You are on page 1of 21

Journal of Intelligent Transportation Systems

Technology, Planning, and Operations

ISSN: 1547-2450 (Print) 1547-2442 (Online) Journal homepage: https://www.tandfonline.com/loi/gits20

A protocol for pedestrian crossing and increased


vehicular flow in smart cities

Sara El Hamdani, Nabil Benamar & Mohmed Younis

To cite this article: Sara El Hamdani, Nabil Benamar & Mohmed Younis (2019): A protocol
for pedestrian crossing and increased vehicular flow in smart cities, Journal of Intelligent
Transportation Systems, DOI: 10.1080/15472450.2019.1683451

To link to this article: https://doi.org/10.1080/15472450.2019.1683451

Published online: 12 Nov 2019.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=gits20
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS
https://doi.org/10.1080/15472450.2019.1683451

A protocol for pedestrian crossing and increased vehicular flow


in smart cities
Sara El Hamdania , Nabil Benamara, and Mohmed Younisb
a
Department of Mathematics and Computer Science, School of Technology, Moulay Ismail University of Meknes, Meknes, Morroco;
b
Department of Computer Science and Electrical Engineering, University of Maryland, Baltimore, MA, USA

ABSTRACT ARTICLE HISTORY


With the technological drive for realizing smart cities, work on Autonomous Intersection Received 26 July 2018
Management (AIM) protocols opts to replace the traditional traffic light system by using Revised 18 October 2019
cooperative Vehicle-to-Vehicle (V2V) communication in order to decrease road congestion Accepted 18 October 2019
and increase vehicular throughput. However, these protocols simply ignore pedestrians as
KEYWORDS
road users and do not provision for safe pedestrian crossing. Basically, a road intersection autonomous traffic
not only could be traffic bottleneck, but also nearly 23% of the total automotive related management; cooperative
fatalities and almost 1 million injury-causing crashes occur at or within intersections every driving; mixed traffic;
year. This article opts to fill such a technical gap. We present a novel system that prioritizes pedestrian crossing; V2V
pedestrians crossing and guarantees safety while preserving the efficiency of AIM based communication
approaches. Our system does not require additional infrastructure or pedestrian-carried devi-
ces, and works for both self-driving and human-derived vehicles. The simulation results
show that our proposed system for Autonomous Pedestrian Crossing (APC) protocol of non-
signalized intersections significantly decreases the vehicle’s delay and pedestrian walk dur-
ation compared to the conventional traffic light-based systems. The effectiveness of APC is
validated for traffic scenarios with (1) self-driving vehicles, and (2) mix of human-based and
self-driving vehicles.

Introduction adverse effect on the economy. In 2014, road conges-


tion caused urban Americans to travel an extra 6.9
Intelligent Transportation Systems (ITS) play a crucial
role in diminishing traffic jams, CO2 emissions, travel billion hours and purchase extra 3.1 billion gallons of
delays and accident rates, while improving road safety, fuel for a congestion cost of $160 billion, which is
traffic flow and passenger comfort. Vehicle-to- expected to grow to $192 billion in 2020 (Schrank
Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) et al.). Furthermore, traffic congestion causes extra
(Yang et al.) communications are considered to be the emission of carbon dioxide, which is environmental
foundations of current ITS. V2I communication ena- and health hazard. In addition, traffic congestion leads
bles wireless exchange of critical safety and traffic data to a high number of vehicle crashes threatening
between vehicles and roadway infrastructure; mean- human life and wellbeing. For example, in 2017 a total
while V2V communication utilizes short-range radio of 1,793 people were killed and 24,831 were seriously
links between vehicles to exchange information such injured due to road accidents in the Great
as motion speed and heading in order to avoid colli- Britain (Britain).
sions. Emerging techniques, such as Vehicle-to- In urban settings, road intersections are considered
Pedestrian (V2P), Vehicle-to-Device (V2D) and to be the traffic management bottleneck. Between
Vehicle-to-Grid (V2G) and Vehicle-to-Everything 2011 and 2014, nearly 28% of fatal crashes in the US
(V2X), opt to grow the scope of interactions and sup- occurred at intersections (Lombardi et al.). Hence,
port a broad range of vehicle types (Shladover, 2017). research on intersection management in the realm of
The main objective of ITS is to mitigate congestion self-driving vehicles has focused on possible elimin-
and boost road safety. Road congestion is responsible ation of traffic signals and instrumenting autonomy
for significant losses of time and fuel, and thus has an among the vehicles in order to increase simultaneity

CONTACT Sara El Hamdani S.elhamdani@edu.umi.ac.ma Department of Mathematics and Computer Science, School of Technology, Moulay Ismail
University of Meknes, Meknes, Morroco.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gits.
ß 2019 Taylor & Francis Group, LLC
2 S. E. HAMDANI ET AL.

of intersection crossing, which in turn will decrease simultaneous passage of pedestrians and vehicles.
the travel delay (Bento et al., 2019). Such a non-sig- Fundamentally, we provision for safe pedestrian cross-
nalized intersection concept proved its effectiveness in ing in autonomous traffic management system while
increasing the intersection crossing rate for vehicles sustaining the efficiency of existing vehicular protocols
(Namazi et al.). The majority of published work pur- for the non-signalized intersections. We use V2V
sued Vehicle to Infrastructure (V2I) and Vehicle-to- communication for coordination between vehicles to
Vehicle (V2V) communications to coordinate between enhance the safety and to avoid unnecessary stopping.
vehicles in order to avoid stopping at a non-signalized Our proposed APC system defines a set of rules for
intersection as much as possible. prioritizing pedestrian (especially vulnerable pedes-
However, the presence of traffic lights enables safe trians) crossing of roads and guarantees safety without
crossing of pedestrians; thus supporting autonomous the need for pedestrians to carry devices such as
vehicle passage is not sufficient justification of the smartphones. APC can be effectively applied for traffic
elimination of signals. In fact, the majority of new involving only self-driving vehicles and for cases
autonomous intersection management protocols where there is a mix of human-driven and self-driven
assumed that only vehicles would use the road and vehicles. We have validated our system using the
ignored the existence of other important road users, SUMO simulator (Krajzewicz et al.). The simulation
namely, bicycles, motorcycles, and pedestrians results show that our APC system outperforms the
(Elhamdani & Benamar, 2017). Such an assumption is conventional traffic light model and significantly
not practical in the broad sense, especially in urban decreases the vehicle’s time loss and pedestrian walk
setups. Therefore, despite the reduced travel time duration due to pedestrian crossing. The effectiveness
advantage, a non-signalized intersection would not be of APC is also confirmed for cases when both self-
deemed as a suitable solution without supporting safe driving and human-driven vehicles are involved.
pedestrian crossing. This is particularly important tak- The remainder of this article is organized as fol-
ing into account that pedestrians accounted for 26% lows. “Related work” section goes over related work.
of Great Britain road fatalities according to reported The system model and approach overview are pro-
road casualties in GB in 2017 (Britain). Moreover, vided in the “System model and approach overview”
according to NHTSA report, 76% of pedestrian fatal- section. The APC protocol is described in detail in
ities in 2016 occurred in urban areas (Brock et al.). both “Pedestrian collision detection” and “Pedestrian
Previous work on pedestrian safety has focused on crossing protocol” sections, covering pedestrian colli-
improving the capabilities of self-driving vehicles for sion detection, and the pedestrian crossing protocol,
detecting pedestrians (Franke et al.), and avoiding respectively. “APC in mixed traffic scenario” section
them using Fuzzy Steering Controller (Fernandez discusses the applicability of APC in mixed traffic
Llorca et al.). However, published solutions lack prac- scenario. The safety and practicality of APC are dis-
ticality since multiple pedestrians usually need to cross cussed in “Protocol safety and practicality” section,
the road simultaneously, and the vehicle cannot avoid while “Performance evaluation” section reports the
them one by one without stopping. Some techniques performance validation results. Finally, “Conclusion
rely on coordination between the vehicle and the ped- and future work” section concludes the article and
estrian through V2P communication (Bagheri et al.). hints some directions for further research.
Basically, instantaneous alarms about coming vehicles
are sent to the phones of pedestrians to alert them. Related work
However, such solution strategy requires pedestrians
to carry smartphones, which may not always be pos- Vehicle collision avoidance at intersections
sible. In fact, the pedestrian could be children, old Contemporary Traffic Light Controller (TLC) based
person, handicapped or simply has a smartphone with intersection management mechanisms may not cope
dead battery. with the high number of vehicles aiming to cross the
Although V2V communication has been exploited intersection. Thus, many researchers have focused on
for autonomous traffic management, to the best of improving the TLC operation by using sensors, as
our knowledge little attention has been paid to sup- well as V2V (Shladover, 2017) and V2I communica-
porting pedestrian crossing. Thus, we believe that this tion (Li & Shimamoto, 2011; Gradinescu, Gorgorin,
article fills an important technical gap tackling the Diaconescu, Cristea, & Iftode, 2007; Murphy, Djahel,
shortcomings of published solutions and presenting a Jabeur, Barrett, & Murphy, 2015). A TLC collects data
complete coordination system for supporting from vehicles and neighboring TLCs in order to
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 3

determine the optimal green light phase time and/or could be influential, pedestrian crossing in the pres-
to reroute vehicles to less congested roads based on ence of driverless vehicle has been studies by
information gathered using Infrastructure to (Rodrıguez Palmeiro et al., 2018). They opt to capture
Infrastructure (I2I) communication (Wang, Djahel, & the pedestrian reaction during road crossing in the
McManis, 2014). These advanced TLC systems have presence of autonomous vehicles with dummies seated
been shown to outperform the traditional ones, yet in the driver seat. The study has found out that
they require expensive infrastructure upgrade (Li & pedestrians are stressed and do not attempt to cross
Shimamoto, 2011; Murphy et al., 2015). the road without establishing an eye-contact with
The abovementioned studies opt to sustain auton- human driver.
omy at the level of TLC operation by providing data The importance of pedestrian-driver gestural com-
to enhance decision making. Some work eliminated munication was a motivation to involve interaction
TLCs altogether. For example, the approaches of between pedestrians and vehicles. To that end, the
(Dresner & Stone, 2008) and (Li, Chitturi, Yu, Bill, & approach of (Florentine, Ang, Pendleton, Andersen, &
Noyce, 2015) apply a centralized strategy where a con- Ang, 2016) consists of two methods of pedestrian
troller manages the safe passage of vehicles in a non- notification: turning on a LED strip when a close
signalized intersection. In this case vehicles send obstacle is detected and broadcasting an audio
requests for crossing the intersection to a controller, notification of vehicle intention.
which grants permissions on a first-come-first-served The aforementioned techniques just focus on detec-
basis. However, the response time and scalability are tion and stopping for pedestrians. An alternative
the major issues of a centralized controller and could methodology is to schedule pedestrian crossing so that
hinder the vehicle’s ability for reacting in real time. the vehicular flow is minimally impacted. Earlier work
On the other hand, the decentralized strategy on pedestrian crossing scheduling relied on the TLC
(Azimi, Bhatia, Rajkumar, & Mudalige, 2015) relies on of signalized intersections. For example, (Skikos,
vehicles to manage the intersection autonomously Machia, & Christopoulos, 1993) aims at improving
using V2V communication. The vehicles share infor- the TLC so that it becomes able to emulate the deci-
mation, e.g., speed, position, and destination, with the sion process of an experienced crossing guard.
aim of detecting and avoiding collision by making Nevertheless, TLC-based methodologies are deemed
timely decisions for accelerating and decelerating. inefficient for autonomous vehicles since it requires
Accordingly, autonomous (self-driving) vehicles can signalized intersections and still cause frequent stop-
cross the intersection without need for stopping, ping. Employing vehicle-centric pedestrian collision
which significantly reduces the congestion rate at the avoidance schemes is deemed more appropriate for
intersection. Nevertheless, these intersection manage- supporting safe crossing while limiting the impact on
ment systems assume that autonomous vehicles are vehicular traffic.
the only road user and ignore the pedestrian’s need of The approach of (Franke et al., 1998) is an example
crossing the road. of vehicle-based solutions for supporting autonomous
pedestrian collision avoidance, where a self-driving
vehicle is assumed to be able to detect pedestrians on
Autonomous vehicle-pedestrian
its travel path. A fuzzy steering scheme is employed,
collision avoidance
where the vehicle changes the lane and returns back
The incorporation of pedestrian detection technologies to the original lane after passing pedestrians.
is a fundamental requirement for autonomous vehicles Nonetheless, this approach considers a pedestrian as
(Dollar, Wojek, Schiele, & Perona, 2012). An example an obstacle that should be avoided on the road and
of these technologies is the monocular pedestrian not as a road user that has a right for safe crossing.
detection based on an intelligent real-time vision sys- In other words, a pedestrian has no privilege on the
tem, e.g., zebra crossing detection (Brandstaetter, road. In addition, the approach could be chaotic since
Yannis, Evgenikos, & Papantoniou, 2011). However, changing a lane could limit a vehicle’s ability in
assuming that autonomous vehicles are able to detect detecting a pedestrian in the new lane in a
and stop for pedestrians would not address the con- timely manner.
cern on vehicular flow and very much depends on Meanwhile, the approach of (Bagheri, Siekkinen, &
pedestrian behavior; this is particular a major concern Nurminen, 2014) is based on cellular communication
in crowded urban areas such as Manhattan in New between the vehicle and the pedestrian smartphones.
York City. To highlight how pedestrian behavior The mobile phones exchange awareness messages
4 S. E. HAMDANI ET AL.

periodically containing information such as speed and aforementioned, AIM systems outperform the TLC-
position. Thus, a vehicle will receive an alert from based management model because a vehicle avoids
pedestrian smartphone. One of the concerns about stopping at the intersection, which wastes time not
this approach is the impact on battery life and pedes- only in waiting but also due to deceleration and accel-
trian position accuracy. (Bachmann, Morold, & David, eration. Nonetheless, current AIM systems support
2017) opted to improve pedestrian position accuracy non-signalized intersections without provision for
and minimizing battery consumption using pedestrian pedestrian crossing. To address this shortcoming, we
context information such as speed, location and direc- enable the pedestrian crossing in the middle of each
tion. Nonetheless, this approach is not practical since road, as illustrated in Figure 1(b). Hence, pedestrian’s
a pedestrian cannot be deprived from safe crossing a road crossing will be decoupled from intersections
road if not holding a smartphone. Unlike published and the efficiency of the AIM system’s handling of
pedestrian collision avoidance schemes, our APC sys- vehicles at the intersection is sustained.
tem does not require any infrastructure or pedes- In addition, a refuge island will be added between
trian devices. every pair of consecutive lanes. To accommodate the
addition of refuge islands, the shoulder of the road
System model and approach overview will be narrower, or even eliminated. If necessary, a
lane may become narrower at the pedestrian crossing
Design goal subject to the minimum width standard (Dixon et al.,
Our proposed APC system opts to manage safe pas- 2015; Karim, 2015).
sage of pedestrians on a road crossing while reducing Each pedestrian aiming to cross the road has to
the vehicle’s travel delay and thus decreasing conges- walk on the pathway following these rules:
tion level mainly is urban highways. APC consists of
two modules, namely, pedestrian crossing detection  A pedestrian crosses the road on the pathway lane
and vehicle motion coordination. The following sum- by lane.
marizes the features provided by our system:  If any vehicle does not occupy the pathway of a
lane, the pedestrian can cross such a lane.
 APC does not require additional infrastructure for Otherwise, the pedestrian will wait safely at the ref-
the road. uge island.
 APC supports autonomous intersection manage-  While in the middle of a crossing lane, a pedes-
ment with no need of vehicle stopping trian has priority and walks freely without stopping
at junctions. for coming vehicles.
 Contrarily to conventional AIM systems (Azimi
et al., 2015), APC considers a pedestrian as a road System model and assumptions
user and provisions for safe pedestrian crossing.
 APC privileges the pedestrian over other road Our proposed system makes the following assump-
users and factors in variations in human’s tions about the road configuration, pedestrians, and
motion skills. vehicle’s capabilities:
 APC leverages on-vehicle sensors to detect pedes-
trians and avoid collision.  Road: We assume bidirectional road segments.
 No devices, smartphone applications or V2P com- Traffic in each direction can be over one or mul-
munication are required for a pedestrian to cross tiple lanes, as illustrated in Figure 2. A vehicle con-
the road. siders some safety distance away from the crossing
 V2V communication is exploited to enable inter- to stop at, if a pedestrian is walking through its
vehicle coordination and minimize vehicle stopping lane; hence, we assume that a line is drawn on the
at pedestrian crossing. road to mark such a safety distance. Lanes are sep-
arated at the crossing with refuge islands where
pedestrian can stand on before crossing next lane.
Proposed pedestrian crossing model
 Pedestrian: The typical walking speed for a healthy
The conventional pedestrian crossing model relies on adult is about 1.3 m/s (Mohler, Thompson, Creem-
the use of TLC, as illustrated in Figure 1(a). Regehr, Pick, & Warren, 2007). However, a pedes-
According to this model, pedestrian crossing is trian can also be an elder, child, or handicapped,
allowed at the end of each road segment. As holding a pet, or pushing a stroller. Therefore, in
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 5

Figure 1. Illustration of pedestrian crossing models: (a) Four-way signalized intersection managed by a TLC. Traditional pedestrian
crossings are surrounding the intersection; (b) The crossing model of the proposed APC system. The intersection is non-signalized,
and pedestrian crossings are located in the middle of each road segment and their number are halved to decrease vehicles flow
on the road and mainly at the intersections areas.

our APC system we consider that pedestrian cross- receive information, i.e., position, speed, decision
ing speed is variable. We do not force the pedes- (cross, accelerate, or decelerate) and actual status
trian to accelerate while crossing the road. (stopping, or moving). In APC, a vehicle takes into
Moreover, we do not require a pedestrian to carry account only the part of the crossing that belongs
a smartphone or any other device. to the lane on which it is traveling.
 Vehicles: We assume that vehicles are equipped  Communication: We assume that vehicles can com-
with front cameras and other sensors to detect municate and broadcast Cooperative Awareness
pedestrians, recognize pedestrian crossing, and Messages (CAMs) (ETSI, 2011) to vehicles in their
safety line marks before arriving at them. Likewise, vicinity. To do so, vehicles are to be equipped with
we assume that a vehicle has access to a digital wireless technologies, e.g., dedicated short-range
map database and is equipped with a Global communication (DSRC) or Wireless Access in
Positioning System (GPS) receiver. Moreover, a Vehicular Environments (WAVE) (SAE, 2009).
vehicle is to have sufficient computation resources
to process the collected data and make decision
Pedestrian collision detection
on whether to stop or move in real time.
Furthermore, the vehicle can interact with other In our APC system, vehicles use the front camera to
vehicles using V2V communication and can share/ detect pedestrians at the designated crossing area
6 S. E. HAMDANI ET AL.

Figure 2. Illustrations of APC road model for a sample configuration with two directions: (a) One lane per direction. (b) the case
of three lanes per direction. The crossing is provisioned in the middle of the segment. A Safety Line (SL) is drawn in each road dir-
ection to mark the Safety Zone (SZ) for coming vehicles. The crossings are provided with Refuge Islands (RI) to allow pedestrian to
stand on them safely.

before arriving at it. A vehicle needs to only detect The crossing is a critical area where the safety of
pedestrians in the crossing area of the lane where it is pedestrians could be risked. Since a vehicular hit of a
currently traveling. On the other hand, we assume pedestrian could be severe and may cause injuries or
that a pedestrian detects an approaching vehicle in the even fatalities, we add a safety zone before the pedes-
lane before stepping inside such a lane; this is more trian crossing area. The width of such safety zone
or less an intuitive precaution that is contemporarily (SZ) is based on the distance a vehicle will travel
followed. Contrariwise, when in the middle of the after activating the brake until it completely stops.
lane, the pedestrian does not take into account any To calculate SZ, first we consider the kinetic
approaching vehicles. energy E:
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 7

line. In such situation, the vehicle applies an emer-


gency brake and stops in the area between the cross-
ing and the safety line and guarantees a safe crossing
for the pedestrian.
APC requires a pedestrian to cross the roads in
strides, lane by lane. As outlined in pedestrian cross-
ing process diagram in Figure 4, a pedestrian waits at
the refuge island between two consecutive lanes and
proceeds if it is safe. Such safety assessment is made
visually, e.g., seeing no approaching vehicle close to
Figure 3. An illustration of emergency braking at the safety
zone due to late stopping decision. The vehicle decides to the safety Line (SL).
cross after arriving to the safety line and stop before the cross- For vehicles, we define the collision region of a
ing safety. lane Li as the part of the pedestrian Crossing Dk–CLi
in Li combined with the safety zone Dk–SZi: (Dk–SZi
E ¼ 1=2 mv2 (1) þ Dk–CLi) which is delineated by SL. In the example
in Figure 5, the collision region for L2 of direction D1
where m is the mass of the vehicle (kg), and v is the
is shaded (D1–SZ2 þ D1–CL2). Thus, a vehicle travel-
speed at the start of braking (m/s). On the other
ing on D1–L2 will stop only if a pedestrian exists in
hand, the work W done by braking is determined by:
the collision region of the current lane. When yielding
W ¼ lmg d (2) to a pedestrian is deemed necessary, a vehicle slows
where l is the coefficient of friction (unit less), g is down to stop and broadcasts a CAM to alert the
acceleration due to gravity (m/s2), and d is the trav- vehicles that follow on the same lane. The CAM con-
eled distance (m). The braking distance given in an tains the vehicle’s ID, direction, the lane number,
initial driving speed v is found by putting W ¼ E, vehicle-speed, vehicle GPS coordinates, and the
from which the width of the safety zone is derived as planned action. When the front vehicle announces its
follows: stop, all CAM recipients check their position rela-
tively, e.g., GPS coordinates, the lane number and
l m g d ¼ 1=2 mv2 (3)
travel direction, and decide on their response
v2 accordingly.
SZ ¼ d ¼ (4)
2lg It is worth noting the effect that a pedestrian cross-
Since the vehicle’s speed could significantly vary in ing may have on vehicular collision. A recent study
practice, we consider the speed limit to calculate the (Arvin, Kamrani, & Khattak, 2019) has pointed out
road safety zone. The coefficient of friction l, also that the probability of rear-end crashes could be
changes according to the road condition, i.e., icy, wet impacted by the intersection geometry and traffic
or normal. Therefore, this coefficient should be calcu- density at the intersection. Specifically, lateral acceler-
lated based on the typical conditions of the concerned ation volatility may contribute to increased rear-end
road. The value of acceleration due to gravity is fixed collision at intersections. However, the pedestrian
and equal to 9.80 (m/s2). crossing proposed in this article is placed in the mid-
In order to avoid collision, the SZ is delineated on dle of a road segment and involve refuge islands;
the road by the Safety Lane (Figure 2) as a landmark therefore, lateral acceleration volatility is not an issue
for both vehicles and pedestrians. If a coming vehicle and only longitudinal control of the vehicle is
Vi decide to stop it will stop at the safety line and not required. Moreover, advanced systems based on
at the crossing. Therefore, Vi has enough space and Variable Speed Limit (VSL) control (Wu, Abdel-Aty,
time to stop before the crossing in critical circumstan- Wang, & Rahman, 2019) could be further employed
ces where a stopping decision is taken late or a cross- to mitigate rear-end collisions.
ing decision is taken simultaneously from the vehicle
and pedestrian. For instance, Figure 3 illustrates the Pedestrian crossing protocol
case when both a pedestrian and a vehicle detect sim-
Detailed APC protocol
ultaneously that the collision region of the crossing is
free and then decide to cross. Therefore, the vehicle APC opts to improve the vehicular throughput and
cannot stop at the safety line because it detects the travel time while supporting safe pedestrian crossing.
pedestrian on the crossing lane after arriving to the The main idea is to determine when a vehicle should
8 S. E. HAMDANI ET AL.

Figure 4. Flow diagram of pedestrian crossing procedure. A pedestrian crosses the road in steps according to the number of lanes.
A stop is made at the Refuge Island (RI) between every pair of lanes to check whether the next Crossing Lane (CL) and Safety
Zone (SZ) area is free before crossing the Lane (L).

Figure 5. Illustration of pedestrian-vehicle collision region on each lane. The first direction of the road (D1) is composed of three
lines (L1, L2 and L3). The collision region in each lane includes the Crossing-Lane (CL) plus the Safety Zone (SZ); for instance,
(D1–SZ1 þ D1–CL1) represents the collision region of the first line of this direction. The collision regions for the other lanes are
marked in the same way.

stop, slow down, or continue traveling when  Passing-Message: this is for a vehicle Vx to inform
approaching a pedestrian crossing. APC is lane-based its neighbors that it is entering the collision region
and relies on the vehicle’s on-board cameras (and/or in its lane without slowing down.
other sensing device on board a vehicle) to detect  Stopping-Message: this CAM is to announce that
existing pedestrians in its current lane. Figure 6 shows vehicle Vx will stop at the safety line.
a flow diagram summary of the steps. First, the
approaching vehicle detects the status of the con- When the next coming vehicle Vy in the same lane
cerned Crossing Lane (Dk–SZi þ Dk–CLi) using the receives the CAM of Vx, Vy reacts based on the pro-
front camera. If there is no pedestrian, the vehicle cedure in Figure 7. Accordingly, Vy reads CAM and
proceeds and crosses the pathway. Otherwise, i.e., the checks first if Vx is traveling in the same Direction
lane-crossing is occupied by pedestrians, the vehicle and Lane (if it is not Vx’s CAMs do not concern Vy).
tries to avoid stopping by decelerating while Then, Vy checks if it is very close to Vx, which means
approaching the Safety Line. If the crossing-lane is that the distance between the two vehicles is equal to
still busy, the vehicle brakes at the Safety Line and the Safety Distance (SD). SD is the minimum gap
stays waiting as long as there are pedestrians in this required between two consecutive vehicles and it is
crossing-lane. calculated based on the same formula described in the
Accordingly, the approaching vehicle broadcasts a previous section. Thus, if Vx and Vy are not close
CAM for every decision it takes; either it is stopping enough; Vy ignores Vx’s actions; hence, it decelerates
or crossing. Generally, a vehicle will need to send and and execute the first algorithm. In the other case, Vy
receive two types of CAM: takes the decision based on Vx’s action and does not
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 9

Figure 6. Flow diagram of the APC protocol execution at a vehicle Vi while approach a pedestrian crossing. The approaching
vehicle detects the crossing state using Front Camera (FC) and check the collision region (Safety Zone þ Crossing Lane), takes
accordingly a decision of crossing or decelerate in the aim of avoiding stopping, and informs the following vehicles by broadcast-
ing a CAM message.

Figure 7. Flow diagram of next coming vehicle Viþ1 process. The next coming vehicle takes its decision (crossing, decelerating or
stopping) based on the approaching vehicle CAM and based on the distance between the two vehicles.
10 S. E. HAMDANI ET AL.

Figure 8. An illustration of different scenarios of pedestrians Figure 9. An illustration of pedestrians causing a blocking situ-
on the crossing, along with various positions of vehicles that ation due to a congested Lane. Pedestrians are standing on
are willing to pass the crossing. Refuge Island (RI) and waiting to cross the congested Lane (L).
Pedestrians that are willing to cross (D1–CL3) and cannot find
a room to stand on Refuge Island will stand on the collision
region (D1–CL2) and thus cause a blocking situation for
need to check the collision region using its own cam- vehicles traveling on D1–L2.
era. Accordingly, Vy crosses after Vx and stops if it is
stopping that it can decrease the delay of passing the Pedestrian freedom from blocking situation
crossing. Vy; in its turn, broadcasts a CAM periodic- Pedestrians could wait for long time at a refuge island
ally for each action it takes to aware other com- if the next lane is experiencing high vehicular traffic.
ing vehicles. According to APC, each vehicle takes the action of
Figure 8 illustrates APC operation for all possible the one ahead if they are close to each other.
scenarios. To simplify the figure reading, we assume Therefore, if we assume that a lane is congested or
that all pedestrians are crossing in one direction slow for any reason, i.e., due to turn left down the
(from D1 to D2). In the first lane, L1, of direction D1, road, vehicles will follow each other, and the pedes-
vehicle V1 is still far from the safety line. When trians will be queuing as shown in Figure 9 and even-
detecting the pedestrian P1, V1 decelerates while tually blocking the other lanes one after the other
approaching to avoid stopping and broadcasts a CAM when they cannot find a room to stand on the Refuge
to announce such deceleration. In lane L2 of the same Island. Thus, the crossing-pathway will be in a block-
direction, V3 has to stop and stays still until pedes- ing situation.
trian P2 crosses the crossing pathway of the lane. V3 According to the manual on Uniform Traffic
broadcasts a “Stopping Message”; hence, V2 will stop Control Devices (UTCD) of U.S. Federal Highway
as well since the two vehicles are close to each other. Administration, the width of a crossing shall be not
On the other hand, V4 will continue and pass through less than 3 m (The Federal Highway Administration
the pedestrian crossing since lane L3 is free of (FHWA), 2010). On the other hand, the average
pedestrians. of men’s shoulder width is about 0.465 m ’ 0.5 m
Likewise, in direction D2, vehicle V5 crosses and (Pamela Buxton, 2015). Thus, about six pedestrians
does not wait for pedestrian P3 because P3 is stand- (3 m  0.5) can stand. If we consider the presence of a
ing on the refuge island RI4. V5 broadcasts a crossing gap between each two pedestrians, five could be the
message; yet V7, which follows V5 on the same lane, maximum number of pedestrians that could wait at a
will not take any action because it is relatively far. P3 refuge island. Accordingly, if the number of waiting
will eventually cross before V7 arrives at the crossing. pedestrians exceeds five, they will be queuing in the
Meanwhile, vehicle V6 is decelerating to avoid stop- middle of a lane. Since pedestrians queuing should
ping while pedestrian P4 is crossing. On the other not block the pathway and consequently the lane,
hand, P5 will cross immediately the last lane and no APC imposes the following rule (1):
action is needed because there are no com- Rule (1): If an AVi detects more than five waiting
ing vehicles. pedestrians (wP) in its lane, it cannot cross the
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 11

pathway and has to stop and wait at the SL until the not 100% CAV and we discuss the limitations of
pedestrians
P cross. human-driver based vehicles in this context.
detected wP > 5 ) Action ¼ “Stop”
Correspondingly, we add rule (2) to the pedes-
Supporting traditional vehicle in APC
trian process:
Rule (2): If the waiting pedestrians are more than APC relies on the on board sensors and autonomy of
five, they can cross the lane without waiting. AVs. Nonetheless, human driver is naturally able to
This rule ensures pedestrians privilege in crossing sense the surrounding environment, to detect possible
the road while enabling vehicular flow. Pedestrians obstacles and to have quick reaction. Furthermore, the
will not pileup as the rule controls the flow pedes- APC algorithms are designed based on what naturally
trians and prevents them from blocking any lane a driver would do to avoid stopping in non-signalized
while trying to cross the next. In addition, pedestrians roads without hurting a pedestrian. Thus, a human-
will not wait indefinitely if the road is congested. It is driven vehicle is able to apply the majority of the
important to note that an autonomous vehicle is typ- instructions of APC protocol except sending and
ically capable of monocular vision, is thus able to receiving messages, which make our solution more
count pedestrians at both side of its lane flexible to cover mixed traffic situations. We note that
(Mitzel, 2013). in the context of our solution CAMs are mainly
APC protocol takes on consideration vulnerable exchanged to reduce traffic delay and minimize stop-
pedestrians that are not capable to stand on RI for a ping rather than exchanging safety and security warn-
long while such as pedestrian with a baby, pedestrian ings. The following discusses the two distinct
carrying a pet or pedestrian using a wheelchair. scenarios related to how a Human-driven Vehicle
Accordingly, APC protocol ensures that such vulner- (HVi) affects the operation of APC.
able pedestrians cross immediately without waiting on
refuge islands. Furthermore, the protocol guarantees Approaching a pedestrian crossing
that strollers or wheelchairs would not congest refuge In the case where APC is applied in mixed traffic
islands. Thus, APC imposes rule (3) for AVs: scenario, the human driver of an approaching would
Rule (3): If a coming AVi detects any object on the visually sense the collision region related to the cur-
two neighboring RIs that is distinctive from pedestrian rent lane (SZiþCLi). Upon detecting pedestrians inside
shape, it has to stop at SL until the RI become free. (SZiþCLi), the driver will decelerate and prepare to
Respectively, we add rule (4) to the pedes- stop at the safety line. Otherwise, the driver continues
trian process: on and avoids stopping. The driver of HVi applies
Rule (4): If the pedestrian is using a wheelchair, rule (2) when seeing a neighboring refuge island fully
pushing a stroller or carrying a pet, they can cross the occupied with pedestrians, and rule (3) in the pres-
lane without waiting. ence of strollers or wheelchairs. In these two cases,
the driver stops immediately despite of the fact that
no pedestrian is present inside (SZiþCLi).
APC in mixed traffic scenario
Nonetheless, HVi may not broadcast CAMs informing
Autonomous vehicles are seen as the future of ITS, about the taken decision unless such a feature is sup-
and is envision to solve many pressing issues related ported. We note that a driver is expected to abide by
to road safety and traffic congestion. APC is designed APC rules as part of the traffic regulation, where vio-
mainly for a Connected and Autonomous Vehicles lation will be subject to penalties.
(CAV) system where each vehicle is equipped with
computational resources and can cooperate based on Effect on successors on a lane
V2V communication. However, transitioning to full When HVi follows any vehicle, whether autonomous
CAV system is expected to be gradual and AVs will or human-driven, no APC rule would specifically
have to share the road with human-driven vehicles apply since until it is the turn for HVi to pass the
(Olia, Razavi, Abdulhai, & Abdelgawad, 2018). pedestrian crossing. The immediate successor of HVi
Therefore, cooperative traffic management systems, on the lane may be either a human-driven vehicle
such as APC, should be applicable to traffic scenario HViþ1 or an autonomous one HViþ1. Regardless the
of mixed autonomous and human-driven vehicles. In type, such a successor will not receive CAM broad-
this section, we discuss the possibility to apply APC casted by HVi, unless HVi is equipped with such tech-
protocol in the special scenario where the system is nology. In case of HViþ1, its driver will see HVi and
12 S. E. HAMDANI ET AL.

will react according what HVi does, e.g., deaccelerate, drawn based on SSD where both kinds of vehicles are
stop, etc. analyze action taken by it. Human driver supposed to stop.
is able to know clearly if the front vehicle is stopping
or moving. Accordingly, when HViþ1 and HVi are
Positives and limitations
very close to each other (d(HViþ1, HVi) ¼ SD), HViþ1
keeps going or stops depending on HVi visually recog- Under APC, a HV can navigate safely in a mixed traf-
nized action. fic with AVs and pass the pedestrian crossing without
If HViþ1 is not close enough to HVi (d(HViþ1, colliding with pedestrians, since a human driver is
HVi) > SD), HViþ1 slows down and approaches to able to detect vehicles and pedestrians visually, analyze
(SZiþCLi) and applies Algorithm (1) as an approach- situations, and react accordingly. Furthermore, a
ing normal vehicle HVi. As for HVi, driver of HViþ1 human driver can recognize full refuge island, strol-
is not required to apply instructions related to lers, wheelchairs and other obstacles better than
receiving and broadcasting CAMs. AVs systems.
Accordingly, normal vehicle is able to simplify the
operation of the APC protocol. Moreover, like AVs, a
Safety distance for mixed traffic human-based vehicle is not obliged to stop at the
Human eye especially at night can detects obstacles crossing as long as its lane is free, even if there are
only at a distance of 75 m compared to AV sensors pedestrians walking in other parts of the crossing.
that can detects up to 250 m afar using long-range However, an HV does not apply instructions
radar (Yan, 2016) which gives less time to human related to sending receiving CAMs. Consequently, it is
driver to react. Moreover, based on the observation of not able to communicate with other vehicles and can-
human driver behavior, 85% of drivers need up to not detect the state of the collision zone when it is
2.5 s as perception-reaction time (Layton & Dixon, not in the front. Such a limitation could slightly
2012). Thus, safety distance should be increased extend the crossing time as a human driver would be
in a mixed traffic to ensure safety of passengers in more cautious.
normal vehicles.
Furthermore, Safety Sight Distance (SSD) is the Protocol safety and practicality
minimum distance available on a highway at any spot
having sufficient length to enable the human driver In this section, we analyze how APC will sustain its
to detect an obstacle, react, e.g., stop, safely without design goal under non-ideal scenarios, namely, when
collision (Layton & Dixon, 2012). In essence, SSD is there is a failure in the vehicle and when pedestrians
the sum of the Lag Distance (LD) and the Braking do not comply with the traffic rules mandated by
Distance (BD) as determined by: APC. The scenarios could be experienced in practice.

SSD ¼ LD þ BD (5)
Pedestrian safety
where LD is the distance the vehicle travels during
the reaction time t and is given by (6), where v is the Safety is the biggest issue related to ITS and
velocity in m=s2. autonomous vehicles, since an error in the system
could affect human wellbeing or even life (ElHamdani
LD ¼ vt (6)
& Benamar, 2018). The presentation so far assumes
BD can be determined using (4), similar to that autonomous vehicles work perfectly and
the safety distance discussed in Section 4. Thus, SSD pedestrians follow the crossing process. However, we
is determined by (7), where l is coefficient of explain in the following how APC remains safe even
friction (unite less), and g is acceleration due to in the absence of these assumptions:
gravity (m/s2):
 Communication Fault: APC relies on V2V commu-
v2
SSD ¼ v t þ (7) nication in order to provide vehicles with the
2lg
necessary information to be able to adjust speed
Since SSD is bigger than SD (SSD > SD þ vt) and and to avoid stopping. Actually, V2V is fundamen-
in order to ensure safety, we increase the safety tally required to realize cooperative driving among
distance in mixed traffic for both AVs and human autonomous vehicles and not just for applying
driver-based vehicles. Thus, the Safety Line (SL) is APC. Although communication failure could affect
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 13

Figure 10. Pictures (a) and (b) of real examples of Refuge Islands “B” installed on Pedestrian Crossing “A”. Pictures are for
a crossing in a road in Ellicott City, Maryland, USA.

the efficiency of the cooperative driving, it does  Standing in the middle of road: In APC, pedestrians
not have any impact on pedestrian safety. It is cross the road lane by lane; thus a pedestrian
worth noting that a self-driving vehicle is equipped would need sometimes to stand within the road.
with sensors, e.g., Lidar, to detect possible APC advocates the addition of refuge islands,
collisions; therefore, communication failure can be which is popular in residential areas and around
mitigated safely by vehicles. schools. They are meant not only to provide a safe
 Faulty Camera: APC relies on the vehicle’s ability waiting spot for pedestrians but also motivate
to detect pedestrians on their way and determine vehicles to slow down. The pictures in Figure 10
their proximity. The front camera on a self-driving show an example of a refuge island on an
vehicle is typically able to detect obstacles up to 80 existing road in Ellicott City, Maryland USA.
m ahead, which allows the vehicle to calculate Installing islands on roads has been demonstrated
the deceleration rate to avoid hitting a pedestrian. to decrease the pedestrian crashes and casualties
A vehicle is often equipped with sensors to com- rate by 57 to 82% (U.S. Department of
plement and mitigate camera failure. For example, Transportation, n.d.).
a vehicle will be able to detect pedestrians by  Pedestrian Misbehavior: One of the possible safety
Thermal Infrared sensor and Lidar (Ger onimo, issues is pedestrian’s compliance with the traffic
L
opez, Sappa, & Graf, 2010). Thus, pedestrian regulations. This issue is of concern even when
safety will not be affected. traffic lights are present. Thus, good pedestrian
14 S. E. HAMDANI ET AL.

behavior is not guaranteed in our system. We motion. In other words, there is no reason for a
argue though that we are not causing increased self-driving vehicle to slow down if no pedestrians
risk of casualties. Fundamentally, our proposed are present at the crossing. Our simulation results
pedestrian crossing process matches the natural confirm that APC does not diminish the through-
reflex of humans. A recent study of pedestrian put when pedestrian traffic is very low, e.g., during
behavior while crossing the street (Rasouli, late night hours.
Kotseruba, & Tsotsos, 2018) has shown that pedes-  Pedestrians Convenience: As pointed out earlier,
trians trend to move their head (checking the the pedestrian crossing in APC matches natural
road) even in the presence of traffic lights. human behavior (Rasouli et al., 2018); a pedestrian
Furthermore, pedestrians are more cautious in will cross as an opportunity arises and is not
busy roads and in wider streets. This study has obliged to wait for a sequence of traffic light
also concluded that pedestrians anticipate the phases in the conventional TLC system. The lane-
vehicle action based on its speed and on the traffic by-lane progress will facilitate crossing wide and
condition. Moreover, the insertion of refuge islands busy roads. Moreover, according to the US Federal
will lower the risk since a pedestrian will have a Highway Administration (U.S. Departement of
safe waiting area after a short walk across a lane. Transportation, n.d.), refuge islands permit pedes-
trians to be concerned with only one direction of
traffic at a time and reduce exposure time on the
Practicality
vehicle travel path. Furthermore, APC system pri-
As discussed, APC relies on the cooperation of oritizes vulnerable pedestrians with a stroller a pet
autonomous vehicles and require limited additional or a wheelchair and allows them to cross the road
infrastructure support. This subsection argues the immediately.
practicality of APC’s assumptions:

 Road Architecture: Our proposed pedestrian cross- Performance evaluation


ing model can be applied to both new and existing We have validated the performance of APC through
roads. For existing roads, small changes are simulation. This section discussed the simulation
required by redrawing crossing lines in the middle setup, performance metrics and the obtained results.
of the road and adding refuge islands between the
lanes. It is worth mentioning that our model ena-
Simulation tools
bles the designation of multiple crossing areas and
not necessarily in the middle of road segments; We have implemented APC using the open source
this particularly useful for large blocks and will be traffic simulator SUMO (Simulation of Urban
advantageous for pedestrians who otherwise have Mobility). SUMO is one the most widely used stand-
to walk all the way to the intersection. It is import- ard simulation platforms that are used for purely
ant to note that adopting autonomous vehicles will microscopic modeling whereby each vehicle is mod-
eliminate traffic lights (Badger, 2015); the changes eled explicitly and moves individually through the
required by APC can be viewed as part of the network. We built our road network within XML files
adjustment in the road infrastructure to support and we implemented APC within Python scripts using
such vehicular autonomy. Traffic Control Interface (TraCI) API. To provide
 Road Width: Shifting lanes at pedestrian crossing is access to SUMO (acting as a server), TraCI uses a
a very popular safety mechanism. The rationale is TCP based client/server architecture.
to force drivers to slow down. Figure 10 shows a As aforementioned, APC protocol is mainly dedi-
picture of one of the crossing in the US, where the cated for urban highways with low pedestrian density.
shoulder becomes narrower to make room for a Thus, we consider a simulation area composed of one
refuge island in the middle of the road. Although road of two directions. In addition, we deactivated
the boundaries of a lane shift at the crossing, the randomization for vehicles speed since in this context
lane does not necessarily become narrower and we rely on speed as a factor of congestion. Thus, we
continues to be consistent with the standard width. have set the maximum allowed speed for vehicles on
For self-driving vehicles, we do not envision even the road to 45 MPH, i.e., 20.11 m/s, which is consist-
any effect of lane shifting on the vehicle speed ent with contemporary urban environments, so the
given the autonomous control of the vehicle vehicle travel with this speed unless it need to slow
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 15

Table 1. Simulation parameters.  Walk Duration: The time that a pedestrian takes to
Parameter Value walk the considered path from start to end. To
Simulator SUMO 0.32.0 qualify the walk duration, it is compared with the
Algorithms Code language Python 3.6.3
Simulation time 3600 s smallest walk time tWmin, which reflects the case
Number of road directions 2 when the pedestrian walks at the maximum speed
Number of lanes in each road direction 3
Vehicle’s Speed 45 MPH ’ 20.11 m/s (value set in simulation parameters) on an empty
Pedestrian’s Speed 1.39 m/s path with no other vehicles or pedestrian traveling.
Road length 200 m
Pedestrian flow (1st set of experiments) 480 p/h  Walk Time Loss: Time wasted due to stopping or
Traffic volume (2nd set of experiments) 600 v/h (per direction) walking with a lower speed than the value set in
Safety Distance 20 m
Stopping Sight Distance (SDD) 70 m
simulation parameters. It equals the Pedestrian
Camera detection distance <80 m Walk Duration – tWmin.

Simulation results
down for a specific reason. Likewise, we define pedes-
trian walking speed as 1.39 m/s (Mohler et al., 2007). We conducted two sets of experiments. In the
Based on the maximum allowed speed (45 MPH) first, we fixed the average pedestrian crossing rate
and the coefficient of friction of 0.8 on a dry asphalt, at 100 p/h and varied the traffic volume. We changed
the safety distance (SD) is set to 20 m. For the third the pedestrian crossing rate in the second set of
set of experiment we set a safety distance equal to experiments while fixing vehicles traffic volume at 600
Stopping Sight Distance (SSD) of 70 m. SSD is the v/h (per direction). We compare the performance of
sum of Braking Distance which is 20 m, and Lag APC to the conventional TLC approach. We consider
Distance 50 m which is calculated as follows in (8): two configurations for the TLC: (i) long green dur-
SSD ¼ BD þ LD; ation (traditional Traffic light model of 30 s) which is
BD ¼ SDfSD  20 m for v ¼ 80 Km=hg; known to be more efficient during heavy traffic, and
(8)
LD ¼ vtfv ¼ 80 Km=h; t is perception time  2:5 sg; (ii) short green duration (traditional Traffic light
SSD ¼ 20 þ 80  2:5 ¼ 70 m model of 10 seconds) which suits low traffic volume,
The vehicles front camera is 360 and assumed to be i.e., more privilege for pedestrians. All results reflect
able to detect pedestrians up to 80 m (Information, 2013). the average over 20 simulation runs. The error bars
In the experiments, we vary traffic flow and cross- for 95% confidence intervals are plotted along with
ing rates of pedestrian to assess the efficiency of our the results.
protocol in term of reducing waiting time of vehicles.
The pedestrian crossing rate reflects the average num- Exp. Set 1: Performance under increased
ber of pedestrians who walk through the crossing traffic volume:
pathway in both directions per time unit. Table 1 In this set of experiments, we fixed the pedestrian
summarizes the simulation parameters. The following density at 480 p/h and increased traffic flow rate from
metrics are used to evaluate vehicles’ experience: 100 to 1200 v/h (per direction). Figure 11(a) shows
that our protocol outperforms traffic light models in
 Vehicle Travel Duration: The time that a vehicle term of vehicle travel duration. Moreover, the effi-
takes to travel the considered road from start to ciency of APC is almost stable and does not decrease
end. To qualify the travel duration, we compare it with traffic flow volume. Meanwhile the performance
with the smallest travel time tTmin, which reflects of the conventional traffic signal model degrades with
the case when the vehicle travels at the maximum increased traffic.
allowed speed on an empty road with no other Figure 11(b) shows the average lost time due to
vehicles or pedestrian crossing. waiting for pedestrians and in essence explains the
 Vehicle Time Lost: Time wasted due to stopping or rise of the duration in traffic light models and the
driving with a lower speed than the maximum steady performance of APC. Time lost in traffic light
allowed. It equals the Vehicle Travel Duration – tmin. models is quite high even in a low traffic and it wor-
 Vehicle Waiting Time: This is the time wasted by a sens with the increase in traffic volume. TLC with
vehicle due to stopping only. long green duration performs better in high traffic
due to the time “wasted” at every signal switching;
To evaluate pedestrians’ experience, we used the Meanwhile TLC with short green duration performs
following metrics: well in low traffic because each road segment is served
16 S. E. HAMDANI ET AL.

more frequently. However, APC outperforms both


traffic light models whether the vehicular traffic is
high or low traffic; Such distinct performance is due
to the negligible waiting time spent when APC is
applied, as shown in Figure 11(c). Consequently, the
time lost in case of APC in Figure 11(b). is domin-
antly due to deceleration and not to stopping.
Figure 12 evaluates pedestrian’s experience under
increased vehicular traffic flow. Figure 12(a) shows
that pedestrian walk duration at the crossing under
APC protocol increases with the traffic flow contrarily
to TLC models. However, the walk duration is very
high under Traffic Light 10 s and higher under Traffic
Light 30 s even when vehicles flow is very low (100 v/
h). Figure 12(b) shows that the walk time loss under
APC protocol is negligible especially in low density
(less than 5 s) and very low in high traffic volume
ranging from 4.62 s at 600 v/h (Per Direction) to
11.76 s at 1200 v/h. Compared to APC, walk time loss
under TLC models is very huge starting from 11.19 s
under Traffic Light 10 s and from 16.03 s under
Traffic Light 30 s in very low traffic.

Exp. Set 2: Performance under growing pedestrian


density:
In this set of experiments, we fixed traffic flow rate at
600 v/h (per direction) and we increased pedestrian
density from 60 p/h to 2400 p/h to evaluate the effect-
iveness of APC compared to Traffic Light 10 s and
Traffic Light 30 s.
Figure 13(a) compares the efficiency of APC to that
with the two baselines in term of vehicle travel dur-
ation. The results in the figure show that the trip time
slightly increases with the density of pedestrians.
Nevertheless, APC outperforms the traffic light mod-
els in all pedestrian density volumes.
As shown in Figure 13(b), APC does not waste the
vehicles time under low pedestrian density and scales
nicely with the rise in pedestrian density. Meanwhile
the traffic light model extends the travel time for
vehicles significantly even with little pedestrian cross-
ing; as expected the travel time grows with the
extended green time for pedestrians.
Figure 13(c) shows that the majority of the time
lost under the TLC models is for waiting due to Figure 11. Comparison of vehicles delays under increased traf-
unnecessary stopping during red light phases. fic volume based on: (a) Average Travel duration, (b) Average
Contrarily, the figure confirms that the waiting time Vehicle Time Lost and (c) Average Waiting Time.
in APC does not grow with increased pedestrian
crossing, where the average waiting time for APC cross one lane in a low traffic volume and a low dens-
varies between 0.63 s in low pedestrian density, and ity volume.
6.03 s in high density. Such waiting time almost equals APC involves vehicle stopping at the crossing and
to the stopping duration for one or few pedestrians to increases the travel duration compare to some AIM
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 17

Figure 12. Comparison of pedestrian delays under increased


traffic volume based on: (a) Average Walk Duration and (b)
Average Vehicle Time Loss.

systems. However, those systems are based on consid-


ering autonomous vehicles as the only road user.
Thus, they could not be relevant for the real word’s
roads in urban areas. Accordingly, we assume that
vehicle stopping on the road should be minimized but
cannot be avoidable. Our approach solves many prob-
lems related to pedestrian management on the road.
On the one hand, it prioritizes the pedestrian and
guarantees safe passage. On the other hand, the sys-
tem decreases vehicle-stopping time at the crossing.
According to the simulation, the TLC model forces
a waiting vehicle to wait 30 s or 10 s even in the
absence of pedestrians in the crossing, which causes
vehicles to pile up. On the other hand, our APC
protocol decreases the waiting time to one lane pedes-
Figure 13. Comparison of vehicles delays under increased traf-
trian crossing time. Thus, the vehicle does not stop fic volume based on: (a) Average Travel duration, (b) Average
unless the pedestrian is crossing its lane. Moreover, Vehicle Time Lost and (c) Average Waiting Time.
18 S. E. HAMDANI ET AL.

the cooperation between the vehicles allows them to


avoid stopping by decelerating.

Exp. Set 3: Performance in mixed traffic scenario:


In this set of experiments, we have run simulation
for three scenarios of mixed traffic. The first,
namely “APC_25%HV”, has 25% of HV vehicles and
75% are AVs. Such a ratio is reversed in third
scenario “APC_75%HV”, where 75% of the vehicles
are human-driven. In the second scenario
“APC_50%HV”, the number of HV and AV vehicles
are equal. Meanwhile, we have fixed the pedestrian
density at 480 p/h and increased traffic flow rate from
100 to 1200 v/h (per direction).
Figure 14(a) compares the travel duration for the
three mixed traffic scenarios to that of a full CAV
configuration and of traffic light 10 s and 30 s. The
results show that performance in mixed traffic is
slightly lower (travel duration is longer) compared to
the full CAV scenario. The travel duration also grows
with the increased HV population. For instance,
the average travel duration under heavy traffic of 1000
v/h (per direction) increases from 16,65 s in “APC”, to
19,9 s in “APC_25%HV”, to 22,5 s in “APC_50%HV”
and to 24,6 s in “APC_75%HV”. The results are very
much expected and is attributed to the decreased level
of coordination as the vehicles approach pedestrian
crossing when more HVs are involved. Nonetheless,
APC still significantly outperforms the traffic signal
models even with 75% of the vehicles are HV.
Figure 14(b) shows the average lost time; the
results are consistent with Figure 14(a) and in fact
explains the difference in travel duration among the
full CAV APC, APC in mixed traffic, and traffic light
models. Basically, the time lost in traffic light
increases significantly, especially with traffic light 10 s,
as the traffic gets heavier. Meanwhile, the performance
of APC is very much independent of vehicle density
for both full CAV and mixed traffic scenarios.
Figure 14(c) shows that the average waiting time
under APC is very small and is independent of the
type of vehicles on the road. Meanwhile traffic lights
models are experiencing a very high waiting time Figure 14. Performance of different mixed scenarios traffic
which is increasing with traffic. Figure 14(c) explains under increased traffic volume compared to full CAV APC and
that the difference in time loss between different the traffic light models based on: (a) Average Travel duration, (b)
Average Vehicle Time Lost and (c) Average Waiting Time.
full CAV and mixed vehicle types scenarios is due to
traveling with lower speed (decelerating) and not to
stopping time. On one hand, the safety distance has force them to decrease speed when approaching the
increased from 20 m in case of CAV to 70 m (SSD) in pedestrian crossing, especially when there is another
mixed traffic which obliges vehicles in mixed traffic to vehicle in front and they are not able to know
decelerate in for longer time. On the other hand, the its decision.
lack of communication among human-driven vehicles
JOURNAL OF INTELLIGENT TRANSPORTATION SYSTEMS 19

In summary, the results for mixed traffic scenarios driving at intersections. Proceedings—IEEE 21st
confirm the APC is still as effective and that the pres- International Conference on Embedded and Real-Time
Computing Systems and Applications, RTCSA 2015
ence of human-based vehicles does not impact the
(pp. 167–175). doi:https://doi.org/10.1109/RTCSA.2015.20
coordination at the pedestrian intersection. Bachmann, M., Morold, M., & David, K. (2017). Improving
smartphone based collision avoidance by using pedestrian
context information. 2017 IEEE International Conference
Conclusion and future work on Pervasive Computing and Communications
AIM systems manage the traffic at the intersection Workshops, PerCom Workshops 2017 (pp. 2–5). doi:
10.1109/PERCOMW.2017.7917507
efficiently and are expected to replace the conven-
Badger, E. (2015). 5 confounding questions that hold the
tional TLC model. Although V2V communication has key to the future of driverless cars. Retrieved from
been exploited for autonomous traffic management, https://www.washingtonpost.com/news/wonk/wp/2015/01/
little attention has been paid to supporting pedestrian 15/5-confounding-questions-that-hold-the-key-to-the-future-
crossing. Thus, this article fills an important technical of-driverless-cars/?noredirect=on&utm_term=.33b1906e93d9
gap in AIM. The goal of this article is to include ped- Bagheri, M., Siekkinen, M., & Nurminen, J. K. (2014).
Cellular-based vehicle to pedestrian (V2P) adaptive com-
estrian crossing in autonomous traffic management. munication for collision avoidance. 2014 International
We have developed APC, a cooperative protocol for Conference on Connected Vehicles and Expo, ICCVE
pedestrian crossing management. APC pursues a new 2014 - Proceedings, November (pp. 450–456). doi:10.
pedestrian avoidance scheme using V2V communica- 1109/ICCVE.2014.7297588
tion. We have also studied and solved the pedestrian Bento, L. C., Parafita, R., Rakha, H. A., Nunes, U. J., Conde,
L., Parafita, R., … Urbano, J. (2019). A study of the
blocking situation on the pedestrian crossing. environmental impacts of intelligent automated vehicle
Compared to TLC models, our protocol diminishes control at intersections via V2V and V2I communica-
the vehicle stopping time due to pedestrian crossing tions control at intersections via V2V and V2I communi-
and thus increases the road throughput and decreases cations. Journal of Intelligent Transportation Systems,
congestion rate in both full CAV system and in mixed 0(0), 1–19. doi:10.1080/15472450.2018.1501272
Brandstaetter, C., Yannis, G., Evgenikos, P., & Papantoniou,
traffic scenarios.
P. (2011). Traffic safety basic facts 2010: Single vehicle
The next research steps will be related to classifying accidents.
pedestrians crossing in clusters of different sizes and Buxton, P. (2015). Metric handbook: Planning and design
estimate how different scenarios will influence delays data (5th ed.). Buxton: Routledge.
for both pedestrians and vehicles. Future work Dixon, K., Fitzpatrick, K., Avelar, R., Perez, M., Ranft, S.,
includes considering more vulnerable road users as Stevens, R., … Voigt, T. (2015). Reducing lane and
shoulder width to permit an additional lane on a freeway
bicycles (ElHamdani & Benamar, 2019) and motor- (Technical Report). 7(2), p. 104. doi:10.3141/2588-10
cycles in autonomous traffic management. Autonomous Dollar, P., Wojek, C., Schiele, B., & Perona, P. (2012).
vehicles should undertake the existence of vulnerable Pedestrian detection: An evaluation of the state of the
users and prioritize their safety while increasing traffic art. IEEE Transactions on Pattern Analysis and Machine
flow as possible as it is. Intelligence, 34(4), 743–761. doi:10.1109/TPAMI.2011.155
Dresner, K., & Stone, P. (2008). A multiagent approach to
autonomous intersection management. Journal of Artificial
Acknowledgments Intelligence Research, 31, 591–656. doi:10.1613/jair.2502
Elhamdani, S., & Benamar, N. (2017). A comprehensive
This work was supported by the Grant Project ITIC- study of intelligent transportation system architectures for
TRANSPORT Moulay Ismail University of Meknes. road congestion avoidance. In International Symposium
on Ubiquitous Networking (Vol. 10542, pp. 95–106). doi:
10.1007/978-3-319-68179-5
ORCID ElHamdani, S., & Benamar, N. (2018). Autonomous traffic
Sara El Hamdani http://orcid.org/0000-0002-5358-835X management: Open issues and new directions. 2018
International Conference on Selected Topics in Mobile
and Wireless Networking (MoWNeT) (pp. 1–5). doi:10.
References 1109/MoWNet.2018.8428937
ElHamdani, S., & Benamar, N. (2019). DBDA: Distant
Arvin, R., Kamrani, M., & Khattak, A. J. (2019). How bicycle detection and avoidance protocol based on V2V
instantaneous driving behavior contributes to crashes at communication for autonomous vehicle-bicycle road share.
intersections: extracting useful information from con- 2019 International Conference on Wireless Technologies,
nected vehicle message data. Accident Analysis & Embedded and Intelligent Systems, WITS 2019 (pp. 1–6).
Prevention, 127, 118–133. doi:10.1109/WITS.2019.8723866
Azimi, R., Bhatia, G., Rajkumar, R., & Mudalige, P. (2015). ETSI. (2011). ETSI TS 102 637-2 vehicular communications;
Ballroom intersection protocol: Synchronous autonomous Basic set of applications; Part 2: Specification of cooperative
20 S. E. HAMDANI ET AL.

awareness basic service. History, 1, 1–18. doi:10.1145/ Murphy, J., Djahel, S., Jabeur, N., Barrett, R., & Murphy, J.
2656877.2656887 (2015). Toward V2I communication technology-based
Florentine, E., Ang, M. A., Pendleton, S. D., Andersen, H., solution for reducing road traffic congestion in smart
& Ang, M. H. (2016). Pedestrian notification methods in cities toward V2I communication technology-based
autonomous vehicles for multi-class mobility-on-demand solution for reducing road traffic congestion in smart
service. Proceedings of the Fourth International cities. International Symposium on Networks, Computers
Conference on Human Agent Interaction – HAI ’16 and Communications (ISNCC), September, 1–6. doi:10.
(pp. 387–392). doi:10.1145/2974804.2974833 1109/ISNCC.2015.7238584
Franke, U., Gavrila, D., Gorzig, S., Lindner, F., Puetzold, F., Olia, A., Razavi, S., Abdulhai, B., & Abdelgawad, H. (2018).
& Wohler, C. (1998). Autonomous driving goes Traffic capacity implications of automated vehicles mixed
downtown. IEEE Intelligent Systems, 13(6), 40–48. doi: with regular vehicles. Journal of Intelligent Transportation
10.1109/5254.736001 Systems, 22(3), 244–262. doi:10.1080/15472450.2017.
Geronimo, D., L opez, A. M., Sappa, A. D., & Graf, T. 1404680
(2010). Survey of pedestrian detection for advanced Rasouli, A., Kotseruba, I., & Tsotsos, J. K. (2018).
driver assistance systems. IEEE Transactions on Pattern Understanding pedestrian behavior in complex traffic
Analysis and Machine Intelligence, 32(7), 1239–1258. doi: scenes. IEEE Transactions on Intelligent Vehicles, 8858(c),
10.1109/TPAMI.2009.122 1–1. doi:10.1109/TIV.2017.2788193
Gradinescu, V., Gorgorin, C., Diaconescu, R., Cristea, V., & Rodrıguez Palmeiro, A., van der Kint, S., Vissers, L., Farah,
Iftode, L. (2007). Adaptive traffic lights using car-to-car H., de Winter, J. C. F., & Hagenzieker, M. (2018).
communication. IEEE Vehicular Technology Conference Interaction between pedestrians and automated vehicles:
(1, pp. 21–25). doi:10.1109/VETECS.2007.17 A wizard of Oz experiment. Transportation Research Part
Information, P. (2013). Mercedes-Benz intelligent drive: The F: Traffic Psychology and Behaviour, 58, 1005–1020. doi:
intelligent car. Retrieved from Daimler Communications, 10.1016/j.trf.2018.07.020
70546 Stuttgart/Germany - Mercedes-Benz - A Daimler SAE. (2009). Dedicated short range communications
Brand website: http://media.daimler.com/marsMediaSite/ (DSRC) message set dictionary. Surface Vehicle Standard,
en/instance/ko/Mercedes-Benz-Intelligent-Drive-The-intelligent- 2735. doi:10.4271/J2735_201603
car.xhtml?oid=9904196 Shladover, S. E. (2017). Connected and automated
Karim, D. M. (2015). Narrower lanes, safer streets. CITE vehicle systems: Introduction and overview. Journal
Conference, June (pp. 1–21). Retrieved from https://www. of Intelligent Transportation Systems, 2450, 1–11. doi:
academia.edu/12488747/Narrower_Lanes_Safer_Streets_ 10.1080/15472450.2017.1336053
Accepted_Paper_for_CITE_Conference_Regina_2015_ Skikos, G. D., Machia, A. V., & Christopoulos, S. A. (1993).
Layton, R., & Dixon, K. (2012). Stopping sight distance - Application of fuzzy logic to the control of wind energy.
Discussion Paper #1. (April). Retrieved from http://cce. Proceedings. Joint International Power Conference Athens
oregonstate.edu/sites/cce.oregonstate.edu/files/12-2-stopping- Power Tech, 2(98), 30–38. doi:10.1109/APT.1993.673945
sight-distance.pdf The Federal Highway Administration (FHWA). (2010).
Li, C., & Shimamoto, S. (2011). A real time traffic light con- Mini-roundabouts. Retrieved from https://safety.fhwa.dot.
trol scheme for reducing vehicles CO2 emissions. The 8th gov/intersection/innovative/roundabouts/fhwasa10007/
Annual IEEE Consumer Communications and Networking fhwasa10007.pdf
Conference – Emerging and Innovative Consumer U.S. Department of Transportation. (n.d.). Traffic Calming:
Technologies and Applications, June (2010, pp. 855–859). 25. Crossing Islands. Retrieved from https://safety.fhwa.
doi:10.1109/CCNC.2011.5766627 dot.gov/saferjourney1/Library/countermeasures/25.htm
Li, Z. (Richard), Chitturi, M. V., Yu, L., Bill, A. R., & Wang, S., Djahel, S., & McManis, J. (2014). A multi-agent
Noyce, D. A. (2015). Sustainability effects of next- based vehicles re-routing system for unexpected traffic
generation intersection control for autonomous vehicles. congestion avoidance. 2014 17th IEEE International
Transport, 30(3), 342–352. doi:10.3846/16484142.2015. Conference on Intelligent Transportation Systems, ITSC
1080760 2014, 2541–2548. doi:10.1109/ITSC.2014.6958097
Mitzel, D. (2013). Taking mobile multi-object tracking Wu, Y., Abdel-Aty, M., Wang, L., & Rahman, M. S. (2019).
to the next level (PhD Thesis), 566–579. Combined connected vehicles and variable speed limit
Mohler, B. J., Thompson, W. B., Creem-Regehr, S. H., Pick, strategies to reduce rear-end crash risk under fog condi-
H. L., & Warren, W. H. (2007). Visual flow influences tions. Journal of Intelligent Transportation Systems, 1–20.
gait transition speed and preferred walking speed. Yan, C. (2016). Can you trust autonomous vehicles:
Experimental Brain Research, 181(2), 221–228. doi: Contactless attacks against sensors of self-driving vehicle.
10.1007/s00221-007-0917-0 doi:10.1145/1235

You might also like