You are on page 1of 10

Modeling Pedestrian Delays at Signalized

Intersections as a Function of Crossing


Directions and Moving Paths
Jing Zhao and Yue Liu

Pedestrian delay is a key performance indicator for evaluating the level one-stage crossings based on uniform arrival rates. Many researchers
of service for pedestrians at signalized intersections. Although much is have modified this classic delay calculation model to enhance its
known about the pedestrian delay of a signalized crosswalk, the existing accuracy. Some factors that were considered include the following:
model in the Highway Capacity Manual 2010 cannot provide the necessary
accuracy for estimating the pedestrian delay of the diagonal crossing 1. Influence of pedestrian–vehicle conflict. Even though motorists
(crossing to the diagonally opposite corner of the intersection in two are legally required to yield to pedestrians at crosswalks under most
stages) and the entire intersection. In this paper, a new pedestrian con- circumstances, actual motorist yielding behavior varies considerably
trol delay model is proposed: the model considers the diagonal crossing and is influenced by many factors, such as the pedestrian crossing
and moving paths. The proposed model is validated on the basis of treatments and roadway geometries (16, 17). Therefore, a pedes-
field measurements. The main factors affecting the pedestrian control trian delay model considering pedestrian–vehicle interaction was
delay during diagonal crossing are discussed. Results reveal that the proposed on the basis of field surveys conducted in metropolitan
proposed model is promising in increasing the estimation accuracy of Sydney, New South Wales, Australia (18). The pedestrian delay is
the pedestrian control delay of the diagonal crossing and the entire considered in two parts: (a) the delay of pedestrians arriving during a
intersection (approximately 20%). The delay of the diagonal crossing vehicular bunch and (b) the delay of those arriving during the random
increases with an increase in the time gap of the green light between flow between bunches. These two parts are considered in combination
the two adjacent crosswalks and an increase in the green time length to obtain the overall delay.
of the crosswalk. 2. Influence of pedestrians’ arrival rate. Griffiths et al. proposed
an upgraded model of delay estimation considering the duration of
each signal phase and the number of pedestrians that arrived at each
Pedestrian delay is a key performance indicator needed to evaluate
phase of the cycle (19). Thus, the irregularity of the entry process was
the level of service for pedestrians at signalized intersections (1).
described. Chilukuri and Virkler pointed out that the nonuniformity of
In traffic engineering, the signal control design of the intersection
pedestrians’ arrival rate may be the result of the signal coordination of
achieves two principal objectives: one is to ensure the safety of all
two adjacent intersections (20). The validation results indicated that
users of the intersection, and the other is to promote the efficient move-
delay calculations for signalized intersections in coordinated systems
ment of all users through the intersection (2). The pedestrian delay is
should recognize the effect of signal coordination. Kruszyna et al.
closely related to the above two objectives. With regard to traffic
further established three new delay estimation models based on field
efficiency, the pedestrian delay is used in many optimization models
of signal control that aim to balance the trade-off between pedes- measurements (21). These models differ in complexity, accuracy, and
trians and vehicles by minimizing the delay for both traffic modes concurrence.
(3–10). With regard to traffic safety, previous research has shown 3. Influence of signal noncompliance. Three pedestrian indica-
that pedestrians will most likely cross on red for delays exceed- tions are normally encountered: “Walk” (pedestrian effective green
ing 30 s and most likely wait for green when the delay is less than time), flashing “Don’t Walk” (pedestrian clearance time), and solid
10 s (11–13). “Don’t Walk” (pedestrian red time). Theoretically, all pedestrians
Analysis of pedestrian delays has been a concern for traffic engi- who arrive during the pedestrian clearance and red intervals should
neers for a long time. One commonly used pedestrian delay model— wait until the beginning of the pedestrian green light. However, as
developed by Pretty (14)—is presented in the Highway Capacity aforementioned, pedestrians may decide to cross on the red light in
Manual (HCM) (15). It is a function of the ratio of the length of the instances of long waiting times. Virkler found that most delay reduc-
pedestrian effective green time and the length of the signal cycle for tions were the result of pedestrians who entered crosswalks during
clearance phases, according to data collected in the central business
district area of Brisbane, Queensland, Australia (22). Li et al. further
J. Zhao, Department of Traffic Engineering, University of Shanghai for Science and
Technology, 516 Jungong Road, Shanghai 200093, China. Y. Liu, Department of Civil
developed a pedestrian delay model with adequate consideration of
and Environmental Engineering, College of Engineering, University of Wisconsin, signal noncompliance during the effective red phases that was based
Milwaukee, P.O. Box 784, Milwaukee, WI 53201. Corresponding author: J. Zhao, on a field study conducted in Xi’an, China (23). Similar conditions
jing_zhao_traffic@163.com. were also found in Indian cities. A pedestrian delay model was devel-
Transportation Research Record: Journal of the Transportation Research Board,
oped by combining the consideration of the nonuniform arrival rate
No. 2615, 2017, pp. 95–104. and signal noncompliance behavior (24). This combined model was
http://dx.doi.org/10.3141/2615-11 also calibrated with data collected in Nanjing, China (25).

95
96 Transportation Research Record 2615

4. Influence of two-stage crossing design. The pedestrian delay of pedestrian delay, whereas the time spent walking in the crosswalk is
a signalized two-stage crossing should not be calculated by simply not counted because the pedestrian would have needed to cross the
adding up the delays from the two stages, because they are inde- street no matter what, which involves the walking time.
pendent. Pedestrian arrivals at the second-stage crossing will be As shown in Figure 1, pedestrian crossing directions could be
affected by the signal timing of the first-stage crossing. Wang and divided into two types, namely, the straight crossing and the diagonal
Tian developed an improved pedestrian delay model for signalized crossing directions. The latter contains two moving paths, that is, the
intersections with a two-stage crossing design (26). The model was clockwise and the anticlockwise paths. The overall computational
validated with a microscopic simulation under a wide range of traffic procedure of the pedestrian control delay of the intersection is shown
scenarios. in Figure 2. In the following, the pedestrian control delays of the dif-
5. Influence of bidirectional pedestrian flow. Researchers found ferent crossing directions and moving paths will be discussed sepa-
that pedestrian walking speed may be reduced when the opposing rately. The pedestrian control delay model of the overall intersection
pedestrian flow is high (27). A sophisticated pedestrian simulation will be established at the end of this section.
model was then developed to estimate pedestrian walking times
and delays, particularly for environments with a heavy opposing
pedestrian flow (28). It was validated with survey data. Results Pedestrian Control Delay Model
showed that the new pedestrian simulation model can provide an for Straight Crossing
accurate evaluation of the changes in walking speed under different
scenarios with particular emphasis on the effects of bidirectional The pedestrian control delay of the straight direction could be
pedestrian flows. determined by the HCM 2010 formula given by

Although much is known about the pedestrian delay of the signal- ( c − g j )2


d(Ti,i ′) = (1)
ized crosswalk, one may wonder how to estimate the average pedes- 2C
trian delay of the overall signalized intersection. Can one calculate the
average intersection pedestrian delay by taking a weighted average where
of the crosswalk pedestrian delays? The walking direction of a pedes-
i and i′ = indexes of intersection corners, as shown in Figure 3
trian at intersections can be either a straight crossing or a diagonal
(i = 1 for northeast corner, i = 2 for southeast corner,
crossing (crossing to the diagonally opposite corner of the intersection
i = 3 for southwest corner, and i = 4 for northwest
in two stages). The existing pedestrian delay models focus mainly on
corner);
the straight crossing and regard the diagonal crossing as two separate
T = set of straight crossing directions (i, i′) ∈ T, as shown
straight crossings. However, that approach is not accurate. The reasons
in Figure 3;
justifying the inaccuracy include the following:
j = (i, i′) = index of crosswalks [( j = 1 for crosswalk at
east leg, j = 2 for crosswalk at south leg, j = 3 for cross-
1. The delay of the diagonal crossing should not be seen simply as
walk at west leg, and j = 4 for crosswalk at north leg);
the sum of the two straight crossings because there are two moving
it is also defined that crosswalk j is left-hand cross-
paths for the pedestrians crossing diagonally. Pedestrians could first
walk for corner i and crosswalk ( j − 1) is right-hand
choose the path that meets the green signal requirements. Therefore,
crosswalk for corner i];
the moving path choice is not random as the existing models have thus
d T(i,i′) = average pedestrian control delay of straight direction
far assumed.
(i, i′) ∈ T (s);
2. Moreover, the pedestrian arrival pattern of the second crosswalk
C = cycle length (s); and
will be affected by the length and signal timing of the first crosswalk.
gj = length of effective green time for crosswalk j (s).
It will also affect the delay of the diagonal crossing.

To remedy this deficiency, this paper aims to develop a pedestrian


Diagonal crossing direction
control delay model in which the diagonal crossing and the mov-
ing path are considered to enhance the estimation accuracy of the
pedestrian delay at signalized intersections.
The rest of this paper is organized as follows. The detailed model
formulation and the computational procedure are described next,
followed by the validation of the proposed model on the basis of field
Anticlockwise path
data. The effect factors of the proposed model are then discussed, Straight crossing
and conclusions and recommendations close the paper. direction

Clockwise path
Model Formulation

In this section, the pedestrian control delay model is established as


a function of the crossing directions and moving paths. Considering
that this paper is discussed on the basis of the HCM’s methodology,
the delay defined in the HCM is used; movements at slower speeds
and stops at the intersection are included. Therefore, the time spent FIGURE 1   Concept of pedestrian crossing directions and
waiting on the intersection corner for a walk signal is counted in the moving paths.
Zhao and Liu 97

Distribution of left-hand
Distribution of pedestrian volume
and right-hand crosswalk
Equations 2 and 4
effective green time

Control delay for one pedestrian


Impact of arrival time
Equations 6 and 7

Average control delay along clockwise


Impact of crossing directions
and anticlockwise paths
and moving paths
Equations 10 and 11

Pedestrian control delay of Pedestrian control delay of diagonal


Impact of signal timing straight crossing crossing
Equation 1 Equation 14

Pedestrian control delay of entire


intersection
Equation 15

FIGURE 2   Computational procedure of pedestrian control delay of intersection.

Pedestrian Control Delay Model e


right-hand crosswalk G (j−1) and the end of the effective green of the
for Diagonal Crossing left-hand crosswalk Gej will choose the clockwise path (see the bold
blue line in Figure 4) because the green signal will appear at the left-
Pedestrians crossing diagonally have two moving path options, either hand crosswalk first. Along the same line, pedestrians arriving during
a clockwise path or an anticlockwise path, as shown in Figure 1. the end of the effective green of the right-hand crosswalk G ej and
e
A three-step procedure is used to compute the pedestrian control delay the end of the effective green of the left-hand crosswalk G (j−1) will
for a diagonal crossing: (a) the distribution of the pedestrian volume choose the anticlockwise path (see the dotted blue line in Figure 4).
between the two moving paths is determined, which is affected mainly Assuming that arrivals are uniform, the proportion of the pedestrian
by the signal timing; (b) the control delay for one pedestrian arriv- volume of the clockwise and anticlockwise paths could be determined
ing at time t is calculated by considering the signal timing relation- with Equations 2 and 4, respectively.
ship of the two crosswalks along the moving path, the length of the
crosswalks, and the walking speed of pedestrians; and (c) the average Ti1
Pi1 = (2)
control delay for the diagonal crossing is determined by combining C
the above two steps.
G j − G( j −1) if G ej ≥ G(ej −1)
e e

Ti1 =  (3)
G j − G( j −1) + C if G ej < G(ej −1)
e e
Distribution of Pedestrian Volume

According to the walking delay minimization principle, the pedes- Ti 2


Pi 2 = (4)
trians arriving at corner i during the end of the effective green of the C

Right-hand crosswalk for Corner 1

i=4 j=4 Corner i = 1

Lj
Straight crossing Left-hand crosswalk
direction (i, i' ) = (3, 4) ´T First crosswalk for Corner 1
for crossing
direction (1, 3)
Crosswalk
j=3 Second crosswalk j=1
for crossing
direction (1, 3)

i=3 i=2
Diagonal crossing
direction (i, i" ) = (1, 3) ´D j=2

FIGURE 3   Schematic layout indicating geometric parameters.


98 Transportation Research Record 2615

Corner i"

Clockwise Path
Corner i + 1
G b(j+1) G e(j+1)

Corner i
G bj G ej G b(j–1) G e(j–1) Time
Ti1 Ti2
Anticlockwise Path

C
Corner i – 1
G b(j–2) G e(j–2)

Corner i"

FIGURE 4   Schematic layout indicating signal timing parameters.

G( j −1) − G j if G(ej −1) ≥ G ej


e e
G bj = start of effective green for crosswalk j (s);
Ti 2 =  (5) Lj = length of crosswalk j (m), as shown in Figure 3; and
G( j −1) − G j + C if G(ej −1) < G ej
e e
v = walking speed of pedestrian (m/s).

where Similarly, any pedestrian crossing diagonally who is traveling along


the anticlockwise path will always receive the green signal at the sec-
Pi1 = proportion between pedestrian volume on clock- b
ond crosswalk at time G (j−2). Therefore, the pedestrian’s control delay
wise path and total volume of pedestrians crossing could be calculated by
diagonally;
Pi2 = proportion between pedestrian volume on anticlock- L( j −1)
wise path and total volume of pedestrians crossing d(Di,i2′′) ( t ) = G(bj − 2) − t − ∀t ∈[G ej , G(ej −1) ] (7)
v
diagonally;
Ti1 and Ti2 = assigned time durations of clockwise and anticlock- D2
where d (i,i″)(t) is the control delay of the pedestrian who is traveling
wise paths chosen by pedestrians, respectively (s),
from corner i to corner i″ along the anticlockwise path and arriving
as shown in Figure 4; and
at the intersection at time t (s).
G ej = end of effective green for crosswalk j (s).

Average Control Delay for Diagonal Crossing


Control Delay for One Pedestrian
Assuming that arrivals are uniform, the average pedestrian control
Any pedestrian traveling diagonally from corner i to corner i″ along
the clockwise path receives the green signal first at the left-hand delay of the diagonal direction along the clockwise and anticlockwise
crosswalk, as shown in Figure 3. Moreover, after crossing the first paths could be determined with Equations 8 and 9, respectively.
crosswalk, the pedestrian will always receive the green signal at the Ti 1
second crosswalk at time G b(j+1), no matter when he or she arrives at 1
d(Di,i1′′) = ∫ d( ( t ) dt
D1
i,i ′′ ) (8)
the intersection, as shown in Figure 4. Therefore, the control delay Ti1 0
of the pedestrian who is traveling along the clockwise path can be
calculated by Ti 2
1
d(Di,i2′′) = ∫ d( ( t ) dt
D2
i,i ′′ ) (9)
Lj Ti 2
d(Di,i1′′) ( t ) = G(bj +1) − t − ∀t ∈[G e
( j −1) ,G e
j ] (6) 0
v
D1
where d(i,i″) is the average pedestrian control delay of the diagonal
where D2
direction along the clockwise path (s) and d(i,i″) is the average pedes-
D = set of diagonal crossing direction (i, i″) ∈ D, as shown trian control delay of the diagonal direction along the anticlockwise
in Figure 3; path (s).
t = arrival time of pedestrian (s); Substitution of Equations 3 and 5 in Equations 8 and 9, respectively,
D1
d (i,i″) (t) = control delay of pedestrian traveling from corner i to allows direct calculation of the average pedestrian control delay of
corner i″ (i, i″) ∈ D along clockwise path who is arriving the diagonal direction along the clockwise and anticlockwise paths
at intersection at time t (s); with Equations 10 and 11, respectively.
Zhao and Liu 99

Lj
d(Di,i1′′) = T (Di,1i ′′) − (10)
v Corner 4

L( j −1) Corner 3
d(Di,i2′′) = T (Di,i2′′) − (11)
v
D1
where T (i,i″) is the time period from the middle of the time period
b D2
Ti1 to G (j+1) (s), which can be calculated with Equation 12, and T (i,i″)
b
is the time period from the middle of the time period Ti2 to G (j−2) (s), Corner 1
which can be calculated with Equation 13.
Corner 2
 b G ej + G(ej −1)
G( j +1) − 2
if G(bj +1) ≥ G ej ≥ G(ej −1)
 (a)
 G j + G(ej −1) − C
e

T (Di,1i ′′) = G(bj +1) − if G < G


e
j
e
( j −1) (12)
 2
 G + G(ej −1)
e
G(bj +1) − j +C if G ej ≥ G(ej −1) ≥ G(bj +1)
 2 Corner 1

Corner 4
 b G ej + G(ej −1)
G( j − 2) − ≥G ≥G
b e e
if G ( j − 2) ( j −1) j
2
 Corner 3
 Corner 2
 G j + G(ej −1) − C
e

T (Di,i2′′) = G(bj − 2) − if G e
( j −1) <G e
j (13)
 2
 G + G(ej −1)
e
G(bj − 2) − j +C if G(ej −1) ≥ G ej ≥ G(bj − 2)
 2
(b)
Therefore, the average pedestrian control delay of the diagonal
direction could be calculated by estimating the weighted average FIGURE 5   Video camera views of survey intersections:
of the delay of the clockwise and anticlockwise paths, as shown in (a) Xinchen Avenue and South Second Ring Road and
Equation 14. (b) Xinchen Avenue and G329 Road.

Ti1 D1 Ti 2 D 2
d(Di,i ′′) = d(i,i ′′) + d(i,i ′′) (14)
C C in Figure 5. The cameras were mounted on the top of buildings near
the two intersections to ensure that the overall intersection was cov-
where d D(i,i″) is the average pedestrian control delay of the diagonal ered. The video cameras readily captured pedestrian arrivals during
direction (s). successive 1-s intervals, and moving paths, for later analyses of the
delay. The video-based data collection was performed from 7:30 to
9:30 and from 16:30 to 18:30 for three consecutive days. The length
Pedestrian Control Delay of the Intersection of each crosswalk and signal timings are shown in Table 1. In total,
The pedestrian control delay of the intersection could be determined
with Equation 15 with the combination of the pedestrian control
delay models for straight and diagonal crossings. TABLE 1   Crosswalk Length and Signal Timings

Cycle Start of Duration End of


∑ q( )d(
( )
i,i ′
T
i,i ′ ) + ∑ q(
( )
d D
i,i ′′ ) (i,i ′′ ) Crosswalk Length Green of Green Green
dp =
i,i ′ ∈T i,i ′′ ∈D
(15) Crosswalk Length (m) (s) (s) (s) (s)
∑q
( ) i,i ′ ∈T
(i ,i ′ ) + ∑ q(
( )
i,i ′′ ∈D
i,i ′′ )
Xinchen Avenue and South Second Ring Road
C1–C2 25 150  0 17  17
where dp is the pedestrian control delay of the intersection (s) and
C2–C3 18 150 66 36 102
q(i,i′) and q(i,i″) are the pedestrian volumes from corner i to corner i′
C3–C4 25 150  0 17  17
((i, i′) ∈ T) and from corner i to corner i″ ((i, i″) ∈ D), respectively
C4–C1 18 150 66 36 102
(pedestrians/hour).
Xinchen Avenue and G329 Road
C1–C2 21 120  0 16  16
Model Validation C2–C3 25 120 60 14  74
C3–C4 21 120  0 16  16
The pedestrian control delay model is further validated on the basis C4–C1 25 120 60 14  74
of field measurements at two signalized intersections in Cixi, China.
The traffic data were obtained with the use of video cameras, as shown Note: C1, C2, C3, and C4 indicate Corners 1, 2, 3, and 4, respectively.
100 Transportation Research Record 2615

13,619 pedestrians were measured. The crossing of the right-turn TABLE 2   Pedestrian Volume
slip lanes was uncontrolled. It is not included in the delay observa-
tions and calculations, nor were all the right-turn pedestrians and Original Destination Pedestrian Original Destination Pedestrian
Corner Corner Volume Corner Corner Volume
pedestrians who violated the red signal.
Xinchen Avenue and Xinchen Avenue
South Second Ring Road and G329 Road
Pedestrian Demand and Arrival Distribution 1 2 568 1 2 691
1 3 472 1 3 596
In total, 13,619 pedestrians (excluding the right-turn pedestrians 1 4 503 1 4 673
and pedestrians who violated the red signal) arrived at the two
2 1 621 2 1 690
intersections during the survey time. The pedestrian demand of
2 3 491 2 3 538
each crossing direction is shown in Table 2. The distribution of
2 4 430 2 4 548
the arrival time is shown in Figure 6. Kolmogorov–Smirnov test
results (see Table 3) show that the pedestrian arrival times obey 3 1 528 3 1 567
uniform distribution ( p-value = .237 > .05 and .175 > .05 for the 3 2 546 3 2 664
two intersections). 3 4 463 3 4 718
4 1 602 4 1 744
4 2 395 4 2 555
Walking Speed Distribution 4 3 493 4 3 523

The average walking speed of each pedestrian crossing the street Note: Total pedestrian volume for Xinchen Avenue and South Second Ring
Road is 6,112. Total pedestrian volume for Xinchen Avenue and G329 Road
was measured. The distribution of the mean walking speeds of is 7,507.
the two intersections is shown in Figure 7. As shown in Table 4,
the walking speed of each intersection is examined for normality
(p-value = .465 > .05 and .148 > .05 for the two intersections) and Validation of Moving Path Distribution
homogeneity of variance ( p-value = .808 > .05) with the Kolmogorov–
Smirnov and Levene’s tests, respectively. One could also observe that In the survey, 4,089 pedestrians crossed the street along the diagonal
the difference between the two intersections in regard to the mean direction. According to Equations 2 and 4, the proportion of the
value of the walking speed is minor. The t-test of independent pedestrian volume of the clockwise and anticlockwise paths could
samples shows that there is no significant difference in the walking be calculated.
speed between the two intersections (t = 0.851, p-value = .395 > .05) A comparison between the results of the proposed model and
(see Table 4). Therefore, the average value (v = 1.225 m/s) is used those of the survey is summarized in Table 5. The percentages of
in the following analysis. differences are within 10% in all cases. Further, paired t-test results

60 50

50 40
Frequency

Frequency

40
30
30
20
20

10 10

0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120
Arrival Time in Cycle (s) Arrival Time in Cycle (s)
(a) (b)

FIGURE 6   Distribution of pedestrian arrival time: (a) Xinchen Avenue and South Second Ring Road
and (b) Xinchen Avenue and G329 Road.

TABLE 3   Kolmogorov–Smirnov Test for Arrival Distribution

Uniform Parameter Most Extreme Differences


Sig.
Intersection Number Minimum Maximum Absolute Positive Negative Z (Two-tailed)

Xinchen Avenue and South Second Ring Road 6,112 0 150 0.013 0.013 −0.007 1.033 .237
Xinchen Avenue and G329 Road 7,507 0 120 0.013 0.013 −0.005 1.103 .175

Note: Z = critical value; Sig. = significance level.


Zhao and Liu 101

2,000 100 2,000 100


Frequency Frequency
1,600 distribution 80 1,600 distribution 80

Frequency (%)

Frequency (%)
Cumulative

Cumulative
Frequency

Frequency
1,200 Cumulative 60 1,200 Cumulative 60
curve curve
800 40 800 40

400 20 400 20

0 0 0 0
0. .5

0. .7

0. .9

1. .1

1. .3

1. .5

1. .7
9
.9

0. .5

0. .7

0. .9

1. .1

1. .3

1. .5

1. .7
9
.9
1.

1.
<0

>1

<0

>1
5–

7–

9–

1–

3–

5–

7–

5–

7–

9–

1–

3–

5–

7–
Walking Speed (m/s) Walking Speed (m/s)
(a) (b)

FIGURE 7   Distribution of pedestrian walking speed.

TABLE 4   Statistical Analysis for Walking Speed

Kolmogorov–
Smirnov Test Levene’s Test t-Test for
Normal for Normal for Equality of Equality of
Parameter Distribution Variances Means

Intersection Number Mean SD Z Sig. F Sig. t Sig.

Xinchen Avenue and South Second Ring Road 6,112 1.228 0.321 0.850 0.465 0.059 0.808 0.851 0.395
Xinchen Avenue and G329 Road 7,507 1.223 0.324 1.367 0.148 0.059 0.808 0.851 0.395

Note: F = F-statistic.

TABLE 5   Comparison of Moving Path Distribution Between Model and Field Survey

Field Survey

Crossing Volume of Moving Volume of


Direction Moving Path Path (Ped.) Direction (Ped.) Proportion Model Results Error (%)

Xinchen Avenue and South Second Ring Road


C1–C3 Clockwise 199 472 0.422 0.433 2.6
Anticlockwise 273 0.578 0.567 −1.9
C2–C4 Clockwise 255 430 0.593 0.567 −4.4
Anticlockwise 175 0.407 0.433 6.4
C3–C1 Clockwise 245 528 0.464 0.433 −6.7
Anticlockwise 283 0.536 0.567 5.8
C4–C2 Clockwise 222 395 0.562 0.567 0.9
Anticlockwise 173 0.438 0.433 −1.1
Xinchen Avenue and G329 Road
C1–C3 Clockwise 331 596 0.555 0.517 −6.8
Anticlockwise 265 0.445 0.483 8.5
C2–C4 Clockwise 256 548 0.467 0.483 3.4
Anticlockwise 292 0.533 0.517 −3.0
C3–C1 Clockwise 292 567 0.515 0.517 0.4
Anticlockwise 275 0.485 0.483 −0.4
C4–C2 Clockwise 275 555 0.495 0.483 −2.4
Anticlockwise 280 0.505 0.517 2.4

Note: Ped. = pedestrian.


102 Transportation Research Record 2615

TABLE 6  Paired t-Test for Moving Path Distribution Validation

Paired Differences

95% Confidence
Interval of Difference
Standard Error Sig.
Pair Mean SD Mean Lower Upper t df (Two-tailed)

Survey—proposed model 0.0091 0.0206 0.00728 −0.0081 0.0263 1.254 7 .250

Note: df = degrees of freedom.

(see Table 6) show no significant difference between the results posed and the HCM 2010 models were compared with the field
estimated by the proposed model and those elicited from the field survey data under each crossing direction at the two intersections,
survey (t = 1.254, p-value = .250 > .05), indicating that the accuracy as shown in Table 7.
of the proposed moving path distribution model is acceptable. One could observe that the proposed model could generally elicit
an error under 10% in all cases. Paired t-test results (see Table 8)
further show no significant difference between results estimated by
Validation of Pedestrian Control Delay the proposed model and those from the field survey (t = −0.252,
p-value = .803 > .05), while the difference between the results esti-
To validate the pedestrian control delay model, the control delay of mated by the HCM 2010 model and those from the field survey is
each pedestrian was measured. The calculation results of the pro- significant (t = −3.724, p-value = .001 < .05).

TABLE 7   Pedestrian Control Delay Comparison Between Model and Field Survey

Proposed Model HCM 2010 Model

Crossing Direction Survey Delay (s) Delay (s) Errora (%) Delay (s) Errora (%)

Xinchen Avenue and South Second Ring Road


C1–C2 59.51 58.96 −0.93 58.96 −0.93
C1–C3 69.28 68.67 −0.88 102.28 47.63
C1–C4 39.71 43.32 9.10 43.32 9.10
C2–C1 60.65 58.96 −2.78 58.96 −2.78
C2–C3 44.25 43.32 −2.11 43.32 −2.11
C2–C4 68.67 68.67 −0.01 102.28 48.94
C3–C1 64.35 68.67 6.71 102.28 49.63
C3–C2 41.53 43.32 4.31 43.32 4.31
C3–C4 60.15 58.96 −1.98 58.96 −1.98
C4–C1 42.64 43.32 1.59 43.32 1.59
C4–C2 68.41 68.67 0.37 102.28 49.50
C4–C3 57.87 58.96 1.89 58.96 1.89
Entire intersection 56.16 56.72 1.01 66.41 18.26
Xinchen Avenue and G329 Road
C1–C2 46.14 45.07 −2.32 45.07 −2.32
C1–C3 67.51 56.17 −8.69 91.89 49.38
C1–C4 47.69 46.82 −1.83 46.82 −1.83
C2–C1 44.44 45.07 1.42 45.07 1.42
C2–C3 46.84 46.82 −0.04 46.82 −0.04
C2–C4 54.84 56.17 2.42 91.89 67.56
C3–C1 65.30 56.17 −9.83 91.89 47.50
C3–C2 44.31 46.82 5.66 46.82 5.66
C3–C4 43.72 45.07 3.10 45.07 3.10
C4–C1 46.20 46.82 1.35 46.82 1.35
C4–C2 52.47 56.17 7.04 91.89 75.11
C4–C3 45.86 45.07 −1.72 45.07 −1.72
Entire intersection 50.01 49.03 −0.56 59.81 21.30
a
The error is calculated with field survey data as a benchmark.
Zhao and Liu 103

TABLE 8  Paired t-Test for Pedestrian Control Delay Validation

Paired Differences

95% Confidence
Interval of Difference
Standard Error Sig.
Pair Mean SD Mean Lower Upper t df (Two-tailed)

Survey—proposed model −0.115  2.328 0.457   −1.055 0.825 −0.252 25 .803


Survey—HCM model −11.414 15.627 3.065 −17.726 −5.102 −3.724 25 .001

Moreover, the calculation accuracy of the proposed model for the ing direction also increases with an increase in the green time of the
diagonal crossing directions, including C1-C3, C2-C4, C3-C1, and crosswalk because the pedestrians always have to wait at the second
C4-C2, is much higher than the HCM 2010 model’s. Therefore, the crosswalk regardless of when they arrive at the intersection.
proposed model outperforms the HCM 2010 model in estimating With the consideration of the pedestrian crossing directions and
the pedestrian control delay of the overall intersection (17.26% and moving paths, the proposed model could estimate the pedestrian
20.74% higher in estimation accuracy for the two intersections). control delay of the overall intersection more accurately. The degree
of the improvement in accuracy depends on the proportion of the
pedestrians crossing diagonally. Figure 9 gives an intuitive illustra-
Discussion of Results tion of its effect. In this analysis, three cases of signal timing are
considered: (a) Case 1, in which the cycle length was 160 s and
The conducted validation underlines the advantage of the proposed the length of the crosswalk green time was 40 s, (b) Case 2, in which
model in estimating the pedestrian control delay of the diagonal the cycle length was 120 s and the length of the crosswalk green time
crossing direction, which is affected mainly by the length of the green was 30 s, and (c) Case 3, in which the cycle length was 80 s and the
of the crosswalks and by the time gap of the green light between the length of the crosswalk green time was 20 s. It can be observed that
two adjacent crosswalks. If the time gap of the green light between the improvement increases from 0% to approximately 40% when
the two adjacent crosswalks is small, the pedestrians could receive the the proportion of the diagonal crossing increases from 0 to 0.8.
green signal at the second crosswalk immediately after crossing the
first crosswalk, thereby allowing them to accomplish the diagonal
crossing more smoothly. Figure 8 is a more intuitive illustration Conclusion
of those effects. In this figure, the two horizontal axes represent the
length of the green of the crosswalk and the time gap of the green A new pedestrian control delay model was proposed in which the
light between the two adjacent crosswalks. The vertical coordinate is diagonal crossing and the moving paths were considered. The pro-
the pedestrian control delay of the diagonal crossing direction. In this posed model was validated on the basis of field measurements. The
analysis, the length of the green time of the four crosswalks is equal; main factors affecting the diagonal crossing pedestrian control delay
the length of the crosswalk is set to be 15 m, and the walking speed were also discussed. From the analysis, the following conclusions
is set to be 1.2 m/s. can be drawn:
It can be observed that the delay of the diagonal crossing direction
increases with an increase in the green time gap between the two 1. This paper has proved that the existing model of HCM 2010
adjacent crosswalks because pedestrians have to wait a longer time could not provide the necessary accuracy in estimating the pedestrian
at the second crosswalk. Moreover, the delay of the diagonal cross- delay of the diagonal crossing. The largest differences between
Delay of the Intersection (s)

0–20 20–40 40–60 60–80 80–100


90
Control Delay of Diagonal

80
Crossing Direction (s)

100
70
80
60
60 50
40 40
20 30
20
0 10
Tim 50
Tw e G 0
50
o A ap 40 40 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80
dja of 30 f
ce Gre 30 no
nt ree (s) Proportion of Diagonal Crossings
Cr en B 20 20 o f G
lk
os
sw etwe 10 gth swa Proposed model (Case 1) HCM 2010 model (Case 1)
alk en Len Cros
s( the Proposed model (Case 2) HCM 2010 model (Case 2)
s)
Proposed model (Case 3) HCM 2010 model (Case 3)

FIGURE 8   Effects of signal timing on diagonal crossing FIGURE 9   Effects of proportion of diagonal crossings
pedestrian control delay. on improvement of accuracy.
104 Transportation Research Record 2615

the calculated delays and the actual field measurements were up to   9. Ma, W., X. Yang, W. Pu, and Y. Liu. Signal Timing Optimization Models
75.11%. Accordingly, the existing model in HCM 2010 could not for Two-Stage Midblock Pedestrian Crossing. Transportation Research
Record: Journal of the Transportation Research Board, No. 2198, 2010,
estimate the pedestrian delay of the overall intersection accurately. pp. 133–144. https://doi.org/10.3141/2198-15.
The largest differences reached 21.3%. 10. Yu, C. H., W. J. Ma, and X. G. Yang. Integrated Optimization of Loca-
2. The proposed model yielded more accurate results in estimat- tion and Signal Timings for Midblock Pedestrian Crosswalk. Journal
ing the control delay of the pedestrians crossing diagonally. The of Advanced Transportation, Vol. 50, No. 4, 2016, pp. 552–569. https://
proposed model could generally minimize the error under 10% in doi.org/10.1002/atr.1360.
11. Houten, R., R. Ellis, and J. L. Kim. Effects of Various Minimum Green
all cases. Moreover, paired t-test results showed that no significant Times on Percentage of Pedestrians Waiting for Midblock “Walk” Signal.
difference exists between results estimated by the proposed model Transportation Research Record: Journal of the Transportation Research
and those elicited from the field survey. It reveals that compared with Board, No. 2002, 2007, pp. 78–83. https://doi.org/10.3141/2002-10.
the HCM 2010 model, the proposed model is promising in increasing 12. Ren, G., Z. Zhou, W. Wang, Y. Zhang, and W. Wang. Crossing Behav-
iors of Pedestrians at Signalized Intersections. Transportation Research
the estimation accuracy of the pedestrian control delay of the diagonal Record: Journal of the Transportation Research Board, No. 2264, 2011,
crossing and the entire intersection (approximately 20%). pp. 65–73. https://doi.org/10.3141/2264-08.
3. Analysis of the sensitivity of the delay showed that the delay of 13. Duduta, N., Q. Zhang, and M. Kroneberger. Impact of Intersection
the diagonal crossing increased with increases in the time gap of the Design on Pedestrians’ Choice to Cross on Red. Transportation Research
green light between the two adjacent crosswalks and with increases Record: Journal of the Transportation Research Board, No. 2464, 2014,
pp. 93–99. https://doi.org/10.3141/2464-12.
in the crosswalk green time. 14. Pretty, R. L. The Delay to Pedestrians and Vehicles at Signalized Inter-
sections. ITE Journal, Vol. 49, 1979, pp. 20–23.
The newly developed model is useful in estimating the pedestrian 15. Highway Capacity Manual 2010. Transportation Research Board of the
control delays of the diagonal crossing and the entire intersection. National Academies, Washington, D.C., 2010.
16. Fitzpatrick, K. Improving Pedestrian Safety at Unsignalized Intersections.
The final estimation model given is based on the uniform arrival ITE Journal, Vol. 77, No. 5, 2007, pp. 34–41.
assumption. Therefore, the pedestrian arrival distribution should be 17. Wei, D., H. Liu, and Z. Tian. Vehicle Delay Estimation at Unsignalised
tested before further applications. Pedestrian Crosswalks with Probabilistic Yielding Behaviour. Transport-
metrica A: Transportation Science, Vol. 11, No. 2, 2015, pp. 103–118.
18. Guo, X., M. C. Dunne, and J. A. Black. Modeling of Pedestrian Delays
with Pulsed Vehicular Traffic Flow. Transportation Science, Vol. 38,
Acknowledgment No. 1, 2004, pp. 86–96. https://doi.org/10.1287/trsc.1030.0025.
19. Griffiths, J. D., J. G. Hunt, and M. Marlow. Delays at Pedestrian
This work was supported by the National Natural Science Foundation Crossings—4. Mathematical Models. Traffic Engineering and Control,
of China. Vol. 26, No. 5, 1985, pp. 277–282.
20. Chilukuri, V., and M. R. Virkler. Validation of HCM Pedestrian Delay
Model for Interrupted Facilities. Journal of Transportation Engineering,
Vol. 131, No. 12, 2005, pp. 939–945. https://doi.org/10.1061/(ASCE)
References 0733-947X(2005)131:12(939).
21. Kruszyna, M., P. Mackiewicz, and A. Szydlo. Influence of Pedestri-
ans’ Entry Process on Pedestrian Delays at Signal-Controlled Cross-
  1. Petritsch, T. A., B. W. Landis, P. S. McLeod, H. F. Huang, S. Challa, and
walks. Journal of Transportation Engineering, Vol. 132, No. 11, 2006,
M. Guttenplan. Level-of-Service Model for Pedestrians at Signalized
pp. 855–861. https://doi.org/10.1061/(ASCE)0733-947X(2006)132
Intersections. Transportation Research Record: Journal of the Trans-
:11(855).
portation Research Board, No. 1939, 2005, pp. 53–62. https://doi.org
22. Virkler, M. R. Pedestrian Compliance Effects on Signal Delay. Trans-
/10.3141/1939-07.
portation Research Record, No. 1636, 1998, pp. 88–91. https://doi.org
  2. Lu, G. X., Y. Zhang, and D. A. Noyce. Intelligent Traffic Signal System
/10.3141/1636-14.
for Isolated Intersections: Dynamic Pedestrian Accommodation. Trans-
23. Li, Q., Z. Wang, J. Yang, and J. Wang. Pedestrian Delay Estimation at
portation Research Record: Journal of the Transportation Research
Signalized Intersections in Developing Cities. Transportation Research
Board, No. 2259, 2011, pp. 96–111. https://doi.org/10.3141/2259-09.
Part A: Policy and Practice, Vol. 39, No. 1, 2005, pp. 61–73. https://doi.org
  3. Ma, W., D. Liao, Y. Liu, and H. K. Lo. Optimization of Pedestrian Phase
/10.1016/j.tra.2004.11.002.
Patterns and Signal Timings for Isolated Intersection. Transportation
24. Nagraj, R., and P. Vedagiri. Modeling Pedestrian Delay and Level of
Research Part C: Emerging Technologies, Vol. 58, 2015, pp. 502–514.
Service at Signalized Intersection Crosswalks Under Mixed Traffic
https://doi.org/10.1016/j.trc.2014.08.023.
Conditions. Transportation Research Record: Journal of the Trans-
 4. Ma, W., Y. Liu, and K. L. Head. Optimization of Pedestrian Phase
portation Research Board, No. 2394, 2013, pp. 70–76. https://doi.org
Patterns at Signalized Intersections: A Multi-Objective Approach. Jour-
/10.3141/2394-09.
nal of Advanced Transportation, Vol. 48, No. 8, 2014, pp. 1138–1152.
25. Ye, X., J. Chen, G. Jiang, and X. Yan. Modeling Pedestrian Level of
https://doi.org/10.1002/atr.1256.
Service at Signalized Intersection Crosswalks Under Mixed Traffic Con-
  5. Ma, W., Y. Liu, H. Xie, and X. Yang. Multiobjective Optimization of
ditions. Transportation Research Record: Journal of the Transportation
Signal Timings for Two-Stage, Midblock Pedestrian Crosswalk. Transpor-
Research Board, No. 2512, 2015, pp. 46–55. https://doi.org/10.3141
tation Research Record: Journal of the Transportation Research Board,
/2512-06.
No. 2264, 2011, pp. 34–43. https://doi.org/10.3141/2264-05.
26. Wang, X., and Z. Tian. Pedestrian Delay at Signalized Intersections with
  6. Yang, Z., and R. F. Benekohal. Use of Genetic Algorithm for Phase Opti-
a Two-Stage Crossing Design. Transportation Research Record: Journal
mization at Intersections with Minimization of Vehicle and Pedestrian
of the Transportation Research Board, No. 2173, 2010, pp. 133–138.
Delays. Transportation Research Record: Journal of the Transportation
https://doi.org/10.3141/2173-16.
Research Board, No. 2264, 2011, pp. 54–64. https://doi.org/10.3141
27. Goh, P. K., and W. H. K. Lam. Pedestrian Flows and Walking Speed:
/2264-07.
A Problem at Signalized Crosswalks. ITE Journal, Vol. 74, No. 1, 2004,
  7. He, Q., K. L. Head, and J. Ding. Multi-Modal Traffic Signal Control with
pp. 28–33.
Priority, Signal Actuation, and Coordination. Transportation Research
28. Lee, J. Y. S., and W. H. K. Lam. Simulating Pedestrian Movements at
Part C: Emerging Technologies, Vol. 46, 2014, pp. 65–82. https://doi.org
Signalized Crosswalks in Hong Kong. Transportation Research Part A:
/10.1016/j.trc.2014.05.001.
Policy and Practice, Vol. 42, No. 10, 2008, pp. 1314–1325. https://doi.org
  8. Roshandeh, A. M., H. S. Levinson, Z. Li, H. Patel, and B. Zhou. New
/10.1016/j.tra.2008.06.009.
Methodology for Intersection Signal Timing Optimization to Simulta-
neously Minimize Vehicle and Pedestrian Delays. Journal of Transpor-
tation Engineering, Vol. 140, No. 5, 2014, pp. 382–398. https://doi.org The Standing Committee on Highway Capacity and Quality of Service peer-reviewed
/10.1061/(ASCE)TE.1943-5436.0000658. this paper.

You might also like