Decision Support Dashboard for Trafﬁc and Environment Analysis of a

Smart City

Jorge Pereira

, Susana Sargento

1,2

and Jos

e Maria Fernandes

1,3

Departamento de Eletr

onica, Telecomunicac¸

oes e Inform

atica, Universidade de Aveiro,

Campus Universit

ario de Santiago, Aveiro, Portugal

Instituto de Telecomunicac¸

oes, Campus Universit

ario de Santiago, Aveiro, Portugal

Instituto de Engenharia Electr

onica e Telem

atica de Aveiro, Campus Universit

ario de Santiago, Aveiro, Portugal

Keywords:

Analytics for Intelligent Transportation, Vehicle Information Systems, Smart City, Data Warehouse, Decision

Support Systems, Trafﬁc Management Reporting Systems.

Abstract:

Porto city has an in-place infrastructure of ﬁxed and moving sensors in more than 400 buses and roadside

units, with both GPS and mobility sensors in moving elements, and with environmental sensors in ﬁxed units.

This infrastructure can provide valuable data that can extract information to better understand the city and,

eventually, support actions to improve the city mobility, urban planning, and environment.

Our system provides a full stack integration of the information into a city dashboard that displays and cor-

relates the data generated from buses and ﬁxed sensors, allowing different visualizations over the trafﬁc and

environment in the city, and decisions over the current status of the city. A good example relates bus speed

variation with possible anomalies on the road or trafﬁc jams. Visualizing such information and getting alarms

on anomalies can be a valuable tool to a city manager when taking urban planning decisions to improve the

city mobility.

1 INTRODUCTION

The smart city vision integrates different areas, such

as environment and transportation, as a way to op-

timize its allocated resources, to plan its preventive

maintenance activities, in the end to support better

city decision making. This is possible by incorporat-

ing in the city monitoring systems and smart sensors

to collect data in real time (Hall et al., ). With the

growing trend that cities are facing every day (Merry,

2017), it also comes the increasing number of vehi-

cles that circulate around their streets. As the city

grows, the trafﬁc congestion increases and becomes

a priority to manage the road infrastructure. Optimiz-

ing road trafﬁc is necessary but demanding, so using

different and better approaches to improve the road

management of the city is ideal.

The same vision also considers the environment

quality, so there is a need to improve the air quality by

ﬁrst identifying the sources of pollution to then take

actions to minimize them. One of the sources that are

always identiﬁed is the car trafﬁc pollution. If there is

a way to relate the trafﬁc information with the actual

environment information, anomalies in the trafﬁc can

be detected and a possible solution can be found.

The current work will focus on the data from the

city of Porto and it will depart from its existing IT /

smart city infrastructure (Lopes and Sargento, 2014).

Porto is one of the cities that is implementing smart

solutions to improve the dynamics and intelligence of

the overall city in a wide variety of ﬁelds, from the

environment to the mobility where distinct data of the

city are captured using an in place base of ﬁxed and

moving sensors in more than 400 buses and roadside

units.

Analyzing road trafﬁc in the city without a digital

solution can be done from the driver’s feedback, but

it is difﬁcult to know with accuracy the exact source

of the problem and identify all the variables that make

that contribution. For example, there could be an ac-

cumulation of trafﬁc in an intersection, but the source

of that problem could be at a trafﬁc light 500 meters

before. A decision support system that gathers trafﬁc

data, transforms, correlates and identiﬁes its anoma-

lies can improve the mobility of the city faster and

less costly.

This work proposes a solution based on a dash-

board that serves as a decision support system for

Pereira, J., Sargento, S. and Fernandes, J.

Decision Support Dashboard for Trafﬁc and Environment Analysis of a Smart City.

DOI: 10.5220/0006707603870394

In Proceedings of the 4th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2018), pages 387-394

ISBN: 978-989-758-293-6

387

a city manager. From the dashboard, a trafﬁc man-

ager can pick a road to be analyzed and, with a cer-

tain detail, visualize the trafﬁc proﬁle and its patterns

with the possibility of comparing it from different in-

stances. Also, the environment of the city can be cor-

related with trafﬁc information in a way of detecting

how trafﬁc affects the pollution in the nearby areas,

providing indications of anomalies. From the results

presented in the dashboard, it is detected bad trafﬁc

zones on the streets in different hours of the day, visu-

alize that a certain event also affects the trafﬁc when

comparing the same street in another date, and ob-

serve that higher levels of trafﬁc have direct impact

on the air quality.

The paper is organized as follows. Section 2

presents the related work in this area and some

dashboards implemented in smart cities. Section 3

presents the system architecture and their compo-

nents, and the algorithms and procedures made when

handling information. Section 4 exposes the ﬁnal so-

lution with use cases for analysis. Section 5 concludes

the results of the paper and presents the future work.

2 RELATED WORK

2.1 Trafﬁc Analysis Approach

TrailMarker (Honda, 2016) is an example of a work

where different approaches are used to analyze real

data sets of vehicular sensor data to ﬁnd outliers, ef-

fectively ﬁnding typical patterns and points of vari-

ation on the road. This work relates to the present

work by having similar objectives of ﬁnding road pat-

terns, but is more centralized on the driver side pat-

terns where acceleration and speed are used to predict

the type of driver (inexperienced, aggressive, careful).

Erdogan (Erdogan and et al., 2008) propose a geo-

graphical information system that has the objective of

assisting the analysis of road accidents based on acci-

dent reports. In the system, the accident locations are

geo-referenced into the highways so it is easier to de-

tect the areas with high rate accidents, in a way to take

precautionary measure and improve the safety in criti-

cal areas. Another important topic is the study of pol-

lution levels related with the road trafﬁc. In an anal-

ysis and evaluation made on the main urban roads in

Beijing (Li et al., 2002), it was used an in place noise

sensors along the road for multiple sampling points

in order to identify the factors inﬂuencing the noise

levels. From their results, where more than 3000 ve-

hicles pass every hour, above their designed capacity,

the noise levels exceeded the national standard dur-

ing daytime hours. Other factors were also pointed,

the road width, surface texture and the types of vehi-

cles in the trafﬁc composition. A proposed model for

the evaluation of urban trafﬁc congestion using buses

(Carli and et al., 2015) uses GPS data from bus on-

board units to probe the position and time trace and

road segments, to associate the GPS traces and calcu-

late the trafﬁc congestion. It offers a map, graphical

and tabular representation of trafﬁc patterns. A work

(Bacon et al., 2011) that uses real time bus data makes

an integration of data from multiple sources (buses,

cameras, OpenStreetMap), and calculates the journey

times depending of the time of the day based on bus

real time positioning. A model proposed (Kerminen

and et al., 2015) uses bus location history data to an-

alyze the trafﬁc ﬂuency and the time spent on driv-

ing, on bus stops and trafﬁc lights, and the frequent

delay areas of the city during the vehicle journey to

understand the causes of delays on bus lines. In (Zhu

et al., 2012) it is studied the road trafﬁc conditions us-

ing data collected from taxis where it was probed the

location and driving speeds during the movement. It

uses an algorithm based on compressive sensing that

achieves low error with missing data.

2.2 City Dashboards

Many cities are starting to make changes in their intel-

ligence in different areas by adding sensors and pro-

viding information through dashboards. The city of

Dublin contains a mobility dashboard

where the traf-

ﬁc can be seen in a global view in the main roads, but

also parking lot locations. Our proposed work can ex-

tend the information presented by Dublin dashboard

by analyzing the roads with greater detail to improve

anomaly detection, with information in all streets.

The dashboard of London

presents many city

variables from points of interest, public transportation

(metro status), environment (weather and pollution

data) and cameras (street view) are viewed in real-

time. Although there is a lot of real time information,

this is only being captured and presented for a gen-

eral consumer, lacking on a more detailed analysis on

trafﬁc congestion and pollution data.

3 PROPOSED SYSTEM AND

ARCHITECTURE

We propose a full stack solution as a system that inte-

grates information from the city to a city level dash-

board as exposed in ﬁgure 1.

http://www.dublindashboard.ie/pages/DublinTravel

http://citydashboard.org/london/

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

388

The information ﬂows from the three available

data sources: the Vehicular Adhoc Network (VANET)

database that feeds the trafﬁc analysis script with ve-

hicular data of the buses, such as GPS and mobil-

ity information, all gathrered through a VANET be-

tween more than 400 buses and road side units; the

OpenStreetMap (OSM) website that adds geograph-

ical roads to the database; and a set of environmen-

tal information from multiple sensors that is available

through the FiWARE

data interface and service.

The processing component consists of an imple-

mentation of algorithms that validate, transform and

prepare the information before being stored in the

database. The Trafﬁc Analysis Script consists of a

python script that is capable to match the bus posi-

tion with geographic roads and calculate speed pro-

ﬁles for each section of the road. The Road Segmen-

tation script is able to slice the OpenStreetMap road

segments dataset to Porto region, thus reducing the

number of road entries in the database. The Environ-

ment Analysis script prepares environment informa-

tion by segmenting by subject and removing errors

and imprecision’s in the measurements.

The storage system is composed by a data ware-

house that holds speed proﬁle metrics from the buses,

a list of geographic roads and environment sensor

metrics. It also has functions that serves the algo-

rithms with logic for making spatial operations.

The Web services provide an interface for the out-

side applications. It is made of three Application pro-

gramming interfaces (APIs): the road processing API

receives instructions for the processing units; the road

trafﬁc proﬁles API serves the dashboard with bus traf-

ﬁc information from the database; and the sensor cor-

relation API provides an endpoint for the dashboard

access, that delivers complemented environment met-

rics and road trafﬁc information.

At the end of the architecture it is located the city

web-based dashboard. It has the purpose to provide

an interface for a city manager to analyze the city ve-

hicle trafﬁc, detect anomalies and environment status

for a city decision support.

3.1 Data Sources

The system currently has three data sources:

• VANET Database: consists of a relational

MySQL database that holds logs from bus related

vehicular data that circulate around Porto city. It

has several variables used for the algorithms: the

GPS time is the exact timestamp of the capture

of latitude-longitude attributes, the node ID is a

https://www.ﬁware.org/

unique identiﬁer for each bus unit, and the speed

is the instant captured speed.

• OSM website: provides geographical information

of points of interest, regions and roads. The roads

are the only dataset selected and have an ID that is

a unique identiﬁer of the segment, a name of the

street and a geometry representation of the seg-

ment.

• FIWARE service: is a RESTful web service that

offers environment information from sensors de-

ployed around Porto city. It contains the identi-

ﬁcation and position of the sensors, the time in-

stant of the captured metrics and the environment

metrics (temperature, noise, carbon monoxide).

The VANET database has a granularity of 15 sec-

onds for each bus, which means that in every 15 sec-

onds it adds a new log entry to the database table if

the bus is active. Having more than 400 active buses

adds a large amount of information that may vary de-

pending on the day and hour. In a single weekday,

it is expected to have around 1 million of entry logs.

Moreover, one of the main challenges of the vehicu-

lar dataset is to deal with GPS imprecisions presented

mainly when the vehicle is moving near tall buildings.

The FIWARE service has a granularity of 1 sec-

ond, depending on the status of the sensor, for more

than 6 deployed environment sensors. One of the

downsides is the downtime and calibration of the sen-

sors which affects the quality and quantity of the re-

sults of our dashboard, and that is out of our reach.

The OSM dataset is made of contributions con-

stantly updating, so it can be difﬁcult to always have

the street segments updated and perfectly corrected.

For the region of Porto, it has more than 40 thousand

road segments to work with, which covers almost all

the real streets.

3.2 Data Storage

Our system relies on a data warehouse structure over

a relational database to feed the dashboard as a deci-

sion support system. The data warehouse stores the

processed vehicle data, environment information and

road segments. This was done because using a typi-

cal database approach would have issues with dealing

with the bulk data from data sources, namely to pro-

vide the different levels of data granularity and tem-

poral segmentation besides the need for some subject-

oriented summaries/data views.

The data warehouse structure is based on a star

schema (Moody and Kortink, 2000) that is made of

two fact tables: one that holds the road speed and de-

lay variables, and the other to store the environment

Decision Support Dashboard for Trafﬁc and Environment Analysis of a Smart City

389

Figure 1: Proposed system ﬂow architecture.

sensor measurements. The remaining tables are the

dimension tables to store the time in multiple granu-

larities, the geographic roads, and the sensors details.

Both fact tables have a connection to the time table

to have the time dimension associated with their en-

tries. The road data fact table also connects to the geo-

graphic roads dimension table to associate each entry

to a speciﬁc road, and the sensors data fact table con-

nects to sensor table, so that sensor data is related to

the details of the sensors. This structure helps to re-

duce the size of the database as it reuses information

from the dimension tables and offers more ﬂexibility

when querying information from different granulari-

ties and multiple types of aggregations.

Figure 2: Database star-schema implementation.

Implemented functions on the database side were

made to ease the access of the information and to sup-

port the processing algorithms with spatial logic. One

of the functions gives the latitude-longitude position

of the road that intersects a certain vehicle, used by

the trafﬁc analysis algorithm. Another relevant func-

tion is the one that splits a date-time into multiple time

variables and granularity to ﬁll the time dimension ta-

ble.

PostgreSQL is the database chosen because it has

no associated cost for the software, is reliable and

has a very good support with geographic data types

and spatial operations. To support the database with

the spatial operations, it is added the PostGIS spatial

functions.

3.3 Dashboard

The Dashboard is the visible part of the system to the

end user. It is implemented as a web application based

dashboard with analytic functionality where the fo-

cus is to have multiple visual dimensions to see trafﬁc

information. One example is the geographical repre-

sentation of roads and positions on top of a map layer,

where it relies on the data provided by the data ware-

house API, namely trafﬁc and sensor information.

All geographical data representation of roads, po-

sitions and other overlays are implemented using

leaﬂet.js GIS JavaScript library

together with a more

analytic and precise chart representation with the help

of the D3.JS library

3.4 Building the Trafﬁc Proﬁle

Currently, the main focus of the system is on mo-

bility information. For that reason, special attention

is given to fetching and processing the mobility data

generated by the vehicular nodes containing its geo-

location. We implemented an algorithm to fetch data

and calculate the delay and speed proﬁles for a spe-

ciﬁc road at different periods of the day. Following

the ﬁgure 4, globally the algorithm 1 works as fol-

lows.

• Using the vehicle coordinates, it maps the vehi-

cle into an existing road. Here it is made a spa-

tial intersection operation, which ﬁnds the road

http://leaﬂetjs.com/

https://d3js.org/

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

390

that intersects the vehicle position from the list

of available Porto roads in the database. After

ﬁnding a match, it gets the identiﬁcation of the

road to associate with the point, but also the frac-

tion of the road where the point is located, for ex-

ample, a fraction of 0.5, which is the middle of

the road. With the help of an implemented road

buffer, many of the real GPS data imprecision’s

are removed as there are fewer chances of a point

in a road being located outside of it.

• After matching a vehicle to a speciﬁc road, all

these data points are associated with the road and

used to calculate each individual speed and delay

in different sections of the road. For this calcula-

tion, we will follow the ﬁgure 4:

– Each bus in the ﬁgure is an entry log in the

database.

– After entering in a road, the ﬁrst position marks

the beginning of the iteration, and the latitude-

longitude and time variables are saved on a list.

At this point, no calculations are made.

– For the following positions, the procedure is

different. Using the Haversive formula (ﬁgure

3), it is calculated the distance from the pre-

vious point using the latitude-longitude coordi-

nates and it is also subtracted the time to be able

to get the speed (distance ÷ time) in that sec-

tion.

– The ﬁrst and third/last bus points are used to

calculate the direction of the calculated vari-

ables between them. Here it is subtracted the

fraction from both, and if the result is negative

it gives one direction, and positive otherwise.

– During this interaction, the following variables

are saved for each point in a list: ID of the bus,

the ID of the road, the sequence of the point,

the time, the speed, the distance, the direction

and the fraction of the road. This also serves

as a staging area before the data being stored in

the database.

Figure 3: The Haversine Formula (Sinnott, 1984).

• The vehicle leaves the road, and the spatial in-

tersection matches a different road, ﬁnalizing the

calculations from the point before. All this list

of calculated variables that belong to the previous

road is persisted to the database.

• The cycle is restarted after being mapped to a new

road.

The information that is resulted contains points as

objects, which includes the ID and speed of the vehi-

cle, distance from the previous point and fraction of

the road. This distance is then accumulated at each

point to give the real distance of the point on the road

instead of having the distance to the previous position.

input : Records: A number of vehicle data records

input : Query: object for querying the database

output: a list of road segment variables

points list ←

o f f points ← 0;

last road id ← 0;

foreach entr y ∈ records do

node id ← entry[0];

gps time ← entry[1];

latitude ← entry[2];

longitude ← entry[3];

road id, road f raction ←

Query. f unction.insideRoad(latitude, longitude);

if road id = 0 or road id ! = last road id then

if o f f points = 3 and points list is not empty

then

if points list[0]. f raction −

points list[−1]. f raction < 0 then

direction ← 0;

end

else

direction ← 1;

end

if points list 2 then

Query.persist(points list, direction);

end

points list ←

last road id ← road id;

end

continue;

end

o f f points ← 0;

if points list is empty then

road point ←

createOb ject(bus, road, sequence, time,

f raction);

points list.append(road point);

end

else

last point ← points list[−1];

sequence ← last point.sequence + 1;

delay ← gps time − last point.datetime;

distance ←

havesineDistance(last point, lat, lon);

road point ←

createOb ject(bus, road, sequence, time, speed,

distance, direction, f raction);

points list.append(road point);

end

Algorithm 1: Road points variables calculation.

After the points be mapped to the real distance,

they now are not aggregated by vehicle, but by dis-

Decision Support Dashboard for Trafﬁc and Environment Analysis of a Smart City

391

Figure 4: A bus trajectory.

tance. As ﬁgure 5 shows, multiple points from dif-

ferent vehicles create a proﬁle through the distance of

the road.

By having the points aggregated by distance, it is

now possible to select multiple sections of the road,

and calculate for each one the 25%, 50% (median)

and 75% quartile of the speed values to give an ap-

proximated curve of the speed in the road where the

majority of the vehicles are.

Figure 5: Example of the road proﬁle with points from 3

different vehicles.

3.5 From Sensors to Correlation

between Trafﬁc and Environment

The impact of city trafﬁc in the environment was the

main use case selected. It allows relating information

that could be useful to city planning and city quality

improvement; this can be achieved by relating road

trafﬁc proﬁles with co-located environment sensors

along time. Relating peaks in both trafﬁc and envi-

ronment readings can help to identify time and space

patterns that may be worth attention and/or localized

intervention because they are affected by themselves.

The calculation of the correlation is made in the

sensors correlation API, which uses the following

steps to make the calculations in real time:

• The service receives a speciﬁc sensor and time

range from dashboard inputs selected by the user.

• It is fetched from the database the average of the

environment measurements (noise, CO) aggre-

gated in the different hours of the day for the spe-

ciﬁc sensor and date range.

• As shown in ﬁgure 6, from a geographic position

of the environment sensor, it is made a radius of

100 meters to ﬁnd the roads that lie inside that

area. To do this, it is used PostGIS spatial func-

tions to create a spatial circle area, to then make a

spatial intersection operation with the list of avail-

able roads, returning the ID of the valid roads

from the database.

• The IDs of the valid/nearest roads are used to

Fetch the average speed values from the multi-

ple sections that lie inside the circle, but now ag-

gregating by hours of the day from the same date

range as the data from the sensors.

• The variables from environment metrics and av-

erage speed are associated to the same hour as a

point of connection, so both proﬁles can be visu-

ally analyzed and correlated.

Figure 6: Approach when choosing roads to correlate with

the environment sensor.

4 EVALUATION

The main objectives when using the a visual repre-

sentation is to see the speed proﬁle of the trafﬁc in

a road segment and understand where are the criti-

cal zones with lower speed and also be able to make

comparisons of the street trafﬁc in different moments,

for example, before and after public road works or

trafﬁc mobilization, to see where the improvements

occurred.

A good use case is trying to ﬁnd outliers and dif-

ferent patterns when comparing a normal day against

a day with special events. In Figure 7 it is presented

a comparison of the trafﬁc proﬁle of two different

weekends at the same hours of the day. Here, in the

beginning of the road segment, the trafﬁc speed is af-

fected by the number of vehicles that tries to access

the road and in the end of the segment because of the

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

392

Figure 7: Comparison of a proﬁle on a normal weekend vs an event weekend.

Figure 8: Correlation of speed of trafﬁc vs CO values through the day.

accumulation of trafﬁc near the event area, while the

remaining of the segment follows the normal behav-

ior. This is a good case study for a city manager to

analyze and make trafﬁc changes at the beginning and

end of the segment when hosting similar events in the

area nearby. Something like divert the trafﬁc to sec-

ondary roads or allow only public transportation to

drive on that road are choices that could be analyzed

by a city manager.

Combining environment information with trafﬁc

data helps to make conclusions on how the trafﬁc af-

fects the local environment and how the environment

affects the trafﬁc. This feature is also present in the

dashboard. As ﬁgure 8 shows, the result of correlat-

ing the trafﬁc and a pollution metric (carbon monox-

ide, CO) is something already expected: the trafﬁc

speed affects the pollution index. In the beginning of

the day, the average speed of the buses is higher, and

the CO levels are low, but as the speed starts to get

lower, the CO level increases. From the results we

can also observe the hours of the day more affected in

the sensors area, motivating on making changes in the

trafﬁc at certain hours of the day.

Decision Support Dashboard for Trafﬁc and Environment Analysis of a Smart City

393

5 CONCLUSIONS

This paper proposed a system dashboard as a solu-

tion for abstracting decision makers / city stakehold-

ers from the actual city smart city infrastructure. The

dashboard provides the ability to visualize informa-

tion of the city trafﬁc from local to city wide scale.

We consider that it can be a very useful tool for a city

manager, namely in supporting conclusions on how

the city is moving, improving it in a way to provide

better behavior of the trafﬁc in the city.

The typical scenarios presented illustrate how the

proposed system can be useful to identify and lo-

cate anomalies and unexpected patterns on the roads.

However, the system still lacks an end-user valida-

tion, and the next steps include making it available.

For that reason, the next step already planned is to

make available to the city managers and bus company

for assessing its usefulness and gather their inputs. By

design the system is open to include more city related

variables and allow a multimodal approach to the city

namely correlating changes in sensors with speciﬁc

events / city behaviours. Future work also includes

the integration with machine learning to add the abil-

ity to predict and/or at least estimate the probability of

atypical events occurrence that may impact the city.

ACKNOWLEDGMENTS

This work was supported by the CMU-Portugal Pro-

gram through S2MovingCity: Sensing and Serving a

Moving City under Grant CMUPERI/TIC/0010/2014.

REFERENCES

Bacon, J., Bejan, A., and Beresford, A. (2011). Using Real-

Time Road Trafﬁc Data to Evaluate Congestion. De-

pend. and Historic Computing, 6875:93–117.

Carli, R. and et al. (2015). Automated Evaluation of Ur-

ban Trafﬁc Congestion Using Bus as a Probe. IEEE

International Conference on Automation Science and

Engineering (CASE).

Erdogan, S. and et al. (2008). Geographical information

systems aided trafﬁc accident analysis system case

study: city of Afyonkarahisar. Accident Analysis and

Prevention, 40(1):174–181.

Hall, R. E., Bowerman, B., Braverman, J., Taylor, J., To-

dosow, H., and Von Wimmersperg, U. The Vision of

A Smart City.

Honda, T. (2016). TrailMarker. In Proceedings of the 2016

on SIGMOD’16 PhD Symposium - SIGMOD’16 PhD,

pages 22–26. ACM Press.

Kerminen, R. and et al. (2015). Analysing trafﬁc ﬂuency

from bus data. IET Intelligent Transport Systems,

9(6):566–572.

Li, B., Tao, S., and Dawson, R. (2002). Evaluation and

analysis of trafﬁc noise from the main urban roads in

Beijing. Applied Acoustics, 63(10):1137–1142.

Lopes, D. and Sargento, S. (2014). Network mobility for ve-

hicular networks. In 2014 IEEE Symposium on Com-

puters and Communications (ISCC), pages 1–7. IEEE.

Merry, H. (2017). Population increase and the smart city -

Internet of Things blog.

Moody, D. L. and Kortink, M. A. R. (2000). From Enter-

prise Models to Dimensional Models: A Methodology

for Data Warehouse and Data Mart Design. pages 5–1.

Sinnott, R. (1984). Virtues of the Haversine. skytel, 68:158.

Provided by the SAO/NASA Astrophysics Data Sys-

tem.

Zhu, Y., Li, Z., and Zhu, H. (2012). A Compressive Sens-

ing Approach to Urban Trafﬁc. Ieee Transactions on

Mobile Computing, pages 1–14.

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

394