Our paper for IRF Conference November 2013 - Roadroid

Roadroid Page 1

17:th IRF World meeting www.roadroid.com

ROADROID

CONTINUOUS ROAD CONDITION MONITORING WITH SMART PHONES

Lars Forslöf

CEO/Inventor

www.roadroid.com / www.roadroid.se

[email protected]

Egnahemsgatan 5

827 35 Ljusdal SWEDEN

Phone: +46-72-2426620

27 June 2013

Abstract

By using the built-in vibration sensor in smartphones, it is possible to collect road roughness

data at Class 2 or 3 level [1] in a very simple and cost efficient way. Data collection can be

done continuously and monitor roughness changes over time. The continuous data collection

can also give early warnings of changes and damage, enable new ways to work in the

operational road maintenance management, and can guide more accurate surveys for strategic

asset management and pavement planning. Data collection with smartphones will not directly

compete with Class 1 [1] measurements, but rather complete them in a powerful way. As

Class 1 data is very expensive to collect it cannot be done often, and advanced data collection

systems also demand complex data analysis and take long time to deliver. Smartphone based

data collection can meet both these challenges.

A smartphone based system is also an alternative to Class 4 – Subjective rating [1], on roads

where heavy, complex and expensive equipment is impossible to use, and for bicycle roads.

The technology is objective, highly portable and simple to use. This gives a powerful support

to road inventories, inception reports, tactical planning, program analysis and support

maintenance project evaluation.

History:

2002-2006 - Research for the Swedish Road Administration using accelerometers and a PC.

2010 - Technology to measure vibrations was built in to a smartphone.

2011 - First Android app for research of how cars, phones and speed are affecting values.

2012 - User discovery/development with managers for road maintenance

2013 - First end user projects and piloting in different parts of the world.

The Roadroid smartphone solution has two options for roughness data calculation:

1) estimated IRI (eIRI) - based on a Peak and Root Mean Square vibration analysis – and

correlated to Swedish laser measurements on paved roads. The setup is fixed but made

for three types of cars and is thought to compensate for speed in 20-100 km/h. eIRI is

the base for a Roadroid Index (RI) classification of single points and stretches of road.

2) calculated IRI (cIRI) - based on the quarter car formula for a narrow speed gap such as

60-80 km/h. cIRI allows the operator to calibrate the sensitivity to a known reference.

Collected data are digitally transferred to internet maps and can be aggregated in preferred

sections (default 100 m), as well as exported to Geographical Information Systems (GIS).

In addition to optimizing road maintenance, the information could be a new kind of input to

road navigators and digital route guides. Digital bump warnings can be presented as detected

bumps to road navigators through standards for Intelligent Transportation Systems (ITS).

http://www.roadroid.com/

http://www.roadroid.se/

mailto:[email protected]

Roadroid Page 2


INTRODUCTION

The International Roughness Index (IRI) is a roughness index commonly obtained from

measured longitudinal road profiles. Since its introduction in 1986, IRI has become

commonly used worldwide for evaluating and managing road systems. Vibrations have been

used since early 1900 for expressing road condition and ride quality [2].

The traditional techniques for measuring roughness may be categorized as special built trucks

or wagons with laser scanners, bump-wagons, and manually operated rolling straight edges.

Special built equipment is expensive, due to heavy and complex hardware, low volume of

production and need of sophisticated systems and accessories. Data gathering and analysis

are often time consuming. Data collection is typically done during the summer then analysed

and delivered to the maintenance management systems in late autumn. It is soon then winter

and spring where the road faces continual frost heave/thaw (a very dramatic period in a road´s

life with extreme changes in roughness). The IRI values that were “exact” almost a year ago

might now not be the same any longer. As it is also so expensive to collect and analyse the

data, that many roads are only covered in one lane direction every 3-4 years.

Smartphone based gathering of roughness data, can be done at a low cost and monitor

changes on a daily basis. For frost heave issues it can tell when and where it is happening and

if the situation is worse than in previous years. It can be used in the winter to determine the

performance of snow-removal and ice-grading. It may advantageously be used in performance

based contracts or research on road deterioration, various environmental effects (as heavy

rains, flooding).

It should be mentioned that smartphone based systems like Roadroid might challenge old

knowledge, standards, procedures and existing ways to procure:

- Pavement planners and road engineers know existing inputs;

- Research organisations, suppliers and buyers have existing ways to work;

- Organizations have invested time, prestige and huge amounts of money to develop

more and more exact and complex data collection and management systems.

As described in [1] it is necessary to understand the difference between four generic

classes of road roughness measuring methods in use:

Class 1 - Precision profiles

Class 2 - Other profilometric methods

Class 3 - IRI by correlation

Class 4 - Subjective ratings

It is natural that scepticism will appear when a Class 2/3 smartphone is compared with a

multi-million dollar Class 1 vehicle. But a smartphone can deliver up-to-date good quality

roughness data to a web page within 24 hours, in contrast to an expensive software client with

the “exact” Class 1 data from last year.

On the other end of the scale – many road inventories and assessments are actually made by

Class 4 subjective ratings over large areas using pen and paper.

Smartphone data collection fills a gap between the class 1 measurements and class 4 ocular

inspections, and we see some early adopters and notice steps of development in the market.

Roadroid Page 3


THE FIRST PROTOTYPES 2002-2006

The Roadroid team has been working with mobile ITS since mid-1990s, particularly with

mobile data gathering, road weather information and road databases. During a visit to the

Transportation Research Board in Washington 2001, a Canadian project was presented that

monitored the speed of timber hauling trucks, simply assuming that where the speed was low

the road quality was poor. Our developed idea was to add vibration measurements.

Together with the Royal Institute of Technology, a first pilot scheme was built in 2002-2003.

At that time we used a high-resolution accelerometer at the rear axle of a front wheel drive

vehicle, connected by cable to a portable PC through a signal conditioner. GPS and GSM

were connected through wired serial connections. Two master students built a first prototype

using an industrial software system for signal analysis.

Från Till Hans Bo S Kr. W Hossein1 Hossein2 Hans2 L-E H Robin Kalle Medel 0 100 1 1 1 1 1 1 2 1 1 1,11

100 200 2 1 1 2 2 2 2 2 2 1,78 200 300 2 1 1 1 1 1 2 1 1 1,22 300 400 2 1 2 2 2 2 2 2 2 1,89 400 500 2 2 2 2 2 2 2 3 2 2,11 500 600 1 1 1 1 1 1 1 1 1 1,00 600 700 1 1 2 1 2 2 2 2 2 1,67 700 800 3 2 3 3 3 3 3 3 3 2,89 800 900 2 2 1 1 2 2 2 2 1 1,67 900 1000 2 2 1 2 3 2 2 2 2 2,00

1000 1100 1 1 1 2 1 1 1 1 2 1,22 1100 1200 1 1 1 2 1 1 1 1 1 1,11 1200 1300 1 2 2 1 1 1 1 1 2 1,33 1300 1400 2 2 3 2 3 2 3 2 3 2,44 1400 1500 2 2 2 3 4 3 2 3 3 2,67 1500 1600 1 1 2 1 2 1 2 1 1 1,33 1600 1700 2 2 2 3 3 2 2 2 2 2,22 1700 1800 2 1 1 3 3 2 1 2 2 1,89 1800 1900 1 1 1 1 1 1 1 1 1 1,00 1900 2000 2 1 1 1 2 1 1 1 1 1,22 2000 2100 2 2 1 1 2 2 2 2 1 1,67 2100 2200 2 2 1 2 2 1 1 1 2 1,56 2200 2300 1 1 1 1 2 1 1 1 1 1,11 2300 2400 2 1 2 1 2 1 2 2 2 1,67 2400 2500 2 2 1 2 2 1 2 2 2 1,78 2500 2600 2 2 2 2 2 1 2 1 2 1,78 2600 2700 2 2 2 2 3 2 2 2 2 2,11 2700 2800 3 2 1 2 3 2 2 2 2 2,11 2800 2900 2 2 1 2 3 2 2 2 2 2,00 2900 3000 2 1 1 1 2 1 1 1 1 1,22 3000 3100 1 1 1 1 2 1 1 1 1 1,11 3100 3200 1 1 1 2 2 1 1 1 1 1,22 3200 3300 1 1 1 2 1 1 1 1 1 1,11 3300 3400 1 1 1 1 1 1 1 1 1 1,00 3400 3500 1 1 1 1 1 1 1 1 1 1,00

Fig. 1 From left 1:st prototype 2002-2003 to right 2:nd prototype, developed 2004-2006

The results were promising and the Swedish National Road Administration financed an R&D

project o further develop and validate the prototype with a focus on gravel roads.

The system was developed into a C++ software for Windows program, and a GIS tool was

implemented for viewing the road quality in different colours.

A validation between ocular inspections and the system´s measurements was performed and

presented at the Transport Forum, Linköping, in 2005. 8 segments of 100 m were individually

assessed according to 4 road condition classes. Module analysis (experimental analysis of

oscillation) was performed on a sample of specific sections of the 4 road condition classes.

Regression analysis was then performed with rules based on: 1) amplitude levels for different

G, 2) RMS (Root Mean Square), 3) The vehicle speed measured and 4) The length in meters.

The analysis showed that a single test run would classify properly to 70% compared to an

average of subjective ocular inspections, while a single ocular inspection varied more from

the average than that. The method was considered objective with very good repeatability.

In 2006 the development stalled. The system was considered relatively cheap and simple to

operate at the time (~7000 USD). In retrospect, it had several limitations; particularly the

sensor mounting and cables exposed in the harsh environment under a car, but also the

windows 98 computer and the cable-connected, and a not very accurate, GPS.

FURTHER DEVELOPMENT 2010-2011

In 2010, the ideas from 2002-2006 were reviewed. A major technical development was now

the appearance of mobile smartphones. Literally all peripherals that were previously

connected by cables were now built into a phone and the limitations of certain components

Roadroid Page 4


were removed by new advances in technology. We knew the answers to some of the questions

from 2002-2006, e.g. the basis for signal analysis and the influence of speed. There were

however new big questions to solve, such as:

- Was it possible to pick up the signals from inside the car?

- We knew different car models would give different signals and how could we handle that?

- Would a lower sampling frequency be enough (100 Hz compared to earlier 512 Hz)?

- Would the accelerometer sensitivity and the G-scale be sufficient (+/-2G)?

- Would different smartphone models return different values (accelerometer sensitivity).

We developed an Android application and an algorithm using the built-in accelerometer

signal. The choice of Android rather than iPhone was made considering the open architecture

and hardware price/performance relation. We started to sample data on different roads with

different types of vehicles, and over constructed obstacles in 2011.

Fig. 3 a) left and b) right – The 3:rd prototype 2011

The obstacles were passed by different vehicle types 5 times in 6 different speeds: 20, 40, 60,

80, 100 and 120 km/h. Data were sampled with different phones, both with our algorithm and

in raw 100 Hz. During the data analysis, we then discovered a number of things:

There are differences between different car models, especially at low speeds. In the

40-80 km/h range, differences are however limited. The tests gave us a model for how

to calculate the speed influence of the signal for the 3 different type vehicles.

There are big differences between different phones, both for the sampling frequency

and the quality of the accelerometer data. It is of great importance to know these

dynamics to achieve comparable data. A phone calibration procedure is required.

It is of great importance to mount the phone correctly in a good mounting bracket,

preferably in a way that enables the phone´s camera lens to be directed at the road.

Most importantly: the trials during 2011 showed that usable data could be delivered!

We now had an Android app analysing 100 vibrations per second and saving several of the

essential road condition values with a GPS coordinate!

The road condition data was divided into 4 different levels for visualization: Green for Good,

Yellow for Satisfactory, Red for Unsatisfactory and Black for Poor.

VIEWING OF DATA

We now had a tool delivering data, and created a viewer –an internet based map tool to

present the data. Data was collected after measurements zipped on the unit, and sent by a file

Roadroid Page 5


transfer service to a webserver in the cloud. Data files from different units were imported by a

daily routine, mapped to road links/geometries such as Open Street Map (OSM) and using

Open Layers on Google map background to present the road condition data on a web map.

The app stored one data value each second, based on a sampling frequency of 100 Hz, but to

get an overview in a larger scale it was more convenient to use road links than second

resolution dots. Depending on the road database, there will be many opportunities to refine

the data and add information such as road width, traffic volumes, etc. Globally we have been

using geometries from Open Street Maps.

Fig 4. A screen shot from Mid-Sweden

USE OF DATA – AND THE ROADROID INDEX

We have undertaken studies of the International Roughness Index (IRI) and have developed a

correlation between our measured road condition data values and the IRI. We have

continually been looking to improve the way we present the comprehensive level of

information being collected – flexibility and scalability are key. We wanted to be able to add

data from several measurements over time and compare results over time in a flexible way.

We also wanted to automatically

generate reports for a road and to

compare a road with other roads and

within whole regions. The solution

was to use a percentage of each road

class for a road or an area, an index

scalable for a part of a road, a whole

road, a city or a region.

Figure 5 to the right: - How to pick

data out from the web tool.

Roadroid Page 6


As we want to do continuous monitoring to view development over time we also needed to

find a way to produce reports. Data collection can be made by the contractor’s road guards/

officers who are doing visual inspections 1- 3 times per week, or by other operators such as

the local newspaper distributor.

The percentage of each class seemed to be a suitable way to also make reports from the data.

Road Condition Change report Q4 - 2012Gävleborg

Hudiksvall Contractor 69,4% 15,5% 7,4% 7,8% 65,8% 14,6% 8,5% 11,0%

1089 Km Phone 010-476 14 07 Q4 - 2012 Helår - 2012

Road no. Traffic Class Length Comments Good Sat Usat Poor TREND Good Sat Usat Poor eIRI avg

E4 14000 1 143 93,9% 4,6% 0,9% 0,5% -3,4% 97,4% 2,0% 0,4% 0,3% 1,8

83 8300 2 167 Salt road 88,9% 7,4% 2,2% 1,5% 3,3% 85,6% 8,0% 3,2% 3,2% 2,6

84 7500 2 210 Salt road 90,9% 6,1% 1,7% 1,3% -1,6% 92,5% 4,8% 1,6% 1,1% 2,9

305 1200 3 105 76,7% 14,4% 5,3% 3,6% -0,6% 77,3% 13,3% 5,2% 4,1% 4,5

307 900 3 75 93,7% 5,2% 0,7% 0,4% 0,4% 93,3% 5,5% 0,8% 0,4% 3,7

539 300 3 33 Gravel road 9,1% 23,2% 24,2% 43,4% 7,5

583 1700 3 89 96,9% 2,6% 0,2% 0,3% 0,0% 96,9% 2,0% 0,6% 0,5% 2,3

660 1850 3 64 88,6% 8,3% 0,6% 2,5% 9,1% 79,5% 9,7% 4,5% 6,3% 6,7

Good for Q4 minus Good for all year.

Figure 6 – A Road condition change report – for performance follow up.

ESTIMATED IRI

We now had a promising and scalable index, but also knew that we needed to correlate this to

the IRI. To find the correlation we gathered:

1) our road condition values in absolute terms and matched them to sections in the

Swedish road data base

2) average data from Class 1 (laser beam) IRI measurements to the same sections.

By comparing hundreds of sections we established a correlation factor and could now

estimate an IRI value (eIRI - usable all the way from our second resolution dots to a road

link). The correlation factor (R2) was up to 0,5 which meant that 75% of the classifications

were correct.

We have noticed some limitation in speed adaptation, rough pavement surfaces and that mini-

cars are quite more sensitive than our reference small car. Research is continuing by different

institutions around the world, as the World Bank, UN OPS, specialized Universities and some

large road companies. Throughout 2014 these organizations will report back to inform our

developments, and to improve our solution. This will include a calculated IRI (cIRI) and the

possibility for an operator controlled calibration.

y = PPPx + 1,35R² = 0,5154

0

0,5

1

1,5

2

2,5

0 0,05 0,1 0,15 0,2 0,25

AVG-IRI

0,00

0,20

0,40

0,60

0,80

1,00

1,20

1,40

1,60

1,80

2,00

1 2 3 4 5 6 7 8 9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

Roadroid

ARAN

Figure 7: Left – from our base correlation study / Right - Autostrade Aran comparisons

CALCULATED IRI

Roadroid Page 7


The cIRI approach is promising and seems to be more dynamic then the eIRI. The quarter car

model uses two swinging weights that can be simulated if chassis movement is available as

one input. The estimation here is for the chassis movement in a scale 1-10 based on the

accelerometer data and. T an operator controlled adjustment/calibration. As mobile

technologies get more advanced, suitable laser distance devices for distance monitoring

between road and chassis might be used in a future.

The needs are different – from operational maintenance on developed road networks to first

fact finding in developing countries. Our internet map function is a good way to view data,

but demands some basic knowledge of GIS and road databases.

We also developed a data aggregator, collecting date/time, coordinates (X, Y), vertical profile

(Z), speed together with eIRI and cIRI in preferred sections 25-1000 m. In the current version,

tab separated text files can be generated and imported into other software. By the use of

digital spreadsheets, data can for example be adapted for import to Road Maintenance

Management Systems.

DateTime Latitude Longitude Distance(m)Speed (km/h)Altitude (m)/10 eIRI cIRI

2013-03-09 07:52 -25,6805 28,20773 100 39,76222 128,2082 12,82082 2,876433 4,512167

2013-03-09 07:52 -25,6807 28,20867 200 55,77875 128,5056 12,85056 2,789588 4,18245

2013-03-09 07:52 -25,681 28,2099 300 50,38667 128,8161 12,88161 3,418956 4,668244

2013-03-09 07:53 -25,6811 28,21115 400 32,34556 129,0784 12,90784 3,684656 4,387678

2013-03-09 07:53 -25,6802 28,21152 500 31,28444 129,0249 12,90249 2,580122 4,295356

2013-03-09 07:54 -25,6798 28,21231 600 52,36143 128,944 12,8944 2,055414 4,060471

2013-03-09 07:54 -25,68 28,21335 700 59,92 128,9084 12,89084 1,987614 4,689414

2013-03-09 07:54 -25,6802 28,21451 800 53,67625 128,9856 12,89856 1,963013 3,592075

2013-03-09 07:54 -25,6803 28,21569 900 32,57444 129,3456 12,93456 2,601222 2,459678

2013-03-09 07:54 -25,6805 28,21677 1000 49,26875 129,4633 12,94633 2,181925 4,3462

2013-03-09 07:54 -25,6807 28,21787 1100 61,1 129,536 12,9536 2,068029 5,782043

2013-03-09 07:55 -25,6809 28,21905 1200 68,56167 129,7245 12,97245 3,317617 6,4562

2013-03-09 07:55 -25,681 28,22019 1300 71,29167 129,731 12,9731 2,331533 5,25215

2013-03-09 07:55 -25,6812 28,22138 1400 73,37833 129,7947 12,97947 2,205783 4,7634

2013-03-09 07:55 -25,6814 28,22258 1500 72,27 129,9207 12,99207 2,71375 7,648117

2013-03-09 07:55 -25,6816 28,22375 1600 72,43833 130,0557 13,00557 2,073817 5,782417

2013-03-09 07:55 -25,6818 28,22493 1700 74,04167 130,0723 13,00723 2,115883 6,597683

2013-03-09 07:55 -25,682 28,22616 1800 75,91333 130,1243 13,01243 2,500767 6,394117

2013-03-09 07:55 -25,6822 28,2274 1900 76,99667 130,1088 13,01088 2,797483 6,799267

2013-03-09 07:55 -25,6824 28,22867 2000 74,73167 130,2387 13,02387 2,6794 5,953483

2013-03-09 07:55 -25,6824 28,22992 2100 73,34167 130,4493 13,04493 2,929717 7,847717

2013-03-09 07:56 -25,682 28,23107 2200 76,36667 130,5108 13,05108 2,376033 6,660283

2013-03-09 07:56 -25,6814 28,23218 2300 78,602 130,3952 13,03952 3,0342 6,54626

2013-03-09 07:56 -25,681 28,23315 2400 79,534 130,3542 13,03542 1,874 5,74178

2013-03-09 07:56 -25,6805 28,23413 2500 77,91667 130,2685 13,02685 2,675967 5,830217

2013-03-09 07:56 -25,68 28,23527 2600 52,68 130,2186 13,02186 2,711089 7,014867

2013-03-09 07:56 -25,6795 28,23649 2700 25,329 130,3062 13,03062 1,81006 1,42253

2013-03-09 07:57 -25,6792 28,2375 2800 45,195 130,2119 13,02119 2,8132 3,569563

2013-03-09 07:57 -25,679 28,23849 2900 62,68333 129,9912 12,99912 2,389 5,633067

0

2

4

6

8

10

12

10

0

26

00

51

00

76

00

10

100

12

600

15

100

17

600

20

100

22

600

25

100

27

600

30

100

32

600

35

100

37

600

40

100

42

600

45

100

47

600

50

100

52

600

55

100

57

600

60

100

62

600

65

100

67

600

70

100

72

600

eIRI

0

20

40

60

80

100

120

140

160

180

Speed (km/h)

Altitude (m)/10

st

art

s

top

s

t

art

s

top

Figure 8 – Example - 72 km test run in South Africa – a plot of eIRI, speed and altitude.

As mentioned, we have seen limited IRI-correlation on roads with rough surfaces (as

chipseals/brickroads), but at the same time promising results on gravel roads in Afghanistan

and Sweden. From what we have seen, it is a mainly a question of filtering the data correctly.

We also expect that our operator controlled adjustment of cIRI will facilitate the right levels.

USE OF THE BUILT-IN CAMERA

As most phones now have a high quality built-in camera and GPS, we have developed a

function to easily take photos and position them on the map as thumbnails. The images are

often of acceptable quality, but are subject to mounting and light conditions. This is

recognized as very good support for ocular inspections, and can also be used to capture

dynamic events, such as certain snow conditions or other maintenance contract issues.

Roadroid Page 8


Figure 9 – Use of the phones build-n camera and support on the web tool.

We have also tested GPS high resolution action video cameras (Contour 2+) with good results

for more precise and demanding video requirements.

FRICTION MEASUREMENTS

Connected to winter road maintenance, friction is of crucial importance for road traffic safety

and to pick up road friction / skidding resistance is simple compared to measuring road

roughness.

< 0.15 0.15-0.3 > 0.3

µ

1) Press friction button2) Break3) Friction value saved4) Data send to server5) Information on map

Data stored to reports

• Simple use• Digital information• Calibration to known reference

For an objective dialogue betweenclient and contractor but also fortraffic information

Extremely slippery Good friction

Figure 10 – Roadroid friction sampler

The Roadorid app calculates the friction level, using the longitudinal force (retardation) and

gives the friction value a GPS coordinate. Data can be sent by the phones SMS if it is an

urgent report or transferred with the other data to be shown on the map.

Roadroid Page 9


USE ON CYCLE PATHS

The cities of the world are facing more and more traffic problems. Traffic management is

optimizing the flow – but far from enough to cope with the car traffic growth. Cycling is one

option to commute instead of being stuck in a traffic jam. Modern cities are building cycle

highways, with increased demands of quality of the cycle network. Cyclists buy expensive

racing bikes with thin wheels – and they drive fast!

The same smartphone based system can be used as a quick and easy monitoring system for

cycle roads.

Figure 11 – Roadroid for bicycle paths

REFERNCES

[1] Sayers M W, Gillespie T D and Queroz C A V (1986a). The international road roughness

experiment: establishing correlation and a calibration standard for measurements. Technical

Paper No. 45. World Bank, Washington DC.

[2] THE DANA AUTOMATIC RECORDING ROUGHOMETER FOR MEASURING

HIGHWAY ROUGHNESS, 1933.

[3] Overseas Road Note 18 - A guide to the pavement evaluation and maintenance of

bitumen-surfaced roads in tropical and sub-tropical countries. TRL 1999.

[4] ARR 295 Interpretations of road profile, ARRB Transport Research, 1996

ACKNOWLEDGEMENTS

The Kartographic Society - Innovation award (Sweden, March 2013)

UN World Summit Award - Global Champion in eGovernance (Abu Dhabi, February 2013)

European Satellite Navigation Competition - Regional winner (Munich, October 2012).

Many thanks to my two leading wizards, making ideas to reality:

Android programming expert, Hans Jones, and

Web/GIS/database system developer Tommy Niittula.

Documents

Our paper for IRF Conference November 2013 - Roadroid