2010 Sixth International Conference on Natural Computation (ICNC 2010)

A Cellular Automaton Model of Unidirectional

Pedestrian Walkway Based On Pedestrian Vision
Field
Guirong Hou1,2
1 School of Traffic and Transportation Engineering,
Central South University, No. 472, Segment 3 of Hurong
Middle Road, Changsha, Hunan Province,China.
2
Hunan
Communications
Research
Institute.
houguirong@126.com.

Mingjun Liao 3
3 School of Transportation & Civil Engineering, Beihua
University, Jilin, Jilin Province,China, 132013.

Abstract Cellular Automaton model (CA) is widely applied in

pedestrian behavior, but there are some shortcomings in CA. In
traditional CA, vision of a cell is near-sighted, only has a value of
a cell. The CA model in the paper as a discrete model borrows
some ideas from continuous models. The vision field is
introduced, and the density within the vision field is applied to
modify the transition possibility. The result of the model shows
the model works well.

and the macro-level activity can be attained too. However,

some simplification is carried in collision detection or
pedestrian intelligence. Some improvements can be made with
the help of humans vision field, and region density.

Keywordspedestrian behavior, vision field, Cellular Automaton

model

I.

INTRODUCTION

Currently, attempts are made at modeling pedestrian

traffic flow. In 1991, Matsushita and Okazaki [1] introduced
Magnetic Force Model into modeling pedestrian behavior. In
the model, pedestrian and obstacles, e.g. Wall, pillar, and
handrail are viewed as plus polar, but destinations,
intermediate destinations or attractive activities are viewed as
minus polar. Collision avoiding is achieved by the theory of
interaction between different types of polar. The motion of
pedestrian is mainly based on the physical motion equation. At
the same time, Helbing, D [2] presented a social force model
alike Magnetic Force Model. Actually, in the social force
model, the forces can be viewed as distance vector from
different direction of pedestrian. Because Matsushita and
Helbing, Ds models are both continuous space models, the
main shortcomings of the two models are their slowly
computing speed. It is necessary that a new method must be
presented to meet large scale scenario. Cellular Automata
model (CA) is suitable for the complex behavior like
pedestrian behavior.
CA model is first a tool or method for artificial life. Then,
it is applied in other fields. In 1986, Crmer et al [3] firstly
applied CA in transportation research. Nagel et al [4] and
Biham et al [5] respectively modeled 1-dimensional freeway
traffic flow and 2-dimenisnal urban traffic flow based on CAs
Wolfram rules. In 1998, Blue, V J [6] presented a CA model to
model pedestrian behavior on a unidirectional multi-lane
walkway. A series of simple rules are used in the model, and
behaviors of pedestrians at the micro-level can be captured,

II.

A. CA basic concepts
CA was first introduced by applied mathematicians John
von Neumann and Ulam in the 1940s, who called them
cellular spaces. CA has been shown capable of yielding
discrete approximations to the solutions of systems of
differential equations, in terms of which much of the
macroscopic physics of our world can be expressed.
CA is suitable for discrete dynamical systems, consisting
of an array of nodes (cells) of n dimension [7]. Each cell is in
one of k different states at a given tick of the clock. At each
discrete tick of the clock, each cell may change its state, in a
way determined by the transition rules of the particular CA.
The transition rules describe precisely how a given cell should
change states, depending on its current state and the states of
its neighbors (see Fig.1).the preferred moving direction can be
expressed with 33 matrix(see in Fig.2), namely, Matrix of
preference.

(a) Von neumann neighbor

(b) Moore neighbor
Fig.1 CA neighbors

Fig.2 Matrix of preference

Sponsored by education department of Jilin province(2009155)

978-1-4244-5961-2/10/$26.00 2010 IEEE

CA MODEL

3106

So according to above definition, Features of CA models

can be concluded [8]: (1) size of the state space, which describes
the scope of moving objects; (2) number of each cells
attributes describing the state of the cell; (3) neighborhood of a
cell, which define the cells view about the other cells; (4) local
rules, which allows translating the interactions between cells
into local ones.
B. pedestrian size and density
the size of cell is determined by the pedestrian size an
density.
The pedestrians 95% size of other countries e.g. western
and some eastern race, and china is listed in table 1.
Table 1 average pedestrian size[9,10]
Other countries
China

Maximum shoulder
width(cm)
51.50
45.35

Chest thickness(cm)
32.50
24.2

The size is measured in the nude status. Considering

humans clothes or load, additive value should be added to the
above size. Fruin made the assumption of adding 3.81cm to
the breadth and width, to allow for heavy clothing. So the end
size may be 57.9cm33cm. G. Keith Still [11] considers that the
size provided by Fruin is measured in the low pedestrian
density flow, and he lost sight of the pedestrian position
adjustment and physical interaction in the high density. So it is
not suitable for the pedestrian model in congest pedestrian
flow.
The Green Guide [12] points out in the chapter about sport
field design that the critical safe density of pedestrian traffic is
4p/m2, when actual density is larger than the density, risk or
safety issue will exists in the crowd, But Milinskii AI [13]
surveyed many pedestrian flows in public assembly type
buildings, i.e. stadium, theatres and universities, and
discovered that Flow densities can vary greatly, with a
maximum flow density of approximately 9p/m2. In order to
study the maximum tenable density, Kopylov [14] also set the
condition to experiment the pedestrian flow of high density.
These measurements were made in an experimental
transforming arena, i.e. a full-scale model of an egress route
with variable corridor and door widths and in a specially
constructed 11m frame. Kopylov found that the most
frequent maximum flow density during movement along the
arena was 9~10/m2. The maximum possible density of people
located within the 11m frame under medical control, and thus
the tenable limit of density, even reaches 1314 p/m2 for
healthy young males(with an average body square 0.085m2).
So the size of cell in the model is 30cm30cm, especially is
suitable for evacuation.
The behaviors of the crowd behave differently in
different condition. In normal condition, the crowd is not
overly congested, so the density emerges in pedestrian flow
will not exceed safe level. But in urgent evacuation condition,
serious conflicts will exist in the crowd, and a strong force will
exert to crowd, so the density will become very high.
According to pedestrian size and the density in normal
condition, the size of 40cm40cm in the model for normal
condition will be enough. In order to include both conditions,

the size of 30cm30cm is more suitable, that is, the

corresponding density is 11p/m2. So the size of cell in the
model is 30cm30cm.
III.

PEDESTRIAN BEHAVIORS MODELING BASED

ON CA

A. Pedestrian Transition Possibility

Due to relatively uncomplicated environment, the goal of
pedestrian is single. So the rule is also simpler than open space
or hybrid space. The preference direction of pedestrian during
moving is mostly determined by the habitual walk behavior.
Some prevalent habit of pedestrian can be discovered by
observation and test. during walking, pedestrian do not easily
change his track; pedestrian prefer straight walking; when
pedestrian need to change his track, he will sidestep to right or
left side based on the crowd condition in front of the
pedestrian.
So in the model, it is assumed that the directions of
pedestrian is limited to left, through, and right. The directions
of diagonal and backward are ignored. According to the
preference of the three directions, the transition possibility is
obtained from the Fig.3. if there are three available directions,
the possibility of through is the largest0.8, and that of the
other two direction is 0.1. If there are two available directions,
the possibility of each one direction is 0.5. If there is only one
available direction, the possibility of the direction is 1.0.
0.1

0.5
0.8

0.5

0.1

0.5
0.5

0.5

0.5

1.0
1.0
1.0

Fig.3 assumed transition possibility

According to the states of current cells neighbor of the

three directions, there are seven choices during walking. The
possibility of transition is initial value at the beginning of
simulation. During simulation, the possibility is updated
according to density of the front traffic of each direction.
B. Collision Detection
Pedestrian need to observe other pedestrian and obstacle
in the neighbor region then ensure the walking direction of the
next step. At the same time, in order to avoiding collision with
other pedestrian or obstacle, the states of the cells in a certain
extent must be determined, then the distances between
pedestrians or between pedestrian and obstacle is computed.
At last, the velocity of current pedestrian is gotten. The
process is the so-called collision detection.
Before implementing the collision detection, the detected
extent must be defined. In order to reduce the time complexity,

3107

and reflect the experience of collision detection in the realistic

world, it is not necessary to take the whole scenario as
detected region. We just consider the interested extent in the
pedestrians vision field. The detected region is closely related
to psychological space. The method is usually applied in
continuous space model. Now, the method is applied in CA.
According to the actual experience, pedestrian not only
judge the states of the cells immediate to the current cell, but
also evaluate the density of interested region (namely,
pedestrian vision field).
In the paper, the pedestrian vision field is taken as a
4.5m4.5m rectangle, and the rectangle is divided into three
sub-regions (see in Fig.4). When the width of walkway is
smaller than 4.5m, or current pedestrian is near the wall or
other obstacles, there is no whole vision field, so some subregions are ignored too.

asynchronous update. So some simplification is made, e.g., the

cross of track of pedestrians is ignored.
Synchronous phenomenon exists in array march. The step
of every person in the array is almost same. So it is easy to
predict. But in most conditions, the behaviors of pedestrian are
random, and hard to predict. Asynchronous update loses sight
of the cross track, so the predict pedestrian traffic flow is
relatively high. However, the parallel arithmetic considers the
cross track and give away for conflict pedestrian, so the
predicted traffic flow is lower than that of Asynchronous
update.
C. update rule s
As above discussed, the parallel update is implemented in
term of asynchronous update. So each iteration simulation
consists of two stages. During the first stage, the collision
detect is finished, and according to transition possibility, lane
change is achieved too. During the second stage, all pedestrian
have their speed in their moving direction. The detail rules
refer to reference [6]. In the paper, an improvement should be
made to the judge of occupied state based on Blue, V J. when
the cell immediately to the left or right is free but the cell two
lanes over to the left or right is within the lattice and occupied
by a pedestrian, the state of the cellsee in Fig.5 to the left
or right can be decided by cast a dice. The states of the latent
conflict point may empty, or occupied by left or right
pedestrian.

Fig.4 pedestrian vision field in CA model

It is assumed that the transition possibilities of the left,

through, and right direction are respectively denoted
by m1 , m2 , m3 , and the density of I, II, and III sub-regions are
respectively denoted by k , k , and k . In the model, a scan
function implements the evaluation of density of the three subregions, at last, the transition possibilities can be modified
with equation (1)~(3):

m1k
m1km2 km3 k

(1)

m2 k
m2 =
m1km2 km3 k

(2)

m1 =

m3 k
(3)
m1km2 km3 k
In real world, pedestrian walk behavior behaves
differently. There are two main forms: synchronous and
asynchronous. Synchronous update means the neighbor
pedestrians make decision in the same time. When pedestrian
hops over several cells, especially involves cross track,
collision inevitably emerges. So we must look for method to
solve the collision. The method is that before conflicted point,
first, let the involved pedestrian all stop, then choose a
pedestrian go first, thus the pass rule turns into asynchronous
update. Due to the difficulty of parallel arithmetic, in the
program, the synchronous decision behaviors are simulated by
m3 =

Fig.5 latent conflict point

IV.

CASE STUDY

The pedestrian walkway is modeled as a closed loop ring

as described by Blue, V J [6]. The pedestrian walkway is
represented as a lattice of width W, length G, and the number
of cells is L=W*G. each cell is denoted by L(i, j ) where
1 i W and 1 j G .The size of cell is 30cm30cm. In the
example, the size of the walkway is 15 cells 50 cells (see in
Fig.6).

Fig.6 Unidirectional Pedestrian Walkway model and simulation scenario

3108

From pedestrian traffic survey in the subway station,

pedestrian speed varies from 0.9m/s to 2.1m/s. in order to
simplify the speed distribution, the speed of pedestrian is
divided into three classes: low, normal, and high. The
maximum speed of low class is 1.2 m/s, and the population is
15% of the total. The maximum speed of normal class is 1.5
m/s, and the population is 75% of the total. The maximum
speed of high class is 1.8 m/s, and the population is 10% of the
total. The simulation step is 1s in the model. So the speed of
pedestrian varies from 0 cell/step to 7 cell/step. The density of
pedestrian is set at start of each simulation, the value of which
is from 0.05 to 0.95 with an increment of 0.05. The simulation
runs for 6000 time steps with the first 1000 time steps used to
initialize the model. The statistics of the start-up period are
discarded.
The flow can be counted by a virtual detector. The
average speed can be computed by the speed of pedestrian of
each time step. So the parameter of fundamental curves can be
obtained. The model is implemented with java (see in Fig.6).
And the fitting result of speed-space fundamental curve is
given in the Fig.7

Average speed ( m / min )

90
80
70
60
50
40
30
20
10
0
0

0.5

1.5

2.5

3.5

4.5

Pedestrian space( m / p )
2

Fig.7 the fitted speed- space curve of simulation

V.

continuous model. In addition, when we are walking, we will

change our direction by evaluate the density of surrounding
crowd. The model is implemented based on Blue, V J rules,
the result of experiment shows that the model works well.
Though The CA model is simple, it can reflect the real world
to some extent.
In real world, environment is complex, and pedestrian is
so intelligent. The paper simplifies many factors, e.g. the
transition possibility, density function of speed and so on.
Those issues will further research in the future.
REFERENCES
[1] Matsushita and Okazaki (1991) A study of Simulation Model for Way
Finding Behavior by Experiments in Mazes. Journal of Architecture,
Planning. Environment Engineering. AIJ, No.429, pp. 51-59.
[2] Helbing, D. (1991) A mathematical model for the behavior of pedestrians.
Behavioral Science 36, 298-310.
[3] Cremer M, Ludwig J. A Fast Simulation Model for Traffic Flow on the
Basis of Boolean Operations [J].Mathematics and Computers in
Simulation,1986,28(4):297303.
[4] Nagel K, Schreckenberg M. A Cellular Automaton Model for Freeway
Traffic [J].J. Physique I.1992, 2(12):22212229.
[5] Biham O, Middleton A A, Levine D. Self-Organization and A Dynamical
Transition in Traffic Flow Models [J].Phys.Rev.A, 1992, 46(10-15):6124
6127.
[6] Blue, V J; Adler, J L, Emergent Fundamental Pedestrian Flows from
Cellular Automata Microsimulation, Transportation Research Record
1644, pp. 29-36,
[7] Wolfram S. (1986), Theory and Applications of cellular automata, World
Scientific, Singapore.
[8] ZHU Yu-xue, Cellular Automaton Model of physics system, Qinghua
University publish house, 2003, 8.
[9] Pheasant. S. Bodyspace. Taylor and Francis. 1998 (ISBN 0748403264).
ErgoBase from the Biomechanics Corporation of America, Inc. is
database of the world population sizes, dimensions and aerobic work from
Bodyspace.
[10] National Standard of the People's Republic of China People(GB1000088)(1988), human dimensions of Chinese adult.
[11]G.K. Still (2000), Crowd dynamics, PhD thesis, Warwick University, UK.
[12]The Green Guide (1997, Fourth Edition). Guide to Safety at Sports
Grounds. HMSO London..
[13]Milinskii AI (1951). The study of egress processes from public buildings
mass use. PhD thesis, Moscow Civil Engineering Institute.
[1] Kopylov VA (1974). The study of people motion parameters under
forced egress situations. PhD thesis, Moscow Civil Engineering
Institute.

CASE STUDY

The main aim in the paper is the introduction of the

factors of vision field and density into CA model. The vision
field is the property of human, which is often applied in

3109