Analysis of Real Time Road Surface and Acoustic Data Processing for

Minimizing the Accident Rate Using Feed Forward Neural Network

Eswaramoorthi R., Velliangiri A., Devaviknesh S., Haribaskar S., Naveen S. and Prasanth D.

Department of Electronics and Communication Engineering, K.S.R. College of Engineering, Tiruchengode, Namakkal,

Tamil Nadu, India

Keywords: Feature Selection, Feature Engineering, Classification, Machine Learning, Ensemble Learning, Anamoly

Detection.

Abstract: Simplified road condition monitoring is essential to maintaining road safety and maximizing transportation

efficiency in smart cities. Effective road surface detection has been significantly improved by the use of

artificial intelligence (AI). Issues with asphalt pavement are the main concern of both developed and emerging

countries for the efficient functioning of everyday commutes. The identification of potholes, which are

dangerous to cars and people and can result in an accident, has been the subject of several research. In order

to identify potholes on edge devices, this study aims to explore the possibilities of deep learning models and

use three outstanding deep learning models. This article proposes a low-cost technology for detecting the

surface qualities of road pavement in real-time. The time-frequency domain processing of the inertial signals

given by on-car sensors is done in order to get information about the condition of the road surface. The

effectiveness of the suggested approach in determining the kind and existence of distress is demonstrated by

the high categorization rates. Following data collection from the road surfaces, machine learning methods like

Multi-Layer Perceptron (MLP) are used for analysis. The outcomes show how well the suggested approach

can distinguish between various road conditions. These findings showed that the MLP had a higher accuracy

of 98.98% when evaluating road conditions. In order to give safe transportation services in smart cities, the

study offers important insights into the creation of a more effective and dependable road condition monitoring

system.

1 INTRODUCTION

A pothole, one of the types of asphalt pavement

failures that occur due to water in the supporting soil

structure and traffic over the used area. First, when

the underlying soil structure is hydrated and water is

added, the supporting soil softens. Traffic wears and

breaks up the poorly-supported wearing pavement in

the affected area. It causes the asphalt and dust

particles underneath to come out and sink, leaving an

empty area in the road due to continuous driving

activity.

Figure 1: A deep pothole.

1.1 Formation of Pothole

Pothole primer A public administrator’s guide to an

understanding and handling of the pothole problem,

U.S. Army Corps of Engineers (Eaton et al.), says it

takes two conditions to make a pothole: water and

traffic. Water weakens the soil under the pavement,

or the pressure of traffic applies stresses to the

pavement that make it crack. Potholes do not form

overnight, but rather start as a sign of fatigue within

the surface of the road. Pavement that breaks up in

this way between fatigue cracks is finally released in

the form of a pothole when progressive fatigue leaves

it unvoiced by adjacent surface stresses, and it is

plucked or driven out of the surface by ongoing tire

stresses. In areas that experience freeze and thaw, this

type of freeze and thaw can damage a pavement and

leave openings for water to enter. The process, made

more acute than ever this year by spring thaw which

saturates and weakens the supporting soil happens

208

R., E., A., V., S., D., S., H., S., N. and D., P.

Analysis of Real Time Road Surface and Acoustic Data Processing for Minimizing the Accident Rate Using Feed Forward Neural Network.

DOI: 10.5220/0013880200004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages

208-213

ISBN: 978-989-758-777-1

when the upper parts of a pavement’s soil structure

cannot melt past lower, still-frozen layers. Though

they rarely exceed a few inches in depth, potholes

can stretch out to several feet in width. If large enough

as shown in figure 2 and figure 3, they are likely to

cause damage to tires, wheels and even car

suspensions. On-road implications. This can lead

directly to serious traffic accidents especially on

freeways with high speeds.

Figure 2: Condition of roads with potholes.

Figure 3: Road surface defects.

As illustrated in the figure, pavement distresses

for asphalt sections are primarily divided into eight

categories, as per the Texas Department of

Transportation's (TxDOT) Pavement Management

Information System Rater's Manual: rutting,

patching, block, alligator, longitudinal and transverse

cracking, ravelling, and potholes. Regular road

surface monitoring and maintenance is crucial since

it may extend a road's lifespan from 15 to over 30

years. The goal of this project is to create a road

classification system that can evaluate the

infrastructure in real time. The creation of novel

artificial intelligence methods that can learn on their

own from acoustic data obtained through an

integrated system is the main goal of this article. The

algorithm is implemented on an electronic board that

is fixed to a car's rim flange. It is connected to a

microphone that is placed within the tire hollow and

has components for transmitting Low Energy (LE)

data with web monitoring that is based on the Internet

of Things.

2 LITERATURE SURVEY

Moazzam et al. (2013) may provide a pavement

distress detection technique that mainly focuses on

detecting and analyzing potholes and cracks

generated due to road damage. The study addresses

the essential issue of accurately predicting the volume

of filler material required to repair a pothole to avoid

any shortages and wastage. The idea is to extract road

surface depth images from a Concrete and Asphalt

roadway using an inexpensive Kinect sensor. These

photographs are rendered in a way to generate meshes

that facilitate the identification of potholes.

S.S. Rode et al. Pot-hole avoidance system for

vehicles. (2009) propose a unique architecture for a

pothole detection and warning system which is Wi-

Fi based. As Wi-Fi-enabled cars drive through the

area covered by the access points, they will receive

the data broadcasted by the access points along the

roadway. Additionally, acknowledgments responses

may be transmitted by front-end nodes (cars) to the

access points where they are dispatched to the back-

end servers.

H. Youquan et al. proposed a method of efficient

detection of the 3D sectional area of pavement

potholes. (2011). The optical imaging principle of

three-dimensional projection transformation is

adopted to take pictures of the cross-section of

potholes during detection. It involves a variety of

digital image processing techniques like error

analysis and compensation, image preprocessing,

binarization, thinning, and 3D reconstruction. These

experiment observations indicate that this method

outperforms traditional methods in many aspects and

enhances pothole detecting accuracy and

performance.

Different from the more common investigation on

crack detection in pavement distress (J. Lin et al.

(2010) focus on the less studied area of pothole

identification. The paper proposed a pothole detection

method which uses histogram to extract the textural

features that refers to a region of an image. A non-

linear support vector machine (SVM) is then used to

classify if the specified target region is indeed a

pothole. Based on experimental data, the proposed

algorithm offers an effective approach for detecting

potholes on pavements, attaining a significantly high

identification rate.

A DNN model for identifying the speaker was

presented by F. Ye et al. (2021). Published in the

Analysis of Real Time Road Surface and Acoustic Data Processing for Minimizing the Accident Rate Using Feed Forward Neural Network

209

journal of Applied Sciences, these researchers

outlined a strategy to effectively and efficiently

increase the accuracy of speaker identification

systems using deep learning methods**. By using

speech characteristics to identify specific speakers,

the model offers a reliable solution for use cases such

as voice-activated systems and security.

Experimental findings demonstrate that the model

achieves good recognition performance for tasks like

speaker identification.

3 EXISTING SYSTEM

The current pothole-maintenance system has a black-

box camera-based pothole detector. The pothole-

detection system uses the camera to gather pothole

information, including size, position, and appearance.

The pothole-maintenance server utilizes the gathered

data for intelligent pothole maintenance, which is

kept in the pothole database. Based on our prior

pothole database system, we created new software for

the pothole-maintenance server, as seen in Figure 1

on the right. A variety of information about potholes

is provided by this software, including video clips,

images, regions, road authorities, road number,

driving direction, lane number, road type, latitude,

longitude, collectors, date of collection, pavement

type, location, shape, size, and comments. With the

help of the gathered GPS data, the pothole's position

is displayed on a digital map. As a result, viewers can

observe the pothole distribution with ease.

Additionally, pothole maintenance expenses in the

chosen region are precisely estimated by the program.

In this manner, the program makes it simple and

precise for transportation officials to create road-

maintenance rules and strategies. Then, potholes can

be intelligently fixed using a pothole-maintenance

system, like our intelligent asphalt repair systems, and

pothole data can be shared with other users and

services through Open API and external connections.

Pothole-maintenance system (Figure 1). Insofar as the

current approach for detecting potholes only employs

one black-box camera. It is possible to swiftly collect

data over a large region and construct a variety of

survey vehicles for pothole identification at a

reasonable cost. Actually, because current pothole-

maintenance methods do not offer reliable pothole

information, the Korean government is unable to

budget for yearly road repair expenses with any

degree of accuracy. This initiative aims to identify

and track road conditions and raise awareness of

anomalies on the road, which is likely to occur in

nations like India.

4 PROPOSED SYSTEM

The goal of this project is to create a road

classification system that can evaluate the

infrastructure in real time. The creation of novel

artificial intelligence methods that can learn on their

own from visual and inertial data obtained through an

integrated system is the main goal of this research.

The algorithm is implemented on an electronic board

that is positioned on the dashboard or onboard unit of

a car. It is connected to a camera and sensors that are

placed within the suspension cavity and has WiFi or

Internet of Things data transmission components.

Three stages of the system's evaluation were

conducted. First, the two CNNs were trained, and

their outputs were compared. After that, the model

with the best performance was chosen and quantified.

Additionally, as a last comparison, the model

accuracy for both the quantized and floating-point

models was computed. Finally, the classifier was

included into the embedded firmware, and an

Arduino NANO board was used to evaluate its

operation. To evaluate the application's performance

on actual hardware, it was then installed on the

specially made board. The development of a new

dataset comprising the inertial data resulting from the

contact between the wheel and road surfaces is a

significant proposal of this project. The data

collection was created in part through data

augmentation and mostly through many measurement

efforts.

5 PROPOSED DESCRIPTION

One really interesting and valuable project to detect

road quality is to build a road surface analyser with

Arduino Nano, MPU6050 accelerometer, HC-SR04

ultrasonic sensor, and limit switch. Below is step-by-

step instructions to implement this system:

The aim of this project is to check road condition

based on surface roughness, potholes and bumps. The

limit switch can also serve to discern specific trigger

events (for example, did the wheel hit something, or

when the system is moving), while the accelerometer

will pick up vibration and bumps of the ground, and

the ultrasonic could be used to measure the distance

to the road surface.

• Inertial sensors: not least accelerometers form

the foundation

of the cheapest road surface

estimation systems. This project aims to

develop a road classification system that can

assess the infrastructure

in real time.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

210

• New low-cost AI based on on-device and GPS

based on real

time location and identification

of problem areas based on MEMS signals

• AI embedded

architectures consider

constraints that an Internet of Things (IoT)

device must operate within memory and

embedded systems execution limits.

5.1 Data Processing

1. Accelerometer Data: To identify sudden

changes that point to bumps, potholes, or

uneven spots, use the accelerometer data,

particularly along the Z-axis.

2. Ultrasonic Distance: Track variations in

distance, which might reveal imperfections

in the road surface.

3. Limit Switch: To initiate particular actions,

such as recording data only when the system

is in motion or a wheel strikes an

obstruction, use the limit switch.

5.2 Data Analysis

As the machine moves across various road surfaces,

gather and record data over time. Examine trends in

ultrasonic and accelerometer data to identify

particular kinds of anomalies in the road. For a more

sophisticated system, think about wirelessly sending

data for distant monitoring and analysis as well as

recording it on a PC for surface study.

5.3 Fine-Tuning and Calibration

Adapt thresholds to the driving conditions and the

accelerometer's sensitivity. To concentrate on more

important occurrences that probably point to poor

road condition, filter away little vibrations.

6 HARDWARE

6.1 Power Supplies

An apparatus or system that provides electrical or

other forms of energy to an output load or collection

of loads is called a power supply (sometimes referred

to as a power supply unit or PSU). Electrical energy

sources are most frequently referred to by this phrase,

followed by mechanical ones and, seldom, others.

When working with digital circuits, this circuit's tiny

+5V power supply is helpful. Any supermarket or

electronics store will sell small, low-cost wall

transformers with changeable output voltage.

Although such transformers are readily accessible,

they often have extremely poor voltage regulation,

which limits their usefulness for digital circuit

experimenters until a better regulation can be found.

The solution to the issue is the circuit that follows.

6.2 Transformer

A transformer is an apparatus that transfers electrical

energy between two (or more) circuits through

inductively coupled wires. The changing current in

main circuit creates a changing magnetic field, which

also induces an alternating voltage in secondary

circuit. Current can be flowed in the transformer

which allows energy to transfer from one circuit to

another or a load applied to the secondary circuit.

One factor (n) which is (or rather should be) equal to

the number of wires turns in each of them, here is

employed to dilute the secondary induced voltage

(VS) from the primary VP: Therefore, a transformer

can step up an alternating voltage if NS > NP or step

it down if NS < NP by selection of number of turns

carefully. A key application of transformers is

lowering the current before transmitting electrical

energy over long distances through wires. Due to

their resistance, the vast majority of wires dissipate

electrical energy as a function of the square of the

amount of current passing through them.

Transformers enable long-distance electricity

transmission, where electrical power is converted to

a high-voltage, and therefore low-current, state for

transmission and then back again. As a result,

transformers have spread through the electrical

supply industry covertly implanting the general

concept of generating far away from demand. By the

time it finally reaches the user, nearly all of the

electrical power in the World has passed through a

series of transformers. Some enormous units are

capable of transmitting 99.75 percent of their input

power to their output, making transformers some of

the most efficient electrical “machines.”

Transformers vary in size from tiny coupling

transformers tucked inside backstage microphones to

giant gigavolt-ampere rated pieces of gear used to

connect sections of national power grids. There are

many different designs for various functions, both

home and industry, but they all operate on the same

basic principles:

Analysis of Real Time Road Surface and Acoustic Data Processing for Minimizing the Accident Rate Using Feed Forward Neural Network

211

7 SOFTWARE

7.1 Arduino

Arduino is an open-source electronics platform based

on simple software and hardware. Arduino boards

receive an input, the light on a sensor, a finger over a

button, a tweet, and they create an output in the form

of blinking an LED, activating a motor and posting

something online. The microcontroller of your board

can be commanded to perform some actions by

feeding instructions you may refer to as a Program.

This is done using the Arduino software (IDE) shown

in figure 4, which is built on Processing, and the

Arduino programming language, which is built on

Wiring. Whether for more simple everyday items or

more complex scientific devices, Arduino is still the

brain behind millions of projects every year. This

open-source platform has borne from it a global hive

of makers (students, hobbyists, artists, programmers

and professionals) sharing knowledge creating an

insane reservoir of useful information well suited for

beginners and experts alike. Arduino was invented as

a simple prototyping tool in the Ivrea Interaction

Design Institute so students without any electronics

or programming background could prototype quickly.

The moment that the Arduino board became widely

adopted, it began to be evolved to tackle new needs

and challenges, from basic 8-bit boards to choices for

wearables to those for embedded environments to 3D

printing and Internet of Things/IoT applications. All

Arduino boards are fully open-source, so you can

solder them together yourself and then modify

according to your requirements. The program is

public and continually builds through donations

from people all over the world.

Figure 4: Arduino IDE.

7.2 Arduino Software

The Arduino Integrated Development Environment

(IDE), also known as the Arduino Software, includes

a text editor for writing code, a message box, a text

terminal, a toolbar with buttons for commonly used

tasks, and several menus. It connects to the Arduino

hardware to upload programs and interact with the

hardware.

8 RESULT

Figure 5: Real-time road hazard detection and risk mapping

system.

Predicted Risk Level: "High Risk" (based on data

from a road segment exhibiting uneven auditory

signals and potholes that may indicate tire sliding).

Notice: "Warning: Hazardous Road conditions

detected, slow down immediately."

The report on accuracy states, "Model accuracy: 92%,

Precision: 89%, Recall: 91%."

Visual Representation: Figure 5 shows the city map

with areas coloured according to the neural network's

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

212

rating (green for safe, yellow for intermediate risk,

and red for high danger).

9 CONCLUSIONS

In this study, we have introduced an AI-based real-

time road surface monitoring application. The

program is intended to be used on a microcontroller

board that has a microphone installed to record noises

within a tire's cavity. According to preliminary test

findings, the gadget can identify the asphalt's quality

on the test set with 91% accuracy. This illustrates how

well the suggested Tiny architecture works for this

use case and how the Mel-inspired spectrogram can

be used as an input to identify the condition of the

road. The suggested strategy makes use of cutting-

edge methods. Actually, the implementation of AI

systems on embedded systems is a state-of-the-art

technology that is the subject of a lot of ongoing

study.

REFERENCES

"Deep neural network model for speaker identification,"

written by F. Ye and colleagues, Appl. Sci., vol. 11, no.

11, p. 5172, 2021.

"Pothole recognition based on texture measures and support

vector machine," by J. Lin et al., in Proceedings of the

IEEE International Conference on Pictures, 2010, pp.

3653-3656.

"Three-dimensional pavement pothole detection based on

optical imaging principle," IEEE Trans. Intell. Transp.

Syst., vol. 12, no. 2, pp. 291-301, June 2011. H.

Youquan et associates.

"Wi-Fi-based architecture for pothole detection and

warning system," by S. S. Rode et al., in Proceedings of

the IEEE International Conference on Networking,

Sensing, and Control, 2009, pp. 672-677

The article "Pavement Distress Detection and Analysis for

Pothole Repair," by I. Moazzam and colleagues, was

published in 2013.

Analysis of Real Time Road Surface and Acoustic Data Processing for Minimizing the Accident Rate Using Feed Forward Neural Network

213