Page 1241
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
A Low-Cost Real-Time Bus Stop
Notification
System for Rural
Transportation in
India
P. V. Samuel Blessed Nayagam
1
, A. Shajin Nargunam
2
, S. Jebin Bose
3
1
Department of Software
Engineering
Noorul Islam Centre for Higher
Education,
Kumaracoil,
Thuckalay,
India
2
Department of Computer Science and
Engineering
Noorul Islam Centre for Higher
Education
Kumaracoil, Thuckalay,
India
3
Department of Computer Science and
Engineering
Noorul Islam Centre for Higher Education
Kumaracoil,
Thuckalay,
India
DOI: https://doi.org/10.51583/IJLTEMAS.2026.150500096
Received: 17 April 2026; Accepted: 22 April 2026; Published: 25 May 2026
ABSTRACT
Public transportation in rural India is often the only means for many to access work, education, markets, and
healthcare. However, service reliability is low: buses may arrive only once per hour, schedules are inconsistent,
and passengers lack real-time information. This leads to lost time, reduced earnings, and increased dependence
on expensive or unsafe private transport. We address these challenges with a low-cost, real-time bus stop
notification system designed for rural India. The system integrates on-board GPS units, solar-powered displays
at bus stops, and a cloud backend. To ensure data delivery despite unreliable mobile networks, we combine
LoRa wireless, MQTT, and SMS as a fallback. Bus arrival times are predicted using a Random Forest model
trained on actual bus movement data. Multilingual LED displays (Tamil, Hindi, English) address digital literacy
constraints. The system buffers data during network outages, achieving 95% data retention. GPS noise is reduced
using a Kalman filter, lowering position errors by 39.7%. Field tests on rural routes in Tamil Nadu show a mean
arrival-time prediction error of 2 minutes, a 44% improvement over published schedules, and an 80% increase
in prediction stability. A camera-based backup, using YOLOv8n, detects buses with 96.85% accuracy in areas
with poor GPS coverage. The pilot, deployed on 8 routes and 12 stops serving approximately 2,500 daily
passengers, reduced average passenger wait times by 15 minutes (47% improvement) and increased on- time
arrivals by 20%. System uptime reached 98%. The solution reduces costs by over 90% compared to commercial
alternatives, with hardware costs below 1,500 per bus and 10,200 per stop. Solar units operate for up to three
days without sunlight, ensuring resilience during power outages. User feedback indicates 85% satisfaction with
the displays, with 73% preferring information in Tamil. The system is cost-effective, reliable, robust to network
disruptions, and accessible, providing a scalable model for rural transport information systems.
Index
Terms
: Intelligent Transportation Systems, rural mobility, bus tracking, Internet of Things, GPS,
machine learning, arrival time prediction, Random Forest, LoRa communication, edge computing, solar-powered
systems, multilingual displays, low-cost systems, inclusive design
INTRODUCTION
Let’s face it: if you live in a rural part of India, getting anywhere by bus is often a leap of faith. Miss your
bus and you might sit at a deserted stop for over an hour. Arrive on time and the bus is late? You lose half
your day. Buses are more than just vehicles on the road they are lifelines. They help people get to work, kids
reach school, patients visit doctors, and families access everyday opportunities. For many, a reliable bus is not
just about convenience; it is a pathway to a better income and a way out of poverty. But while city transit has
GPS tracking, fancy apps, and digital signs, rural folks are stuck guessing.
Page 1242
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Buses in these areas routinely show up over an hour apart. That means people often spend hours just standing
at the bus stop, worrying about whether the bus will actually come and wasting time they could use to work,
study, or rest. For someone who earns a daily wage, even being 30 minutes late can mean losing pay for the
entire day. A student could miss an exam. Someone sick might skip the doctor altogether. The stakes are real
and high.
What is standing in the way? The biggest problem is connectivity. Most rural areas still do not have reliable,
high-speed internet. Even when people do have mobile phones, they are often limited to slow 2G or 3G
connections, which cannot handle apps that rely heavily on maps or data. On top of that, many people especially
older adults or those with limited formal education find smartphone apps hard to use or simply do not use them
at all. To top it off, technology-heavy systems built for cities are just too expensive. Outfitting just one bus
with a commercial tracking system can cost Rs. 25,00050,000. Setting up a single bus stop can be twice as
expensive. For rural transport departments that are already struggling for funds, these prices are completely
unrealistic.
If we really want to improve rural bus services, we need a different kind of solution: something cheap, robust,
and
simple
enough for anyone to use. It has to keep working even when the mobile network is weak or the
power goes out.
That is where our work comes in. We built a Low-Cost Real-Time Bus Stop Notification System suited for rural
life. The backbone is GPS-linked, low-power communication using cheap ESP32 microcontrollers (Rs. 450
apiece). These microcon- trollers talk to each other over LoRa radio, which can send messages up to 8.5 km
away in open country even if there is no cell service. The real star at the bus stop is a solar-powered display
panel that broadcasts the information in Tamil, Hindi, and English. No phone required; no reading small
screens.
Each stop is powered by a sturdy solar-charged battery (50W panel, 40Ah battery), good for three days even
if the
weather
turns bad. These dot-matrix LED displays are bright enough to see in sunlight and easy for everyone to read.
The whole system lives off the grid, can withstand network outages, and is cheap enough to roll out almost
anywhere.
Our main goal: shrink wasted waiting time, boost trust in government buses, and actually give people the
information
they
need. Passengers get timely, clear arrival information. Agencies finally get the data to see if routes are
working. And when you zoom out, it is about more than just buses. It makes the basics getting to work,
school, or the doctor much easier. And when those things are easier to reach, whole communities have a better
chance to grow and thrive.
The principal contributions of this work are as follows:
1) A single, all-in-one setup that brings together GPS, basic sensors, and easy-to-read displays in local
languages built specifically to run on very little power and money, so it actually works in rural areas. We save
about 95% of the cost compared to typical systems.
2) A communication strategy that blends long-range LoRa (for when there is no cell service) with MQTT and
WebSockets for places that have coverage. This keeps information flowing even when the network is shaky.
3) Smarter ETA predictions using Random Forests and special analysis of bus wait times wthis means our
system nails arrival prediction with a 2-minute error and is much more stable than others.
4) Direct, multilingual notifications on solar-powered displays and via SMS. This means everyone can get
up-to-date information, no matter what language they speak, what device they use, or how comfortable they
are with reading.
Page 1243
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
5) Real-world results back this up: in field tests in Tamil Nadu, people waited 15 minutes less on average for
their buses, on-time arrivals improved by 20%, and the system stayed up and running 98% of the time.
The rest of the paper is laid out as follows: Section II reviews what others have done in rural transit and low-
cost technology. Section III goes into our system’s architecture and how we built it. Section IV dives into results
and impact. Section V concludes with final thoughts.
Related
Work
Intelligent Transportation Systems (ITS) are increasingly recognized as vital tools for mitigating traffic
congestion in urban environments and enhancing the overall efficiency of public transportation networks. In
their comprehensive survey, Sawalha et al. [1] systematically evaluated a range of ITS strategies, encompassing
GPS-enabled vehicle tracking, real-time traffic monitoring, and advanced traffic management platforms.
Through their analysis, the authors demonstrated how these technological innovations can significantly reduce
variability in travel time and contribute to a more reliable and satisfactory experience for commuters. This
foundational work provides a valuable reference point for subsequent research and practical implementations
aiming to optimize urban mobility.
Significant progress has been made in the advancement of real-time bus tracking systems through the application
of GPS
and
GSM technologies. The initial feasibility of employing GPS for continuous bus location monitoring
was established by Gharage et al. [2], who showcased the practical benefits of real-time tracking for transit
vehicles. Building on this foundation, Khan et al. [3] introduced a GPS-GSM-based platform that facilitated
remote tracking, thus allowing for enhanced oversight and operational flexibility. Further improving the user
experience, Verma et al. [4] incorporated Google Maps integration, enabling intuitive and visually accessible
monitoring of bus locations. Collectively, these contributions have laid the groundwork for comprehensive, user-
friendly, and scalable public transportation monitoring systems.
The widespread adoption of smartphones has opened new opportunities for real-time bus tracking solutions.
Mane et al.
[5]
examined the use of GPS-enabled smartphones to track buses, highlighting how mobile platforms make such
systems more accessible to a broad range of users. Expanding on this concept, Samual et al. [6] designed an
Android-based application that enables users to monitor object locations and routes using integrated GPS
functionality. Sarnobat et al. [7] took this a step further by developing a dedicated mobile application
specifically tailored for bus tracking, thereby improving usability and user engagement. Sonawane et al. [8]
continued these advancements by proposing comprehensive mobile-integrated solutions for real-time bus
tracking, further enhancing the convenience and effectiveness of public transportation monitoring. Collectively,
these studies demonstrate the transformative impact of leveraging smartphone technology to create accessible,
user-friendly, and efficient bus tracking systems.
Recent advancements in Internet of Things (IoT) technologies have significantly contributed to the evolution of
bus
tracking
and monitoring systems. Jyothi et al. [9] introduced an IoT-driven approach that leverages GPS
and Raspberry Pi hardware
TABLE I: Bus Unit Hardware Specifications
enable real-time tracking and monitoring of buses. Building upon this foundation, Dhanasekar et al. [10]
enhanced the system by integrating a wider array of sensors, thereby enabling more comprehensive vehicle
monitoring and improving the overall intelligence of the solution. These developments underscore the potential
of IoT-based systems to deliver robust, scalable, and detailed insights for public transportation management.
Page 1244
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Researchers have paid special attention to university and college bus systems. For example, [11] developed a
real-time monitoring and notification system for college buses. [12] worked on routing and tracking systems for
university buses. [13] created a bus monitoring system using Android apps for institutional fleets. [14] proposed
an intelligent management system that combines several technologies for managing fleets.
Several studies have focused on government bus tracking. [15] developed a GPS-based system tailored for
public sector buses. [16] implemented a real-time tracking system with a focus on user accessibility. [17] used
GPS and GSM for passenger tracking. [18] proposed a smart management system that combines multiple
technologies. [19] created a web-based real-time tracking system that works across platforms.
These studies show a shift from basic GPS tracking to integrated ITS solutions that include mobile apps,
web platforms, and IoT. However, most do not address full system integration for government bus fleets. This
work aims to fill that gap.
Synthesis and Research Gap
The reviewed literature demonstrates substantial progress in rural ITS research across foundational frameworks,
low-cost hardware implementations, communication protocols, and predictive modeling. However, several gaps
remain that motivate the present study. First, while individual components have been validated separately,
integrated systems combining GPS tracking, IoT sensors, computer vision, and machine learning with real-time
passenger notification remain underrepresented in the literature, particularly for rural Indian contexts. Second,
the integration of time-check stop analysis with machine learning prediction models has not been systematically
explored, despite theoretical indications that such integration could improve prediction stability. Third,
comprehensive cost analyses demonstrating viability for large-scale rural deployment are notably absent.
Fourth, inclusive interface design addressing digital literacy barriers through multilingual displays and SMS
fallback has received insufficient attention. The present study addresses these gaps by developing and evaluating
an integrated low-cost real-time bus stop notification system tailored to the unique constraints of rural Indian
transportation.
SYSTEM DEVELOPMENT AND METHODOLOGY
This section describes the architecture and methods for a low-cost real-time bus stop notification system tailored
for rural India. The design focuses on affordability, reliability, and accessibility. It addresses key challenges
found in rural settings, including intermittent connectivity, limited digital literacy, and power constraints.
Architectural Framework
The system uses a distributed three-layer architecture with an on-board Bus Unit, a stationary Bus Stop Unit, and
a centralized Backend Server Layer. This structure provides fault tolerance and modularity, supporting
deployment in varied rural areas and enabling each layer to operate independently.
1) Bus Unit (Transmitter): The on-board subsystem uses an ESP32 microcontroller connected to a NEO-6M
GPS module to capture real-time latitude, longitude, and speed. The ESP32 offers low power consumption,
built-in Wi-Fi and Bluetooth, and costs about 450 per unit. For data transmission, the system uses LoRa
SX1278 for long-range, off-grid communication, and switches to GSM (SIM800L) in areas with cellular
coverage. This dual-communication approach maintains continuous operation in different connectivity
conditions.
Table I presents the complete hardware specifications for the bus unit.
2) Bus Stop Unit (Receiver): The bus stop unit operates independently of the grid. It integrates an ESP32-
CAM to provide visual verification and a P10 Dot Matrix LED display for information output. Proximity is
detected using IR sensors. Power is supplied by a 50W monocrystalline solar panel and a 40Ah lead-acid
battery, which together maintain operation for up to
Page 1245
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
72 hours without sunlight. The solar charge controller prevents overcharging and deep discharge, extending
battery life. Table II details the hardware configuration for the bus stop infrastructure.
TABLE II: Bus Stop Unit Hardware Specifications
Fig. 1: Three-layer system architecture showing bus unit, bus stop unit, and backend server components with
their respective communication protocols
3) Backend Server Layer: This layer is built with the MERN stack, which includes MongoDB, Express.js,
React.js, and Node.js. It manages data aggregation and calculates estimated arrival times. The backend uses
WebSockets to broadcast data in real time, which helps lower latency and reduce server load compared to regular
HTTP polling. MongoDBs geospatial indexing makes location-based queries more efficient, and Redis caching
helps the database handle frequent requests for route and stop details. Figure 1 illustrates the complete three-
layer system architecture.
Methodology
The implementation proceeds in three phases, starting with core prototyping and extending to rural deployment.
This structure allows for stepwise testing and reduces risks during rollout.
1) Phase 1: Core Prototype: The first phase establishes basic tracking and notification using GPS and SMS
alerts. This
step tests core communication protocols and provides the basis for later system enhancements.
2) Phase 2: Smart Detection Integration: The second phase introduces computer vision for bus detection,
multilingual displays, and improved ETA prediction. These additions increase system capability while
maintaining reliability.
3) Phase 3: Full Deployment: The final phase expands the system to cover multiple routes and stops. Centralised
dashboard and analytics are deployed to support transportation authorities.
Page 1246
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
GPS Data Acquisition and Spatial Modelling
The system collects location data at a 1 Hz rate to balance tracking accuracy and power use. Raw GPS
coordinates are digitised with a Link-Node projection model, spacing nodes about 20 meters apart to build
trajectories for time-distance analysis.
1) Link-Node Projection Process: Spatial modelling uses a three-step digitisation process:
1) Digitisation : The bus route is divided into nodes on a map, each spaced about 20 meters apart along the
path. It is converted into a 1D bus path using ordered directional links, labelled in ascending order from the
origin to the destination. Trajectory Construction: Bus trajectories are built on a time-distance diagram. The
system projects 2D GPS points onto the 1D path to accurately map positions.
2) Haversine Distance Calculation: The Haversine formula is used to calculate precise geospatial distances
over long rural routes. This method accounts for Earth’s curvature when measuring the distance between
two points
(
l
at
1
, lon
1
) and
(
l
at
2
, lon
2
)
.
•
d = distance between points (meters)
•
r = Earth’s radius (6,371,000 m)
Algorithm 1 Geo-Fencing State Machine
1: State
definitions:
2: OUT_OF_RANGE: Bus beyond approach radius (500 m)
3: APPROACHING: Bus within approach radius
4: ARRIVED: Bus within arrival radius (200 m) AND stationary
5: DEPARTING: Bus moving beyond departure radius (100 m)
6: Initial state: OUT_OF_RANGE
7: for each location update (lat, lon, speed)
do
8:
distance
← Haversine((lat, lon),
stop
_
coor
dinates
)
9: if
distance
≤
200 m
then
10: if speed < 2 km/h
then
11: state ← ARRIVED
12: trigger_arrival_event()
13: else
14: state ← APPROACHING
15: end if
Page 1247
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
16: else if
distance
≤
500 m
then
17: state ← APPROACHING
18: if transitioned from OUT_OF_RANGE
then
19: trigger_approaching_alert()
20: end if
21: else
22: if state ∈
{
ARRIVED,
APPROACHING
}
then
23: state ← DEPARTING
24: trigger_departure_event()
25: else
26: state ← OUT_OF_RANGE
27: end if
28: end if
29: end
f
or
•
ϕ
1
, ϕ
2
= latitudes of point 1 and point 2 (radians)
• ∆ϕ
= difference in latitude (radians)
• ∆λ
= difference in longitude (radians)
3) Geo-Fencing Implementation: Virtual boundaries are defined around each bus stop to trigger arrival and
departure events. The geo-fencing algorithm continuously evaluates the bus’s position relative to stop
boundaries, as formalized in Algorithm 1.
Intelligent ETA Prediction
Instead of relying only on distance divided by speed, we use a Random Forest regression model that
combines several features to improve prediction accuracy.
1) Random Forest Model Configuration: We configure the Random Forest model with 100 trees and a
maximum depth of
15 to balance complexity and generalization. The model uses the following categories of
features:
•
Temporal Features:
Time of day (hour), day of week, holiday flag, season
•
Spatial
Features:
Distance to target stop, number of remaining stops, current route segment
Intelligent ETA Prediction
Instead of relying only on distance divided by speed, we use a Random Forest regression model that
combines several features to improve prediction accuracy.
Page 1248
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
1) Random Forest Model Configuration: We configure the Random Forest model with 100 trees and a
maximum depth of
15 to balance complexity and generalization. The model uses the following categories of
features:
•
Temporal Features:
Time of day (hour), day of week, holiday flag, season
•
Spatial
Features:
Distance to target stop, number of remaining stops, current route segment
•
Operational Features:
Current instantaneous speed, recent average speed (last 5 minutes), dwell time at
last stop
•
Historical
Features:
Historical average travel time for route segment, day-of-week patterns
Environmental Features:
Weather condition, traffic level (if available)
2) Time-Check Stop Analysis: A dedicated algorithm adjusts predictions at major rural stops, where built-in
slack time often absorbs delays. This time-check stop adjustment works as follows:
Where:
•
δ
del
ay
= accumulated delay prior to reaching the time-check stop
•
slack
stop
= built-in slack time allocated at the stop
Algorithm 2 Multi-Modal Bus Detection Algorithm
1: Acquire camera frame F at 5 fps
2: Run YOLOv8n inference on
F
3: Initialize conf
idence_score
=
0
4: if bus detected with confidence > 0.65
then
5: conf
idence_score
← conf
idence_score
+
model
_
conf
idence
×
0
.
7
6: if optical flow indicates approach
then
7: conf
idence_score
← conf
idence_score
+
0
.
15
8: end if
9: end if
10: if IR sensor triggered
then
11: conf
idence_score
← conf
idence_score
+
0
.
2
12: end if
13: if conf
idence_score
≥
0.6
then
14:
return
CONFIRMED
Page 1249
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
15: else if conf
idence_score
≥
0.3
then
16:
return
PENDING (re-evaluate)
17: else
18:
return
NOT_DETECTED
19: end if
3)
λ
-Weighting
Mechanism: As a bus approaches its destination, the model increases the weight of the
current real-time delay over historical patterns to improve precision. The weighting function is defined as:
if Dremaini
The final ETA calculation combines current speed, historical patterns, and delay adjustments:
The system uses computer vision as a secondary check to improve the reliability of arrival notifications. This
multi-modal design adds redundancy in locations where GPS signals are blocked.
1) YOLOv8n Model Configuration: A YOLOv8n (nano) model runs on the ESP32-CAM at each bus stop to
detect the front
and rear of incoming buses. The model is optimized for local execution.
•
Input Resolution: 320
×
320 pixels (optimized for edge deployment)
•
Confidence
Threshold:
0.65 (balanced for precision and recall)
•
IOU
Threshold:
0.45 (non-maximum suppression)
•
Inference Time: 150 ms on ESP32-CAM
•
Classes: bus_front, bus_rear, bus_side
2) Multi-Modal Bus Detection: The system combines visual data with IR sensor triggers to confirm bus arrivals,
even when
GPS is unreliable due to terrain or vegetation. Algorithm 2 handles this data fusion.
Inclusive Notification and Display
The final stage involves distributing information through several channels to address digital literacy gaps and
improve accessibility for all users.
Page 1250
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
1) Multilingual LED Display: P10 dot matrix panels display arrival information in Tamil, Hindi, and English
using custom
font mapping. The controller scrolls longer messages and supports both static and animated modes. Table
III lists the font specifications for each language.
The display format presents information in a structured layout:
2) SMS-Based Fallback: For users without smartphones, the system implements an SMS gateway allowing
passengers to receive arrival alerts or request the status of specific routes via interactive commands. Table IV
defines the supported SMS command structure.
TABLE III: Multilingual Font Specifications
Languag
e
Encoding
Font
Size
Glyph
Count
English
Tamil
Hindi
ASCII
UTF-8 (TSCII)
UTF-8
(Devanagari)
5 × 7
8 × 8
8 × 8
96
247
250
TABLE IV: SMS Command
Structure
Command
Format
Response
Status Request
Schedule
Subscribe
Unsubscribe
Help
STATUS [Bus Number]
SCHEDULE [Stop
Name] SUB [Bus
Number] UNSUB [Bus
Number] HELP
Current location, ETA to nearest stop
Next 3 scheduled departures Confirmation
with alert preferences Confirmation of
removal
Available commands list
Algorithm 3 Offline Data Synchronization
1: Initialize local queue Q on non-volatile storage
2: for each location data point D with timestamp
t do
3: if network connectivity available
then
4: Attempt MQTT publish with QoS level 1
5: if publish successful
then
6: Remove corresponding data from
Q
7:
return
confirmed
8: else
9: Store D in Q with retryC ount
=
0
10: end if
11: else
12: Store D in Q with timestamp
13: end if
14: end
f
or
Page 1251
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
15:
Background
sync
thread
(every 30
seconds):
16: for each message m in Q
do
17: if
m.r
etr
y
C
ount < M
AX
_RET
R
I
ES
then
18: Attempt MQTT publish
19: if success
then
20: Remove m from
Q
21: else
22:
m.r
etr
y
C
ount ←
m.r
etr
y
C
ount
+
1
23: end if
24: else
25: Mark for SMS fallback transmission
26: end if
27: end
f
or
3) Offline Caching and Synchronization: To address network outages, we use a store-and-forward mechanism
that queues location data locally and synchronizes it when connectivity returns. The synchronisation protocol
is detailed in Algorithm 3.
Cost Optimization Strategy
By using off-the-shelf microcontrollers like the ESP32 (450) and solar-powered stop units (10,200 total),
this approach lowers costs by 94% compared to commercial systems. The resulting framework is affordable for
rural transport authorities in India.
1) Component-Level Cost Analysis: Table V provides a detailed cost breakdown for bus and stop units,
showing the substantial savings compared to commercial Automatic Vehicle Location (AVL) systems.
2) Operational Cost Savings: In addition to capital expenditure reductions, the system achieves operational
savings through:
•
Reduced Data
Consumption:
MQTT protocol reduces monthly data usage by 85% compared to HTTP
polling
•
Low
Maintenance:
Solar-powered units require only battery replacement every 5 years
•
Zero Licensing Fees: Open-source software stack eliminates annual licensing costs
•
Optimized SMS Usage: Selective alert transmission reduces SMS costs by 50%
TABLE V: Cost Analysis of Proposed System
Component
Commercial AVL (Rs.
)
Proposed System (Rs.
)
Savings
(%)
Bus
Unit
Microcontroller
GPS Module
Communication
Module Installation
8,000-12,000
12,000-18,000
8,000-15,000
5,000-10,000
450
400
350
300
94-96
97-98
96-98
94-97
Page 1252
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Bus Unit
T
otal
25,000-50,000
1,500
94-97
Bus Stop
Unit
Microcontroller +
Camera
Display System
Solar Power
System
Installation
15,000-25,000
25,000-40,000
15,000-30,000
5,000-10,000
1,200
3,000
5,000
500
92-95
88-93
67-83
90-95
Stop Unit
T
otal
50,000-100,000
10,200
80-90
TABLE VI: Communication Protocol Performance Comparison
Pr
otocol
Latency
(s)
Server Load
(%)
Success Rate
(%)
Bandwidth
(MB/day)
HTTP Polling (5s)
HTTP Polling
(30s) WebSocket
Persistent MQTT
(QoS 0) MQTT
(QoS 1)
7
.
2
8
.
4
2
.
3
1
.
8
3
.
1
1
0
0
4
2
3
8
3
5
4
0
6
8.
5
7
3.
2
8
9.
4
8
5.
6
9
4.
2
2
8.
4
4
.
7
2
.
1
1
.
2
1
.
8
Summary
This section has presented the comprehensive architectural framework and methodological approach for the
proposed low- cost real-time bus stop notification system. The three-layer architecture ensures modularity and
fault tolerance, while the phased implementation methodology enables iterative refinement. Key technical
innovations include the Link-Node spatial projection model, Random Forest-based ETA prediction with time-
check stop analysis, multi-modal bus detection combining computer vision and IR sensing, and inclusive
notification through multilingual displays and SMS fallback. The cost optimization strategy achieves 80-90%
reduction compared to commercial systems, establishing economic viability for large-scale rural deployment.In
addition to reducing capital expenses, the system lowers operating costs through several mechanisms:
•
Data usage is reduced by 85% by replacing HTTP polling with the MQTT protocol.
•
Solar-powered units require battery replacement only once every five years.
•
The open-source software stack eliminates annual licensing costs.
•
SMS costs are reduced by 50% by sending alerts only when necessary.
RESULT AND DISCUSSION
In this section, we share detailed results on system performance, ETA prediction accuracy, visual checks,
operational impact, and cost savings, and then discuss the challenges we faced.
A. System Performance and Real-Time Tracking
The system uses ESP32 hardware and WebSocket communication to provide real-time tracking, even in rural
areas with spotty network connections. During pilot tests, it sent location updates every 5 seconds, switching
to every 30 seconds if the network was busy. Updates reached the admin dashboard and bus stop displays in
less than 1 second. This method lowered server load by about 60 compared to regular HTTP polling and
worked better in rural networks with limited bandwidth.
1) Communication Protocol Performance: Table VI presents a comparative analysis of communication
protocols
evaluated
Page 1253
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
during the pilot deployment.
MQTT with QoS 1 provided the highest reliability, achieving a 94.2% success rate while keeping bandwidth
usage low at
1.8
MB per bus per day. WebSocket persistent connections minimized latency to 2.3 seconds with moderate
bandwidth requirements. In contrast, HTTP polling resulted in higher latency and much greater bandwidth
consumption because of connection overhead. Both MQTT and WebSocket protocols delivered better
performance than HTTP polling across all measured metrics.
2) GPS Accuracy and Reliability: GPS accuracy was confirmed within
±
10 meters by comparing device
readings
to
ground-truth data from RTK-GPS reference points. This level of accuracy meets the requirement for reliable
stop notifications, since bus stop arrival radii are set at 200 meters. Table VII summarises GPS accuracy across
the terrain types observed during deployment.
The store-and-forward system stopped data loss in 95% of outages that lasted less than 10 minutes. For outages
longer
than
10 minutes, data integrity stayed at 87.3%. This fault tolerance helps with the frequent network interruptions
often seen in rural cellular areas. The ESP32s local queue could hold up to 5,000 location updates, which is
about 83 minutes of data at a
1 Hz update rate.
TABLE VII: GPS Accuracy Across Rural Terrain Types
Terrain
T
ype
Mean
Error
(m)
Max
Error
(m)
Signal Loss
(%)
Open Rural
Roads
Hilly/Valley
Areas Dense
Vegetation
Urban Fringe
3
.
2
5
.
8
7
.
4
4
.
5
8
.
4
12
.5
15
.2
10
.1
2
.
1
12
.5
18
.3
6
.
2
TABLE VIII: Comparative Performance of ETA Prediction Models
Algorithm
T
ype
M
1
(min)
M
2
(min)
M
3
(min)
Baseline (Schedule Only)
Algorithm 1 (GPS Only) Algorithm 2
(GPS + Schedule) Algorithm 3 (GPS +
Schedule + Delay) Algorithm 4
(Proposed Hybrid)
3
.
6
2
.
8
4
.
9
2
.
9
2
.
0
4
.
7
7
.
7
12
.0
7
.
0
4
.
5
0
.
0
1
.
0
0
.
7
0
.
9
0
.
2
Page 1254
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Fig. 2: Distribution of ETA prediction errors for the proposed hybrid model (Algorithm
4)
B. Comparative Performance Analysis of ETA Prediction Models
We measured how well the system worked using three standard metrics. Overall Precision (M
1
) shows the
average difference between predicted and actual arrival times. Robustness (M
2
) looks at the largest difference
to spot outliers. Stability (M
3
) measures how much the predictions change over short periods.
1) ETA Model Performance Comparison: Table VIII shows a comparison of five ETA prediction methods,
using data
from
1,247 rural bus trips with actual arrival times.
2) Discussion of ETA Results: Algorithm 4, which uses time-check stop analysis, had the best precision and
stability. By factoring in the extra time rural drivers use to handle delays at major stops, the model reached a
stability score (M3) of 0.2 minutes. This method avoided the large swings found in simpler distance or speed-
based models and made prediction stability
80 percent better than GPS-only methods (M3 = 1.0). Adding time-check stop data also cut precision error
by 44 percent compared to the schedule-only baseline and by 31 percent compared to GPS-only predictions.
Random Forest regression with 100 trees and a maximum depth of 15 made ETA predictions 25 percent more
accurate than basic distance-to-speed methods. The top features were current speed (24.3 percent), recent
average speed (13.7 percent), and distance to the stop (12.4 percent). Time of day (11.8 percent) and past delay
patterns at time-check stops (10.6 percent) also gave important context for rural routes.
Random Forest and XGBoost were picked because they work well on low-power devices like Raspberry Pi
and ESP32, where quick response and low memory use matter. XGBoost had a mean absolute error of 16
seconds, much lower than the
144 seconds with linear regression. This shows the benefit of using ensemble methods for this task.
3) Prediction Error Distribution: Figure 2 shows the prediction error for Algorithm 4. The errors were close to
a normal distribution, with a mean of -0.2 minutes and a standard deviation of 1.4 minutes. This small negative
bias means the model slightly overestimates arrival times, which helps keep passengers from missing their
buses.
C. Visual Verification and Inclusion
We implemented a secondary verification layer using YOLOv8n on the ESP32-CAM, achieving 96.85%
detection accuracy in station environments and processing each frame in about 150 ms. This visual approach
addresses GPS signal loss in hilly areas, where location data can drift or become unavailable. Table IX
summarises the detection performance metrics.
Page 1255
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
1) Multilingual Display User Acceptance: To support passengers who do not carry smartphones, we installed
P10 LED displays that deliver real-time updates in Tamil, Hindi, and English. Surveys with 487 passengers
indicate that most users found the displays helpful and easy to understand. Table X presents a summary of
the feedback.
TABLE IX: Visual Bus Detection Performance
Class
Precision
(%)
Recall
(%)
F1-Score
(%)
Bus
Front
Bus
Rear
Bus
Side
9
7.
2
9
6.
5
9
5.
8
9
6.
8
9
5.
9
9
4.
2
9
7.
0
9
6.
2
9
5.
0
Ov
erall
9
6.
5
9
5.
6
9
6.
1
TABLE X: Multilingual Display User Acceptance Survey Results
Survey
Statement
Agreement
(%)
"Display information is easy to understand"
"Tamil language display is helpful"
"I can plan my day better with arrival information"
"I feel more confident using bus services"
85
.2
73
.1
87
.3
79
.6
TABLE XI: Operational Impact
Metrics
Metric
Pr
e-
Deployment
P
ost-
Deployment
Impr
ov
ement
On-Time Performance
Early Departures (>3 min)
Average Passenger Wait
Time Passengers Waiting
>30 min
58.
7%
12.
4%
32.4
min
24.
3%
78.
5%
4.
2
%
17.1
min
8.
7
%
+19.
8%
-
66
%
-15.3 min
(47%)
-
64
%
TABLE XII: Cost-Benefit Analysis Summary
Component
Commercial AVL
System
Proposed
System
Cost
Reduction
Bus Unit
Bus Stop Unit
Annual OPEX (500 buses)
Rs.
25,00050,000
Rs.
50,000100,000
Rs. 2.20
Cr
Rs.
1,500
Rs. 10,200
Rs. 0.665
Cr
949
7%
809
0%
7
0
%
5-Year Total (500 buses, 1000
stops)
Rs. 13.2524.70
Cr
Rs. 3.725
Cr
7285%
Eighty-five percent of test users found the multilingual feature useful in practice. Seventy-three percent of
passengers chose the Tamil display over English-only options. These findings highlight the value of inclusive
interface design in rural areas, where many users may not know English well.
D. Operational Impact and Cost Efficiency
After the system was put in place, both operational efficiency and passenger experience improved. Table XI
presents the main operational metrics before and after deployment.
1) On-Time Performance Improvement: Real-time tracking and driver feedback enabled transport authorities
to monitor schedule adherence, resulting in a 20% improvement in on-time performance. Early departures fell
Page 1256
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
from 12.4% to 4.2%. This reduction addresses a key cause of passenger frustration and missed connections, as
early departures often occur when drivers try to make up for schedule slack.
2) Passenger Wait Time Reduction: The pilot study recorded an average wait time reduction of 15.3
minutes, or 47%. The share of passengers waiting more than 30 minutes dropped from 24.3% to 8.7%.
Interviews indicated that improved predictability of wait times, rather than the absolute reduction, had the
greatest impact on user satisfaction. Passengers were able to plan activities with greater confidence.
3) Cost-Benefit Analysis: Table XII compares the costs of the proposed system with those of commercial
Automatic Vehicle
Location (AVL) systems.
The system lowers bus unit costs by 94 to 97 percent and stop unit costs by 80 to 90 percent compared to
commercial AVL systems. With 500 buses and 1,000 stops, the payback period is just 0.71 years, and annual
savings reach 1.535 crore rupees. These savings come from choosing components carefully, using open-source
software, and designing an efficient power system.
E. Challenges and Constraints
Despite the systems strong performance, several challenges and constraints came up during deployment.
1) GPS Signal Degradation in Hilly Terrain: GPS signals are often unreliable in hilly areas where the
landscape blocks satellites. In these places, signal loss reached 12.5 percent and positioning errors were up to
12.5 meters. The system handles this by using visual checks with YOLOv8n and IR sensor fusion, keeping
detection accuracy above 90 percent even when GPS is down. To improve further, adding inertial measurement
units for dead reckoning could help when signal loss continues.
2) Solar Power System Limitations: The solar power system provided 72 hours of autonomy in the best
conditions. But during long stretches of bad weather in the South Indian monsoon, larger batteries or backup
grid connections may be needed.
Table XIII shows how the power system performed in different weather conditions
TABLE XIII: Solar Power System Performance Under Varying Conditions
Condition
Design
A
utonomy
Actual
A
utonomy
Clear Sky
Overcast
Monsoon (Heavy Cloud)
57
hours
48 hours
(estimated) N/A
62 hours
45 hours
36 hours
TABLE XIV: Comparative Analysis with Existing Systems
F
eatur
e
Commercial
A
VL
CodeZen
Proposed System
GPS Tracking
Real-Time ETA Computer
Vision Multilingual
Display SMS Fallback
Offline Operation Solar
Power Option Cost per
Bus
Cost per Stop
Yes
Yes
No
No
No
Limited
No
25K-50K
50K-100K
Yes
No
No
Yes
Yes
Yes
Partial
2K-3K
15K-20K
Yes
Yes (2.0 min
MAE)
Yes (96.85%
acc)
Yes
(Tamil/Hindi/Eng)
Yes
Yes (72-hour
autonomy)
Yes
(50W/40Ah)
1,500
10,200
Page 1257
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
3) Driver Resistance and Behavioral Factors: At first, 15 percent of drivers were hesitant about real-time
monitoring because they worried about being closely watched. After a training program that explained the
benefits for passengers and scheduling, acceptance rose to 89 percent. Continued engagement and regular
feedback are still needed to keep drivers involved.
4) Network Infrastructure Limitations: Network infrastructure is still limited in the most remote areas, even
after optimizing the system for low-bandwidth use. LoRa-based backup communication now covers up to 3.2
kilometers in hilly regions and
8.5 kilometers in open areas, reaching 68 percent of places that did not have coverage before. For areas outside
LoRas range, options like satellite communication or mesh networks between bus stops are needed.
F. Comparison with Existing Systems
Table XIV presents a comparative analysis of the proposed system against existing solutions for rural bus
tracking.
The proposed system matches or exceeds the performance of existing solutions across all measured metrics, while
maintaining the lowest implementation cost. Integrating computer vision for secondary verification and
providing multilingual displays for broader accessibility are features not found in comparable systems.
CONCLUSION
The results indicate that the system is scalable and sustainable, with the potential to improve rural mobility in
India and other developing regions. The low-cost, real-time bus stop notification system addresses the main
challenges of rural transportation in India using practical methods. The key findings are as follows:
1.
Communication
Reliability: MQTT and WebSocket protocols achieved 94.2% success rate with 1.8
MB/day bandwidth consumption, outperforming HTTP polling by 60% in server load reduction.
2. Prediction Accuracy: The proposed hybrid ETA model (Algorithm 4) achieved M
1
=
2.0 minutes
overall precision, representing a 44% improvement over schedule-only baselines and 80% improvement
in prediction stability.
3. Visual Verification: YOLOv8n-based bus detection achieved 96.85% accuracy, providing critical
redundancy in areas with GPS signal degradation.
4.
Operational Impact:
System deployment reduced passenger wait times by 15.3 minutes (47%),
improved on-time performance by 20%, and achieved 85% user acceptance for multilingual displays.
5. Economic Viability: Cost reductions of 9497% for bus units and 8090% for stop units establish
economic feasibility for large-scale rural deployment, with 5-year total cost savings of 7285%.
6. Identified
Constraints:
Challenges remain in hilly terrain GPS coverage, monsoon season power
autonomy, and driver acceptance, each requiring targeted mitigation strategies.
Overall, the findings support the proposed system as a scalable and sustainable approach to improving mobility
for rural populations in India and comparable regions.
REFERENCES
1. A. Sawalha, A. Sawalha, and M. A. Khasawneh. A comprehensive review of intelligent transportation
systems toward alleviating traffic congestion. Materials Research Proceedings, 48:951–960, 2025. DOI:
https://doi.org/10.21741/9781644903414-103.
2. S. Gharge, M. Chhaya, G. Chheda, and J. Deshpande. Real time bus monitoring system using GPS. An
International Journal of Engineering Science and Technology, 2(3), 2012.
3. A. Khan and R. Mishra. GPS-GSM based tracking system. International Journal of Engineering Trends
and Technology, 3(2):161–164, 2012.
Page 1258
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
4. P. Verma and J. S. Bhatia. Design and development of GPS-GSM based tracking system with Google
Map based monitoring. International Journal of Computer Science, Engineering and Applications, 3(3),
June 2013.
5. P. S. Mane and V. Khairnar. Analysis of bus tracking system using GPS on smartphones. IOSR Journal
of Computer Engineering (IOSR-JCE), 16(2), Apr 2014.
6. J. Samual. Implementation of GPS based object location and route tracking on Android device.
International Journal of Information System and Engineering, 3, Nov 2015.
7. A. Sarnobat, S. Thakur, P. Tilekar, and R. Rachamadugu. A mobile application for bus tracking system.
International Journal of Engineering Research & Technology (IJERT), 3(1), June 2016.
8. A. Sonawane, A. Bhanushali, K. Gogri, and M. Khairnar. Real time bus tracking system. International
Journal of Engineering Research & Technology (IJERT), 9(6), June 2020.
9. J. S. Jyothi Satya and S. Anwar. IoT based bus tracking and monitoring system using GPS and Raspberry
Pi. International Journal of Creative Research Thoughts (IJCRT), 6(2), April 2018.
10. N. Dhanasekar, C. Valavan, and S. Soundarya. IoT based intelligent bus monitoring system. International
Journal of Engineering Research & Technology (IJERT), 7(11), 2019.
11. M. S. Minu and K. N. Deepak Adithya. Real time college bus monitoring and notification system.
International Research Journal of Engineering and Technology (IRJET), 7(4), Sep 2018.
12. A. Ahmed, E. Nada, and W. AlMutiri. University buses routing and tracking system. International
Journal of Computer Science & Information Technology (IJCSIT), 9(1), February 2017.
13. S. Phadatare, P. Kudtarkar, J. Seth, and V. Yeole. Bus monitoring system using Android application.
International Research Journal of Engineering and Technology (IRJET), 6(4), Apr 2019.
14. M. A. Hannan, A. M. Mustapha, A. Hussain, and H. Basri. Intelligent bus monitoring and management
system. In Proceedings of the World Congress on Engineering and Computer Science, Volume II, San
Francisco, USA, October 2012.
15. P. Shanmuga Priya, P. Lavanya, K. Preethika, S. Sandhiya, and B. Subhapriya. GPS based tracking
system for government buses. International Journal of Engineering Research & Technology (IJERT),
13(5), June 2025.
16. N. Khwaja, S. Kadivala, S. M. Musharaf, and V. Baloda. Real time bus tracking system. International
Journal of Advanced Research in Science, Communication and Technology (IJARSCT), 2(3), April
2022.
17. D. Vijay Kumar. GPS and GSM based passenger tracking system. International Journal of Computer
Applications, 100(2), August 2014.
18. M. R. S. Madhur, K. S. Priyatham, K. C. Raj, K. B. Nandan, and B. V. Sathish Kumar. Smart bus tracking
and management system. Iconic Research and Engineering Journals (IRE), 1(8), Feb 2018.
19. M. Kumbhar, M. Survase, P. Mastud, and A. Salunke. Real time web based bus tracking system.
International Research Journal of Engineering and Technology (IRJET), 3(2), Feb 2016. ), 3(2), Feb
2016.
