2025道路质量监测机器学习技术指南（英）.docx

GUIDEBOOK ON MACHINE LEARNING TECHNIQUES FOR ROAD QUALITY MONITORING
MARCH 2025

ASIAN DEVELOPMENT BANK

GUIDEBOOK ON MACHINE LEARNING TECHNIQUES FOR ROAD QUALITY MONITORING
MARCH 2025

Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO)

© 2025 Asian Development Bank
6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel [phone]; Fax [phone]
www.adb.org
Some rights reserved. Published in 2025.
ISBN 978-92-9277-230-7 (print); 978-92-9277-231-4 (PDF); 978-92-9277-232-1 (ebook)
Publication Stock No. TCS250091-2
DOI: http://dx.doi.org/10.22617/TCS250091-2
The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies of the Asian Development Bank (ADB) or its Board of Governors or the governments they represent.
ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned.
By making any designation of or reference to a particular territory or geographic area in this document, ADB does not intend to make any judgments as to the legal or other status of any territory or area.
This publication is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound by the terms of this license. For attribution, translations, adaptations, and permissions, please read the provisions and terms of use at https://www.adb.org/terms-use#openaccess.
This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed to another source, please contact the copyright owner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material.
Please contact [e-mail] if you have questions or comments with respect to content, or if you wish to obtain copyright permission for your intended use that does not fall within these terms, or for permission to use the ADB logo.
Corrigenda to ADB publications may be found at http://www.adb.org/publications/corrigenda. Note:
In this publication, “$” refers to United States dollars and “SLRs” refers to Sri Lanka rupees.
Cover design by Mike Cortes.

Contents

Tables and Figures v

iii

Foreword viii
Abbreviations x
I. Introduction 1
Conventional Approaches of Evaluating Quality of Road Pavements 1
Challenges with Using Conventional Approaches of Evaluating Quality of Road Pavements 2
Potential of Innovative Data and Frontier Technologies 2
II. A Review of Satellite Imagery-Based Methods in Road Quality Assessment 4
Overview of Satellite Imagery Datasets 5
Some Similar Applications 7
Feasibility of Remote Sensing Techniques Using Satellite Imagery 8
III. Application of Machine Learning Algorithms on Satellite Imagery for Road Quality Monitoring 11
Neural Networks 12
Generative Models 13
IV. Hardware and Software Requirements and Setup 16
Hardware 16
Software Requirements and Installation 16
V. Data Preparation 28
Acquiring Shapefiles 28
Acquiring Road Bounding Boxes 30
Uploading the Output to Google Drive 32
Processing Bounding Boxes 35
Downloading Satellite Imagery 39
VI. Training the Super-Resolution Model (REAL-ESRGAN) 50
Real-ESRGAN and Basic-SR 50
VII. Classification Using Satellite Imagery 56
Downloading the Philippines’ Road Sections 56
Training the Classification Model 64
iv Contents
VIII. Alternative Method: Smartphone-Based Pavement Condition Assessment 72
Smartphone Technology Used for Pavement Condition Assessment 72
Roughness Evaluation Using Smartphone-Based Data 75
Smartphone-Based Data for Pavement Condition Evaluation: State of the Practice 82
Expanding Smartphone-Based Data to Crowdsourced Data for Pavement Condition Assessment 88
Feasibility of Adopting Smartphone-Based Assessment Methods 100
IX. Moving Forward: Integrating Different Techniques in Road Condition Monitoring 107
Comparison of Pavement Condition Monitoring Techniques: Traditional versus Frontier Techniques 107
Summary of Financial and Technical Resource Requirements for Implementing These Techniques 108
Discussion on Integrating Different Techniques in Road Condition Assessment 110
X. Summary and Conclusion 112
Appendix 114
References 134

Tables and Figures

Tables
1 Smartphone Sensors Used to Evaluate Pavement Condition Metrics 73
2 Summary of Different Influencing Factors on Simulated Vehicle Body Acceleration 74
3 Vehicle Operation Speeds for Different Smartphone Sampling Frequencies 77
4 Summary of Machine Learning Techniques Used for Roughness Evaluation 79
5 Summary of Current Smartphone Technologies to Measure Roughness 85
6 Effects of Driving Regimes on Simulated Grms 99
7 Comparison of Cost Values for Pavement Roughness Measurement Compared to the Laser Profiler 101
(Class I), Bump Integrator (Class III) Illustrated from the Practice in Sri Lanka
8 Case Studies Conducted to Evaluate the Applicability of Road Condition Monitoring Using 102
Smartphone-Based Approaches
9 Cost Comparison for Smartphone-Based and Satellite-Image-Based Monitoring for
Illustrative Example 109
A.1 Overview of Pavement Condition Metrics Measured Through Laser Profilers 117

A.2 Overview of Pavement Condition Metrics Measured Through LiDAR 119
A.3 Overview of Pavement Condition Metrics Measured Through Video and Image Processing 123
A.4 Overview of Pavement Condition Metrics Measured Through GPR 128
A.5 Overview of Pavement Condition Metrics Measured Through VehicleResponse-Based Measurement 129
A.6 Comparison Between Automated and Manual Data Collection 131
A.7 Important Characteristics of Recommended Equipment for Data Collection 132

Figures

1 Road Map of Methodology for Predicting Road Quality Using Satellite Imagery
4

2 Comparison Between ESRGAN and Low-Resolution Images
14

3 QGIS Download Page
17

4 CUDA Toolkit
20

5 Command to Install PyTorch Built with CUDA
21

6 Creating a New Project
23

7 Enabling the Earth Engine API
24

8 Getting Started with Earth Engine
24

9 Earth Engine Sign-Up Page
25

10 Registering a Noncommercial or Commercial Project
26

11 Unpaid Usage for Earth Engine
26

12 Creating a new Google Cloud Project
27

13 Mainland US Primary and Secondary Roads Shapefiles Loaded into QGIS
28

14 Merge Vector Layers Tool in QGIS
29

15 Reprojecting Layer in QGIS
29

16 How to Generate Points Along Lines
30

v

vi Tables and Figures
17 This Map Shows the Region New Jersey (NYC) with the US Tiger Lines Primary Roads 30
18 How to Export Data 31
19 Exported Data in Local Directory 31
20 How to Navigate to Google Drive 32
21 Google Drive 33
22 Available Cloud Storage 33
23 How to Upload File 34
24 How to Upload a File 34
25 How to Upload a Jupyter Notebook from Local Directory to Colab 35
26 How to Initialize Colab Environment 36
27 The Location of the Yaml File on Google Drive 51
28 Editing the Yaml File 52
29 Location Where the Pretrained Models Should be Placed 52
30 Location of Trained Model 53
31 Sample of Super-Resolved Imageries 54
32 Confusion Matrix 71
33 Z-Axis Acceleration vs. Time Graph Representing a) Smooth Road b) Rough Road 73
34 Plot for ÖPSD versus IRI After Removing the Effect of Distress 76
35 Different Acceleration Graphs for Various Acceleration Components 83
36 Average Vertical Acceleration of a Journey 83
37 The Relationship Between Bump Integrator IRI versus Roadroid IRI 87
38 The Relationship Derived Between IRI versus PCI 87
39 A Typical Crowdsourced Data Analytics Scheme 89
40 Framework to Convert Crowdsourced Data into Roughness Estimation 90
41 An Illustrative Example of Predicting IRI and Pothole Using Crowdsourced Acceleration Data 91
42 An Example of Pavement Data Collection Can Be Achieved with the High-Resolution Camera 92
43 Representative Road Link with Excessive Road Dust Levels Identified Using Roadroid
Image Analysis Module at Auckland and Waikato, New Zealand 93
44 The Automated Cracking and Pothole Identification Module (RoadData) 94
45 Street-Smart Cloud-Based Platform to Report Distresses, Webster, Texas 94
46 TotalPave Portal Home Screen, Montana, United States 95
47 Roadbotics Platform to Assess Distress Condition, Pittsburgh, United States 96
48 Comparison of FFS Variation with Increasing of IRI 97
49 Comparison of V50 Variation with Increasing of IRI 97
50 A System Diagram of a Typical Smartphone-Based Data Collection for Road Condition Monitoring 100
A.1 Typical Front Layout of a Laser Profiler 116
A.2 Automated Pavement Distress Detection Steps 118
A.3 Distress Patterns Captured Through Color or Elevation Profiles 119
A.4 Process of Image Processing (Pothole Used as an Example) 121
A.5 Raw Images Taken into the Analysis and the Processed Image for Identification 122
for Different Distress Types (Cracking Used as an Example)
A.6 An Image of Pavement from UAV 124
A.7 A Typical Pavement Distress Detection Result Using YOLOv3 Model: 125
a) True-Positives of Crack, b) True-Positives of Repair, c) True-Positives of Pothole,
d) False-Positive Results, and e) Typical False-Negative Results
A.8 Typical GPR System 126
A.9 Line Scan of Pothole Recorded with 10, 20, 50 and 100 Scan/m (from left to right) 127

Tables and Figures vii
A.10 Different Patterns from GPR to Identify the Different Distresses 127
A.11 Detection of Pothole with Thermal Mapping a) Thermogram of Pothole on Pavement; 130
b) Digital Photo of the Same Pothole on Pavement
A.12 Detection of Pothole with Thermal Mapping a) Thermogram of Crack on Pavement; 131
b) Digital Photo of the Same Crack on Pavement

Foreword

Road quality monitoring is essential for sustainable development, as it directly affects economic growth, social equity, and environmental resilience. Well-maintained roads build crucial connections between rural and urban areas, giving access to education, health care, markets, and job opportunities, while also making communities more resilient to disasters triggered by natural hazards. Yet in many regions, particularly in rural areas of developing countries, road conditions remain poor, and road quality data are often limited, thus constraining effective infrastructure planning and investment.
The Rural Access Index RAI) serves as a valuable metric in addressing these challenges. It evaluates the access of rural populations to reliable transportation and highlights accessibility gaps that must be bridged for sustainable and inclusive development. But a significant data availability gap limits the RAI’s potential, especially in remote areas, where traditional road quality surveys can be logistically challenging and costly. Developing innovative, cost-effective methods for monitoring road quality is
therefore critical to improving rural access and contributing directly to the United Nations Sustainable Development Goals (SDGs). SDG 9 in particular is centered on building resilient infrastructure, promoting inclusive and sustainable industrialization, and encouraging innovation.
To accelerate these efforts, the Japan Fund for Prosperous and Resilient Asia and the Pacific (JFPR), in cooperation with the Asian Development Bank, has emerged as a pivotal partner in driving sustainable rural development across the region. Committed to addressing urgent infrastructure and connectivity needs, the JFPR has mobilized resources and expertise to promote resilient and accessible rural areas for improved prosperity and sustainability. By supporting innovative solutions tailored to the unique needs of diverse communities, the JFPR backs initiatives that place rural areas within reach and give countless people across Asia and the Pacific a better quality of life.
With support from this fund, in collaboration with the World Data Lab, this guidebook provides a walk- through of different approaches to road quality monitoring. Outlined here is a step-by-step procedure for using geospatial data (particularly satellite imagery) and machine learning techniques to make the indicators for road quality and access more granular and timely. This procedure includes, among others, the steps outlined in the methodology used in the paper Application of Machine Learning Algorithms
on Satellite Imagery for Road Quality Monitoring: An Alternative Approach to Road Quality Surveys. This
guidebook was written by H.R. Pasindu, Aaron Thegeya, Thomas Mitterling, Clifford Njoroge, Arturo Martinez Jr., Joseph Albert Niño Bulan, Ron Lester Durante, Jayzon Mag-atas, and Oshean Lee Garonita, under the overall direction of Elaine Tan. Data and valuable technical input were supplied by the Philippine Statistics Authority and the country’s Department of Public Works and Highways;
Thailand’s National Statistical Office, Department of Land Transport, and Department of Rural Roads; and participants in various workshops organized under this project. Michael Anyala, Mac Cordel, Yohan Iddawela, and Jiwoo Kang also provided insightful feedback. Manuscript editing was performed by Sean Crowley and Mary Ann Asico. Typesetting was conducted by Jonathan Yamongan. The cover was designed by Mike Cortes. Proofreading was conducted by Lawrence Casiraya. Administrative support was provided by Rose Anne Dumayas and Iva Sebastian-Samaniego.

viii

Foreword ix

The guidebook outlines alternatives to traditional, resource-intensive road surveys. As satellite imagery becomes more available, this methodology can help make road quality monitoring more efficient and cost-effective—particularly in resource-constrained countries.
This publication illustrates how embracing innovative technology and data-driven solutions can help accelerate progress toward a more connected and prosperous future for all.

Albert Park
Chief Economist
Asian Development Bank

Abbreviations

AI
artificial intelligence
AMD
Advanced Micro Devices
AOI
around the bounding box
API
application programming interface
ASTM
American Society for Testing Materials
CNN
convolutional neural network
CPU
central processing unit
CSV
comma-separated values
FFS
free flow speed
GB
gigabyte
GEE

Google Earth Engine
GPR
ground penetrating radar
GPS
global positioning system
GPU
graphics processing unit
Grms
root-mean-square acceleration or vehicle body acceleration
IRI
international roughness index
LiDAR
light detection and ranging
PCI
pavement condition index
PDI
pavement d