Transportation Research Part A 39 (2005) 6173

www.elsevier.com/locate/tra

Pedestrian delay estimation at signalized intersections

in developing cities
Qingfeng Li *, Zhaoan Wang, Jianguo Yang, Jinmei Wang
Department of Industrial Automation, School of Electrical Engineering, Xian Jiaotong University,
West Xianning Road 28, Xian 710049, China

Abstract
This paper proposes a pedestrian delay model suitable for signalized intersections in developing cities, on
the basis of a eld study conducted in Xian, China. The eld study consisted of two parts: Part I involved
only one crosswalk, and the signal cycle was divided into 13 subphases; Part II involved 13 crosswalks, but
the signal cycles were only divided into green phases and non-green phases. It was found that pedestrian
arrival rates were not uniform throughout cycles; pedestrians arriving during green phases might also
receive delays; pedestrian signal non-compliance was so severe that delays were greatly reduced, but
non-complying pedestrians might still receive delays; and for pedestrians walking dierent directions,
though the relationships between average delay and arrival subphase were dierent, the overall average
delays were almost the same. On the basis of the eld study results, some assumptions are made about
the relationship between average pedestrian delay and arrival subphase, and a new model is developed
to estimate pedestrian delays at signalized intersections. The model is validated using the eld data, and
the validation results indicate that in Xian the new model provides much more accurate estimation than
the existing models.
2004 Elsevier Ltd. All rights reserved.
Keywords: Pedestrian delays; Signalized intersections; Developing cities; Signal non-compliance; Walking directions

Corresponding author. Tel.: +86 29 82664911; fax: +86 29 82667858.

E-mail address: [email protected] (Q. Li).

0965-8564/$ - see front matter 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tra.2004.11.002

62

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

1. Introduction
Delay has been widely accepted as the key performance index for signalized intersections, thus
it is necessary to estimate delays accurately for all kinds of trac participants at signalized intersections. Till now, much research has been conducted on vehicle delays at signalized intersections
under a broad range of conditions, but the research on pedestrian delays is much less. In Highway
Capacity Manual (HCM), sophisticated models are provided to estimate vehicle delays at signalized intersections, but the provided pedestrian delay model is quite simple (Transportation Research Board, 2000). This might because at signalized intersections vehicle delays are usually
regarded as more important than pedestrian delays. Another reason might be the diculty of estimating pedestrian delays accurately due to the exibility and diversity of pedestrians.
At signalized intersections, usually there are three kinds of phases for pedestrians: green phases,
clearance phases and red phases. Pedestrians are allowed to walk during green phases, but not allowed to walk during red phases. And during clearance phases, pedestrians are not allowed to enter
crosswalks, and those already on crosswalks are required to speed up. In western countries, the signals for clearance phases are usually ashing red or ashing Dont Walk. In China, a clearance
phase usually includes two parts: the signal is ashing green during the rst part, and the signal
is all-red during the second part, i.e., both the signals for pedestrians and conicting vehicles are red.
Presently, the most frequently used model to estimate pedestrian delays at signalized intersections is as follows:
C  G2
1
2C
where d is average pedestrian delay, C is cycle length, and G is green time. The model was developed under the assumptions of uniform pedestrian arrival rate, complete signal compliance, xed
cycle length, and no pedestrian actuation (Braun and Roddin, 1978). And models similar to Eq.
(1) were also developed in several other studies (Pretty, 1979; Rouphail et al., 1998).
Considering that some pedestrians may violate trac signals, another model was suggested:
d

C  G2
2
2C
where F is the fraction of pedestrians who arrive during non-green phases and comply with trac
signals. This equation assumes that pedestrians receive no delay if they violate trac signals
(Braun and Roddin, 1978).
In a study conducted in Australia, it was noted that most delay reductions were due to pedestrians who entered crosswalks during clearance phases. And the following model was proposed
(Virkler, 1998):
dF

C  G 0:69A2
3
2C
where A is clearance time.
The basic idea of Virklers model is to estimate eective red time more accurately by subtracting
several seconds from clearance time. Models similar to Virklers model were also developed in the
other two studies. In the model developed in another study conducted in Australia, eective red
d

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

63

time is estimated by subtracting only two seconds from clearance time; and in the model developed in a study conducted in USA, however, eective red time is estimated by subtracting ve
seconds from clearance time (Rouphail et al., 1998).
In UK, the models to estimate pedestrian delays at three kinds of crosswalks were developed
(Griths et al., 1984a,b,c, 1985). However, the crosswalks chosen in the study were those faraway
from intersections. Because of the inuence of vehicle turning movements, crosswalks at signalized intersections are somewhat dierent.
And according to the trac conditions in Cardi, UK, a microsimulation system was built to
study pedestrian delays at signalized intersections. And a pedestrian delay model was developed
through the analysis of simulation results. In addition to signal congurations, the model also
takes into accounts conicting vehicle ow rate and eective road width (Hunt and Al-Neami,
1995). However, since the model was developed using the simulation results, the applications of
the model might be limited.
All the above existing models were developed on the basis of the studies conducted in developed
cities. However, in developing cities like Xian, China, trac conditions are signicantly dierent.
Firstly, at signalized intersections in Xian, if there is no person specially assigned to manage
pedestrians, pedestrians arriving during non-green phases usually violate trac signals. And since
vehicles start and run slower, and drivers get used to the signal non-compliance of pedestrians,
pedestrians usually can cross successfully during non-green phases. Secondly, the existing models
ignore the delays received by pedestrians arriving during green phases, but in Xian, these pedestrians might also receive delays, because usually there are some conicting vehicles even if the signals for pedestrians are green. Thirdly, in Xian, signal congurations are also somewhat dierent,
e.g., signal cycles usually are longer (more than 100 s).
Because of the above dierences between developed cities and developing cities like Xian, the
models developed according to the trac conditions in developed cities might not be applicable to
developing cities. The ignorance of some inuence factors might cause only negligible estimation
errors in developed cities, but big errors in developing cities. Therefore, it is necessary to carry out
a study on pedestrian delays at signalized intersections in Xian.
Considering that walking directions might inuence pedestrian delays, pedestrians are divided
into two categories: downward-to-upward (D2U) pedestrians, who encounter the downward vehicle ow rstly, and upward-to-downward (U2D) pedestrians, who encounter the upward vehicle
ow rstly. Here the downward vehicle ow is the vehicle ow exiting an intersection, and the
upward vehicle ow is the vehicle ow entering an intersection.
In this paper, on the basis of a eld study conducted in Xian, a pedestrian delay model suitable
for signalized intersections in developing cities is proposed. The eld study consisted of two parts.
Part I was carried out only at one crosswalk, and the signal cycle was divided into 13 subphases.
The objective was to study the pedestrian arrival rate for each subphase and the delays of pedestrians arriving during each subphase. Part II of the eld study involved 13 crosswalks, but signal
cycles were only divided into green phases and non-green phases. The objective was to see if the
ndings of Part I of the eld study are widely applicable, and study the relationship between average pedestrian delay and conicting vehicle ow rate. On the basis of the eld study results, some
assumptions are made about the relationship between average pedestrian delay and arrival subphase, and a new pedestrian delay model is developed. Finally, the new model is validated using
the eld data, and compared with the existing models.

64

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

2. Field study, Part I

Part I of the eld study was carried out at only one crosswalk, and the signal cycle was divided
into 13 subphases. The objective was to study the pedestrian arrival rate for each subphase and
the delays of pedestrians arriving during each subphase. In the study, rstly the videos of a crosswalk were taken using a digital camcorder. And then original data were extracted manually from
the videos. Finally, statistics were obtained by processing the original data.
The intersection under study was the intersection of West Xianning Road and South Xingqing
Road in Xian, China, as shown in Fig. 1. And the crosswalk on the eastern leg of the intersection
was selected for study. To facilitate data collections, the segment under study was between lN and
lS, which were the outer lines of the dividing strips used to separate motorized vehicle lanes and
non-motorized vehicle lanes, and the length of the segment was 17.6 m. In this way, the measured
pedestrian delays would not include the delays occurring at roadside waiting areas and nonmotorized vehicle lanes. However, eld observations indicated that most delays occurred at the
segment under study, because pedestrians seldom waited at the roadside waiting areas, and
non-motorized vehicle ow rate was low when the videos were taken.
The phase plan adopted by the intersection was pre-timed two-phase control, and right-turnon-red (RTOR) was permitted at the intersection. The signal congurations of the crosswalk
are shown in Fig. 2. The cycle length was 123 s. For pedestrians using the crosswalk, the green
time was 45 s, the ashing green time was 2 s, and the red time was 76 s. And the red phase for
conicting vehicles began 2 s ahead of the green phase for pedestrians, but ended 2 s later. Therefore, the all-red time was 2 s, and the clearance time was 4 s. And then the green phase and the
ashing green phase were divided into 5 subphases (PG1-PG5), 10 s for each subphase, except that

Fig. 1. Intersection of West Xianning Road and South Xingqing Road in Xian, China.

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

65

Fig. 2. Signal congurations of crosswalk.

7 s for PG5 (the 2-s ashing green phase was included in PG5). The red phase was divided into 8
subphases (PR1-PR8), 10 s for each subphase, except that 6 s for PR1.
In the eld study, rstly 5 h were chosen to take videos in a tall building near the intersection.
The hours were selected within the non-peak periods (3:00 p.m.5:00 p.m.) of ordinary workdays
in July 2003, under the sunny or cloudy weather conditions.
And then, the arrival time, the departure time, and the nish time of each pedestrian were recorded manually by playing back the videos. The areas corresponding to the dividing strips were
dened as waiting areas. The time when a pedestrian arrived at a waiting area was dened as arrival time; the time when a pedestrian leaved the waiting area and started to walk was dened as
departure time; and the time when a pedestrian leaved another waiting area was dened as nish
time. Totally 1050 D2U pedestrians and 634 U2D pedestrians were recorded.
Next, the arrival subphase and the travel time of each pedestrian, and whether he/she violated
trac signal were determined by processing the original data. Since the phase plan was pre-timed,
arrival subphase could be determined according to arrival time. And travel time was taken as the
dierence between nish time and arrival time. If a pedestrians departure time fell into non-green
phases, he/she was regarded as a trac signal violator.
And delay is dened as the additional travel time a trac participant experienced (Transportation Research Board, 2000). In this study, a pedestrians delay was taken as the dierence between
his/her actual travel time and ideal travel time. The ideal travel time was determined by measuring
the average travel time of 30 randomly selected pedestrians, who were not inuenced obviously by
conicting vehicles. Conicting vehicles inuence pedestrians in two forms: they either cause delays or force pedestrians to increase walking speeds. For that reason, pedestrians running across
the crosswalk were excluded because usually they were forced to run. The measured ideal travel
time was 12.7 s.
The delay measurement method adopted in the study is dierent from the method adopted by
others (Virkler, 1998; Griths et al., 1984a). In others studies, usually only the roadside stopped
delays of pedestrians were measured. However, in the eld study, it was noted that a large percentage of delays occurred during crossings. Therefore, considering the trac conditions in Xian, the
delay measurement method adopted in this study is more accurate.
Finally, statistics were obtained by aggregating the data of all pedestrians. The pedestrian arrival
rate for each subphase, the mean and the standard variance of the delays of pedestrians arriving
during each subphase are presented in Fig. 3 (D2U pedestrians and U2D pedestrians are presented
respectively). And it was measured that 89.4% of D2U pedestrians arriving during non-green
phases and 74.4% of U2D pedestrians arriving during non-green phases violated trac signals.

66

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

Fig. 3. (a) Pedestrian arrival rate for each subphase, (b) average delay of pedestrians arriving during each subphase and
(c) standard deviation of delays of pedestrians arriving during each subphase.

Several conclusions could be drawn after analyzing Fig. 3. Firstly, pedestrian arrival rates were
not uniform throughout cycles, and arrival rates were the highest during PG1 and PR8. This
might because pedestrians could adjust their walking speeds according to trac situations before
entering the waiting areas. Secondly, pedestrians arriving during green phases might also receive
some delays. The average delay of these pedestrians was 1.9 s. Thirdly, delays were greatly reduced
because of the high signal non-compliance rate of pedestrians arriving during non-green phases,
but pedestrians violating trac signals might still receive delays.
And there were some dierences between the delays received by D2U pedestrians and those
received by U2D pedestrians. D2U pedestrians arriving during PR1 received big delays, while
U2D pedestrians arriving during PR1 received small delays. And the standard variance of the
delays of D2U pedestrians arriving during PR1 was big, because some pedestrians crossed immediately, but others had to wait long time; but the standard variance of the delays of U2D pedestrians arriving during PR1 was small, because most pedestrians crossed immediately. In
addition, the signal non-compliance rate of D2U pedestrians was 15.0% higher than that of
U2D pedestrians.
Nevertheless, for pedestrians walking dierent directions, the overall average delays were almost the same. The measured overall average delay was 8.5 s for D2U pedestrians, and 8.4 s
for U2D pedestrians.

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

67

Judging from the eld study results, it is obvious that the existing pedestrian delay models are
not applicable to the crosswalk under study. The estimated average delay is 24.7 s by Eq. (1), 2.6 s
for D2U pedestrians and 6.3 s for U2D pedestrians by Eq. (2), and 23.0 s by Eq. (3). The big errors
occur because these models were developed on the basis of several assumptions. These assumptions might be reasonable in developed cities, but they are not reasonable as the crosswalk under
study is concerned. Therefore, the existing models cannot provide accurate estimation in this case.

3. Field study, Part II

Part II of the eld study was also carried out in Xian. The study involved 13 crosswalks at 9
intersections adopting pre-timed two-phase control, but the signal cycles were only divided into
green phases and non-green phases. The studys major objective was to see if the conclusions
drawn in Part I of the eld study are widely applicable. In addition, since most pedestrians in
Xian are willing to violate trac signals if there are acceptable gaps, conicting vehicle ow rate
was expected to be the key factor inuencing pedestrian delays. Therefore, the relationship between average pedestrian delay and conicting vehicle ow rate was also studied.
The pedestrian ow rates at all selected crosswalks were within low to medium range (the highest one was 620 pedestrians/h). That is because the signal non-compliance of pedestrians and bicycle riders is so severe that the Xian city authorities have to assign persons to manage pedestrians
and bicycles at intersections where pedestrian and bicycle ow rates are high. If there are persons
managing pedestrians at signalized intersections, Eq. (1) might be able to estimate average pedestrian delay accurately.
Pedestrian data were collected using the method same with that used in Part I of the eld study.
Firstly a digital camcorder was used to take videos. At the crosswalk studied in Part I of the eld
study, two hours of videos were taken. One was taken during peak periods, and the other was
taken during non-peak periods. At the other 12 crosswalks, one hour of video was taken at each
crosswalk. And then, by playing back the videos, the arrival time, the departure time, and the nish time of each pedestrian were recorded. Next, the arrival subphase and the delay of each pedestrian, and whether he/she violated trac signal were determined by processing the original data.
Finally, by aggregating the data of all pedestrians, the number, the average delay, and the signal
non-compliance rate of pedestrians arriving during green phases and non-green phases were determined respectively. Totally 2404 D2U pedestrians and 2189 U2D pedestrians were recorded.
To record the arrival time, the departure time, and the nish time of each pedestrian, it was
necessary to dene waiting areas for each crosswalk. If there were dividing strips to separate lanes
for motorized vehicles and non-motorized vehicles, and crosswalk was near to the ends of the
dividing strips, the areas corresponding to the dividing strips were dened as waiting areas, or else
the ends of crosswalks were dened as waiting areas.
And pedestrian delay was still taken as the dierence between actual travel time and ideal travel
time. Actual travel time was the dierence between nish time and arrival time, and ideal travel
time equaled to crosswalk length divided by the ideal walking speed on crosswalks.
In principle, the ideal walking speed should be determined for each pedestrian respectively, but
that is not realistic. So it was assumed that the ideal walking speeds of all pedestrians were the
same. And the average walking speed of pedestrians not obviously inuenced by conicting

68

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

vehicles was taken as ideal walking speed. Four crosswalks at signalized intersections in Xian
were selected. And the conicting vehicle ow rates were high (about 0.6 vehicle/s) at two crosswalks, but low (about 0.2 vehicle/s) at the other two crosswalks. Thirty pedestrians were randomly
selected at each crosswalk. And the measured average walking speed of 120 pedestrians was
1.33 m/s.
Conicting vehicle data were collected using hand-held event recorders. Since all the intersections selected were controlled by pre-timed trac lights, a vehicles crossing subphase could be
determined by recording its time crossing the crosswalk. And then, by counting the numbers of
vehicles whose crossing subphases belong to green phases and non-green phases, the conicting
vehicle ow rate during green phases and non-green phases were determined respectively. Note
that bicycles and motorcycles were not recorded.
And two sets of data were extracted from each 30-minute-video. One set of data was for D2U
pedestrians, and the other was for U2D pedestrians. In this way, totally 56 sets of data were obtained. And the data characteristics are shown in Table 1.
Part II of the eld study indicated that the conclusions drawn in Part I of the eld study were
widely applicable to signalized intersections in Xian. Firstly, pedestrian arrival rates were not uniform throughout cycles. The average number of pedestrians arriving during green phases was
46.3% of the average overall number of pedestrians, but the average green time was 41.0% of
the average cycle length. This is similar to the results of the study conducted in Cardi, UK (Hunt
and Al-Neami, 1995). Secondly, pedestrians arriving during green phases might also receive delays. If green ratio is high, these delays might take a signicant portion of overall delays. Thirdly,
delays were greatly reduced because of the high signal non-compliance rate of pedestrians arriving
during non-green phases, but pedestrians violating trac signals might still receive delays.

Table 1
Characteristics of data collected in Part II of eld study
Parameter

Max.

Min.

Mean

Median

Std. Dev.

Cycle length (s)

Green time (s)
Clearance time (s)
Number of pedestrians
arriving during green phases
Number of pedestrians arriving
during non-green phases
Average delay of pedestrians
arriving during green phases (s)
Average delay of pedestrians
arriving during non-green phases (s)
Signal non-compliance rate of pedestrians
arriving during non-green phases (%)
Conicting vehicle ow rate during
green phases (vehicle/s)
Conicting vehicle ow rate during
non-green phases (vehicle/s)

149
70
15
101

90
32
4
11

120
49
8
38

122
46
8
34

18
12
3
20

93

44

39

24

6.2

0.5

2.1

1.9

1.6

31.1

0.3

12.3

12.0

7.6

100.0

41.9

86.7

91.0

14.2

0.46

0.02

0.22

0.22

0.11

1.19

0.27

0.63

0.70

0.28

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

69

Part II of the eld study conrmed that the overall average delays of pedestrians walking different directions were almost the same. The mean of overall average delays was 7.8 s for D2U
pedestrians, and 7.7 s for U2D pedestrians. And the dierence between the signal non-compliance
rates was also not signicant (averagely 88.0% for D2U pedestrians arriving during non-green
phases, and 85.4% for U2D pedestrians arriving during non-green phases).
Judging from the eld study results, it is clear that the assumptions made by the existing models
are not reasonable in Xian, thus the models might not be applicable here. In fact, in developing
cities like Xian, pedestrian crossings at signalized intersections could be regarded as unsignalized,
because most pedestrians have no respect toward trac signals at all. If a pedestrian is waiting, in
most cases he/she is waiting for an acceptable gap, but not waiting for a green signal. However,
though pedestrians do not respect trac signals, vehicles do, thus pedestrians conicting vehicle
ow periodically changes. Unfortunately, the existing delay models for unsignalized pedestrian
crossings were developed on the basis of the assumption that conicting vehicle ow is stable
(Rouphail et al., 1998). Therefore, it is necessary to build a new model to estimate pedestrian delays at signalized intersections in developing cities like Xian.
4. Model development
On the basis of the eld study on the average delay of pedestrians arriving during each subphase, it is assumed that the relationship between average pedestrian delay and arrival subphase
follows certain patterns, as shown in Fig. 4. The average delay of pedestrians arriving during eective green phases is constant. And at the beginning of eective red phases, average pedestrian delay is the biggest, and then linearly decreases.
And then, considering that pedestrian arrival rates are not uniform throughout cycles, the following model can be established to estimate the overall average delay by integrating the average
delay of pedestrians arriving during each subphase:
k NU kR2E
4
2C
where dG is the average delay of pedestrians arriving during green phases, kNU is the adjustment
factor for non-uniform arrival rate, k is the absolute value of the decreasing lines slope in Fig. 4,
and RE is eective red time.
To use the model, it is necessary to estimate dG, kNU, k, and RE. And kNU can be estimated
using the following equation:
d dG

Fig. 4. Relationship between average pedestrian delay and arrival subphase.

70

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

k NU

CnT  nG
nT C  G

where nG is the number of pedestrians arriving during green phases, nT is the overall number of
pedestrians. And by using the data collected in Part II of the eld study, kNU is estimated to be
0.92.
In Eq. (4), it is obvious that the accurate estimation of RE is critical for the accurate estimation
of average pedestrian delay. In eld studies, RE can be estimated by measuring the percentage of
no-signicant-delay pedestrians arriving during clearance phases. At the 13 crosswalks studied in
Part II of the eld study, 67% of pedestrians arriving during clearance phases received delay less
than 3 s. Therefore, RE can be estimated using the following equation:
RE C  G 0:67A

Note that in China, A includes both ashing green time and all-red time. The result is quite similar
to the results of several previous studies (Virkler, 1998; Rouphail et al., 1998).
And dG is expected to be mainly dependent on the conicting vehicle ow rate during green
phases, which is denoted by qG, but the assumption is surprisingly not supported by eld data
analysis. The calculated correlation between dG and qG is only 0.29. This might because the conicting vehicle ow rate during green phases is usually within low to medium range. Within this
range, though the increase of qG might increase pedestrian stopped delays, pedestrians usually
walk faster, thus overall delays might not increase. Besides, the average delay of pedestrians arriving during green phases is quite small, so it can be quite sensitive to other inuence factors, such as
the percentage of elderly pedestrians. This further weakens the correlation between dG and qG.
Since the correlation between dG and qG is rather weak, in using Eq. (4), according to the results
of Part II of the eld study, dG can be directly set to be 2.1 s.
k is expected to be mainly dependent on the conicting vehicle ow rate during non-green
phases, which is denoted by qNG. And since the eld study indicated that the correlation between
qNG and the overall conicting vehicle ow rate, which is denoted by q, was as high as 0.98, but q
is much easier to be estimated, the independent variable is changed from qNG to q.
According to Eq. (4), there is
k

2Cd  d G
k NU R2E

In Eq. (7), let kNU be 0.92, dG be 2.1 s, and then the k for each set of the data collected in Part II
of the eld study can be determined. And the following equation can be obtained by regression
analysis:
k 0:08 0:90q

where q 2 [0.15, 0.81]. The unit of q is vehicle/s. Note that bicycles and motorcycles should not be
counted in determining q.
And k is also expected to be dependent on the percentage of pedestrians willing to violate trac
signals if there are acceptable gaps, which is denoted by PW. If PW = 0, then k = 1. And when PW
is set, the bigger q, the closer k to 1; the smaller q, the closer k to 0. To satisfy these constraints,
k 1  1  0:08 0:90qP W

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

71

In a certain city, PW can be set at a xed value. In Xian, PW can be estimated to be 1; in other
cities, PW can be estimated by measuring the signal non-compliance rate when conicting vehicle
ow rate is quite low.
Note that though Eq. (4) is widely applicable, the equations to estimate the intermediate variables in Eq. (4) have been obtained using the eld data. Since the eld study only involved crosswalks where pedestrian ow rates were within low to medium range, the estimation provided by
the new model might be inaccurate at crosswalks with high pedestrian ow rates. In developing
cities like Xian, when there is no person specially assigned to manage pedestrians, pedestrians
are easier to establish precedence if pedestrian ow rates are high, thus average pedestrian delay
might be lower than the estimation provided by the new model.

5. Model validation and comparison

Actually, dierent pedestrian delay models were developed on the basis of dierent assumptions
about the relationship between average pedestrian delay and arrival subphase. Among the exisiting models, Eq. (1) can be obtained by assuming dG = 0, RE = C  G, and k = 1; Eq. (2) can be
obtained by assuming dG = 0, RE = C  G, and k = F; and Eq. (3) can be obtained by assuming
dG = 0, RE = C  (G + 0.69A), and k = 1.
The proposed new model has several advantages over the existing models, such as considering
non-uniform pedestrian arrival rate by introducing kNU, estimating RE according to the trac
conditions in developing cities like Xian, considering the delays of pedestrians arriving during

Fig. 5. Comparison of measured average delay and estimated average delay provided by (a) Eq. (1), (b) Eq. (2), (c) Eq.
(3) and (d) new model.

72

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

green phases by introducing dG, considering the delay reductions caused by signal non-compliance
during eective red phases by introducing k. However, if pedestrians arriving during green phases
receive no delay, pedestrian arrival rates are uniform throughout cycles, and signal compliance is
complete, the new model is identical to Eq. (1).
And then, the data collected in Part II of the eld study are used to validate the models, as
shown in Fig. 5. To facilitate comparison, the line where the measured average delay equals to
the estimated average delay is superimposed on each gure. And it is found that the estimation
provided by Eq. (1) is much higher, because it does not consider the delay reductions caused
by signal non-compliance at all. The estimation provided by Eq. (2) is much lower, because it does
not consider the delays received by non-complying pedestrians. And the estimation provided by
Eq. (3) is also much higher, because it does not consider the delay reductions caused by the signal
non-compliance during eective red phases. Comparatively, the estimation provided by the new
model is much more accurate.

6. Conclusions
A eld study on pedestrian delays was carried out at signalized intersections in Xian. The eld
study indicated that pedestrian arrival rates were not uniform throughout cycles; pedestrians
arriving during green phases might also receive delays; pedestrian signal non-compliance was
so severe that delays were greatly reduced, but non-complying pedestrians might still receive delays; and for pedestrians walking dierent directions, the relationships between average delay and
arrival subphase were dierent, but the overall average delays were almost the same.
And on the basis of the eld study results, a new pedestrian delay model has been developed.
The models most important improvement is the adequate consideration of the signal non-compliance during eective red phases. In addition, the new model also considers non-uniform pedestrian arrival rate, estimates eective red time according to the trac conditions in developing
cities like Xian, and considers the delays of pedestrians arriving during green phases. The model
has been validated using the eld data, and the validation results indicate that in Xian the new
model can provide much more accurate estimation than the existing models.
Note that the eld study did not include intersections with high pedestrian ow rates, because
persons are specially assigned to manage pedestrians at those intersections in Xian. At intersections with high pedestrian ow rates, if there are management persons, Eq. (1) might be able to
provide accurate estimation; if there is no management person, average pedestrian delay might be
lower than the estimation provided by the new model.

Acknowledgements
The authors thank Xuefeng Li, Lin Feng, Jian Li, Li Fang, Qiwu Ran, Jingzhou Cao, Wankun
Lu, Zhigang Zhai, and Haoming Deng for their great help in data collections, thank Haili Hao
and Quan Wang of Xian Trac Police Department for their assistance in data collections, and
thank Jinjun Liu, Mark R. Virkler, Shuguang Li, Min Yang, He Zhang, and Junfeng Zhang
for helpful discussions.

Q. Li et al. / Transportation Research Part A 39 (2005) 6173

73

References
Braun, R., Roddin, M., 1978. NCHRP Report 189: Quantifying the Benets of Separating Pedestrians and Vehicles.
Transportation Research Board, National Research Council, Washington, DC.
Griths, J.D., Hunt, J.G., Marlow, M., 1984a. Delays at pedestrian crossings: 1. Site observations and the
interpretation of data. Trac Engineering and Control 25, 355371.
Griths, J.D., Hunt, J.G., Marlow, M., 1984b. Delays at pedestrian crossings: 2. The development and validation of a
simulation model of a zebra crossing. Trac Engineering and Control 25, 505509.
Griths, J.D., Hunt, J.G., Marlow, M., 1984c. Delays at pedestrian crossings: 3. The development and validation of a
simulation model of a pelican crossing. Trac Engineering and Control 25, 611616.
Griths, J.D., Hunt, J.G., Marlow, M., 1985. Delays at pedestrian crossings: 4. Mathematical models. Trac
Engineering and Control 26, 277282.
Transportation Research Board, 2000. Highway Capacity Manual. National Research Council, Washington, DC.
Hunt, J.G., Al-Neami, A.H.K., 1995. Eectiveness of pedestrian facilities at signal controlled junctions. Proceedings of
the Institution of Civil Engineers, Transport 111 (4), 268277.
Pretty, R., 1979. The delay to pedestrians and vehicles at signalized intersections. ITE Journal 49 (5), 2023.
Rouphail, N., Hummer, J., Milazzo II, J., Allen, P., 1998. Capacity Analysis of Pedestrian and Bicycle Facilities:
Recommended Procedures for the Pedestrians Chapter of the Highway Capacity Manual. Federal Highway
Administration, Washington, DC.
Virkler, M., 1998. Pedestrian compliance eects on signal delay. Transportation Research Record 1636, 8891.