Transportation Research Part A 37 (2003) 789810

www.elsevier.com/locate/tra

A generalised function for modeling bi-directional ow

eects on indoor walkways in Hong Kong
William H.K. Lam *, Jodie Y.S. Lee, K.S. Chan, P.K. Goh
Department of Civil and Structural Engineering, The Hong Kong Polytechnic University,
Hung Hom, Kowloon, Hong Kong
Received 1 April 2002; received in revised form 16 October 2002; accepted 9 June 2003

Abstract
This paper examines the relationships between walking speed and pedestrian ow under various bidirectional ow conditions at indoor walkways in Hong Kong. The eects of bi-directional pedestrian ows
are investigated empirically with particular emphasis on their eects on walking time for dierent directions
of ow at pedestrian walkways in Hong Kong. Flow measurements were conducted at selected indoor
walkways in urban areas. A generalized walking time function that takes bi-directional ow distributions
(or ow ratios) into account is proposed for these pedestrian facilities and calibrated for various ow
conditions ranging from free-ow to congested-ow (at-capacity) situations. The bi-directional ow eects
on free-ow walking speed, eective capacity and at-capacity walking speed are validated with observed
data. It was found that the bi-directional ow ratios have signicant impacts on both the at-capacity
walking speeds and the maximum ow rates of the selected walkways but not on the free-ow walking
speeds. The ndings and study methodology provide better insight into the eects of bi-directional pedestrian ow characteristics and will assist engineers/planners in improving the design and operation of
pedestrian facilities not only in Hong Kong, but also in other countries as well.

2003 Elsevier Ltd. All rights reserved.
Keywords: Bi-directional ow eects; Walking speeds and maximum ow rates

1. Introduction
The need to study pedestrian ow characteristics for developing assignment or simulation
models for planning is recognized by researchers and authorities world-wide. Over the years,

Corresponding author. Tel.: +852-2766-6045; fax: +852-2334-6389.

E-mail address: [email protected] (W.H.K. Lam).

0965-8564/$ - see front matter
2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0965-8564(03)00058-2

790

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

many scientists have looked into the problem of pedestrian ow characteristics. Some early examples include Older (1968), who examined the pedestrian ow characteristics on footways (or
outdoor walkways) near shopping streets in London, United Kingdom, Oeding (1963), who
studied speed-ow relations in Germany, and Pushkarev and Zupan (1975), who compared the
ow characteristics of shoppers in London against that of commuters in New York, United
States. These early studies reported that observed speed-ow relations for pedestrians diered by
the location (Older, 1968), pedestrian type (Oeding, 1963) and trip purpose (Pushkarev and
Zupan, 1975).
Later studies on the pedestrian ow problem revealed that planners and engineers should take
into account the local pedestrian ow characteristics when planning for pedestrian facilities
(Seneviratne and Morrall, 1985a,b). This was further supported when comparisons of pedestrian
ow characteristics between Western (Canada) and Asian cities were made (Morrall et al., 1991).
Researchers looking into the ow characteristics in Singapore (an Asian city) suggested that
Asians tend to require less personal space and are more tolerant when this personal space is invaded when compared to their Western counterparts (Tanaboriboon et al., 1986). Tanaboriboon
and Guyano (1991) further recommended that local standards are required for the design of
pedestrian facilities in Asian countries. Local authorities or governments should thus not adopt
Western pedestrian design standards directly, as inherent dierences between Western and Asian
cultures as well as particular problems faced by individual countries may inuence pedestrians
ow characteristics.
The Hong Kong Government (Transport and Trac Survey Division, 1985) has recognized the
need to study local pedestrian characteristics since 1985. The need for the study of local pedestrian
characteristics is especially critical in Hong Kong, where the high population density and heavy
reliance on public transit modes by the majority of the population fuels the need for good estimates of pedestrian ow characteristics for planning and assignment purposes. A number of
studies on pedestrian ow characteristics had since been completed to fulll the need to understand pedestrian ow characteristics under various conditions and in dierent pedestrian facilities
in Hong Kong. This serves to recognize the fact that walking facilities, such as stairways have
dierent characteristics and thus, results obtained from one facility should not be generalized
without conducting additional studies.
Lam et al. (1995) conducted a comprehensive study on pedestrian ow characteristics in Hong
Kong. Pedestrian speed-ow relationships for a number of pedestrian facilities were obtained for
walkways (indoor and outdoor), signalized crosswalks, Light Rail Transit (LRT) crosswalks and
stairways in Hong Kong. It was found that mean walking distances for LRT and Mass Transit
Railway (MTR) in urban areas are shorter than those for the Kowloon Canton Railway (KCR).
A comparison between the mean walking speeds of various international cities showed that the
pedestrian characteristics in Hong Kong are similar to those in Singapore and other Asian cities,
but diered from those in Western cities (walking speeds were lower than those in London and
Calgary under normal conditions, but vice versa under congested conditions).
A subsequent study by Lam and Cheung (2000) extended the pedestrian speed-ow relationships to include dierent types of pedestrian facilities such as indoor and outdoor walkways,
signalized crosswalks (with and without mid-block pedestrian medians), crosswalks leading to
LRT stations (with and without signalization), and ve other types of walking facilities found at
railway stations. The pedestrian walking time functions at these facilities were calibrated and

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

791

pedestrian ow characteristics on these facilities compared. It was found that pedestrians walked
faster on outdoor than on indoor walkways, and walking speeds in commercial areas tended to be
higher than those in shopping areas. Pedestrians on crosswalks with mid-block pedestrian medians generally walked slower than those on crosswalks without mid-block pedestrian medians
while pedestrian ow characteristics for walking facilities were similar in KCR and MTR stations.
These dierences were partially attributed to dierences in the inherent characteristics particular
to the type of facility as well as the characteristics of the pedestrians using them. However, despite
the wide coverage of the aforementioned studies it should be noted that they were mainly conned
to studying pedestrian characteristics under uni-directional ow conditions.
The problem of bi-directional pedestrian ows and its eects on pedestrian ow characteristics
is one that has been gaining increasing recognition in pedestrian research. Unlike roadways where
vehicle ow is separated by direction, pedestrian walkways tend to operate in a bi-directional as
opposed to a uni-directional manner. This means that solutions derived from uni-directional ow
assumptions may not be able to properly describe pedestrian ows when bi-directional ow
conditions are prevalent. Examples of some recent studies that investigated the eects of bidirectional pedestrian ows included the study by Blue and Adler (1999), where the bi-directional
emergent fundamental pedestrian ows was simulated by using the cellular automata microsimulation method. Daly et al. (1991) calibrated the pedestrian speed-ow relationships for
walking facilities in London Underground (LU) stations and proposed eective capacities for
dierent bi-directional ow distributions (or directional distributions of pedestrian ows).
However, these proposed eective capacities have not been validated or supported by empirical
data. An empirical study of the bi-directional pedestrian ow characteristics in MTR stations had
also been conducted by Cheung and Lam (1997). They investigated the relationships between the
reduction of eective capacity on pedestrian facilities (for individual direction) and the directional
distribution of pedestrian ows (or ow ratios as dened in Section 2.4). They also studied the
eects of ow ratio on the reduction of at-capacity walking speed.
To extend the previous research on pedestrian ow characteristics in Hong Kong, this paper
investigates the pedestrian walking speed-ow relationships under various bi-directional ow
conditions ranging from free-ow to congested-ow (at-capacity) situations. It is believed that the
ndings could be used as a reference for design and improvement of the pedestrian walkways in
Hong Kong and/or other Asian cities with similar pedestrian ow characteristics, and can also be
extended to include any city in the world after calibrating the function for local conditions. A
generalized walking time function was proposed and calibrated for various ow conditions
ranging from free-ow to congested-ow (at-capacity) situations with the purpose of developing
an analysis tool for assessing the long-term impacts of pedestrian ows in walking facilities for
strategic planning.
This paper is organized as follows. Section 2 describes the data collection on selected sites in
Hong Kong and the data extraction process adopted. The eects of the bi-directional ows on a
number of pedestrian facilities are reviewed and briey discussed in Section 3. In Section 4, a
generalized walking time function that takes into account bi-directional ow distributions (or ow
ratios) for the pedestrian facilities mentioned in Section 3, is proposed and calibrated for various
ow conditions. The bi-directional ow eects on free-ow walking speed, eective capacity and
at-capacity walking speed are then investigated empirically. Finally, the conclusions are drawn in
Section 5 together with recommendations for further study.

792

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

2. Data collection and extraction

Hong Kong, with a land area of only 1095 km2 and a population of 6.73 million (2001 gures),
is one of the most densely populated cities in the world. With such a high population density,
there exists a great potential for conict between vehicles and pedestrians if eorts were not actively made to separate pedestrian and vehicular trac. It is therefore imperative for authorities in
Hong Kong to provide a safe walking environment for pedestrian movement in an urban area. In
order to do so, it is particularly important to study pedestrian preference and movement behavior.
However, the emphasis on strategic planning in this study means that variable selection, namely
ow ratio (a property of ow volume) and land use, is more or less pre-determined. Thus, less
emphasis is placed on individual variables governing pedestrian ows (like age, gender composition, etc.), which are aggregated and implicitly accounted for during the variable selection/calibration process.
Data was collected from two selected indoor walkways in Hong Kong within commercial and
shopping areas during peak hours at the study location. Table 1 shows the locations and land use
of the two selected sites together with the physical characteristics of the measurement sections of
the two walkways. A time-lapse photography technique was adopted from previous studies
(Cheung and Lam, 1997, 1998) to collect information on pedestrian walking times (or walking
speeds) and the corresponding pedestrian ows. With the use of video recording equipment, the
walking time and the corresponding pedestrian ow data were extracted from video records in the
laboratory with the aid of a computer (see Sections 2.1 and 2.2). To increase the precision of
the measurement, a time code with a precision of 0.04 s was mapped onto the video images before
data extraction. Three reference marks were made on the ground to form the pre-determined
measurement section. The measurement of pedestrian speed within the section is illustrated in Fig.
1. Photographs of the two selected indoor walkways (Shatin: shopping area and Wan Chai:
commercial area) are shown in Figs. 2 and 3 respectively.
2.1. Pedestrian ow data
To extract the pedestrian ow proles for the indoor walkways, the system clock of the
computer was synchronized to correspond to the time code mapped on the video records. This
was achieved with the aid of a computer program which was developed to complement the data
extraction process. The pedestrian ow for each minute was extracted by counting the number of
pedestrians passing through the centre line of the measurement section. The pedestrian ow
proles of the selected walkways were used for subsequent analysis.

Table 1
Physical characteristics of the selected indoor walkways
Facility

Location

Land use

Eective width
(m)

Length
(m)

Eective area
(m2 )

Indoor walkway

Shatin
Wan Chai

Shopping
Commercial

4.20
2.80

6.12
8.80

25.704
24.640

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

Entry Line

Time

0:00:30:1

Center Line

0:00:35:0

793

Exit Line

0:00:39:2

Fig. 1. A sample of a pedestrian walking through the measurement section.

Fig. 2. Indoor walkway in Shatin (shopping area).

Fig. 3. Indoor walkway in Wan Chai (commercial area).

2.2. Walking time data

A systematic sampling technique was used to collect the walking time data at a statistically
signicant level. A pedestrian was chosen every 2 s and the corresponding walking time of the

794

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

pedestrian was extracted from the tape. This was achieved by recording the entry and exit times of
the chosen pedestrian as he or she crosses the pre-determined measurement section of the walkway. The measurement method of the walking time for the chosen pedestrians and their corresponding pedestrian ows is presented in Fig. 4. The walking time prole was extracted with
respect to the corresponding ow prole and then stored in the computer le for subsequent
analysis.
2.3. Maximum pedestrian ow
The maximum pedestrian ow, which is also known as capacity, can be dened as the rate of
ow that can be repeatedly achieved during peak periods for which sucient demand exists and

Entry Line

Time Code

Center Line

0:00:30:15

0:00:35:02

Time

Pedestrian
Flow

31

50

32

57

33

53

34

51

35

61

N+29

38

N+30

41

Exit Line

0:00:39:24

Pedestrian
flow
profile
database

Fig. 4. Measurement of the walking times for the chosen pedestrians and their corresponding pedestrian ows.

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

795

that can be achieved on facilities with similar characteristics (HCM, 2000). Inherent limitations of
observational studies have made it necessary for researchers to develop a number of techniques to
estimate this value from eld observations. A few examples include the bimodal distribution
method (Cohen, 1983), expected extreme value method (Hyde and Wright, 1986) and the selected
maxima method (Minterhound et al., 1997). In this study, the selected maxima principle, which
uses the maximum ow rates measured over the observation period to estimate capacity, was
chosen because it proved to be the most suitable choice among the various methods considered.
The procedure involved the ranking of all per-minute ow rates which then allowed the
maximum ow rates (or observed capacities) for indoor walkway facilities to be determined (Lam
and Cheung, 2000). The exact procedure was aided by the use of two computer programs. The
rst program determined the observed maximum pedestrian ow rates from the ow prole
obtained at the indoor walkways, while the second program generated the rst ow rate from the
rst to 60th second; the second ow rate from the second to 61st second and so on. This procedure
was then continuously repeated to generate the minute-ow rates in 1-s increments.
In order to ensure that the maximum ow rates were indeed measured within the observation
period, care was taken to ensure that marked slowdowns as well as queue formation were indeed
observed within the boundaries of the walkway during the observation period. The slowdown in
walking speed within the walkway is a particularly important phenomenon as the maximum ow
of the facility (eective capacity) must occur at a point before queues form (as can be roughly
described by the fundamental ow diagram on trac ow) or shortly after observed walking
speed starts to decrease rapidly. In addition, many researchers (e.g. Fruin, 1987; Lam and Cheung, 2000) have also observed the pedestrian walking speeds remain fairly constant and then
decreased rapidly once a certain density has been reached. It can therefore be inferred that situations where pedestrian ows experience marked slowdowns, queue formation as well as a
corresponding drop in measured ow volumes are prime indicators that maximum capacity had
already been reached. Since a range of ow conditions ranging from free ow to short queue
formations were observed, it can be concluded that the maximum ow rate (or at least a relatively
close estimate) was derived in the study especially since they were observed to take place within
the boundaries of the walkway.
2.4. Flow ratio
The ow ratio (r) is the percentage of one-way pedestrian ow out of the total two-way pedestrian ow. The one-way ow is the number of pedestrians passing through a pre-determined
measurement section of a walking facility per meter width per minute in one direction. It is also
necessary to dene the major and minor ow directions of the pedestrian ow. Since the eects of
pedestrian ow in either direction of the walkway should be similar when the walkway is at with
no gradient change, the major ow direction would be the one where the ow ratio is greater than
or equal to 0.5. Correspondingly, the direction is said to be the minor ow direction when the ow
ratio (r) of that direction is less than 0.5. When the volume of pedestrian ow in both directions of
the walkway is known, the ow ratio (r) of individual ow direction can be dened as follows:
v1
P 0:5; for major flow direction
1
r
v1 v2

796

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

v2
< 0:5;
v1 v2

for minor flow direction

where v1 and v2 are the major and minor ows on the walkway respectively.

3. The bi-directional pedestrian ow eects on pedestrian facilities

The rst challenge of modeling a pedestrian network is that, unlike roadways where vehicle
ow is separated by direction, bi-directional walkways are the norm rather than the exception.
The reduction in eective capacity and walking speed due to opposing pedestrian ow in a bidirectional link was reported by Navin and Wheeler (1969), but not supported by empirical data.
These concepts of capacity and speed reduction seem to be supported by microsimulation experiments (Blue and Adler, 2001). Particularly, Blue and Adler (2001) used a cellular automata
approach and simulated the bi-directional ow mechanism according to a prescribed set of behavioral rules. They reported the reduction of walking speed and pedestrian volume over density
for various ow rates under the situations of separated ow, interspersed ow and dynamic
multiple lane ow. A basic nding in their experiments is that, for interspersed ow and dynamic
multiple lane ow, the reduction of walking speed and the reduction of pedestrian volume for a
given density decreases as the dierence between the directional split (ow ratio) approaches zero,
and reaches its minimum when the opposing ows on a bi-directional walkway are split 5050.
However, in the case of separated ow, ow rates seem to have little eect on the reduction of
walking speed and pedestrian volume over density.
Interestingly, these observations concurred with the empirical study by Lam and Cheung
(2000), who videotaped pedestrian movements at various walking facilities in Hong Kong.
Nonlinear regression analysis over the collected data showed that curves of capacity and speed
reduction for minor ows (with a less than 50% ow rate) argue for the following behavioral
rationale: People in a minor ow direction constantly need to switch positions with heading individuals of a large crowd. Conversely, pedestrians in the minor ow direction also have less
freedom in choosing directions to avoid conict. The disadvantage of traveling in a minor ow,
especially at small ow rates means that it is harder to form a stream and be separated from the
major opposing ow. When the minor ow rate increases from a very small percentage to 50%,
the chance for people from a minor ow direction to gain eective capacity by forming streams
increases, giving rise to the gradual dissipation of the interspersed ow condition to a dynamic
multiple lane ow, and eventually, to a separated ow condition.
Cheung and Lam (1997) observed the eects of bi-directional ow on passageways and stairways with an emphasis on ows under at-capacity condition. When there is bi-directional pedestrian ow on a two-way passageway, the eective capacity of the passageway for each direction
can be observed to correspond to the directional distribution of pedestrian ows (i.e. the ow
ratio, r). When the directional distribution of the pedestrian ow is 50:50 (i.e. r 0:5), the effective capacity of the passageway for each direction is approximately half the two-way capacity.
It implies that the pedestrians share the width of the passageway equally. Cheung and Lam (1997)
also reported that reductions in the walkway capacity and the at-capacity walking speeds in the
minor ow direction were observed. This reduction increased as the imbalance of directional split

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

797

of pedestrians increased. The bi-directional pedestrian ow eects on the stairways are more
signicant than those on the passageways under at-capacity conditions.
This paper aims to investigate the eects of bi-directional pedestrian ow on indoor walkways
for various ow conditions ranging from free-ow to congested-ow (at-capacity) situations.
There is an observed phenomenon of the bi-directional eects on the minor ow direction that
can be described as follows: When the number of pedestrians on a walkway is very low, they face
little opposing ow and have more freedom to choose their walking speeds and easily overtake
other people on the walkway. However, when pedestrian ow in the opposing direction is heavy,
they then have less freedom to choose their walking speeds as it is not easy to bypass other pedestrians. Thus, the eective capacity of a walkway and the walking speeds of the pedestrians will
be reduced when the opposing pedestrian ow increases. The bi-directional ow eects are varied
and become signicant when the pedestrian ows are close to the capacity of the two-way
walkway.

4. Calibration results
4.1. Speed-ow relationships for indoor walkways
In the previous research, Lam and Cheung (2000) investigated the relationships between
walking speed and pedestrian ow for dierent types of walking facilities in Hong Kong (including indoor walkways). These relationships are mainly conned to uni-directional ow conditions. Pedestrian walking speed can be found once the walking times and the corresponding
pedestrian ows at the selected pedestrian facility are known using the following equation:
sv

60
tv

where v is the pedestrian ow (ped/m/min); tv is the unit walking time (seconds) at pedestrian
ow v; and sv is the pedestrian walking speed (m/min) at ow v.
The well-known BPR function (Bureau of Public Roads, 1964) has been widely used for the
estimation of link travel times on road networks. It is also consistent with the function employed
in the PEDROUTE (Halcrow Fox and Associates, 1994)a pedestrian simulation software. The
formula is given below and is dened as the walking time function for a pedestrian link (or
walkway):

v n
4
tv t0 B 
C
where B and n are the parameters to be estimated; v is the pedestrian ow (ped/m/min); C is the
observed capacity of walkway under uni-directional ow condition (ped/m/min); t0 is the unit freeow walking time (seconds); and tv is the unit walking time (seconds) at ow v.
It should be noted that C is constant or xed under uni-directional ow conditions. In the
existing approach, such as the one adopted in the PEDROUTE software, the BPR function (Eq.
(4)) was used to estimate the walking times of passengers under uni-directional ow conditions. It
was assumed that there is no reduction in the capacity of the walking facility in the major ow

798

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

direction (i.e. 0:5 6 r 6 1:0). However, it was found that there were reductions in the capacities of
walking facilities in the minor ow direction (i.e. 0:0 < r < 0:5). Hence, walking speeds of passengers in the minor ow direction should be adjusted accordingly due to the reduction of capacities of walking facilities (Daly et al., 1991). Under bi-directional ow conditions, a step
function was incorporated in PEDROUTE to calculate the resulting eective capacities and
walking speeds in the minor ow direction (i.e. 0:0 < r < 0:5).
4.2. A generalised walking time function with bi-directional pedestrian ow ratio (GBPR)
The bi-directional ow eects on pedestrian facilities were described in Section 3. It should be
noted that the reduction in eective capacity and walking speed at capacity are mainly due to
conicts with the opposing pedestrian ow particularly for the minor ow direction (i.e. ow
ratio, r < 0:5). When the pedestrian ow for each direction of a two-way walkway is known, the
ow ratio (r) can be calculated using Eq. (1). To investigate the bi-directional ow eects on
walkways, Navin and Wheeler (1969) showed that the eects of even a small opposing ow will
considerably reduce the capacity of sidewalks. Subsequently, Fruin (1971, 1987) restated that the
reduction of the eective capacity on walkways should follow a step function with increasing
imbalance of directional split of pedestrians. This step function has been incorporated in the
pedestrian simulation modelPEDROUTE (Halcrow Fox and Associates, 1994). As a result of
this capacity reduction, the walking time increased as the walking speed decreased. However, the
concept of capacity reduction in the minor ow direction of the walkway has not been validated
with empirical data (Fruin, 1971, 1987).
Cheung and Lam (1997) used video cameras to collect observed data at various walking facilities in Hong Kong. On the basis of the collected data, they calibrated the polynomial regression
relationships between the reduction of eective capacity and the directional distribution of pedestrian ows (in terms of ow ratios) for passageways, stairways in ascending and descending
directions. They also proposed polynomial regression relationships between the reduction of atcapacity walking speed in the minor ow direction and the ow ratios for these facilities.
However, their research was only focused on the bi-directional ow eects on passageways and
stairways for the minor ow direction under congested-ow (or at-capacity) conditions. In this
paper, a new generalised walking time function with dierent bi-directional pedestrian ow ratios
(GBPR) was proposed to account for the bi-directional ow eects on pedestrian facilities under
various ow conditions (ranging from free-ow to congested-ow situations). To incorporate the
eects of ow ratio (r) on walking time, a preliminary statistical analysis was performed to determine the GBPR function (i.e. GBPRr; v=Ceff ). A statistical package, SPSS (Norusis, 1994),
was used to calibrate the GBPR functions. The parameters of the GBPR functions were estimated
using the non-linear regression technique. Initially several functional forms were adopted (i.e.
polynomial regression and power functions). However, these were eventually discarded as the
calibration results did not t well since the R2 values of the discarded functional forms were all less
than 0.5 (note that R2 is the coecient of determination, which reects the accuracy of the model
equation adopted). Correspondingly, a model that is able to obtain a R2 value that approaches 1
represents one that has better goodness of t.
Therefore, a generalized GBPR function was proposed for modeling the bi-directional ow
eects on pedestrian facilities at various ow conditions ranging from free-ow to congested-ow

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

799

situations. This function does endogenously consider the eects of the reduction of eective capacity under dierent directional distributions of pedestrian ows. The form of this newly proposed GBPR function is written as below:



n
v
v
m
5
GBPR r;
t0 B1  r 
Ceff
Ceff
where
Ceff r a0 a1  r1 a2  r2 a3  r3
B B1  r

t0 is the unit free-ow walking time (seconds); r is the ow ratio (0 6 r < 1) of walkway; v is the
one-way pedestrian ow (ped/m/min); Ceff r is the tted eective capacity of walkway at ow
ratio r; B1 , m and n are the parameters to be estimated; and GBPRr; Cveff is the unit walking time
(seconds) for pedestrian walking on walkway at ow v and ow ratio r.
Table 2 summarizes the calibration results of the GBPR functions for each of the selected
indoor walkways. The R2 values are all greater than 0.9. The calibration results and their eects
are described in the following. It should be noted that v is the one-way pedestrian ow. C is the
observed eective capacity of walkway under uni-directional ow while Ceff is the tted eective
capacity of walkway at ow ratio r (see Eq. (6)). When the ow ratio equals to 1 (i.e. r 1), then
Ceff approaches C. It should be noted that when the ow ratio equals to 1 (i.e. r 1), the
GBPRr; Cveff function (see Eq. (5)) is equal to the BPR function shown in Eq. (4) (as
m
B B1  r ), which is the function for the uni-directional ow condition.
Finally, it should be pointed out that there are four merits of this GBPR function:
The GBPR function takes account the bi-directional ow eects on walkways at various ow
conditions ranging from free-ow to congested situations;
The GBPR function also focuses on the bi-directional ow eects on walkway at dierent ow
ratios but not only in the minor ow direction at capacity;
The GBPR function takes a logical and simple form;
The GBPR function can be incorporated into pedestrian simulation models. In so doing, it allows the model to estimate pedestrian walking time under various ow conditions with dierent
bi-directional ow ratios, which represents an improvement to current models.

Table 2
The GBPR functions for indoor walkways
Walkway

Total
sample size

Parametersa
t0

B1

Indoor shopping
Indoor commercial

2829
3951

0.9978
0.9015

0.3595
0.2547

)0.2835
)0.2131

7.5453
10.001

a
b

R2b
0.9279
0.9166

t0 , B1 , m and n are the parameters dened in Eq. (5).

R2 is the coecient of determination, which is a measure to reect the accuracy of the function adopted.

800

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

4.2.1. Eects of the free-ow walking time t0

When pedestrian ow is very small (i.e. v=Ceff  0), the value of the GBPR function should then
be close to the free-ow walking time t0 (i.e. GBPRr; Cveff t0 ). Obviously, the free-ow walking
speed decreases as the free-ow walking time increases. Table 2 shows that free-ow walking speed
for indoor walkways in commercial area is relatively higher than that in shopping area. A number
of reasons can be advanced for this observation. Pedestrians in commercial areas are likely to have
a high motivational component to travel to and from work. Conversely, shoppers may not be
operating under such high time pressures and may choose to walk at a leisurely pace and/or be
slowed down by window-shopping in shopping areas. The trip characteristics of pedestrians in
commercial areas are also likely to be more homogenous than those in shopping areas as commuters in commercial areas tend to use the same route every day and are thus, more familiar with
the walking environment than those in shopping areas. In addition, inherent dierences in the
pedestrian ow characteristics in terms of the age and gender of pedestrians (a larger proportion of
child, elderly as well as female pedestrians are found in shopping areas), may also play a part in the
dierences in the observed walking speed. In addition, cultural norms (e.g. Asians tend to require
less space than their Western counterparts) and physiological dierences (e.g. stride-length) may
also play a part when comparisons between dierent countries are considered.
4.2.2. Eects of the ow ratio r
From Table 2, it can be seen that the value of the power term m of the ow ratio is negative (i.e.
rm 1=rm ). It implies that when the imbalance of directional pedestrian split increases
(0 < r < 1), the pedestrian walking time increases. Therefore, the walking speed decreases with the
increasing imbalance of directional pedestrian split. The bi-directional ow eect on the indoor
walkways in shopping areas (i.e. the power terms m of the ow ratio is )0.2835) is signicantly
greater than that in commercial areas.
4.2.3. Eects of the v=Ceff ratio
The value of the power term n of the v=Ceff ratio is positive. It implies that when pedestrian ow
increases, pedestrian walking time also increases. Therefore, walking speed decreases with the
increase of pedestrian ow. The curvature n of the GBPR function for walkways in the commercial area (i.e. n 10:001) is comparatively more non-linear than the one in the shopping area.
4.3. A generalised walking speed function
In this paper, a generalised walking speed (GWS) function is proposed for the selected walkways in commercial and shopping areas, which can be calibrated with the observed data. The
functional form of the GWS function is shown as below:


v
60




8
GWS r;
Ceff
GBPR r; Cveff


v
60

n
GWS r;
9

Ceff
t0 B1  rm  Cveff

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

801

where: GBPRr; Cveff is the unit walking time (seconds) for pedestrian walking on walkway at ow
v and ow ratio r; t0 is the unit free-ow walking time (seconds); r is the ow ratio (0 < r 6 1) of
walkway; v is the one-way pedestrian ow (ped/m/min); Ceff r is the tted eective capacity of
walkway at ow ratio r; B1 , m and n are the parameters to be calibrated; and GWSr; Cveff is the
walking speed (m/min) for pedestrian walking on walkway with ow v and ow ratio r.
The observed and tted relationships between pedestrian ow (one-way) and walking speed
under various ow ratios for the indoor walkway (shopping area) are presented in Figs. 5 and 6
respectively. The following observations could be made after a careful analysis of these two gures. As ow ratio decreases from 1.0 to 0.1, the at-capacity walking speed decreases. Moreover,
as the imbalance between the directional split of the ows increased (i.e. smaller ow ratio), the
maximum one-way pedestrian ow also decreased (i.e. the maximum pedestrian ow rate in the
minor ow direction).
Walking Speed
(m/min)

Walking Speed
(m/min)

80

80

70

70

60

60

50
50
40
0

40
1
10

0.8
20

0.6

30
40

One-way
Pedestrian Flow
(ped/m/min)

0.4

50
60

Flow Ratio
(r)

0.2
70
0

Fig. 5. Observed relationship between pedestrian ow and walking speed for various ow ratios at Shatin indoor
shopping area.

Walking Speed
(m/min)

Walking Speed
(m/min)

80

80

70

70

60

60

50
50
40
0

40
1
10

0.8
20

0.6

30

One-way
Pedestrian Flow
(ped/m/min)

40

0.4

50
60

Flow Ratio
(r)

0.2
70
0

Fig. 6. Fitted relationship between pedestrian ow and walking speed for various ow ratios at Shatin indoor shopping
area.

802

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

4.4. Bi-directional eects on indoor walkways in commercial and shopping areas

When pedestrians in a walkway face heavy opposing pedestrian ows, they usually weave
through the opposing pedestrians and would have little freedom in choosing their walking speeds.
Thus, the eective capacity of the walkway and the pedestrian walking speeds would be reduced
signicantly in the minor ow direction.
Therefore, the eects of bi-directional ow at walkway under various ow ratios (r) that would
be studied in this section are:
the free-ow walking speed;
the eective capacity of walkway; and
the at-capacity walking speed.
Generally, the bi-directional ow eects are signicant in the minor ow when the ow ratio on
a two-way walkway is much less than 0.5. When the ow ratio is close to 0.5, pedestrians in both
directions will eventually distribute themselves to minimize conict. However, when the ow ratio
approaches to 0.1, the width shared by pedestrians in the minor ow direction will be a proportion
that is less than 10% of the eective width of walkway. It is because at this ow ratio, pedestrians
in the minor ow direction would experience greater conict and have less freedom to choose their
walking speeds due to the heavy opposing ow. As a result, both walking speed and eective
capacity for the walkway will be reduced signicantly for the minor ow direction.
4.4.1. Free-ow walking speed
Fig. 7 shows, in a graphical manner, the variations of the free-ow walking speed for indoor
walkways for shopping and commercial areas under dierent ow ratios. It can be observed that
the free-ow walking speed at indoor walkways in commercial areas is higher than those in
shopping areas. It can also be seen that there is no signicant dierence between the free-ow
walking speeds under various ow ratios (r). Therefore, the ow ratio or the bi-directional pedestrian ow does not aect the free-ow walking speed, particularly when the pedestrian ow is
very low.
4.4.2. Eective capacity
The observed and estimated maximum ow rates (or eective capacities Ceff ) of these selected
walkway facilities under dierent ow ratios are summarized in Tables 3 and 4. It can be observed
that the eective capacities of the selected walkways decreased with the increasing imbalance of
the directional pedestrians split (i.e. smaller ow ratio). For example, it can be seen in Tables 3
and 4 that the eective capacities of indoor walkways in shopping and commercial areas are 68.0
and 75.0 ped/m/min respectively when the ow ratio is equal to 1.0. These eective capacities are
reduced to 56.1 and 66.7 ped/m/min respectively when the ow ratio is 0.1. The reduction of the
eective capacity with increasing imbalance of the directional pedestrian split can be partially
explained by the observed phenomenon (see Section 3) between the minor ow and the heavy
opposing ow. As a result, the eective capacity of the walkway is reduced when the ow ratio is
deceased from 1.0 to 0.1 particularly in the minor ow direction.

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

Indoor Shopping (Observed)
Indoor Shopping (Fitted)

803

Indoor Commercial (Observed)

Indoor Commercial (Fitted)

Free Flow Walking Speed (m/min)

70
68
66.09

66.59

66.84

66.97

67.33

66.93

67.46
66.85

67.00

66.41

66
64
62

61.78

61.11

60.21

60

59.62

59.73

59.98

59.82

59.31

58.89

58.58

58
56
0.0

0.1

0.2

0.3

0.4

0.5
0.6
Flow Ratio (r)

0.7

0.8

0.9

1.0

Fig. 7. The free ow walking speed against ow ratio.

Table 3
Observed and estimated eective capacities of indoor walkway in shopping area under various ow ratios
Flow ratio

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
R2a
MAEb
MAPEc
a

Observed eective
capacities

Estimated eective capacities

Newly proposed
polynomial regression function

Polynomial regression
function by Lam and
Cheungs (2000)

Step function
proposed by Fruin
(1971, 1987)

56.1
61.3
63.3
65.0
66.8
67.2
67.4
67.6
68.0
68.0

57.3
60.9
63.5
65.4
66.6
67.3
67.7
67.8
67.8
67.8

59.0
62.7
65.5
67.2
67.9
68.0
68.0
68.0
68.0
68.0

58.1
61.2
64.6
64.6
68.0
68.0
68.0
68.0
68.0
68.0

0.9900
0.3412
0.55%

0.9692
1.5327
1.85%

0.9673
0.6772
1.08%

R2 is the coecient of determination, which is a measure to reect the accuracy of the function adopted.
P
jestimatedobservedj
b
MAE is the mean absolute error
.
10
P jestimatedobservedj
100%

c
observed
MAPE is the mean absolute percentage error
.
10

804

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

Table 4
Observed and estimated eective capacities of indoor walkway in commercial area under various ow ratios
Flow ratio

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
R2a
MAEb
MAPEc

Observed eective
capacities

Estimated eective capacities

Newly proposed
polynomial regression function

Polynomial regression
function by Lam and
Cheungs (2000)

Step function
proposed by Fruin
(1971, 1987)

66.7
69.2
71.2
73.4
73.8
74.3
74.6
74.8
74.8
75.0

66.6
69.4
71.5
73.0
74.0
74.5
74.7
74.8
74.8
74.8

65.1
69.1
72.2
74.2
74.8
75.0
75.0
75.0
75.0
75.0

64.1
67.5
71.3
71.3
75.0
75.0
75.0
75.0
75.0
75.0

0.9941
0.1750
0.24%

0.9703
0.5975
0.84%

0.9493
0.9250
1.31%

R2 is the coecient of determination, which is a measure to reect the accuracy of the function adopted.
P
jestimatedobservedj
b
MAE is the mean absolute error
.
10
P jestimatedobservedj
100%

c
observed
MAPE is the mean absolute percentage error
.
10

4.4.3. At-capacity walking speed

Table 5 shows the observed and estimated at-capacity walking speeds of the selected walkways
under dierent ow ratios. It can be observed that the at-capacity walking speeds of these facilities
decreased with the increasing imbalance of the directional pedestrian split. For example, the atcapacity walking speeds for indoor walkways in shopping and commercial areas are 43.0 and 51.0
m/min respectively when the ow ratio is equal to 1.0. They are reduced to 35.9 and 42.5 m/min
respectively when the ow ratio is 0.1. The reduction of at-capacity walking speed, corresponding
to the increasing imbalance of the directional pedestrians split, can be partially explained by the
fact that the smaller volume of minor ow is dominated by the heavy opposing ow. As a result,
pedestrians in the minor ow are forced to give way to the oncoming pedestrian stream. Hence,
the at-capacity walking speed is decreased signicantly in the minor ow direction when the ow
ratio is reduced from 0.5 to 0.1.
It was found that the bi-directional ows (or the ow ratios) have signicant impacts on both
at-capacity walking speeds and observed maximum ow rates (or the eective capacities) of the
selected indoor walkways but not the free-ow walking speeds. The eective capacity and atcapacity walking speed decreases with increasing imbalance of the directional split of pedestrians,
i.e. the ow ratio from 1.0 to 0.1. These eects become more signicant in the minor ow direction
when the ow ratio is close to 0.1. This can be partially due to the observed phenomenon (see
Section 3) between the minor ow and the heavy opposing ow. As a result, pedestrians in the
minor ow direction do not have adequate space to weave through the heavy opposing ow.

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

805

Table 5
Observed and estimated at-capacity walking speeds under various ow ratios
Flow ratio

At-capacity walking speeds at indoor

shopping walkway

At-capacity walking speeds at indoor

commercial walkway

Observed

Estimated

Observed

Estimated

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0

35.9
39.0
40.7
41.9
42.3
42.4
42.7
42.9
42.9
43.0

37.2
38.8
40.3
41.3
41.7
41.9
42.7
43.4
43.8
43.9

42.5
45.8
48.5
49.9
50.5
50.7
50.8
50.9
51.0
51.0

43.4
46.1
48.7
49.5
50.0
50.6
50.7
51.2
51.5
51.9

R2a
MAEb
MAPEc

0.9063
0.5920
1.46%

0.9726
0.4204
0.87%

R2 is the coecient of determination, which is a measure to reect the accuracy of the function adopted.
P
jestimatedobservedj
b
MAE is the mean absolute error
.
10
P jestimatedobservedj
observed 100%

c
MAPE is the mean absolute percentage error
.
10

Therefore, the eective capacity of walkway and at-capacity walking speed have been reduced
signicantly when the ow ratio is close to 0.1.

5. Comparisons
Table 3 shows the observed and tted eective capacities of indoor walkway in shopping area
under various ow ratio conditions. The tted values were estimated by the following three
functions: (1) newly proposed polynomial regression function; (2) adapted polynomial regression
function proposed by Cheung and Lam (1997); and (3) step function proposed by Fruin (1971,
1987). The R2 , MAE and MAPE of the estimates of these three functions are also given in Table 3.
Note that R2 is the coecient of determination, which reects the accuracy of the model equation
adopted (a better t is obtained the closer R2 is to 1). MAE is the mean absolute error between the
estimated and observed data, where a better t is obtained as MAE value approaches 0. MAPE is
the mean absolute percentage error between the observed and estimated data. Once again, a value
of MAPE that approaches 0% represents a better tting model.
It can be seen that R2 of these three functions are 0.9900, 0.9692 and 0.9673 respectively.
Obviously, the R2 of the newly proposed polynomial regression function is higher than that obtained for the other two functions. This indicates that the observed eective capacities of indoor
walkways in shopping areas in Hong Kong are better tted with this newly proposed function

806

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

(Eq. (6)). It is also shown that the smallest values of both MAE and MAPE of are obtained when
the newly proposed polynomial regression function was used (i.e. 0.3412 and 0.55% respectively).
In view of the above, it may be concluded that the newly proposed polynomial function for estimating eective capacity of indoor walkway in Hong Kong shopping area is comparatively more
accurate than the other two functions.
Table 4 shows the tted eective capacities of indoor walkway in Hong Kong commercial area
obtained using these three functions. It can be seen that R2 values of these three functions are
0.9941, 0.9703 and 0.9493 respectively. Similar to the above results for indoor walking in shopping
area, the R2 of the newly proposed polynomial regression function (Eq. (6)) is the largest for the
indoor walkway in Hong Kong commercial areas. The MAE and MAPE of the newly proposed
polynomial regression function are also the smallest (i.e. 0.1750 and 0.24% respectively). This
implies that the newly proposed polynomial function for estimating eective capacity of indoor
walkways in Hong Kong commercial areas is also comparatively more accurate than the other
two functions.
The observed and estimated eective capacities of indoor walkways in shopping and commercial
areas under various ow ratios (Hong Kong) are superimposed with the step function derived by
Fruin and the polynomial regression function proposed by Cheung and Lam (1997) in Fig. 8. The
latter is a polynomial regression function for estimating the eective capacities of passageways in
Hong Kong MTR stations under dierent ow ratios, which has been adapted for indoor

Step Function (Fruin, 1971 & 1987)

Polynomial Regression Function (Shopping Area)
Indoor Shopping Walkway (Observed)

Polynomial Regression Function (Cheung & Lam, 1997)

Polynomial Regression Function (Commercial Area)
Indoor Commercial Walkway (Observed)

76

Ceff = 52.61 + 52.36 r 60.18 r 2 + 23.02 r 3

Effective Capacity (ped/m/min)

72

Ceff = 75 1 (0.19 0.65 r + 0.26 r 2 + 0.44 r 3 + 0.61 r 4 0.99 r 5 + 0.33 r 6 )

68

64

Ceff = 62.98 + 40.97 r 46.87 r 2 + 17.66 r 3

60

56

52
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Flow Ratio (r)

Fig. 8. Comparison on eective capacities of indoor walkways under various ow ratios.

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

807

walkways in shopping areas. It is pointed out that this paper also proposes a new polynomial
regression function (Eq. (6)) for modeling the eective capacity of indoor walkways in Hong
Kong.
When the ow ratio falls in the range of 0:0 < r < 0:5, it can be seen in Fig. 8 that the proposed
tted curve for eective capacity of indoor walkways in shopping areas is similar in shape to the
adapted function based on the one proposed by Cheung and Lam (1997) for passageways in MTR
stations. However, the step function suggested by Fruin (1987, 1971) may lead to over-estimation
or under-estimation of the eective capacities for ow ratios between the steps.
When ow ratio r 0:5, it can be observed in Fig. 8 that the eective capacities estimated by
these three curves are very close to each other. However, when the ow ratio is greater than 0.5
and falls in the range of 0:5 < r 6 1:0, the eective capacity estimated by Fruins function is very
close to that predicted by the polynomial regression function (Cheung and Lam, 1997). It is
because the eective capacity reected by the two curves for ow ratios ranging from 0:5 < r 6 1:0
was more or less constant. Nevertheless, it was found that some reductions in the eective capacity
of indoor walkways occurred in both shopping and commercial areas even when the ow ratio
was in the range of 0:5 < r 6 1:0.
Table 6 gives a comparison of the observed maximum ow rates and at-capacity walking speeds
obtained for indoor walkways in Hong Kong and those in the MTR, KCR and LU Stations. The
observed maximum ow rates of indoor walkways in both shopping and commercial areas in
Hong Kong are 68 and 75 ped/m/min respectively, which are the lowest among the passageways in
MTR, KCR and LU stations (i.e. 92, 88 and 86 ped/m/min respectively). This can be partially
explained by the fact that MTR, KCR and LU, which are major transport modes, may function
as a special case where pedestrians exhibit a greater amount of tolerance to the invasion of
personal space while inside the stations, like in the case of passengers in lifts where space invasion
rules are temporarily suspended (Fruin, 1987). The observed maximum ow rate of passageway in
Hong Kong MTR station is the largest while that of indoor walkways in shopping area in Hong
Kong the lowest. On the other hand, the at-capacity walking speeds of the Hong Kong indoor

Table 6
International comparison of observed maximum ow rates and at-capacity walking speeds
City

Observed maximum ow rate

(ped/m/min)

At-capacity walking speed

(m/min)

68.00
75.00

43.00
51.01

92.00

36.75

Hong Kong
Kowloon-Canton Railway

88.00

35.98

United Kingdom
London Underground

86.00

36.00

Hong Kong
Shopping area
Commercial area
Passageway
Hong Kong
Mass transit railway

808

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

walkways for shopping and commercial areas are 43.00 and 51.01 m/min respectively. They are
much higher than those of the passageways in MTR, KCR and LU stations (i.e. 36.75, 35.98 and
36.00 m/min respectively). This can be explained by the observed maximum ow rates of the Hong
Kong indoor walkways, which are smaller than those observed in passageways in these railway
stations. It was found that the at-capacity walking speed of passageway in KCR is the lowest
while that of Hong Kong indoor walkway in commercial area is the highest.

6. Conclusions
The walking speed of pedestrians is of prime importance in modeling pedestrian movements
and designing of pedestrian facilities. In this paper, the bi-directional eects on free-ow walking
speed, eective capacity and at-capacity walking speed on indoor walkways are investigated
empirically. A new generalized walking time function with bi-directional pedestrian ow ratio
(GBPR) is calibrated for various ow conditions ranging from free-ow to congested-ow (atcapacity) situations, on the basis of the data observed at the indoor walkways in Hong Kong
urban areas.
The estimated eective capacities were compared to those obtained using the step function
derived by Fruin (1971, 1987). It was found that the proposed generalized function is better than
the step function theoretically because it is a continuous function which can be more accurate. In
particular, the eective capacity estimated by the proposed generalized function increases gradually and represents an improvement to the step function, which is prone to estimation errors
when ow ratios are found between incremental steps.
Analyses conducted on the eects of bi-directional pedestrian ow on indoor walkways revealed that no signicant impact of bi-directional ow was detected on the free-ow walking
speed, which agreed with the observations by Fruin (1971, 1987). However, the eective capacity
as well as the at-capacity walking speeds decreased with increasing imbalance of directional split
of pedestrians. In particular, these bi-directional eects are particularly signicant in the minor
ow direction. Bi-directional pedestrian ow eects were also found to be more signicant at
indoor walkways in shopping areas than in commercial areas.
The proposed generalized function allows bi-directional ows to be modeled yet retains the
ability to represent uni-directional ow eects (by setting the ow ratio r 1). The proposed
generalized walking time function, which is meant for the purpose of strategic planning, is expected to nd applications in the planning and design of pedestrian networks, which classies
pedestrian characteristics by land use but not by age or other factors and represents a evolutionary contribution to an active eld of research and is a resource for those who are developing
pedestrian models in practice. It is believed that similar studies should be carried out at dierent
walking facilities so as to enhance the pedestrian modeling for the design and operation of
walking facilities. Findings derived from this research are expected to be directly applicable in
Hong Kong or in other Asian Cities with similar pedestrian characteristics. In addition, the approach used is also expected to be applicable to any city in the world once local conditions have
been accounted for. Note that factors such as the composition of ow with respect to factors such
as age and gender are considered implicitly during the calibration of the generalized ow function
under the land use variable.

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

809

Acknowledgements
This research is jointly supported by a granted from the Research Grant Council of the Hong
Kong Special Administration Region awarded to the Hong Kong Polytechnic University (project
no. PolyU 5043/99E), a Postgraduate Studentship from the Research Committee of The Hong
Kong Polytechnic University and 5046/00E.

References
Blue, V.J., Adler, J.L., 1999. Bi-directional emergent fundamental pedestrian ows from cellular automata microsimulation. Transportation and Trac Theory, 235254.
Blue, V.J., Adler, J.L., 2001. Cellula automata microsimulation for modeling bi-directional walkways. Transportation
Research B 35, 293312.
Bureau of Public Roads, 1964. Trac Assignment Manual. US Department of Commerce, Urban Planning Divisions,
Washington, DC.
Cheung, C.Y., Lam, W.H.K., 1997. A study of the bi-directional pedestrian ow characteristics in the Hong Kong mass
transit railway stations. Journal of the Eastern Asia Society for Transportation Studies 2 (5), 16071619.
Cheung, C.Y., Lam, W.H.K., 1998. Pedestrian route choices between escalator and stairway in Hong Kong mass transit
railway stations. Journal of Transportation Engineering, ASCE 124 (3), 277285.
Cohen, S., 1983. Methodes destimation de la Capacite des Autoroutes. INRETS, rapport de recherche 62 (in French).
Daly, P.N., McGrath, F., Annesley, T.J., 1991. Pedestrian speed/ow relationships for underground stations. Trac
Engineering and Control 32 (2), 7578.
Fruin, J., 1971. Designing for Pedestrians: A Level of Service Concept. Metropolitan Association of Urban Designer
and Environmental Planners Inc., New York, USA.
Fruin, J., 1987. Pedestrian Planning and Design, second ed. Elevator World Inc., New York, USA.
Halcrow Fox and Associates, 1994. User Guide to PEDROUTE 3.15. Halcrow Fox, London.
Highway Capacity Manual, 2000. Transportation Research Board, National Research Council, Washington, DC.
Hyde, T., Wright, C., 1986. Extreme value methods for estimating road trac capacity. Transportation Research B 20,
125138.
Lam, W.H.K., Cheung, C.Y., 2000. Pedestrian speed/ow relationships for walking facilities in Hong Kong. Journal of
Transportation Engineering, ASCE 126 (4), 343349.
Lam, W.H.K., Morrall, J.F., Ho, H., 1995. Pedestrian Flow Characteristics in Hong Kong. Transportation Research
Record, 1487. Transportation Research Board, National Research Council, Washington, DC, pp. 5662.
Minterhound, M.M., Botma, H., Bovy P.H.L., 1997. Assessment of Roadway Capacity Estimation Methods.
Transportation Research Record, 1572. Transportation Research Board, National Research Record Council,
Washington, DC, pp. 5967.
Morrall, J.F., Ratnayaka, L.L., Seneviratne, P.N., 1991. Comparison of Central Business District Pedestrian
Characteristics in Canada and Sri Lanka. Transportation Research Record, 1294. Transportation Research Board,
National Research Record Council, Washington, DC, pp. 5662.
Navin, P.D., Wheeler, R.J., 1969. Pedestrian ow characteristics. Trac Engineering, 3036.
Norusis, M.J., 1994. SPSS Advance Statistics 6.1. Chicago, IL.
Oeding, D., 1963. Trac Loads and Dimensions of Walkways and Other Pedestrian Circulation Facilities. Strassembau
and Strassenverkehrstechnik, No. 22, Bonn.
Older, S.J., 1968. Movement of pedestrians on footways in shopping streets. Trac Engineering and Control 10, 160163.
Pushkarev, B., Zupan, J.M., 1975. Urban Space for Pedestrian. The MIT Press.
Seneviratne, P.N., Morrall, J.F., 1985a. Analysis of factors aecting the choice of route of pedestrians. Transportation
Planning and Technology 10, 147159.
Seneviratne, P.N., Morrall, J.F., 1985b. Level of service on pedestrian facilities. Transportation Quarterly 39 (1), 109
123.

810

W.H.K. Lam et al. / Transportation Research Part A 37 (2003) 789810

Tanaboriboon, Y., Guyano, J.A., 1991. Analysis of Pedestrian Movements in Bangkok. Transportation Research
Record, 1294. Transportation Research Board, National Research Record Council, Washington, DC, pp. 5662.
Tanaboriboon, Y., Hwa, S.S., Chor, C.H., 1986. Pedestrian characteristics study in Singapore. Journal of
Transportation Engineering ASCE 112 (3), 229235.
Transport and Trac Survey Division, 1985. Data Record No. 393, Model for Pedestrian MovementLiterature
Review. The Government of the Hong Kong Special Administrative Region.