Towards Assessing Cycleway Pavement Surface Roughness Using an

Action Camera with IMU and GPS

Muhammad Hassam Baig

1,2 a

, Jeziel Antonio Ayala Garcia

, Waqar Shahid Qureshi

and Ihsan Ullah

1,2 d

School of Computer Science, University of Galway, Galway, Ireland

Data Science Institute, University of Galway, Galway, Ireland

Keywords: Action Camera, Cycleways, Corrected Roughness Index, Greenways, International Roughness Index,

Pavement Roughness, Quarter Car Model, Rolling Variance, Stability.

Abstract: This paper introduces an autonomous and cost-effective method for assessing cycleway pavement roughness,

using an action camera equipped with high-resolution sensors including an Inertial Measurement Unit (IMU)

and a Global Positioning System (GPS). The methodology utilizes simplified quarter car model for bicycles,

without manual intervention, to calculate International Roughness Index (IRI) for cycleway surface quality

evaluation. It utilizes our novel approach to determine stable section from which average acceleration

orientation vector is computed. For analysis we propose a corrected-roughness index (CRI), which is a

quantized version of IRI. Experiments conducted on asphalt cycleways in Ireland revealed strong correlations

between vehicle vibration and surface roughness. Results further demonstrate the consistency of the proposed

model across different bikes through comparative analysis. Observations indicate bias in vibration data,

influenced by different tire sizes and the mechanical features of the bicycles.

1 INTRODUCTION

Pavement surface roughness is a key indicator of

pavement quality, directly affecting user comfort. As

greenways wear down over time, irregularities

negatively impact cycle performance, increase

maintenance costs and compromise safety.

Measuring and analysing pavement roughness is

crucial for maintaining cycleways, guiding

infrastructure investments, and ensuring user

satisfaction.

Pavement roughness has been previously assessed

through various methods such as Present

serviceability index, IRI, Mean ride index etc., but

recent advancements in technology have led to more

cost-effective approaches (Hettiarachchi, Yuan,

Amirkhanian and Xiao, 2023).

Process of measuring pavement roughness

changed a lot in the last few decades, growing rapidly

with the help of advanced tools and technologies, due

https://orcid.org/0009-0003-3153-9527

https://orcid.org/0009-0006-7829-9948

https://orcid.org/0000-0003-0176-8145

https://orcid.org/0000-0002-7964-5199

to the demand of accurate, time-effective and cost-

effective methods to calculate road roughness. Table

1 enlists some of the instruments, sensors,

measurement methods along with their pavement

roughness measurement year.

Initially, pavement roughness was measured with

mechanical systems such as profilograph, that

measure vertical deviations on a road's surface by

means of wheels and beams. This was later replaced

by non-contact profilers which measure vehicle

responses to roughness that yield indirect estimates of

the profile (Woodstrom, 1990).

One of the most significant changes occurred in

the 1986 when Sayers, Gillespie and Queiroz (1986)

from Federal Highway Administration, established

the use of the IRI for roughness calculation as a

standard. The authors utilized quarter car model, a

simplified representation of a car’s suspension

system, to compute IRI.

Baig, M. H., Ayala Garcia, J. A., Qureshi, W. S. and Ullah, I.

Towards Assessing Cycleway Pavement Surface Roughness Using an Action Camera with IMU and GPS.

DOI: 10.5220/0013504900003941

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2025), pages 247-255

ISBN: 978-989-758-745-0; ISSN: 2184-495X

247

Around 1990s, inertial profilers were utilizing

accelerometers and lasers for increased accuracy and

efficiency, before the implementation of laser-only

devices. Advances continued into the following

decades with the use of 3D imaging systems mounted

with scanning lasers and reflectors to produce

detailed digital profiles of pavement surfaces for

evaluations of road conditions, such as roughness,

rutting, and cracking (Fares and Zayed, 2023).

Since the 2010s and beyond the research has been

moved towards automated low-cost systems. The

studies show common use of available technology

such as smartphones and IoT systems for the

measurement of pavement roughness (Kumar, Tallam

and Kumar, 2022; Zhang and Wenjiang, 2022; Yu,

Fang and Wix, 2022; Alatoom and Obaidat, 2022).

The accuracy and reliability of such measurements

have been greatly improved by the integration of

advanced analytical techniques along with machine

learning models. The validation of these

methodologies against professional instruments

demonstrates their potential for increasing

accessibility to a wide user group.

Today, devices equipped with sensors such as

lidar, laser, IMU and GPS offer efficient solutions to

measure road conditions. These devices capture

precise data regarding the surface deviations of roads

and can also be used for cycleways, which are often

overlooked in traditional assessments.

Zang, Shen, Huang, Wan and Shi, (2018)

developed a new methodology for computing IRI

using simplified quarter car model for bicycles. They

collected road surface data by mounting smartphone

over the handlebar of the bicycle and captured sensor

data. Compared the results with laser pavement

scanner, a professional instrument, and reported

significant associations with it.

However, their methodology involves manual

interpretation of average acceleration vector. For this

purpose, authors had to ask the bicycle riders to

maintain the speed and posture as stable as possible

for initial 5 seconds.

Similarly, Rizelioğlu and Yazıcı, (2024) used a

mountain bike to obtain road roughness measurement

using the quarter car model, by sensors,

accelerometer (MPU-6050) and GPS (NEO-6), then

compared the measurements with reference, laser

profilometer device results. The results were close to

reference IRI values. To develop the method

considerably, they suggest further studies into

alternate wheel characteristics and types of roads. It

adds weight to the idea that sensor-equipped bicycles

can be practically useful for evaluating road

roughness.

To generate real time IRI maps, integration of

accelerometers and GPS sensors with

microcontrollers was also under consideration. The

validity of this system is confirmed by correlation

with smartphone data (Hafizh, Abdullah, Ateeq,

Majeed, Isaac, and Hu, 2023). It suggested that

adding variety of roads and environment condition

would help evaluate the accuracy of measurements,

emphasizing the flexibility of IoT systems.

This study presents a forward-looking approach

for autonomously monitoring cycle path surface

roughness using readily available components. With

cycling infrastructure expanding globally and a

growing emphasis on sustainability and active

transportation, the need for efficient and accurate

surface monitoring technologies becomes

increasingly crucial.

Our research builds upon the framework

established by Zang et al. (2018) for calculating

pavement roughness using bicycles. We further

develop their approach to autonomously calculate IRI

to assess surface quality of cycleways without manual

intervention. The system’s automation eliminates

human variability in posture and speed, providing

consistent and accurate measurements. This approach

enables large-scale deployment and continuous

monitoring without manual intervention, making it

cost-effective and user-friendly.

The core of our methodology is a refined

technique for determining the average acceleration

orientation vector by finding stable section,

autonomously. To further enhance the analytical

robustness of our assessments, we introduce the CRI,

quantized version of the traditional IRI. This new

Table 1: Provides list of the instruments, sensors, measurement methods with year of pavement roughness measurement.

Instrument Senso

Metho

Yea

Mays Mete

Acceleromete

Bump Integrato

1962

Profilograph Inclinometer, Acceleromete

Profile Index 1966

Contact Profiling Device Inertial Measurement units Pavement Condition Index 1973

stic

Laser, Inclinomete

Profilo

h Index 1980

South Dakota Road Profiler Laser, Inertial Measurement Units,

Global positioning system

International roughness Index 1986

Non-contact

rofilin

device Laser, O

tical Half Car rou

hness Index 1986

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

248

index provides generic assessment of surface

roughness specifically tailored to cycleways. In the

subsequent sections, we discuss the mathematical

model of the simplified quarter car model adapted for

bicycles, the rationale behind finding stable section,

detailed descriptions of our proposed methodology,

results and conclusion.

2 MATHEMATICAL

BACKGROUND

First, we will present the mathematical model as

described for roughness calculation by Zang, Shen,

Huang, Wan and Shi, (2018). They developed the

quarter car model for bicycles to calculate IRI,

providing all the necessary mathematical proofs. The

researchers explain that the pavement's signature is

determined from the longitudinal profile of the path,

with the profile’s vertical displacement being

quantified in equation 1.

𝐷



ℎ



−ℎ



for all 𝑖≥1

(1)

Where 𝐷 is vertical displacement, 𝑖 is the

sampling time, ℎ is height. Vertical displacement

data captured by action camera is prone to huge errors

thus we cannot rely on it. Thus, we calculate vertical

displacement using accelerometer by utilizing the

formulas from physics as shown in equation 2.

𝑉



𝑑𝐷

𝑑𝑡

𝛼















(2)

Where 𝑉



is vertical speed and 𝛼



is vertical

acceleration. This leads to vertical displacement,

equation 3.

𝐷= 

𝛼



𝑑𝑡

)







(3)

Since orientation of bicycle and action camera are

variable. The vertical acceleration

(

𝛼



)

can appear in

either of three dimentional acceleration data. Which

means vertical acceleration measurement cannot be

taken directly through the accelerometer data.

According to rules of physics the only force the

accelerometer gets in the stable condition is of

gravity, which is downward, with a value equal to 1

g. This condition is fulfilled mathematically as in

equation 4.

𝐴





∗

𝐴





𝐴





∗

𝐴





𝐴





∗

𝐴





(4)

Where 𝐴





𝐴





and 𝐴





are average acceleration

values of x, y and z axes in stable section. Using these

average acceleration values we can derive vertical

acceleration projection as in equation 5.

𝛼



𝐴



∗

𝐴





𝐴



∗

𝐴





𝐴



∗

𝐴





(5)

IRI quantifies smoothness of pavement surface

based on the response of moving vehicle. It is

expressed in meters per kilometre (m/km). Based on

its definition, its formulation can be expressed as

shown in equation 6.

𝐼𝑅𝐼=

∬

𝛼







(

𝑑𝑡

)



𝑆

(6)

Where 𝐼𝑅𝐼 is International Roughness Index,

𝑡𝑠𝑡𝑎𝑟𝑡 is starting time, 𝑡𝑠𝑡𝑜𝑝 is the stopping time and

𝑆 is the distance.

3 FINDING STABLE SECTION

The stable section refers to the duration of ride in

which the cycle remains steady, for five seconds.

Meaning the only force acting on the device in this

duration is gravity. Identifying this stable section is

crucial because the vertical acceleration component

cannot be directly obtained from the accelerometer

data due to the variable orientation of both the bicycle

and the action camera. Thus, the authors (Zang, Shen,

Huang, Wan and Shi, 2018) had to request the bicycle

rider to keep the speed and posture as stable as

possible for initial 5 seconds. Therefore, we propose

finding this stable section autonomously.

We started with rolling variance and created an

algorithm to find stable section. Rolling variances

were calculated, over a window of 500 samples of the

data from gyroscope using the equation 7.

𝑅𝑉=

𝑁−1



(

𝑥



−µ



)

















(7)

Where 𝑅𝑉 is rolling variance, 𝑁 is the size of the

rolling window i.e the number of observations

considered at each step, 𝑥



is the value of the time

series at time 𝑖. µ



is the rolling mean average at time

𝑡 which is calculated using the equation 8.

Towards Assessing Cycleway Pavement Surface Roughness Using an Action Camera with IMU and GPS

249



𝑁

𝑥





  

(8)

Use of algorithm 1, provides variance details of

the most stable section of the ride, excluding the part

in which cyclist stops and rides with speed less than

11 km/h. From this stable section we calculate

average acceleration values of x, y and z axes, which

are used to calculate the vertical acceleration.

Define Parameters: Set the rolling window size

W=500 for variance calculations.

Compute Rolling Variances: For each

gyroscope axis (-Y [rad/s], -X [rad/s], and Z

[rad/s]), calculate the rolling variance over the

window W.

Combine the computed variances and calculate

the average variance for each window.

Sort by Average Variance: Sort the dataset

based on the average variance in ascending order

to prioritize the lowest variances.

Iterate and Identify the First Significant

Variance: Initialize a flag found_first = False.

For i=1 till total_rows:

Compute the starting and ending indices of

the rolling window.

Calculate the cumulative sum of distances

within this window.

If the cumulative distance exceeds

D_threshold (cycle needs to be moving) and no

significant variance has been found.

Record the corresponding details: Rank,

index range and average variance value.

Set found_first to True and exit the loop.

Output Results: Return the first significant

variance's details.

Algorithm 1: Finding stable section.

4 METHODOLOGY

The methodology describes a systematic process for

assessing pavement surface. The flowchart of our

methodology is shown in figure 1. Initially,

pavement surface data is collected as the bicycle

moves along the cycleway. Following, metadata is

extracted from video file and unnecessary data (video

and audio) is removed to abide by GDPR. The

remaining data undergoes data cleaning and

processing, including the removal of redundant

values and interpolation for consistency. Later, the

autonomous simplified quarter-car model for bicycles

is applied to analyze pavement roughness by

processing sensor data.

In this study, GoPro Hero 9 action camera is used

due to its compact design, integrated sensing

capabilities and ease of deployment. It combines

high-frequency inertial and positional data

acquisition in a portable and cost-effective device.

The primary data for roughness assessment comes

from the Bosch BMI260 IMU, operating at a

sampling rate of 200 Hz. This IMU comprises a three-

dimensional accelerometer, which measures linear

accelerations, and a three-dimensional gyroscope,

which captures angular velocities. Positional data is

recorded via the UBlox UBX-M8030-CT GPS

module, which works at a frequency of 10 Hz (Gopro,

2024). This allows precise geotagging of IMU data,

mapping surface roughness to specific locations

along the cycleway.

Figure 1: Shows flowchart of methodology.

The placement and orientation of the device have

a direct impact on the quality of data collection.

Misplaced sensors can result in misaligned data,

thereby reducing pavement roughness accuracy.

Mounting the device on the centre of the bicycle

handlebar maintains a perfect symmetry between the

stability and sensor data collection (Westerhuis and

Waard, 2016). Thus, it is extremely important to

mount it at optimal position, as shown in figure 2.

The action camera produces video file that

contain embedded metadata. This metadata was

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

250

extracted using the GoPro Metadata Format parser

available on GitLab (GoPro, 2024). Useful metadata

includes timestamps, accelerometer, gyroscope and

GPS data.

We used Haversine Formula (Sinnott, 1984) to

compute distance using GPS locations. It calculates

the distance between two geographic points,

incorporating the curvature of the Earth. Its formula

is shown in equation 9. Where S is distance, 𝜑



and

𝜆



are latitude and longitude of point 1, 𝜑



and 𝜆



are

latitude and longitude of point 2, and R is the Earth

radius (mean radius = 6371 km).

The cubic spline interpolation method was used

to up-sample GPS location data from 10 Hz to 100

Hz. This method generates a smooth curve that passes

through the original data points, ensuring a

continuous and natural trajectory between recorded

locations. By interpolating additional data points at

higher frequency intervals, it helps maintain

smoothness in the time-series data, which is essential

for sensor fusion and further calculations.

The same cubic spline interpolation technique

was also applied to down-sample accelerometer and

gyroscope data from 200 Hz to 100 Hz. Instead of

simple decimation (which removes excess samples),

interpolation was used to construct a new 100 Hz

signal by fitting a smooth curve through the original

200 Hz data points. This ensures that the reduced

dataset maintains continuity and minimizes aliasing

and loss of critical motion information.

By applying interpolation in both up-sampling

and down-sampling, the time-series data remains

smooth and well-aligned across different sensor

modalities, improving the accuracy of sensor fusion

and subsequent analysis.

Using the gyroscope data, equation 7, equation 8,

and algorithm 1 we determine the most stable part of

the ride from which we calculated average

acceleration vector. After computing this vector, we

used it in equation 5 to find the true vertical

accelerations.

Previous computation provides all the necessary

data to compute IRI. Thus, the IRI is calculated using

equation 6. The calculation results were divided in 3

parts. Invalid, unable to calculate and the actual

reading. Invalid section is the one in which either the

cyclist stopped or moved at speed less than 11 km/h.

The unable to calculate part is the one in which there

was missing data reading from any of the sensors. The

actual reading part was utilized for further analysis.

IRI is a common way to measure road roughness,

but it has some limitations when used for cycleways.

Cyclists naturally move while riding, which creates

sudden fluctuations in IRI values that do not reflect

actual pavement roughness. Moreover, different

vehicles react differently to the same road surface,

direct IRI values can sometimes give misleading

results when analysing cycleway conditions.

For better analysis, we utilized CRI. First, a

moving median filter is applied to smooth the IRI

values. This filter slides over the data, replacing each

value with the median over a 5-meter window. By

reducing sharp fluctuations caused by human

movement, it ensures that only meaningful roughness

variations remain.

The second step is quantization (

𝑄(𝐼𝑅𝐼



)), where

the smoothed IRI (𝐼𝑅𝐼



) values are assigned to fixed

levels based on specific intervals, as shown in

equation 10. This process simplifies roughness

variations, ensuring more consistent and comparable

measurements. Quantization ensures that roughness

measurements from different vehicles can be

compared more effectively by eliminating minor

differences caused by vehicle dynamics. Through

these two steps, CRI provides a more stable and

interpretable measure of cycleway pavement

roughness. Additionally, setting a threshold of 8.5

helps filter out extreme variations that might not be

relevant to actual surface roughness.

𝑄(𝐼𝑅𝐼



⎩

⎪

⎨

⎪

⎧

0, 0 ≤ 𝐼𝑅𝐼



≤0.5

1, 0.51 ≤ 𝐼𝑅𝐼



≤1.5

2, 1.51 ≤ 𝐼𝑅𝐼



≤2.5

3, 2.51 ≤ 𝐼𝑅𝐼



≤3.5

4, 3.51 ≤ 𝐼𝑅𝐼



≤4.5

5, 4.51 ≤ 𝐼𝑅𝐼



≤5.5

6, 5.51 ≤ 𝐼𝑅𝐼



≤6.5

7, 6.51 ≤ 𝐼𝑅𝐼



≤7.5

8, 7.51 ≤ 𝐼𝑅𝐼



≤8.5

(10)

5 EXPERIMENTS AND RESULT

Two experiments were performed to evaluate the

effectiveness of the proposed model assessing

cycleway pavement roughness. The first experiment

analyses performance across diverse tracks, ensuring

its capability to capture roughness under different

conditions. The second experiment compare model

consistency and reliability when applied to different

S=2∗R∗arcsin



sin





𝜑



−𝜑





+cos

(

𝜑



)

∗ cos

(

𝜑



)

∗sin





𝜆



−𝜆





(9)

Towards Assessing Cycleway Pavement Surface Roughness Using an Action Camera with IMU and GPS

251

types of bikes, highlighting its adaptability to various

vehicle configurations.

Bikes used for these experiments were

manufactured from renowned companies. They were

selected based on market availability and their

suitability for use on paved roads. They were in their

original condition, without any modifications.

Data was collected by adult volunteers who

consented to process their personal data. Cyclists

wore safety equipment and collected data in

compliance with GDPR. The bikes were ridden on

asphalt cycleways. Ethical approval for data

collection was obtained beforehand.

5.1 Experiment 1

The 1

experiment was conducted to evaluate the

performance of the proposed model across diverse

cycle tracks. The assessment focused on data

accuracy, reliability, and result consistency to

comprehensively analyse the system’s overall

effectiveness.

A volunteer, aged between 25 and 50 years, rode

a manual bicycle across three cycle tracks in the

northern region of Donegal, Ireland. A total of 11.66

kilometres of path data was processed. The specific

track routes were as follows:

Track-1: From 54.906389° N, -8.309722° W to

54.930833° N, -8.318889° W.

Track-2: From 54.896111° N, -8.308611° W to

54.855278° N, -8.310556° W.

Track-3: From 54.834722° N, -8.332500° W to

54.855278° N, -8.310000° W.

This dataset provides a basis for analysing the

model’s response to varying track conditions. Table 2

presents a summary of the key characteristics of each

cycle track. Track 1, the shortest, spans 2,890.95

meters and features three intersections, two potholes

and vegetation encroaching along the path. In

contrast, Track-2, the longest at 5,374.33 meters,

consists of eight intersections, 20 potholes, and

significant vegetation coverage. Track 3, measuring

3,399.71 meters, lacks vegetation but includes five

intersections and ten potholes, contributing to a

varied cycling environment. The information was

captured by manually looking the video after data

capture from the GoPro camera. The selected tracks

encompass a diverse mix of urban and rural terrains,

incorporating steep gradients, smooth and uneven

surfaces, potholes, sharp turns, intersections, and

extended straight segments. These variations ensure

that the evaluation considers a broad range of real-

world cycling conditions.

Table 2: Show track statistics summary.

Track Number of

Intersections

Number

Potholes

Presence

Vegetation

Distance

(m)

1 3 2 Yes 2890.95

2 8 20 Yes 5374.33

3 5 10 No 3399.71

Bike used for this experiment was a manual bike,

Trek 7100 Bike. Figure 2 shows this bike along with

action camera. It is built with a lightweight

aluminium frame having front suspension. It is

equipped with 27.5 x 1.38-inch tires and powered by

a 21-speed drivetrain, allowing it to adapt to various

terrains with ease.

Figure 2: Trek 7100 Bike with action camera.

5.1.1 Results of Experiment 1

The analysis focuses on the CRI values as a function

of distance along each track, considering variations in

track features such as length and track surface. All

three tracks exhibited CRI values within the range of

0 to 5.

Figure 3 shows the results for track 1, where CRI

values remain relatively stable at 1 for most of the

track, indicating a smooth surface. The maximum

CRI value recorded on this track was 5. The average

CRI value for the entire track was 1.3. Minor spikes

in the CRI values are observed indicating isolated

sections of increased roughness. This increased

roughness could be due to potholes and intersections.

From these results, it is hard to conclude the actual

effect of intersections, potholes and vegetation.

However, it gives a holistic view of the ride

roughness and quality.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

252

Figure 3: Track 1 results.

Track 2 exhibits similar fluctuations to those

observed in track 1, as seen in figure 4, with CRI

values mostly at 1, suggesting a smooth segment.

Only one spot shows a CRI value of 5. Despite being

the longest track, it recorded the lowest average CRI

value of 1.2, which indicates a high-quality asphalt

surface.

Figure 4: Track 2 results.

Figure 5 shows the results for track 3, which

exhibits frequent variations in CRI values. The

average CRI value for this track was 1.45. The

smoother

sections

track

indicate

similar

paving

Figure 5: Track 3 results.

standards as of track 1 and track 2. Despite having

fewer intersections and potholes than Track 2, the

quality of Track 3 creates noticeable roughness at

certain stretches along the track.

5.2 Experiment 2

The 2

experiment was conducted to analyse the

behaviour of the proposed model across different

bicycles and evaluate its response to asphalt

pavement characteristics. The study involved two

cyclists, aged between 25 and 50 years, riding distinct

bicycles on the same dedicated cycle track in

Connemara, Ireland, under dry weather conditions.

The total distance covered was 3.40 kilometres,

starting from 53.45482° N, -9.86668° W and ending

at 53.45681° N, -9.82077° W.

The experiment was conducted to compare the

proposed model behaviour on different bikes. The

assessment focuses on evaluating the asphalt

pavement characteristics. The route represents a rural

cycling environment with a combination of even and

uneven terrain, vegetation coverage, straight

stretches, six sharp turns, and two intersections.

The e-bike and e-scooter were equipped with a

folding mechanism, pneumatic tires with different

radius, disc brakes and a digital speedometer. Figure

6, include the Mirider One Folding Electric Bike (e-

bike) and the Xiaomi Electric Scooter 4 Pro (e-

scooter).

Figure 6: Mirider One Folding Electric Bike and Xiaomi

Electric Scooter 4 Pro.

The Mirider One Folding Electric Bike features a

magnesium alloy frame and a rear suspension system,

designed for improved comfort on varying terrain. It

is fitted with 16 × 1.75-inch pneumatic tires and

of up to 25 km/h. The Xiaomi Electric Scooter 4 Pro

is a lightweight aluminium alloy scooter equipped

with 10-inch pneumatic tires. It is powered by a

350W front hub motor, capable of reaching a top

speed of 25 km/h. These two distinct vehicle

configurations provide a basis for comparing model

performance, contributing to a more comprehensive

understanding of road surface interactions.

Towards Assessing Cycleway Pavement Surface Roughness Using an Action Camera with IMU and GPS

253

5.2.1 Results of Experiment 2

The results were divided into 250-meter sections, to

do comprehensive analysis of the cycleway

roughness experienced by two cycles: e-bike and e-

scooter. CRI values were calculated for both bicycles

to evaluate the impact of road surface irregularities.

Across all sections, e-bike exhibited higher CRI

values, mostly clustering between 3 and 6, indicating

a rougher ride. These values indicate that the e-bike

was more sensitive to road surface irregularities.

On the other hand, e-scooter resulted with lower

CRI values, clustering between 0 and 2, signifying a

smoother ride. These lower values reflect the e-

scooter's ability to handle surface unevenness more

effectively, providing a smoother ride.

Varying degrees of bias in results is observed, as

illustrated in figure 7, due to the tyre size and

mechanical differences of bicycles. It compares CRI

values for both cycles across section 1. In the plot,

multiple distinct peaks are also visible at the same

distances for both e-bike and e-scooter. Thus,

indicating the reliability of the simplified quarter car

model without requiring manual intervention, in

identifying pavement surface conditions using

bicycles.

Figure 7: Section 1 results.

Furthermore, additional factors can introduce

variability in roughness assessment. One potential

source of error is the mounting position of the action

camera, slight changes in placement alter vibration

measurements. Environmental factors such as wind

also play a crucial role in introducing bias in CRI

measurements.

6 CONCLUSION AND

LIMITATION

This research was conducted to evaluate the

effectiveness of a simplified quarter car model for

cycles, without manual intervention, in assessing

cycleway surface quality. The methodology utilizes

proposed method to find stable section in order to

calculate average acceleration vector.

For analysis, two experiments were carried out on

paved asphalt cycleways. Cyclists rode action camera

mounted bicycles and collected GPS and IMU sensor

data. This data was processed using proposed model,

and the results were further processed through CRI

methodology.

In Experiment 1, the results showed that CRI

values varied significantly with track features. The

CRI values for all three tracks ranged from 0 to 5.

Among three tracks, track 2 had the smoothest surface

indicating a high-quality asphalt surface, while Track

3 exhibited the highest roughness and was assessed as

having low quality asphalt surface. Track 1 had

noticeable surface irregularities and was rated as

having moderate asphalt quality surface.

In Experiment 2, the behaviour of the proposed

model was assessed by riding different bicycles on the

same track, focusing on evaluating asphalt pavement

roughness. Two types of bicycles were used: an e-bike

and an e-scooter. The results indicated bias such that

the e-bike consistently experienced higher CRI values

(between 3 and 6), and the e-scooter experienced

lower CRI values (between 0 and 2).

Consistency in results indicate model’s ability to

identify rough patches. Its self-sufficiency nature

points towards the method’s tendency for scalability.

Through which, on large scale, insights can be

extracted to improve cycling infrastructure and

enhance ride comfort.

While this study provides valuable insights into

roughness measurements, certain limitations should

be acknowledged. The experiments were conducted

without considering camera data. This data would

have helped to conclude deep insights about relation

between track features and CRI.

Additionally, the absence of ground truth data for

validation, limits the ability to directly assess the

accuracy of the proposed model against established

benchmarks. Another limitation is that the study was

restricted to paved asphalt cycleways, excluding other

surface types such as gravel paths, which could

exhibit different roughness characteristics.

7 PRACTICAL

IMPLEMENTATIONS

The proposed surface roughness assessment system is

well-suited for real-world deployment due to its

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

254

reliance on commercially available action camera.

The system can be integrated into smart city

frameworks, allowing transportation departments to

monitor cycleway conditions efficiently. By linking

roughness data to county council infrastructure

management systems, authorities can prioritize

maintenance efforts, improving cycling safety and

experience. Additionally, a web-based dashboard and

mobile application could facilitate access to

roughness metrics, enabling cyclists to make

informed decisions about their routes. The

autonomous nature of this system makes it scalable

for city-wide deployment, reducing the need for

manual intervention while ensuring continuous

monitoring of cycling infrastructure.

ACKNOWLEDGEMENTS

This research is conducted with the financial support

of the EU commission Recovery and Resilience

Facility under the Research Ireland OurTech

Challenge Grant Number 22/NCF/OT/11220 and the

support of Science Foundation Ireland under Grant

number [SFI/12/RC/2289\_P2] the Insight SFI

Research Centre for Data Analytics. The authors

acknowledge support from Transport Infrastructure

Ireland and Katleen Bell-Bonjean (Societal Impact

Champion from GORTCYCLETRAILS.ie).For the

purpose of Open Access, the author has applied a CC

BY public copyright license to any Author Accepted

Manuscript version arising from this submission.

REFERENCES

Alatoom, Y. I., & Obaidat, T. I. (2022). Measurement of

street pavement roughness in urban areas using

smartphone. International Journal of Pavement

Research and Technology, 1-18.

Fares, A., & Zayed, T. (2023). Industry-and academic-

based trends in pavement roughness inspection

technologies over the past five decades: A critical

review. Remote Sensing, 15(11), 2941.

GoPro. (2024, October 11). gpmf-parser [GitHub

repository]. GitHub. https://github.com/gopro/gpmf-

parser

Hafizh, H., Abdullah, R., Ateeq, M., Majeed A., Isaac, M.,

Hu, B. (2023). Measurement and Analysis of

International Roughness Index using IoT-based

System. IEEE Symposium on Wireless Technology &

Applications, 10249899.

Hettiarachchi, C., Yuan, J., Amirkhanian, S., & Xiao, F.

(2023). Measurement of pavement unevenness and

evaluation through the IRI parameter–An overview.

Measurement, 206, 112284.

Sinnott, R. W. (1984). Virtues of the Haversine. Sky and

Telescope, 68(2), 159.

Kumar, L., Tallam, T., & Kumar, C. N. (2022, March).

Assessment of ride quality and road roughness by

measuring the response from a vehicle mounted

Android smartphone. In IOP Conference Series: Earth

and Environmental Science (Vol. 982, No. 1, p.

012062). IOP Publishing.

Rizelioğlu, M., & Yazıcı, M. (2024). New approach to

determining the roughness of bicycle roads.

Transportation research record, 2678(1), 781-793.

Sayers, M. W., Gillespie, T. D., & Queiroz, C. A. V. (1986).

The international road roughness experiment: A basis

for establishing a standard scale for road roughness

measurements. Transportation research record, 1084,

76-85.

Westerhuis, F., & De Waard, D. (2016). Using commercial

GPS action cameras for gathering naturalistic cycling

data. Journal of the Society of Instrument and Control

Engineers, 55(5), 422-430.

Woodstrom, J. H. (1990). Measurements, specifications,

and achievement of smoothness for pavement

construction (Vol. 167). Transportation Research

Board.

Yu, Q., Fang, Y., & Wix, R. (2022). Pavement roughness

index estimation and anomaly detection using

smartphones. Automation in construction, 141, 104409.

Zang, K., Shen, J., Huang, H., Wan, M., & Shi, J. (2018).

Assessing and mapping of road surface roughness

based on GPS and accelerometer sensors on bicycle-

mounted smartphones. Sensors, 18(3), 914.

Zhang, Z., & Wenjiang, L. V. (2022, February). Research

on theoretical evaluation method of road roughness

based on smart-phone and whole vehicle model. In

Sixth International Conference on Electromechanical

Control Technology and Transportation (ICECTT

2021) (Vol. 12081, pp. 339-347). SPIE.

Towards Assessing Cycleway Pavement Surface Roughness Using an Action Camera with IMU and GPS

255