A vehicular network–based intelligent transport system for smart cities

Abstract

Smart cities and the Internet of Things have enabled the integration of communicating devices for efficient decision-making. Notably, traffic congestion is one major problem faced by daily commuters in urban cities. In developed countries, specialized sensors are deployed to gather traffic information to predict traffic patterns. Any traffic updates are shared with the commuters via the Internet. Such solutions become impracticable when physical infrastructure and Internet connectivity are either non-existent or very limited. In case of developing countries, no roadside units are available and Internet connectivity is still an issue in remote areas. In this article, we propose an intelligent vehicular network framework for smart cities that enables route selection based on real-time data received from neighboring vehicles in an ad hoc fashion. We used Wi-Fi Direct–enabled Android-based smartphones as embedded devices in vehicles. We used a vehicular ad hoc network to implement an intelligent transportation system. Data gathering and preprocessing were carried on different routes between two metropolitan cities of a developing country. The framework was evaluated on different fixed route-selection and dynamic route-selection algorithms in terms of resource usage, transmission delay, packet loss, and overall travel time. Our results show reduced travel times of up to 33.3% when compared to a traditional fixed route-selection algorithm.

Keywords

Internet of Things intelligent transport system route selection vehicular ad hoc networks

Introduction

With advances in ubiquitous wireless communication and commodity sensors, Internet of Things (IoT) has become an emerging area of research.¹ In particular, the area can play a crucial role in developing technologies for smart homes, intelligent transportation systems (ITSs), smart cities, and so on. The entire concept of IoT-based applications relies on the physical infrastructure deployed. It is worthwhile to mention that research challenges are completely different in developing countries with very limited supporting infrastructure.² In developed countries, roadside units (RSUs) are used to facilitate daily commuters by re-routing them to avoid congestion. With people driving to and back from work using similar routes and around the same hours. In urban areas, during such peak hours, commuters face traffic congestion resulting in economic loss, pollution, and even deaths due to traffic accidents with help not arriving in time after road traffic injuries (RTIs). In developing countries like Pakistan, the number of vehicles on roads is increasing at a higher rate. Alone in Karachi (Pakistan), almost 16,562 new vehicles are on roads every month³ leading to an economic loss of almost US$2.8 billion per year due to traffic congestion.⁴ Similarly, in India, the economic loss due to traffic congestion is around US$8.4 billion per year.⁵ More alarmingly, air pollution caused due to this congestion has an adverse impact on health.⁶ Arguably, even with limited resources, the main focus of these countries is to build new road infrastructure to alleviate the problem, somewhat neglecting the use of ITSs.

Problem statement

Road accidents are one of the leading causes of deaths around the world with 3500 deaths every day and about 1.27 million deaths every year.⁷ Notably, 90% of the deaths occur in low- and middle-income countries.⁸ For instance, in one massively populated city of Pakistan (Rawalpindi), traffic jams led to increased death rates due to first responders failing to reach the location of the accident within the golden hour.⁹ Here, the use of ITSs for efficient route selection can help rescue services avoid traffic congestion in real time. However, limited resources, that is, no roadside sensors and limited Internet connectivity in developing countries, is the main focus of our study. We have initiated a way for developing countries to be the part of the world of smart cities without any costly infrastructure or with using resources easily available.

Proposed framework

In this work, we propose an intelligent transport (iNET) framework to overcome the aforementioned challenges, mainly focused on developing countries to facilitate drivers making timely decisions in the absence of Internet connectivity and other RSUs. The framework disseminates vehicular data via an inter-vehicular communication network, where every vehicle in the network shares its information to compute a traffic congestion index. The index is used to select a suitable path to the destination.

Main contributions

The main contributions of this work are briefly listed:

iNET is a data sharing framework designed to support intelligent transportation network in smart cities. Using ad hoc communication for information sharing, the proposed framework can work in the absence of Internet and other physical infrastructure.

The framework computes an index measuring congestion based on the information received from other vehicles. This index is incorporated into multiple single-source shortest path algorithms to select the least congested path to the destination, recalculated in real time at every upcoming intersection.

The framework is designed to work on low power devices, subsequently, evaluated on resource-limited android handheld devices in terms of resource usage, transmission delay, packet loss, and total travel time.

The rest of the article is organized as follows: section “Literature review” revisits the existing literature on vehicular communication. Section “Proposed intelligent network (iNET) framework” presents the proposed iNET framework. The performance evaluation is presented in section “Performance evaluation.” Finally, section “Conclusion” concludes the article.

Literature review

Vehicular communication is an emerging IoT technology that includes vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), and vehicle-to-pedestrian (V2P) communication. More recently, the term vehicle-to-everything (V2X) incorporates all the aforementioned specific types of communication involving vehicles. Figure 1 illustrates a V2X communication. In recent literature, many opportunities and challenges have been identified to enable IoT for developing countries, especially in the field of transportation, safety, agriculture, environmental monitoring, and social security management. The core challenges to support such IoT-based applications are smooth Internet connectivity, device resources, data center connectivity, device reliability, financial challenges, and privacy issues.

Figure 1.

V2X communication scenario involving V2V, V2I, and V2P communications.

Vehicular communication technologies

Over the last decade, lightweight commodity devices using cellular technology have gained interest for V2V communication.¹⁰ The devices gather and forward relevant information to a server without driver intervention. In developed countries, many of the works done on V2V use either dedicated short-range communications (DSRCs) or cellular technology. In Abboud et al.,¹¹ a hybrid architecture—a combination of DSRC and cellular technology—is used for an efficient transportation system. Chang et al.¹² introduce near-field informatics (NFI) based on vehicular infrastructure. The solution is compared against technologies like Bluetooth, ZigBee, Wi-Fi, and RFID for ad hoc communication in terms of range, speed, and built-in handset support. The results show that Wi-Fi and Bluetooth are suitable for ad hoc communication.

More recently there is a shift from cellular to vehicular ad hoc networks referred to as to Internet of Vehicles (IoV). One such study¹³ uses smartphones as Wi-Fi hotspots connecting clients and other hotspots. The evaluation results demonstrate a significant reduction in data transfer delay using the hotspot approach. Similarly, another study implements a decentralized traffic management approach, especially for developing countries.¹⁴ The approach establishes a communication link for information sharing when the vehicles are in range for V2V communication. Other similar studies explore V2V, V2I, and infrastructure-to-infrastructure (I2I) communication in terms of big data.

ITSs

Researchers at Nissan conducted a study regarding fuel economy in cars, and they observe an improved fuel economy up to 7.8% for routes to destination without congestion.¹⁵ Based on these findings, many studies propose different congestion detection algorithms for ITS applications. RoadRunner¹⁶ is an in-vehicle V2V-based application providing voice-over instructions to drivers, while SimMobility¹⁷ demonstrates a simple token-based strategy to overcome traffic congestion by limiting the number of vehicle on road.

Meneguette et al.¹⁸ develop a neural network–based intelligent protocol for congestion detection. They use a multi-layer perceptron with an input layer comprising two neurons, one for vehicle speed while the other for vehicle density corresponding to the number of neighboring vehicles on the road. The system is simulated using Simulation of Urban Mobility (SUMO) with a map of urban and highway scenarios integrated with a model for co-emission and fuel consumption. The authors report a reduction in average trip time, co-emission, and fuel consumption. However, such data-driven approaches are compute-intensive. Neal et al.¹⁹ investigate the energy consumption of the data-driven approach and propose an energy-efficient architecture specifically laid out for mobile platforms. Backfrieder et al.²⁰ suggest a V2X communication scheme based on predictive congestion minimization with $A *$ (PCMA*) algorithm to minimize consumption of fuel and total travel time. The scheme is simulated using TraffSim to detect, predict, and re-route vehicles to avoid traffic congestion, subsequently reducing fuel consumption up to 47.3% and total travel times up to 71.8% compared to simple re-routing algorithms. Furthermore, an improved model in Parrado and Donoso²¹ suggested a solution to overcome the congestion problem for individual vehicular trips but taking into account the entire vehicular network for the global congestion (GC) algorithm. Similarly, energy-efficient routing and direction-aware forwarding protocol for vehicular ad hoc networks are discussed in Baker et al.²² and Haider et al.²³

Information sharing

Often end-to-end route selection is done based on heuristics, local congestion metrics, and distances traveled with no communication among vehicles. In literature, there are limited works related to cooperative V2V, for instance, in Froehlich and Krumm,²⁴ route selection is done based on global positioning system (GPS) data gathered during vehicle past trips. They evaluate the viability of this approach using location-based data of 250 drivers collected for Microsoft multi-person location survey (MSMLS).

More complex selection strategies use trajectories from the past vehicle paths.²⁵ An extension of such work is a dynamic data-driven approach.²⁶ The evaluation results using the T-Drive data set comprising GPS trajectories of 10,000 taxicabs in Beijing demonstrate an improved route prediction based on the current location of a vehicle without knowing its final destination. Wan et al.²⁷ propose a model to detect traffic congestion using mobile crowd sensing (MCS) by forwarding data from vehicles onto traffic clouds. Simulated experiments show an efficient path selection in terms of time of arrival at the destination.

Edge/fog computing for vehicular networks

Vehicular edge computing (VEC) is a three-layered architecture: the top most is cloud layer, next is edge layer, and the last is smart vehicular layer. The cloud layer is used for batch processing or high-level processing tasks. The edge cloud layer is to ensure connection and real-time interaction between cloud layer and the smart vehicular layer. However, the smart vehicular layer comprises group of vehicles that share storage and computing resources through wireless network.²⁸ Hou et al.²⁹ proposed the concept of utilizing smart vehicles for infrastructure-as-a-service for computations. This strategy is termed as vehicular fog computing (VFC). It is a cost-effective solution providing real-time communication and geographically distributed support. Furthermore, the differences between mobile edge computing (MEC) and fog computing (FC) for vehicular communication and architecture are highlighted in Borcoci et al.³⁰

Social Internet of Vehicles

Ning et al.³¹ emphasize the use vehicular social network (VSN) to enhance trust and quality of transportation system. The authors define VSNs as the combination of social networks and IoVs as VSNs contain the human factor in addition to V2V and V2I communication. Similarly, the concept of social Internet of Vehicles (SIoVs) is covered in Ning et al.³² Furthermore, the authors consider SIoV as the emerging field to guarantee road safety and to enhance quality of driving. However, the focus of the referred study is to design a cooperative quality-aware service access (CQS) system for SIoVs. Another study³³ declared SIoV as a promising field for traffic management and road safety in smart cities. The authors investigate MCS techniques to overcome end-to-end delay in store-carry-forward-based vehicular networks. Furthermore, the authors propose a multi-tier device-to-device (D2D) framework to enable crowdsensing-based real-time traffic management.

Summary of literature review

Potentially, similar ITS-based applications can facilitate rescue services as demonstrated in a study by Mercedes-Benz.³⁴ They deploy a V2V-based collaborative model to facilitate emergency services saving lives. The effectiveness of the model is backed using statistical data analysis with avoidance of 80% of the road accidents. Currently, emergency services, for instance, eCall in Europe and OnStar in the United States, are using V2I. Future improvements can be the combination of both V2V and V2I to further reduce first responder arrival time.

Notably, many of the existing works focus on route selection using sensors and the Internet. In this article, we propose a framework specifically designed to work in the absence of RSUs and Internet connectivity. The framework used a decentralized approach to compute the least congested route. In contrast to other offline routing algorithms, it provides real-time route guidance by dynamically adapting to changing traffic conditions. The motivation is to make the transportation system more efficient, and provide a cooperative and safe driving experience via sharing of near real-time congestion information, and hence, alleviating economic loss and environmental pollution. The next section covers the proposed framework in detail.

Proposed intelligent network (iNET) framework

The proposed intelligent network (iNET) framework is implemented using a modular approach. The core modules of framework are maps manager, visualization module, storage manager, communication module, and route selection module. The architectural diagram of iNET framework is illustrated in Figure 2. The functionality of each module is explained below:

1. Maps manager: the maps manager module is responsible to store and manage maps. The module used previously acquired maps to prepare offline map tiles. These tiled maps are stored and later used for handheld devices.

2. Visualization module: the visualization module is used to present the current location of the vehicle using offline map tiles. The visualization module acquires the map from the maps manager using a location-based query. The visualization module also displays congestion on nearby road segments. The current location of a vehicle is important for route mapping and selection, which is achieved using a built-in GPS available in every embedded device. Using the current device location, the module mapped all possible routes to the destination. Each route to the destination is composed of multiple road segments each labeled with a congestion index calculated based on the information received from other vehicles in near real time. This index was afterward used by the route selection module to avoid traffic congestion.

3. Storage manager: the storage manager module stored traffic information including all possible tracks, congestion index, and travel history, in a repository. The repository was later used for route selection.

4. Communication module: the communication module established an ad hoc communication link with nearby vehicles using Wi-Fi Direct. A dynamic transient network was formed to share traffic information. Algorithm 1 explains the formation of such real-time traffic network based on neighboring vehicles. Here, the state of the vehicle is represented by the four-dimensional vector $ϑ = (ϕ, θ, α, β)$ consisting of latitude, longitude, MAC address, and status, respectively.

Algorithm 1. iNET formation
1 $ConnectVehicle$ ( $ϑ = (ϕ, θ, α, β)$ );
Input: Traffic information from neighboring vehicles
Output: iNET formation
1. if $β = TRUE$ then
$2 . T_{stat} \leftarrow add (ϑ,$ client )
{Already GO of another group, acts as client for this group, and relay for its group members}
3. else if $β = FALSE$ then
4. exit()
{Already client of some group}
5. else
6. $T_{stat} \leftarrow add (ϑ,$ client)
{Not connected, can be client of this group}
$7 . T_{loc} \leftarrow add (ϕ, θ)$
{Save latitude/longitude of new peer}
$8 . T_{addr} \leftarrow add (α)$
{Save MAC address of new peer for future correspondence}
9. end if

Figure 2.

Proposed iNET framework architecture.

For communication between vehicles, the link layer service discovery was achieved through a generic advertisement protocol (GAS). All peers share a MAC address, device name, and connection status to confirm their availability. Once a connection is established, the group formation process gets initiated. All devices within a group support single-hop communications. However, to establish multi-hop communication support for handheld Android devices, we used Wi-Fi Direct with dynamic mobile ad hoc network (MANET) topology.³⁵ The setup maintained routing tables including route information among the peers. Furthermore, routing in the framework was achieved using MAC addresses providing openness to support multiple connections simultaneously.

5. Route selection module: one of the main components of the iNET framework is the route selection module. It takes input from other modules and suggests the least congested route to the destination. The route selection was based on traffic information received from neighboring vehicles connected directly or via ad hoc mechanisms. The information received was used to compute a congestion index of every road segment from the vehicle’s current location. Based on the index, the least congested path to the destination is suggested at the next intersection using well-known shortest path algorithms such as Dijkstra shortest path (DSP), modified Dijkstra shortest path (MDSP), heuristic-based $A *$ search, and optimal uniform cost search (UCS) algorithm.

The next section covers the implementation details of iNET framework.

Implementation of iNET Framework

The proposed iNET framework was designed to work on handheld devices, or onboard units installed in vehicles. The framework allows vehicles to dynamically build a network and share traffic information in real time from nearby vehicles. This information was used to recommend the least congested route to the destination. In addition, each installed onboard device stored its limited travel history. A weighted graph from source to the destination was built using the gathered traffic information. The route costs were calculated using classical single-source shortest path algorithms. Note that the edge weights were initialized as the ratio of the road segment length $(L)$ to the maximum speed on the road segment $(υ^{\max})$ . The cost function $Θ$ used to assign weights to road segments in the graph is

Θ = \frac{L}{υ^{\max}}

(1)

Figure 3(b) illustrates all possible paths from source to destination. The blue-colored polyline represents the shortest path computed using the DSP algorithm. In this study, we considered this computed path as a baseline for comparison with other route selection algorithms. It is evident that traffic conditions are continuously changing and static weights computed initially result in lower results. Therefore, we incorporated changing road segment costs at regular time intervals. This was achieved by maintaining a congestion index for each road segment based on real-time traffic information from neighboring vehicles. Further, a speed averaging method was used to compute the speed of a road segment as a ratio of the average velocity of vehicles to the total number of vehicles on the road segment. The average speed $\bar{υ}$ is calculated as

\bar{υ} = \frac{\sum_{k} υ}{k}

(2)

where $k$ is the total number of vehicles on a road segment. The updated congestion index for the road segment denoted as $Θ$ is given as

Θ = \frac{\bar{υ}}{υ^{\max}}

(3)

where $υ^{\max}$ is the maximum speed of the road segment. This index is used to select the next road segment with minimum congestion. Figure 3(b) illustrates a vehicle’s movement on the least congested route to the destination.

Figure 3.

iNET framework showing the shortest route calculated using Dijkstra algorithm (blue color), while the vehicle move on least congested path determined using traffic information: (a) all possible routes showing in different colors and (b) vehicle selected the less congested route, vehicle movement is shown.

Apart from classical DSP algorithm, we extended it to incorporate changing costs of road segments. The result is a real-time congestion-aware route selection algorithm referred to as MDSP. In the proposed framework, the computation was performed at every intersection to find the next shortest and least congested road segment. The MDSP algorithm is presented in Algorithm 2. Keeping generality, a road segment $R_{i}$ is represented with road segment id $Ω (i)$ , and $R_{Ω (i)}^{k}$ represents $k$ vehicles on the road segment. Moreover, $ξ_{uv}$ represents the distance between two vehicles $u$ and $v$ , which is used to compute a congestion index denoted as $Θ_{i}$ for the road segment.

Algorithm 2. Congestion-based MDSP algorithm
Input: Road segment $R_{i}$ with vehicle $u$
Output: Selected path $ρ$
1. ${υ \| υ \in R_{Ω (i)}^{k}} - u$
2. ${\bar{υ}}_{i} = \sum_{k} υ_{i} / k$
3. $υ_{i}^{\max} = max_{x \to k} (υ_{i}^{x})$
4. $Θ_{i} \leftarrow {\bar{υ}}_{i} / υ_{i}^{\max}$
5. for each $v \in υ$ do
6. $ϕ_{v} \leftarrow Θ_{i} + ξ_{uv}$
7. end for
8. $ρ_{u} \leftarrow \underset{x \to k}{\arg \min} (ϕ_{x})$

The route selection module also includes an implementation of $A *$ and UCS³⁶ algorithms for dynamic route selection based on congestion. At every intersection, the module adopts a route with the minimum congestion index. Algorithm 3 illustrates the implementation details of the $A *$ algorithm where $h_{x}$ is a heuristic function that estimates the cost of the cheapest path from $x$ to the destination.

Algorithm 3: Congestion-based $A *$ algorithm
Input: Road segment $R_{i}$ with vehicle $u$
Output: Selected path $ρ$
1. ${υ \| υ \in R_{Ω (i)}^{k}} - u$
2. ${\bar{υ}}_{i} = \sum_{k} υ_{i} / k$
3. $υ_{i}^{\max} = max_{x \to k} (υ_{i}^{x})$
4. $Θ_{i} \leftarrow {\bar{υ}}_{i} / υ_{i}^{\max}$
5. for each $v \in υ$ do
6. $ϕ_{v} \leftarrow Θ_{i} + ξ_{uv} + h_{v}$
7. end for
8. $ρ_{u} \leftarrow \underset{x \to k}{\arg \min} (ϕ_{x})$

The fourth algorithm implemented in the proposed framework is UCS, another variant of Dijkstra algorithm. The UCS is useful for large or infinite graphs.³⁶ The algorithm explores options in every direction requiring no information about the objective. Instead of inserting all nodes or road segments into a priority queue, only the source node is inserted followed by the insertion of other nodes one by one when needed. In the proposed framework, the algorithm found the best route based on the congestion and cumulative cost of a path from source to destination. The algorithm ensures that the cumulative cost of the selected path remained minimum. Algorithm 4 illustrates the implementation details of the algorithm where $π (u)$ represents the cost of the cheapest path taken by the vehicle $u$ to reach its current position.

Algorithm 4: Congestion-based UCS algorithm
Input: Road segment $R_{i}$ with vehicle $u$
Output: Selected path $ρ$
1. ${υ \| υ \in R_{Ω (i)}^{k}} - u$
2. ${\bar{υ}}_{i} = \sum_{k} υ_{i} / k$
3. $υ_{i}^{\max} = max_{x \to k} (υ_{i}^{x})$
4. $Θ_{i} \leftarrow {\bar{υ}}_{i} / υ_{i}^{\max}$
5. for each $v \in υ$ do
6. $ϕ_{v} \leftarrow Θ_{i} + ξ_{uv} + ϕ_{π (u)}$
7. end for
8. $ρ_{u} \leftarrow \underset{x \to k}{\arg \min} (ϕ_{x})$

Performance evaluation

The proposed iNET framework was developed in Java for embedded devices, and all tests were performed using handheld devices. The characteristics of the devices are listed in Table 1.

Table 1.

Handheld devices characteristics.

	Samsung S3	Samsung A3	Samsung Note 4
Battery	2100 mAh	1900 mAh	3220 mAh
RAM	1 GB	1.5 GB	3 GB
OS	Android 4.3	Android 5.0	Android 6.0

The devices being inherently resource constraint, and hence, we benchmarked the proposed data-driven framework in terms of resource usage. Instrumentation was performed to measure CPU, memory, and network usage alongside packet loss and transmission delay. Furthermore, travel time was returned from the route selection algorithms.

Experimental setup

To evaluate the proposed iNET framework, it is more practical to develop a traffic generator rather than using a large number of vehicles with onboard computing units installed. Therefore, to generate traffic patterns from the generator, we gathered traffic information of the twin metropolitan cities of Pakistan (Islamabad and Rawalpindi). Since there are no traffic data sets available for the cities and with no RSUs installed to monitor and share traffic information, we compiled a data set between two points: X interchange (point A) and Y university (point B) between 8 a.m. and 4 p.m. Note that the university is located within Islamabad with a large number of staff and students coming in and out every day from nearby towns and cities. There are 13 different bi-directional routes from the source to the destination; however, during peak office hours, the main routes are usually jam-packed. We used GPS location data (acquired using third-party solutions like GPS Logger (https://gpslogger.app/) and Backitude GPS Location Tracker (https://www.gpsies.com/backitude.do)) to record the location of a vehicle, speed, and travel time between source A and destination B in the morning. Similarly, the steps were repeated from location B to A at 4 p.m. The final data set comprised 6018 records of bi-directional tracks between X university and Y interchange. The raw data, containing GPS points, time stamps, and distances, were transformed to tuples corresponding to individual trips with attributes including average speed, maximum speed, time duration, track length, start time, and source and destination locations.

To assess the quality of the collected data set, we used clustering to measure data consistency and underlying variance. The features used for quality assessment were average speed, maximum speed, time duration, track length, source, and destination locations. We used two clustering techniques: k-means and hierarchical clustering, with the objective to identify different clusters in the data set. We used cluster tendency with silhouette coefficient of $0.85$ for the former, whereas the cophenetic correlation coefficient of $0.69$ for the latter. It turns out that there was one cluster per track, leading to the conclusion that data under assessment was consistent with minimum variance.

The traffic generator module used the aforementioned compiled data set to generate traffic accordingly. The module being executed on different devices injected traffic information to other devices using Wi-Fi Direct, all acting as onboard units installed in a vehicle and emulating traffic information received from neighboring vehicles. Moreover, the device also sent the received traffic information to other connected mobile nodes. After the expiration of a time interval $δ t$ , the vehicle onboard unit updated its position read from the available tracks in the compiled data set. In a real scenario, GPS can be used to provide such position updates. The position updates were followed by sharing of traffic information with other vehicle units. Furthermore, at every intersection, a route was selected based on the recommendation of one of the multiple route selection algorithms implemented in the proposed framework. Table 2 summarizes the parameters used for the experimental setup.

Table 2.

Simulation parameters.

Parameters	Values
Connectivity protocol	Wi-Fi Direct
Map layout	$4 \times 5$
Network type	Ad hoc
No. of vehicles	Variable
No. of road segments	37
Transmission range	60–100 m
Distance between source and destination	45 km
No. of routes between source and destination	13

iNET profiling

Due to a limited resource on handheld devices, the iNET framework is benchmarked in terms of resource usage such as CPU, memory, and network. The device profiled was the Samsung Note 4, and the results were obtained after multiple runs of the proposed strategies. Furthermore, the framework is also tested in terms of transmission delay, packet loss, and total travel time:

CPU usage: the proposed framework used classical DSP-based shortest path algorithm as the baseline. The other three algorithms made use of the traffic information received from neighboring vehicles to compute a congestion index, which was later used to select a path with the least congestion. Figure 4(a) shows CPU usage at intervals for the four route-selection algorithms. We observe that $A *$ , MDSP, and UCS algorithms consumed more CPU cycles due to computing the index at every upcoming road segment and using it to select a route. However, traditional DSP returned the shortest path at the start of the journey, which remained unchanged irrespective of the traffic information received along the journey. Keeping in mind the use of battery operated devices, CPU usage is an important factor determining the suitability of an algorithm for such devices. Furthermore, the results show that $A *$ consumed more CPU cycles compared to the other algorithms. However, all algorithms consumed only $2 % - 3 %$ of the device battery when traveling from source to destination.

Memory usage: Figure 4(b) shows a comparison of memory consumption for different route selection algorithms. We observe that all algorithms resulted in moderate memory usage averaging around 140 MB, considerably lower than many social media applications, for instance, Facebook uses approximately 340 MB. We used Trepn profile (Trepn Profiler application; https://developer.qualcomm.com/software/trepn-power-profiler) for profiling. Furthermore, 24.5 MB was used to store offline map tiles, while 320 KB was used to store the tracks. The usage can increase depending on how far back route history is maintained. Moreover, with an increasing number of vehicles, the computational intelligence (CI)-based algorithms consumed more energy, as they processed more information.

Network usage: Figure 5 shows the total network usage on a device. The proposed framework sent and received packets through MAC address binding denoted as $T_{x}$ and $R_{x}$ , respectively. Moreover, the device transmit and receive rates are the same due to the ad hoc mechanism used. We observe that the maximum network usage during the entire trip was around 230 B/s. Note that classical DSP algorithm neither sent nor received any traffic information during the course of the trip. However, in the case of $A *$ , MDSP, and UCS, $T_{x}$ and $R_{x}$ depend on the number of vehicles in a vehicular cluster and the distance between vehicles. As we emulated the scenario using a traffic generator, therefore, all algorithms are evaluated using a similar set of conditions.

Travel time: we also compared the algorithms in terms of average travel times. The basic DSP algorithm used a precomputed path without considering any incoming traffic information, not the case for $A *$ , UCS, and MDSP algorithms. Figure 6 shows the average travel times from source to destination. We observe that $A *$ resulted in the highest average travel time, as the algorithm used two cost functions—the actual and approximate cost—to compute the cumulative congestion index to determine a congestion-aware shortest route. As a result, the selected route was longer than the actual route initially selected. Moreover, its performance degraded when the graph became larger. In contrast, MDSP resulted in the least average travel time, a significant improvement over traditional DSP. Moreover, the MDSP algorithm performed better compared to UCS, the latter returning routes with smaller cumulative costs and congestion index but relatively larger distances, in turn, increasing total travel time.

Figure 4.

Profiling results for the four route selection algorithms: (a) CPU usage when DSP, UCS, MDSP, and A* algorithms used for route selection and (b) memory usage on handheld devices while different route selection algorithm in use.

Figure 5.

Network usage reported using proposed framework.

Figure 6.

Average travel time using different route selection algorithms.

Communication using Wi-Fi Direct

The transmission delay $D_{T}$ incurred due to Wi-Fi Direct was based on the transmission rate of group owner (GO), clients, and retransmission rates. It is computed using the equation

D_{T} = D_{T}^{O} + D_{T}^{C} + D_{RT}^{O}

where $D_{T}^{O}$ is the total delay from GO, $D_{T}^{C}$ is the total delay for transmission from the client, and $D_{RT}^{O}$ is the total delay for retransmission from GO.

Figure 7(a) shows the number of vehicles and its effect on transmission delay. As the number of vehicles in a group increased, retransmissions also increased. This contributed to a delay as the GO spent most of its time retransmitting messages. Moreover, we analyzed packet loss by varying the network $T_{x}$ and $R_{x}$ . Figure 7(b) shows this loss while varying the distance between vehicles.

Figure 7.

Communication benchmark between devices using Wi-Fi Direct: (a) number of vehicles versus transmission delay in iNET and (b) packet loss while communication between vehicles at various distances.

We used Wi-Fi Direct for communication among vehicles, as it provides better data rates compared to Bluetooth and near-field communication (NFC). Irrespective of it being a recent technology, we observed higher packet drop rates due to moving vehicles, around $60 % - 86 %$ packets dropped when vehicles moved within a 57 m radius. Figure 7(b) shows the time to transmit a 71 bytes data packet between vehicles at variable distances. Here, we observe that the data easily transferred among nearby vehicles but transmission times increased with increasing distances, mainly due to hardware constraints of Wi-Fi Direct. Nevertheless, the technology provides a substitute for RSUs and cellular data for information exchange among neighboring vehicles, and hence, making it feasible for use in the proposed framework.

Discussion

The proposed framework was designed for vehicular ad hoc networks with vehicles communicating via Wi-Fi Direct. The incoming information at every node was processed separately with relay nodes or GOs capable of sending messages to other groups.

Comparative analysis

We analyzed the differences between centralized, decentralized, and distributed approaches for V2V communication from the literature^{14,19,27,37–39} and compared them to our approach. The core attributes analyzed include manageability, security, mobility, cost-effectiveness, and scalability.³⁸ A detailed comparative analysis is presented in Table 3.

Table 3.

Comparison of proposed framework with other techniques.

Property	Proposed iNETframework	Mobile crowdsensing²⁷	Vehicular cloud service^36,37	V2V mobile routers¹⁴
Networking	IoV	IoV + cloud	IoV + cloud	IoV
Interaction	V2V	V2V + V2I +V2H + V2S	V2V + V2I	V2V
Network	VANET	MCS	SVN	VANET
Model	SaaS	VCS	VCS	SaaS
Internet protocol	Wi-Fi Direct for VANET	Wi-Fi	Wi-Fi + cellular	OLSR
Range	100–110 m	60 m	100 m	100 m
Built-in GPS	Yes	Yes	Yes	No
Built-in Wi-Fi	Yes	Yes	Yes	Yes
Cost (US$)	No additional cost	High	High	70–80
Max internal/externalstorage	60 MB/256 MB	1 GB/Unlimited	1 GB/Unlimited	32 MB/32 GB
Real-time routeprediction	Yes	Yes	Yes	Yes
Implementation	Smartphones	Smartphones +onboard sensors	Smartphones	V2V mobilerouters
Onboardequipment (OBE)	Not required	Yes	Yes	No
Roadsideequipment (RSE)	Not required	Yes	Yes	No
Vehicular cloud	Yes	Yes	Yes	Yes
Internet cloud	Not required	Yes	Yes	No
Reliability	Medium	High	Low	Medium
Contents of service	Abundant	Very abundant	Very abundant	Abundant
Improvement	Cost effective+low traveltime+less CO₂ emissions	Less distance traveled+lowtravel time	Less distance traveled+notifications	Low traveltime
Data collection for area	Yes	No	No	No

VANET: vehicular ad hoc network; MCS: mobile crowd sensing; SVN: social vehicular network; VCS: vehicular cloud services; OLSR: optimized link state routing protocol.

Often centralized systems are accurate compared to others approaches.²⁷ This is due to the fact that centralization allows a system to process more data patterns and features in order to take informed decisions. In contrast, decentralized ad hoc frameworks use limited information to make such decisions. However, the response time and costs of centralized systems are higher, as a centralized server is responsible for all the processing and sharing of information among requesting vehicles,³⁷ thus become a single point of failure.¹⁹ Moreover, such systems are not scalable, their performance degrades with increasing number of vehicles.

Goyal³⁹ analyzes different architectures in different scenarios. Their findings indicate that in the absence of existing infrastructure and Internet connectivity, the only feasible option is to use a distributed architecture. The architecture is highly scalable and more stable compared to a centralized one. No doubt the latter is easier to maintain and more secure but at the same time more vulnerable to failures. In this study, the proposed iNET framework is decentralized and based on the P2P model, thus highly scalable.

In this work, road congestion is the only context for decision-making in route selection, which may lead to the selection of a road with quite long driving time. Since there exists the possibility that the selected route could be chosen by too many vehicles, the “optimal” selection then becomes the “worst” choice in a future moment, especially when many vehicles move toward the same direction and take the same route suggestion. Such key issue is not considered in this article.

Sustainable development goals

This section elaborates the importance of adoption of intelligent transportation frameworks for sustainable development. As the proposed framework is designed to work with no physical infrastructure and Internet availability, it can help achieve a number of development goals, especially in developing countries:

Reduction in $C O_{2}$ emission and pollution effects. As mentioned earlier, traffic congestion leads to more air pollution increasing $C O_{2}$ emissions, subsequently, affects health.⁶ Silva et al.⁴⁰ evaluate these real-world situations via simulation, and they observe that due to several stops in traffic congestion more fuel is consumed resulting in more carbon emissions. Furthermore, Backfrieder et al.²⁰ identified the scenarios of $C O_{2}$ emission, fuel consumption in detail. They note that around 30% $C O_{2}$ is generated from vehicles. Therefore, signal scheduling can play an important role in $C O_{2}$ reduction. Importantly, to alleviate this pollution caused by congestion, a controlled vehicular flow on streets can improve the health conditions.^21,41 Therefore, in under developed countries, the proposed scheme can help reduce $C O_{2}$ emissions using real-time route suggestions.

Reduction in economic loss by reducing fuel consumption and travel time. Traffic congestion can lead to high fuel consumption. It has been reported that there is a loss of Rs. 1 million per day in the arterial area of Karachi (Karachi—the most populous city in Pakistan) city due to fuel consumption.⁴² This rise in economic loss is mainly due to the mismanagement of traffic during peak hours. Therefore, countries with limited resources require the use ITSs, similar to one proposed in this work, to achieve sustainable development.

Reduction in deaths after RTIs. Similar frameworks to the one proposed can reduce death rates after RTIs. As per the research conducted in Rawalpindi (Rawalpindi—the fourth largest city in Pakistan by population), traffic congestion has an impact on emergencies; long waiting hours not only cause problems for rescue services but also leads to increase in anxiety and stress.⁴³ Moreover, studies confirm that there is an increase in death rates is due to traffic congestion. Several studies suggested various mechanisms to provide intelligent transportation to rescue services.⁹ The proposed framework is designed for developing countries and to facilitate communities in providing support to rescue services.

Urban planning. As mentioned earlier, the proposed framework uses ad hoc communication, that is, the data are stored on vehicles. This can be gathered for effective town planning, such as road expansion and flyovers. Moreover, such data can be useful in planning public transport routes to facilitate maximum number of commuters, as well as to monitor traffic congestion to help take decisions for instance trees plantation.

In summary, the focus of this study is to provide a workable approach for urban scenarios with no Internet connectivity. Most of the existing works on route selection are developed using sensors and RSUs, and the communication is achieved through Internet. Thus, based on already gathered data from externally placed sensors and received information, the selected routes are disseminated to the end users using Internet. In similar scenario, very less attention is paid to for areas where no such physical devices and Internet are available. Therefore, iNET is a novel framework that is designed to cater developing countries and also the developed countries where Internet connectivity is not available on highways.

In the future, a hybrid approach can be evaluated using RSUs and/or a central server to store traffic data with the help of Internet and physical infrastructure. The data can be further analyzed using the central server to enhance its informativeness. Surely, this would help eliminate frequent message retransmissions by eliminating redundant messages in the vehicular network. That is, messages from the central server can be transmitted only to the GO in the Wi-Fi Direct network, which afterward forwards the message to other vehicles within the group. Moreover, the central server can be deployed to verify message validity to detect misinformation and to minimize other related security issues.

Conclusion

In this study, iNET framework is presented with supported route selection in the absence of physical devices and Internet connectivity. The proposed framework provides implementations of well-known route selection algorithms for congestion-aware traffic routing. A detailed analysis performed in terms of time, energy, and memory consumption shows that the framework takes limited resources with maximum CPU usage around 25% and utilizes less memory and network bandwidth. Moreover, the travel times are also reduced by 33% using the proposed congestion-aware framework.

Footnotes

Appendix 1

Handling Editor: Sergio Toral

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Asad Waqar Malik

Anis Ur Rahman

References

Safe cities: a revolution driven by new ICT (IHS—safe city white paper), October 2016, http://e.huawei.com/mediafiles/MarketingMaterial/EBG/en/af592e6a-cc1b-44f5-982c-bfdefad01f75.pdf

Wikipedia. List of countries by research and development spending. Wikipedia—The Free Encyclopedia, 2017, https://en.wikipedia.org/wiki/List_of_countries_by_research_and_development_spending

Pakistan Today, 3 February 2017, http://www.pakistantoday.com.pk/2011/12/24/city/karachi/in-karachi-16562-more-vehicles-hit-the-roads-each-month/

Dawn News, 29 January 2017, http://www.dawn.com/news/1083354

The Arabian Gazette, 4 January 2017, http://www.arabiangazette.com/india-traffic-congestion-losses/

Levy

Buonocore

von Stackelberg

. Evaluation of the public health impacts of traffic congestion: a health risk assessment. Environ Health 2010; 9(1): 65.

Peden

. World report on road traffic injury prevention, 2004, https://www.who.int/violence_injury_prevention/publications/road_traffic/world_report/summary_en_rev.pdf

World Health Organization. Global status report on road safety2015, http://www.who.int/violence_injury_prevention/road_safety_status/2015/en/

Martinez

Toh

C-K

Cano

J-C

, et al. Emergency services in future intelligent transportation systems based on vehicular communication networks. IEEE Intel Transp Sy 2010; 2(2): 6–20.

10.

Toutouh

Alba

. Light commodity devices for building vehicular ad hoc networks: an experimental study. Ad Hoc Netw 2016; 37: 499–511.

11.

Abboud

Omar

Zhuang

, et al. Interworking of DSRC and cellular network technologies for V2X communications: a survey. IEEE T Veh Technol 2016; 65(12): 9457–9470.

12.

Chang

T-H

Yang

L-K

Hsieh

B-C

. Building iCaution and traffic game in smart cities. In: Hsu

C-H

Wang

Zhou

, et al. (eds) International conference on Internet of vehicles. Berlin: Springer, 2016, pp.82–90.

13.

Liu

, et al. Towards automatic phone-to-phone communication for vehicular networking applications. In: IEEE conference on computer communications (IEEE INFOCOM 2014), Toronto, ON, Canada, 27 April–2 May 2014, pp.1752–1760. New York: IEEE.

14.

Kanchanasut

Boonsiripant

Tunpan

, et al. Internet of cars through commodity V2V and V2X mobile routers: applications for developing countries. KSCE J Civ Eng 2015; 19(6): 1897–1904.

15.

Deguchi

Kuroda

Shouji

, et al. HEV charge/discharge control system based on navigation information. SAE technical paper 2004-21-0028, 2004.

16.

Gao

Peh

L-S

. RoadRunner: infrastructure-less vehicular congestion control. In: Proceedings of the 21st intelligent transport systems world congress, DSpace@MIT, Massachusetts Institute of Technology, 2014. Intelligent Transport Systems, https://dspace.mit.edu/bitstream/handle/1721.1/89791/Peh_Roadrunner%20infrastructure-less.pdf?sequence=1&isAllowed=y

17.

Azevedo

Soh

, et al. SimMobility short-term: an integrated microscopic mobility simulator. Transport Res Rec: J Transport Res B 2017; 2622: 13–23.

18.

Meneguette

Filho

GPR

Guidoni

, et al. Increasing intelligence in inter-vehicle communications to reduce traffic congestions: experiments in urban and highway environments. PLoS ONE 2016; 11(8): e0159110.

19.

Neal

Fujimoto

Hunter

. Energy consumption of Data Driven traffic simulations. In: 2016 winter simulation conference (WSC), Washington, DC, 11–14 December 2016. New York: IEEE.

20.

Backfrieder

Ostermayer

Mecklenbräuker

. Increased traffic flow through node-based bottleneck prediction and V2X communication. IEEE T Intell Transp 2017; 18(2): 349–363.

21.

Parrado

Donoso

. Congestion based mechanism for route discovery in a V2I-V2V system applying smart devices and IoT. Sensors 2015; 15(4): 7768–7806.

22.

Baker

García-Campos

Reina

, et al. GreeAODV: an energy efficient routing protocol for vehicular ad hoc networks. In: Huang

Gromiha

Han

, et al. (eds) Intelligent computing methodologies (ICIC 2018. Lecture notes in computer science), vol. 10956. Cham: Springer, 2018, pp.670–681.

23.

Haider

Abbas

, et al. DABFS: a robust routing protocol for warning messages dissemination in VANETs. Comput Commun 2019; 147: 21–34.

24.

Froehlich

Krumm

. Route prediction from trip observations. SAE technical paper 2008-01-0201, 2008.

25.

Pecher

Hunter

Fujimoto

. Past and future trees: structures for predicting vehicle trajectories in real-time. In: Proceedings of the winter simulation conference 2014 (WSC), Savanah, GA, 7–10 December 2014. New York: IEEE.

26.

Pecher

Hunter

Fujimoto

. Data-driven vehicle trajectory prediction. In: Proceedings of the 2016 annual ACM conference on SIGSIM principles of advanced discrete simulation, Banff, AB, Canada, 15–18 May 2016. New York: ACM.

27.

Wan

Liu

Shao

, et al. Mobile crowd sensing for traffic prediction in Internet of vehicles. Sensors 2016; 16(1): 88.

28.

Raza

Wang

Ahmed

, et al. A survey on vehicular edge computing: architecture, applications, technical issues, and future directions. Wirel Commun Mob Com 2019; 2019: 3159762 (19 pp.).

29.

Hou

Chen

, et al. Vehicular fog computing: a viewpoint of vehicles as the infrastructures. IEEE T Veh Technol 2016; 65(6): 3860–3873.

30.

Borcoci

Vochin

Obreja

. Mobile edge computing versus fog computing in Internet of vehicles, 2018, https://www.iaria.org/conferences2018/filesAFIN18/BorcociVochinObreja_MobileEdgeComputing_Slides.pdf

31.

Ning

Xia

Ullah

, et al. Vehicular social networks: enabling smart mobility. IEEE Commun Mag 2017; 55: 49–55.

32.

Ning

Chen

, et al. A cooperative quality-aware service access system for social Internet of vehicles. IEEE Internet Things 2018; 5(4): 2506–2517.

33.

Wang

Ning

, et al. A city-wide real-time traffic management system: enabling crowdsensing in social Internet of vehicles. IEEE Commun Mag 2018; 56(9): 19–25.

34.

Mercedes-Benz. Mercedes-Benz TV: the new car-to-X technology (Video file; YouTube, 27 August 2013, Web 3 November 2013, YouTube 13 May 2014), https://www.youtube.com/watch?v=FwBgJn-S3EA

35.

Liu

Shen

Yin

, et al. Development of mobile ad hoc networks over Wi-Fi Direct with off-the-shelf android phones. In: 2016 IEEE international conference on communications (ICC), Kuala Lumpur, Malaysia, 22–27 May 2016. New York: IEEE.

36.

Felner

. Position paper: Dijkstra’s algorithm versus uniform cost search or a case against Dijkstra’s algorithm. In: Fourth annual symposium on combinatorial search, 2011, https://www.aaai.org/ocs/index.php/SOCS/SOCS11/paper/viewFile/4017/4357

37.

Kwak

Liu

Kim

, et al. Seeing is believing: sharing real-time visual traffic information via vehicular clouds. IEEE Access2016; 4: 3617–3631.

38.

Guthikonda

Chitta

Tekawade

, et al. Comparative analysis of IoT architectures. Technical report, 2014, https://pdfs.semanticscholar.org/5639/73338b85d2cd9ed1527c3e3709f70a2c809e.pdf

39.

Goyal

. Centralized vs decentralized vs distributed, 20 July 2017, https://medium.com/@bbc4468/centralized-vs-decentralized-vs-distributed-41d92d463868

40.

Silva

Farias

Frey

, et al. Evaluation of numerical models for simulation of real-world hot-stabilized fuel consumption and emissions of gasoline light-duty vehicles. Transport Res D: Tr E 2006; 11(5): 377–385.

41.

Chen

L-W

Sharma

Tseng

Y-C

, et al. Dynamic traffic control with fairness and throughput optimization using vehicular communications. IEEE J Sel Area Comm 2013; 31(9): 504–512.

42.

Ali

Adnan

Noman

, et al. Estimation of traffic congestion cost—a case study of a major arterial in Karachi. Procedia Engineer 2014; 77: 37–44.

43.

Nasir

Chaudhry

Ahmad

, et al. Urban traffic jams, health emergencies and architectural perspective. Science International 2015; 27: 611–613.