving to
the right and left, respectively. The number in each circle represents each individuals serial number. (a) Pedestrian distribution in the simulation. (b) Pedestrian
distribution in the corresponding real-life scenario.

Fig. 13. Comparison of the density-speed fundamental diagram between the

simulation, the real-life scenario and the empirical results made by Weidmann.

be more urgent and have a higher speed in the real life for
crossing the road within limited green time.
Moreover, another quantitative evaluation indicator, namely,
the relative distance error [44], was calculated to better illustrate
the reliability of the developed approach. The relative distance
error at time step t for pedestrian is defined by:
Fig. 12. Comparison of pedestrian trajectories. The blue and red lines represent
pedestrians moving to the right and left, respectively. (a) Pedestrian trajectories
in the simulation. (b) Pedestrian trajectories in the corresponding real-life
scenario.


 simulated
P
(t + t, ) P tracked (t + t, )
E(t, ) =
P tracked (t + t, ) P tracked (t, )
(7)

3168

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

Fig. 14. Frequency distribution of the relative distance error.

where P tracked (t, ) represents the actual coordinate of pedestrian a at time step t in the real-life scenario; P tracked (t +
t, a) represents the actual coordinate of pedestrian a after one time step t in the real-life scenario; the values of
P tracked (t, ) and P tracked (t + t, a) can be extracted from
the recorded video clip; P simulated (t + t, a) represents the
predicted position of pedestrian at time step t + t; and
the value of P simulated(t + t, a) can be calculated using the
developed approach in terms of actual input parameters at
time step t. The relative distance error reflects in some sense
the inconsistency between the real and simulation trajectories.
Here, the relative distance error was calculated over each time
step for each pedestrian. Fig. 14 shows the corresponding frequency distribution in histogram form. The mean of the relative
distance error was calculated to be 0.322. This value indicates
that the accuracy of the simulation trajectory was acceptable.
Thus far, the comparison on pedestrian distributions, erratic trajectories, and the density-speed fundamental diagram
and quantitative relative distance error calculation over the
pedestrian movement trajectory between the simulation and the
real-life scenario were conducted. To validate the developed
approach perfectly, we also investigated the possibility of reproducing pedestrian self-organized phenomena. Fortunately, the
typical self-organized phenomena (lane-formation) that existed
in real life were successfully reproduced in our simulation
experiments. Fig. 15(a) shows the lane-formation phenomenon
in the simulations. Several narrow pedestrian walking lanes
consist of a sequence of same directional pedestrians can
be observed. This phenomenon revealed that the pedestrians
tended to follow the pedestrian ahead in the same direction to
obtain free moving space and avoid the potential for collision
from pedestrians moving in the opposite direction. Fig. 15(b)
shows the corresponding video screenshot in the real-life scenario when the lane formed. This finding further indicates that
the developed approach is capable of simulating bidirectional
pedestrian flow.
IV. C ONCLUSION
In this paper, we presented a novel microscopic pedestrian
flow simulation approach based on an ANN. The approach

Fig. 15. Lane-formation phenomenon. (a) Lane-formation phenomenon in

simulation. The black and white circles represent pedestrians moving to the
right and the left, respectively. The number in each circle represents the serial
number of that individual. The red and blue arrows indicate the direction of the
lane. (b) The corresponding video screenshot in the real-life scenario.

comprised two steps: a learning and a feedback stage. In

the learning stage, knowledge of and experiences in how to
move in crowds were accumulated and encapsulated into the
network via a learning process based on the real-life data. In
the feedback stage, the learned ANN was fed back into the simulation environment to direct virtual pedestrians in responding
reasonably in the subsequent moving direction and at the expected speed in terms of the accepted environmental stimuli.
A fundamental test was conducted to reveal whether the
learned ANN approach was capable of handling the basic
pedestrian collision avoidance behavior. The varying of output
speed against the horizontal relative distance and speed were
analyzed using the control variate method. The varying curves
revealed that the virtual pedestrian can engage in reasonable
forward deceleration and sideways acceleration to avoid colliding with other pedestrians. A critical parameter, the reacting
distance, was determined to be 2.5 m. This value represents
the boundary of the realistic collision buffer zone. In general, a
pedestrian will not rapidly change his/her current moving state
to avoid other pedestrians when the horizontal distance between
them is larger than 2.5 m.

MA et al.: ARTIFICIAL INTELLIGENCE-BASED APPROACH FOR SIMULATING PEDESTRIAN MOVEMENT

The approach was validated by simulating the pedestrian

counter flow on a crosswalk. The quality of the simulation was
evaluated based on the aspects of the qualitative comparison on
pedestrian distributions, trajectories, and the density-speed fundamental diagram, the quantitative evaluation indicator calculation and the capability analysis regarding whether the model
could reproduce typical self-organized phenomena. First, the
qualitative comparison results show that pedestrian distributions, trajectories, and the density-speed fundamental diagram
in simulations were basically consistent with the real-life scenario. Then, the chosen indicator, namely the relative distance
error, was calculated to evaluate the error of pedestrian trajectory between simulation and real-life scenario. The mean was
calculated to be 0.322, indicating that the simulation results
were acceptable. Finally, a typical self-organized phenomenon
(lane-formation) was observed in the simulation. These results
indicate that the developed approach is capable of simulating
the bidirectional pedestrian flow.
The approach is very promising to serve as a simulation
tool to analyze useful statistical quantities for pedestrian traffic
problems and to provide auxiliary information for designing a
safe, comfortable and efficient pedestrian traffic environment in
a transportation system. Besides, it can be speculated that the
ANN can be generalized and applied into various pedestrian
flow scenarios, such as unidirectional flow, bidirectional flow,
multi-directional flow, pedestrian evacuation and etc., given
that the ANN is a powerful nonlinear simulation tool that can
reproduce various nonlinear system behaviors, if enough hidden
layers and neurons are introduced and the quantity and quality
of the training samples are large and accurate enough.
Although the developed approach can simulate the pedestrian movement very well, two possible limitations about the
approach itself should be still noted. One is that, the increase
in the numbers of input parameters and training samples can
increase the performance of the approach theoretically, but the
resulting high computational cost will be disadvantageous to
the conduction of the network training in reality. According to
our experience, the training time would be significantly slowed
down if taken one more neighbor into account in the input layer
(i.e., 6 neighbors). Another is that, the centralized network system may lead to the mechanized movement in the simulations
since all pedestrians share the same perceptions. Overcoming
these limitations will be the focus of our future work.
Besides, the current pedestrian movement data set used to
train the ANN only extracted from one scene. More data from
different crosswalks that considered different times, kinds of
the group and other features is necessary in the future works.
R EFERENCES
[1] X. Zheng, T. Zhong, and M. Liu, Modeling crowd evacuation of a building based on seven methodological approaches, Build. Environ., vol. 44,
no. 3, pp. 437445, Mar. 2009.
[2] N. Bellomo, B. Piccoli, and A. Tosin, Modeling crowd dynamics from
a complex system viewpoint, Math. Models Methods Appl. Sci., vol. 22,
Aug. 2012, Art. no. 1230004.
[3] R. L. Hughes, A continuum theory for the flow of pedestrians, Transp.
Res. B, Methodol., vol. 36, pp. 507535, 2002.
[4] C. Xie, D. Y. Lin, and S. T. Waller, A dynamic evacuation network optimization problem with lane reversal and crossing elimination strategies,
Transp. Res. B, Methodol., vol. 46, no. 3, pp. 295316, May 2010.

3169

[5] S. B. Liu, S. M. Lo, J. Ma, and W. L. Wang, An agent-based microscopic

pedestrian flow simulation model for pedestrian traffic problems, IEEE
Trans. Intell. Transp. Syst., vol. 15, no. 3, pp. 9921001, Jun. 2014.
[6] D. Helbing and P. Molnar, Social force model for pedestrian dynamics,
Phys. Rev. E, Stat. Phys. Plasmas Fluids Relat. Interdiscip. Top., vol. 51,
no. 5, p. 4282, May 1995.
[7] C. Burstedde, K. Klauck, A. Schadschneider, and J. Zittartz, Simulation
of pedestrian dynamics using a two-dimensional cellular automaton,
Physica A., vol. 295, no. 3/4, pp. 507525, Jun. 2001.
[8] M. Muramatsu, T. Irie, and T. Nagatani, Jamming transition in pedestrian
counter flow, Physica A., vol. 267, no. 3/4, pp. 487498, May 1999.
[9] M. Asano, T. Iryo, and M. Kuwahara, Microscopic pedestrian simulation model combined with a tactical model for route choice behaviour, Transp. Res. C, Emerging Technol., vol. 18, no. 6, pp. 842855,
Dec. 2010.
[10] V. J. Blue and J. L. Adler, Cellular automata microsimulation of bidirectional pedestrian flows, Transp. Res. Rec., vol. 1678, pp. 135141,
1999.
[11] S. P. Hoogendoorn and W. Daamen, Pedestrian behavior at bottlenecks,
Transport. Sci., vol. 39, no. 2, pp. 147159, 2005.
[12] D. Parisi and C. Dorso, Microscopic dynamics of pedestrian evacuation,
Physica A, vol. 354, pp. 606618, Aug. 2005.
[13] S. Liu, L. Yang, T. Fang, and J. Li, Evacuation from a classroom considering the occupant density around exits, Physica A., vol. 388, no. 9,
pp. 19211928, May 2009.
[14] J. Dzubiella, G. P. Hoffmann, and H. Lwen, Lane formation in colloidal
mixtures driven by an external field, Phys. Rev. E, Stat. Nonlinear Soft
Matter Phys., vol. 65, 2002, Art. no. P021402.
[15] S. Hoogendoorn and P. H. L. Bovy, Simulation of pedestrian flows by
optimal control and differential games, Opt. Control Appl. Methods,
vol. 24, no. 3, pp. 153172, May/Jun. 2003.
[16] D. Helbing, L. Buzna, A. Johansson, and T. Werner, Self-organized
pedestrian crowd dynamics: Experiments, simulations, and design solutions, Transp. Sci., vol. 39, pp. 124, 2005.
[17] M. Chraibi, U. Kemloh, A. Schadschneider, and A. Seyfried, Force-based
models of pedestrian dynamics, Netw. Heterog. Media, vol. 6, no. 3,
pp. 425442, Sep. 2011.
[18] Z. Zarita and E. A. Lim, Intelligent exit-selection behaviors during a room evacuation, Chin. Phys. Lett., vol. 29, no. 1, 2012,
Art. no. 018901.
[19] J. Castro, M. Delgado, J. Medina, and M. Ruiz-Lozano, An expert
fuzzy system for predicting object collisions. Its application for avoiding
pedestrian accidents, Expert Syst. Appl., vol. 38, no. 1, pp. 486494,
Jan. 2011.
[20] P. T. Szemes, H. Hashimoto, and P. Korondi, Pedestrian-behavior-based
mobile agent control in intelligent space, IEEE Trans. Instrum. Meas.,
vol. 54, no. 6, pp. 22502257, Dec. 2005.
[21] C. Toth, D. A. Grejner-Brzezinska, and S. Moafipoor, Pedestrian tracking
and navigation using neural networks and fuzzy logic, in Proc. IEEE Int.
Symp. Intell. Signal Process., 2007, pp. 16.
[22] X. P. Zheng and M. T. Liu, Forecasting model for pedestrian distribution
under emergency evacuation, Reliab. Eng. Syst. Safe., vol. 95, no. 11,
pp. 11861192, 2000.
[23] P. Shao, A more realistic simulation of pedestrian based on cellular
automata, in Proc. IEEE Int. Workshop OSSC, 2009, pp. 2429.
[24] L. Zhao and C. Thorpe, Stereo and neural network-based pedestrian
detection, IEEE Trans. Intell. Transp. Syst., vol. 1, no. 3, pp. 148154,
Sep. 2000.
[25] G. K. H. Pang, K. Takahashi, and T. Yokota, Adaptive route selection for
dynamic route guidance system based on fuzzy neural approaches, IEEE
Trans. Veh. Technol., vol. 48, no. 6, pp. 20282041, Nov. 1999.
[26] X. Cheng, C.-X. Wang, B. Ai, and H. Aggoune, Envelope level crossing rate and average fade duration of nonisotropic vehicle-to-vehicle
Ricean fading channels, IEEE Trans. Intell. Transp. Syst., vol. 15, no. 1,
pp. 6272, Feb. 2014.
[27] X. Cheng, Wideband channel modeling and intercarrier interference
cancellation for vehicle-to-vehicle communication systems, IEEE J. Sel.
Areas Commun., vol. 31, no. 9, pp. 434448, Sep. 2013.
[28] M. Gardner and S. Dorling, Artificial neural networks (the multilayer
perceptron)A review of applications in the atmospheric sciences,
Atmos. Environ., vol. 32, no. 14/15, pp. 26272636, Aug. 1998.
[29] G. Zhang, B. E. Patuwo, and M. Y. Hu, Forecasting with artificial neural
networks: The state of the art, Int. J. Forecast., vol. 14, pp. 3562, 1998.
[30] E. W. M. Lee, R. K. Yuen, S. Lo, K. Lam, and G. Yeoh, A novel artificial
neural network fire model for prediction of thermal interface location
in single compartment fire, Fire Safety J., vol. 39, no. 1, pp. 6787,
Feb. 2004.

3170

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 11, NOVEMBER 2016

[31] J. G. Juang and K. C. Cheng, Application of neural networks to disturbances encountered landing control, IEEE Trans. Intell. Transp. Syst.,
vol. 7, no. 4, pp. 582588, Dec. 2006.
[32] P. C. M. Leung and E. W. M. Lee, Estimation of electrical power consumption in subway station design by intelligent approach, Appl. Energ.,
vol. 101, pp. 634643, Jan. 2013.
[33] K. Hornik, M. Stinchcombe, and H. White, Multilayer feedforward
networks are universal approximators, Neural Netw., vol. 2, no. 5,
pp. 359366, 1989.
[34] W. Wang, S. Lo, S. Liu, and H. Kuang, Microscopic modeling of
pedestrian movement behavior: Interacting with visual attractors in the
environment, Transp. Res. C, Emerging Technol., vol. 44, pp. 2133,
Jul. 2014.
[35] F. Martinez-Gil, M. Lozano, and F. Fernndez, MARL-Ped: A multiagent reinforcement learning based framework to simulate pedestrian groups, Simul. Modell. Pract. Theory, vol. 47, pp. 259275,
Sep. 2014.
[36] H. Gotoh, E. Harada, and E. Andoh, Simulation of pedestrian contra-flow
by multi-agent DEM model with self-evasive action model, Safety Sci.,
vol. 50, no. 2, pp. 326332, Feb. 2012.
[37] D. Yanagisawa et al., Introduction of frictional and turning function for
pedestrian outflow with an obstacle, Phys. Rev. E, Stat. Nonlinear Soft.
Matter Phys., vol. 80, 2009, Art. no. 036110.
[38] L. Luo et al., Agent-based human behavior modeling for crowd simulation, Comput. Animation Virtual Worlds, vol. 19, no. 3/4, pp. 271281,
2008.
[39] NeuroShell 2 Manual, Ward Syst. Group Inc., Frederick, MA, USA, 1996,
Ward.
[40] M. Ballerini et al., Interaction ruling animal collective behavior depends
on topological rather than metric distance: Evidence from a field study,
Proc. Nat. Acad. Sci., vol. 105, pp. 12321237, 2008.
[41] C. Saloma, G. J. Perez, G. Tapang, M. Lim, and C. Palmes-Saloma, Selforganized queuing and scale-free behavior in real escape panic, Proc.
Nat. Acad. Sci., vol. 100, no. 21, pp. 11 94711 952, 2003.
[42] J. L. Deneubourg, J. Halloy, J. M. Am, C. Rivault, and C. Detrain, Selforganised choice based on inter-attraction: The example of gregarious
animals, in Proc. Pedestrian Evac. Dyn., N. Waldau, P. Gattermann,
H. Knoflacher, and M. Schreckenberg, Eds., 2007, pp. 455463.
[43] J. Ma, W. G. Song, J. Zhang, S. M. Lo, and G. X. Liao, k-NearestNeighbor interaction induced self-organized pedestrian counter flow,
Physica A, vol. 389, pp. 21012117, 2010.
[44] A. Johansson, D. Helbing, and P. K. Shukla, Specification of the social force pedestrian model by evolutionary adjustment to video tracking
data, Adv. Complex Syst., vol. 10, pp. 271288, 2007.
[45] K. Teknomo, Microscopic pedestrian flow characteristics: Development of an image processing data collection and simulation model,
Ph.D. dissertation, Department of Human Social Information Sciences,
Tohoku Univ., Sendai, Japan, 2002.
[46] P. Werbos, Beyond regression: New tools for prediction and analysis in
the behavioral sciences, Ph.D. dissertation, Department of Mathematics,
Harvard Univ., Cambridge, MA, USA, 1974.
[47] S. L. zesmi and U. zesmi, An artificial neural network approach to
spatial habitat modelling with interspecific interaction, Ecol. Modell.,
vol. 116, pp. 1531, 1999.

[48] J. Zhang, W. Song, and X. Xu, Experiment and multi-grid modeling

of evacuation from a classroom, Physica A, vol. 387, pp. 59015909,
2008.
[49] U. Weidmann, Transporttechnik der fussgnger, in Transporttechnische Eigenschaften des Fussgngerverkehrs, ser. Schriftenreihe des IVT
Nr. 90, 2nd ed. Zurich, Switzerland: ETH Zrich, Mar. 1993.

Yi Ma received the B.E. and M.E. degrees in

transportation from Shijiazhuang Tiedao University,
Shijiazhuang, China, in 2008 and 2012, respectively.
He is currently working toward the Ph.D. degree
with the Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon,
Hong Kong.
His research interests include artificial intelligence, pedestrian behaviors, crowd movement, and
evacuation analysis in transportation stations.

Eric Wai Ming Lee received the B.Eng. (Hons.)

degree in building services engineering and the Ph.D.
degree in fire engineering from the City University
of Hong Kong, Kowloon, Hong Kong, in 2000 and
2004, respectively.
Before pursuing his Ph.D. studies, he was an
Executive Engineer in one of the leading consultant
firms in Hong Kong. He is currently an Associate
Professor with the Department of Architecture and
Civil Engineering, City University of Hong Kong.
His research interests include the application of soft
computing techniques in building engineering.
Dr. Lee is a Registered Professional Engineer in Hong Kong.

Richard Kwok Kit Yuen received the B.Sc. (Hons.)

and M.Sc. degrees in mechanical engineering from
The University of Hong Kong, Pokfulam, Hong
Kong, in 1981 and 1991, respectively, and the Ph.D.
degree in mechanical engineering from the University of New South Wales, Kensington, Australia,
in 1998.
He is currently a Professor with the Department of
Architecture and Civil Engineering, City University
of Hong Kong, Kowloon, Hong Kong. His research
interests include building energy conservation; building fire modeling; combustion; pyrolysis; computation fluid dynamic applications in building design; and heating, ventilation, and air-conditioning system
simulation.
Dr. Yuen is a Registered Professional Engineer in Hong Kong.