Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

IoT-Enhanced Smart Surveillance System for Wildlife

Collision Prevention on Sri Lankan Roads
Amerasinghe N.D.K1, Udara I.W.A.S.2, Somabandu B.P.S.3, Jayakody J.A.B.U.4, Nelum Amarasena5, Rivoni De Zoysa6
Faculty of Computing, Sri Lanka Institute of Information Technology
New Kandy Rd, Malabe, Sri Lanka
WildLife Department Baththaramulla, SriLanka
Wildlife department Minneriya, Sri Lanka

Abstract:- This research paper presents a mobile The Sri Lankan government and conservation
application-based solution for real-time road safety organizations have implemented measures to reduce animal
aimed at mitigating animal-vehicle conflicts in Sri collisions, such as building wildlife crossings and promoting
Lanka. The proposed system is developed in response to road safety awareness. Drivers must also be aware of the
the increasing incidents of animal-vehicle collisions, potential for AVCs and take steps to reduce the risk, such as
which pose significant risks to both human safety and slowing down and being alert for wildlife on the road.
wildlife conservation efforts. By combining knowledge-
based case studies and crowd-sourcing techniques, the The primary focus will be on conducting a knowledge-
application aims to identify animal habitats and based case study to determine the habitat and behavior of
behaviors based on user location, enabling drivers to animals in various locations [1]. This study will involve
take proactive measures to prevent such accidents. gathering and retrieving data to inform passengers, utilizing
Image processing is used to identify objects using crowd-sourcing technologies to obtain additional details and
YOLOv7 technology. Ultrasonic sensors, a insights from people. Additionally, an RSS feed will be used
microcontroller, the Doppler Effect, and relative velocity to obtain news and weather information for specific areas,
calculations are used to segment the signal-to-noise ratio enabling the notification of users in different locations when
and alert the driver of any nearby animals. The research necessary.
also provides guidelines for implementing methods to
reduce animal-vehicle collisions and raises awareness The second aspect of the study aims to capture animals
among the public regarding the importance of road from moving vehicles, identifying their behaviors, and
safety in wildlife-rich areas. Upon detection of the conveying information to users. It will focus on clear, high-
animal, the system emits an anti-frequency range quality footage of different animal types, enabling
tailored to specific species, effectively deterring them identification of direction and distance. Computer vision and
from approaching the vehicle. Through the mobile image processing techniques will be employed to overcome
application, users can access real-time alerts, receive human observation limitations and improve animal
feedback from others, and contribute to a collective detection.
effort in ensuring safer roads for both humans and
animals. The third aspect, Unseen identification is highly
effective in dark environments [19][20], as a signal is
Keywords:- AVC (Animal Vehicle Collision), Animal transmitted to the environment and processed to identify
Behavior and Habitat, Crowdsourcing, YOLOV7, Doppler animals within a specified distance. Research has focused on
Effect, Anti-Frequency. frequency, frequency emission, and rollback frequency
recovery after being hit by animals. Key factors include
I. INTRODUCTION vehicle vibration, Doppler effect, moving object laws, and
time.
Animal-vehicle collisions (AVCs) are a significant
threat to biodiversity, human safety, and economic costs in The study proposes improved solutions for animal-
Sri Lanka [1]. These collisions occur due to the close vehicle collision mitigation, including guidelines for
interaction between human and wildlife habitats, resulting in implementing techniques and public education. It involves
numerous accidents and environmental impacts. In Sri projecting anti-frequency waves onto animals and providing
Lanka, an estimated 26,489 motor vehicle and animal instructions for responding in encountered scenarios. User
accidents occur annually, causing significant human deaths, feedback and experience are crucial, aiming to prevent
injuries, property damage, and other costs [2]. The lack of conflicts and minimize tragic incidents [17].
public awareness and negligence towards AVCs exacerbates
the problem, leading to a lack of effective mitigation
measures. Road networks often fragment animal habitats
and home ranges, leading to a reduction in animal
population size due to behavioral changes.

IJISRT23OCT1283 www.ijisrt.com 1203

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. LITERATURE REVIEW tired, distracted, or exhausted, limiting effectiveness and
accuracy in animal detection. Computer vision can help
Sri Lankan government and conservation organizations reduce these limitations by allowing 24-hour rest for image
have implemented measures to reduce conflicts between processing.
humans and animals. These include building wildlife
crossings and raising road safety awareness. Drivers should The third point emphasizes the effectiveness of unseen
be aware of potential collisions and take steps to reduce identifications in low-light or dark environments. This
risks, such as slowing down and being alert for wildlife. innovative approach uses ultrasonic waves to transmit a
Solutions cover various parameters to prevent human- specific distance, followed by processing and analyzing
animal conflict. variations in frequencies to accurately calculate the animal's
distance. The calculation considers the Doppler effect and
As first we are having A knowledge-based case study moving object equations, ensuring timely notification of
explores identifying animal habitats and behavior using distance and animal identification. This system ensures
crowd-sourcing technologies. The study discusses data essential information is relayed efficiently.
collection and retrieval methods for driver notification,
utilizing crowd-sourcing technologies to identify dispersion Enhanced alerting systems have been investigated to
factors and behavior. Data will be collected anonymously improve driver awareness and response to animal presence
from the stakeholders as well as from targeted audience (Sri on the road. Studies have shown the effectiveness of
Lankan government agencies, NGOs, local communities, different alerting mechanisms in reducing AVCs. Visual
drivers and commuters, researchers and academics) alert systems, audible alerts and vibrations can increase
whenever an incident is found or through past history and driver anticipation and response to animal presence.
research. The varying factors like climate changes, news Comprehensive guidelines for unexpected animal
information that may help for user to be precautious will be interference are crucial in educating drivers on how to
notified through OpenWeatherMap API based the GPS respond to unpredictable behavior [16].
locations and News API with RSS feeds. The gathered data
will be analyzed and visualized in a statistical manner to III. METHODOLOGY
ensure the application of technology to the proposed case
study to alert the user in necessary circumstances. As illustrated in Fig. 1, the system uses geofencing,
YOLOv7, and ultrasonic microcontroller technologies to
Geofence a specific area by drawing a virtual fence on detect potential collisions between stakeholders, researchers,
a map and using geotargeting to send notifications to people and drivers. The mobile application collects and processes
meeting specific criteria and within a defined radius. data in four main sections: geofencing using knowledge
base case studies and crowdsourcing; seen identification;
As the second point, we developed a motionless or unseen identification; and the fourth and final section,
moving animal identification concept to capture and identify generating timely and relevant notifications and alerts to
its behavior in a moving vehicle. The concept also focuses notify drivers on how to avoid collisions based on the
on conveying the captured message to the user and driver, identified animal.
ensuring effective communication. Human eyes can become

Fig 1 Overall System Diagram of the Proposed Solution

IJISRT23OCT1283 www.ijisrt.com 1204

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
A. Knowledge base case Study on Animal Habitat / Behavior and Application of Geo Fencing

 Study Area of the Knowledge-based Case Study

The Central and North Central Provinces of Sri Lanka have experienced significant animal-vehicle collisions since 2011 [1].
Key protected areas include Kaudulla (69 km2) and Minneriya (88.894 km2) National Park as depicted in Fig 2. The region's
climate varies with 650mm annual rainfall occurring during wet seasons [4], and its habitat is dominated by trees, supporting a
diverse wildlife population including 24 mammals, 170 birds, 8 reptiles, and 2 amphibians [3].

The study reveals a rise in AVCs in the region since 2011, likely due to increased road traffic and human settlements in
wildlife areas. These vehicles can cause severe harm to animals and humans, resulting in death or serious injury.

Fig 2 Identification and Classification of Wildlife Protected Areas in Srilanka

A study in Sri Lanka reveals high vehicle speeds, small-bodied animals, and nighttime driving are major factors contributing
to animal-vehicle collisions (AVCs). AVCs result in 161 accidents annually, killing 40 species of animals.

Table 1 Causes of Elephant Deaths from 2010 to 2023.05.17

Causes of Elephant Deaths from 2010 to 2023.05.17
Year Train Accidents Road Accidents
2010 -
2011 -
2012 9 -
2013 7 -
2014 10 -
2015 12 -

IJISRT23OCT1283 www.ijisrt.com 1205

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
2016 12 -
2017 7 -
2018 16 -
2019 8 5
2020 3 3
2021 6 1
2022 14 3
2023.05.17 4 1
Total 108 13

The study utilized road surveys, driver surveys, and  Users are notified with relevant and proper notifications
interviews with wildlife experts to identify drivers' main using the processed information.
causes of AVCs. They believed high vehicle speeds were
the main cause, with small-bodied animals more likely to be Data collection is time-consuming and labor-intensive,
hit. Nighttime visibility made it difficult for drivers to but crucial for accurate results. Geofence can be created
identify animals, potentially causing accidents. Also, it using technologies like GPS, RFID, Wi-Fi, or cellular data.
highlights the importance of understanding the causes and Customized notifications include relevant information, such
factors of AVCs to develop strategies for wildlife as animal type, sighting time, and location.
conservation and road safety [1]. Measures to reduce AVCs
in Sri Lanka include reducing vehicle speeds, raising driver  Process Information of News and Weather
awareness, installing wildlife crossing signs, and educating The data is analyzed to identify factors affecting
the public about the importance of wildlife conservation. animal habitats and behaviors, enabling predictions about
future movements. The methodology uses the
 The proposed smart system can identify animal behavior OpenWeatherMap API and News API to alert users of
and interaction within roads. The system uses a climate change and news events affecting animal behavior.
knowledge-based case study methodology, which Statistical methods are used to identify trends and patterns.
involves the following steps:
B. Seen Identification using Camera trap with YOLOv7
 Identify the research problem and objectives. The goal is Algorithm.
to identify animal habitat and behavior according to The smart system detects animals using devices, image
different locations for future predictions. processing, YOLOv7, and human vision [8]. It uses relative
 Gather anonymous data through crowd-sourcing velocity to identify movements. The camera trap setup is
platforms, by collecting information from stakeholders designed based on angle and distance, and the system uses
whenever an incident is reported, or through past history Yolov7 algorithm and image processing to detect animals.
and research. The captured output message is sent to the user via an alert
 Analyze the gathered data in a statistical manner to system.
identify the dispersion factors of animal habitat,
behavior, and varying factors like climate changes and  Data Collection
news information. A study in the Habarana area identified elephants and
deer as the most needed animal classes due to accidents. A
The proposed system uses a knowledge-based case diverse set of images was collected, covering angles,
study methodology to analyze animal behavior and lighting conditions, backgrounds, and object poses, to train a
interaction on roads. By identifying potential risks and robust model. For data collection purposes, we used a 1000
taking precautions, drivers can be more aware of potential image set to train.
dangers and take necessary precautions. The results can help
drivers be more cautious while driving. Images were categorized into training (80%-90%),
validation (20-30%), and testing (10-20%) sets for model
 Methodology for Application of Geofencing and training, validation, and final evaluation. Training sets
Geotargeting: trained the model, validation monitored performance, and
Data from various sources is collected to identify testing set evaluated.
animal habitats and behaviors. This data creates a geofence,
allowing users to receive notifications based on targeting  Training the Detection Models
criteria within a defined radius. To train YOLOv7 detection models for recognizing
animals in images or videos, a series of stages are involved.
 The Methodology for Animal Habitat and Behavior Using the robowflow, bounding boxes and class labels were
Identification can be Summarized as follows: added to the obtained dataset. To guarantee a fair
assessment, the dataset is carefully selected and segregated
 Data is collected from a variety of sources. into training, validation, and test sets. The YOLOv7
 A geofence is created around a specific area. architecture is set up, including class sizes and input image
 Notifications or filters are delivered to users who meet size. The model is initialized using pre-trained weights
specific targeting criteria within the defined radius. based on Darknet53, and data preparation techniques are

IJISRT23OCT1283 www.ijisrt.com 1206

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
used to expand the dataset [1]. The loss function is
optimized using gradient descent methods, and parameters
are updated over 100 training epochs.

Performance evaluations are conducted on validation The average precision and mean AP values of each
and test sets, measuring mean average precision (mAP) [9]. category are combined to get the mean average precision, or
The YOLOv7 model is now ready for deployment and mAP. The following is the calculating formula [12]:
inference in practical applications.
∑𝑆𝑗=1 𝐴𝑃(𝑗)
 Use of Measure of Appropriate Angle and Distance from 𝑚𝐴𝑃 = (4)
𝑆
the Vehicle.
The inverse perspective transformation method Where S is the total number of categories and APs for
converts two-dimensional images into three-dimensional all categories are added to form the denominator.
space images with a perspective using camera parameters.
Real-world coordinates are calculated to estimate the C. Unseen Identification using Ultrasonic Waves.
distance between the camera and detected vehicle pixels in
the transformed three-dimensional space.  Data Gathering
Firstly, A literature review explores the Doppler effect
 Design the camera setup and convey the captured output in animal identification, and a controlled environment test
message to user through the camera setup. setup measures the effect caused by moving animals.
The camera setup, the CPU converts the video into
frames, pre-processes images, and applies an animal Equation 1: Doppler Effect Equation for Frequency
detection algorithm. Edge detection technique obtains Shift [22]. The Doppler effect equation for frequency shift
animal outline, turning images into grayscale [10]. can be expressed as:
YOLOV7 object recognition algorithm compares data with
pre-processed animal photos, finds animals, and measures 𝑓′ = (𝑣 + 𝑣𝑑) / (𝑣 + 𝑣𝑠) ∗ 𝑓 (5)
distance before applying electromagnetic braking with
varying strengths. Where: f' is the observed frequency, f is the emitted
frequency, v is the speed of sound, vd is the velocity of the
 Evaluation Metrics detector (vehicle), vs is the velocity of the source (animal).
The evaluation indices utilized in this study to measure
the algorithm's performance were precision (P), recall (R),
Equation 2: Doppler Effect Equation for Velocity
mean average precision (mAP), and F1 score [12]. Calculation. The Doppler effect equation can also be used to
calculate the velocity of the moving animal. It can be
Precision is the percentage of positive samples in expressed as:
samples with positive prediction outcomes. The following is
the calculating formula: 𝑣 = (𝑓 − 𝑓′) ∗ 𝑣 / (𝑓 + 𝑓′) (6)
𝑇𝑃
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (1) Where: v is the velocity of the moving animal, f is the
𝑇𝑃 + 𝐹𝑃 emitted frequency, f' is the observed frequency, v is the
speed of sound.
The percentage of the actual positive samples in the
positive samples to the positive samples in the entire sample
Research investigated various signals for identifying
is how recall depicts the prediction outcome. The following
animals in darkness, developed test setups to measure
is the calculating formula:
effectiveness, and conducted experiments to compare their
𝑇𝑃 effectiveness.
𝑅𝑒𝑐𝑎𝑙𝑙 = (2)
𝑇𝑃 + 𝐹𝑁 Research examined animal identification devices,
identified key features, and conducted experiments to
The model's precision in identifying negative samples
compare their performance in sending and receiving signals.
is influenced by its ability to discriminate them. Higher
precision indicates stronger identification, while recall
Experiments analyzed data to determine Doppler
measures its ability to locate positive samples. The F1 score,
effect, effective signal, and best device for animal
combining these two, indicates a model's reliability.
identification, leading to the development of an algorithm.
The average precision (AP), which is often determined
For the fourth objective, mathematical equations
independently for each category, is the average value of the
developed using Doppler effect and other theorems for
greatest precision under various recall situations. The
moving objects, considering external variables like vehicle
following is the calculating formula [12]:
speed, vibration, wind, and time. Implemented in an
algorithm, accurate outputs were provided to the driver,
tested using the chosen device and signal.

IJISRT23OCT1283 www.ijisrt.com 1207

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Statistical Analysis D. Solutions and Alternatives to Reduce AVC.
The second and third objectives analyzed data to The proposed solution aims to intelligently detect and
identify the most effective signal and suitable device for identify animals on the road using various methods,
animal identification, while the first objective developed an including visual fences, tracking, images, movements, and
algorithm for animal identification. Doppler effect analysis. Upon detection, the system alerts
users and provides guidelines based on case studies.
 Implementation Frequency waves are used as a deterrent, and alerts are
The algorithm for animal identification tested accuracy delivered through audible, vibration, and visual
using an Arduino-Uno and ultrasonic sensor, utilizing notifications.
equations for the fourth objective and ensuring accurate
outputs for the driver. This methodology combines literature  Data Collection
review, experimental testing, data analysis, and algorithm Data on animal-vehicle collisions was collected
development to create a mobile app for animal detection and through surveys, interviews, and observational studies, with
identification in dark environments. primary data from drivers and stakeholders and secondary
data from government reports and academic literature.
 System Architecture
The proposed smart system aims to detect animals near Surveys and stakeholder interviews were conducted to
roads during dark hours and daylight hours using unseen gather information on drivers' experiences with animal-
identification methodology. Ultrasonic waves are emitted to vehicle collisions, awareness of alerting systems, and
a specific distance, and the difference between outgoing and guidelines for reducing AVCs. Stakeholder interviews were
incoming waves is used for calculations. A Passive Infrared conducted to obtain insights from wildlife experts, police
(PIR) sensor is used to improve accuracy by identifying heat officers, transportation authorities, and road safety
bodies in the vehicle's path. Principles related to moving professionals.
objects, such as Doppler effect and relative velocity, are
used to improve accuracy. External factors like vehicle  Development of Enhanced Alerting Systems
vibration, speed, and wind are considered to provide the The phase involved designing and developing alerting
most accurate output. The system then compares the results systems to enhance driver awareness and response to animal
to a database and uses proper methods to alert drivers with presence on the road. Various mechanisms, including
accurate information, such as distance, animal type, and audible, voice, vibration, and visual, were explored. Sensors
precautions. and cameras were used to detect and trigger alerts. The
design process involved prototyping, testing, and refinement
The ultrasonic waves emitter in a vehicle emits to ensure effectiveness and user-friendliness.
frequency waves while traveling, while the receiving device
detects the incoming signal after being hit by an animal. An  Evaluation of Guidelines for Animal Interference
Arduino-Uno and ultrasonic sensors are used to capture Guidelines were created to guide drivers on handling
these signals and calculate the distance to the animal based unexpected animal interference on the road. They were
on differences in frequencies and other factors. evaluated through simulation exercises and driver feedback,
assessing their clarity, practicality, and usefulness.
Ultrasonic sensors emit waves into the environment,
bouncing back when hit by animals. The receiver calculates The research investigated high-intensity animal-vehicle
distance based on frequencies and other factors, determining accidents in Battaramulla, identifying common animals and
the animal's proximity to the vehicle [18]. identifying the species involved. The following animals
were selected for further analysis based on their frequency
of occurrence: elephants (94.4%), deer (83.3%), wild
buffalos (27.8%), wild boars (22.2%), porcupines (22.2%),
dogs (50%), and cats (38.9%).

After confirming the presence of these common

animals, our research focused on collecting guidelines for
drivers to follow when encountering them. These guidelines
are intended to enhance road safety and mitigate the risk of
Fig 3 HC-SR04 Ultrasonic Sensor Module collisions.

IV. RESULTS AND DISCUSSION

The data was analyzed using SPSS, with descriptive

statistics for demographics. A Chi Square goodness of fit
test was used to identify the primary respondents' taxon
groups, AVCs, and their impact on roads. The study also
examined drivers' perceptions of AVCs, their perception of
road safety, and their awareness of necessary actions.
Fig 4: PIR Sensor

IJISRT23OCT1283 www.ijisrt.com 1208

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Spearman's rank correlation test was applied to determine Southern Motorway. Elephants, water buffalo, and deer are
the relationship between species' body mass and AVC risk, the most frequently involved animals. However, Sri Lanka
despite the small sample size and select species tested [4]. has a larger diversity of animal species involved in vehicle-
animal accidents, with collisions involving over 40 distinct
Over 1,700 animal-vehicle accidents have been species (Figure 4).
documented in Sri Lanka since 2011, primarily on the

Fig 5: AVC Collision Statistics

Birds accounted for 58% of the total population, AVCs caused significant anxiety among drivers, with
followed by mammals (24%), reptiles (11%), and 65% stating they were a big problem and 35% not. Common
amphibians (7%). Drivers felt mammals were most affected causes include animal behaviors, darkness, sluggish
by AVCs (91.30%), followed by reptiles (3.40%), birds movement, adverse weather, and high vehicle speeds. The
(9.50%), and amphibians (1.70%). Changes may occur at least likely causes were intentional collisions and poor
other times of the year, but the study was conducted during roads.
the wet season, assuming maximum species abundance.
Geo fencing in mobile applications can mitigate
wildlife vehicle collisions by creating wildlife corridors,
connecting different habitats and allowing animals to move
safely. This allows drivers to be alerted of animals' presence,
allowing them to slow down or take other precautions to
avoid collisions.

Geo fencing can help mitigate wildlife vehicle

collisions by creating warning systems using wildlife
cameras, motion sensors, and other devices. These systems
alert drivers of potential collisions when animals are
detected, reducing the risk of accidents. A knowledge base
case study will create a database of potential AVC locations
and animal corridors, enabling geofences. An alert will be
sent to a monitoring system when an animal enters or leaves
a geofence, reducing animal-vehicle collisions by alerting
drivers.
Fig 6: Taxonomic Division of the Observed AVC Geo fencing is a promising technology for mitigating
WVCs. Geo fencing is a cost-effective method to warn
The majority of the 100 drivers—85%—indicated that drivers of wildlife presence, reducing collisions, injuries,
AVCs happened more frequently at night, while just 10% and fatalities in wildlife-related areas.
reported that they happened during the day. The remaining
5% reported that AVCs happened both during the day and at
night (2 = 77.523, df = 2, p = 0.001).

IJISRT23OCT1283 www.ijisrt.com 1209

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
For the YOLOV7, model was trained using 1000
image configurations, two classes, 100 steps per epoch, and
a minimum detection confidence of 0.9. The supplied image
dimensions were 1024 × 1024. Following training, images
were inferred on the webcam for detection and the animals
were properly recognized via the webcam.

The Doppler effect and ultrasonic sensors were used

for unseen animal detection and identification in the dark.
Experiments revealed the frequency shift of received signals
was proportional to animal velocity. Ultrasonic signals
effectively detect animals in the dark due to Doppler effect
frequency differences. Arduino-Uno microcontrollers and
ultrasonic sensors are chosen for accurate distance
Fig 7: Geofencing Identification calculations and precise outputs. An algorithm with
mathematical equations is developed, enhancing driver
The YOLOV7 algorithm-based Python was used for safety by accurately identifying animals in the dark.
animal recognition, achieving 96.4% training. When the
dataset comprises numerous classes, a single number is The proposed methodology for animal detection and
required to evaluate the model's performance. The mAP identification in darkness is confirmed through experimental
value was determined as a consequence. The mAP value is and data analysis. The combination of Doppler effect,
the mean average of the computed average precision and F1 ultrasonic sensors, and the algorithm results in a reliable and
[11]. accurate system alerting drivers to animals near the road.
The Arduino-Uno microcontroller and ultrasonic sensors are
Fig 7 shows precision curve with recall, while Fig 8 suitable for implementing the system architecture.
shows F1 curve with confidence, evaluating trained models.
The proposed system alerts drivers to potential animal
The experimental findings are displayed in Figure 3.1, hazards in low-light conditions, improving road safety.
and the metrics were recalling rate, [email protected], and Future research should refine and optimize the system by
[email protected]:0.964. expanding the database of animal characteristics and
incorporating additional sensors or technologies.

This study investigated the effectiveness of a system to

reduce animal-vehicle collisions by providing alerts to
drivers upon animal detection using audible, visual, and
vibration cues. The system also emitted anti-frequency
waves. Advanced sensors and algorithms accurately
identified animals near the roadway, providing timely
warnings and capturing drivers' attention to potential animal
hazards.[17]. Anti-frequency waves were used to deter
animals from crossing roads, but their effectiveness varies
depending on species and individual animals. Further
research is needed to explore the specific frequencies that
are most effective in deterring different animal species.

Fig 8: Precision Curve Diagram for Trained Data The results section presents research findings on
alerting system effectiveness, driver behavior impact, anti-
frequency wave projection deterrence potential, and user
feedback insights.[17]

V. CONCLUSION AND FUTURE WORK

In Conclusion, this study proposes a mobile application

in Sri Lanka to reduce animal-vehicle collisions. The
application uses methods like knowledge-based case studies,
geofencing, and geotargeting to identify areas where
animals are likely to be present and notify drivers of
potential hazards. The Sri Lankan application aims to
improve road safety for humans and animals by using
YOLOv7s for detecting wild animals in natural habitats.
Fig 9: F1 Curve Diagram for Trained Data This is the first system to detect wild animals during dark

IJISRT23OCT1283 www.ijisrt.com 1210

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
hours and mitigate collisions. The application aims to [10]. S. Köpke, S. Withanachchi, R. Pathiranage and C. R.
accurately and timely identify animals, with future Withanachchi, "Human—Elephant Conflict in Sri
expansion. The study highlights findings, contributions, Lanka: A Critical Review of Causal Explanations,"
potential applications, and benefits, while providing August 2021. [Online]. Available:
research directions and recommendations for enhanced https://www.researchgate.net/publication/353683505
alerting systems. The study emphasizes the importance of _Human-Elephant_Conflict_in_Sri_Lanka_A_
enhanced alerting systems and guidelines in reducing Critical_Review_of_Causal_Explanations.
animal-vehicle collisions and suggests policy interventions, [11]. Kokhan, "Animal vehicle collisions," February 2018.
technological advancements, and public awareness [Online]. Available: https://www.researchgate.net/
campaigns for future work in this area. publication/323228119_Animal_vehicle_collisions.
[12]. J. Kioko, C. Kiffner, P. M. Phillips and C. Patterson-
REFERENCES Abrolat, "Driver knowledge and attitudes on vehicle
collisions in Northern Tanzania," June 2015.
[1]. Yusman, A. Finawan and Rusli, "Design of Wild [Online]. Available: https://www.researchgate.net/
Animal Detection and Rescue System with Passive publication/279491842_Driver_knowledge_and_attit
Infrared and Ultrasonic Sensor based udes_on_vehicle_collisions_in_Northern_Tanzania#:
Microcontroller," 4 May 2018. [Online]. Available: ~:text=We%20monitored%20the%20patterns%20an
https://www.emerald.com/insight/content/doi/10.110 d%20frequency%20of%20AVCs,the%20drivers%20t
8/978-1-78756-793-1-00042/full/html#s8. o%20be%20the%20most%20common%20AVC..
[2]. M. Tan, W. Chao, J.-K. Cheng and M. Zhou, [13]. J. C. Kilgo, J. I. Blake, T. E. Grazia and A. Horcher,
"Animal Detection and Classification from Camera "Use of Roadside Deer Removal to Reduce Deer–
Trap Images Using Different Mainstream Object Vehicle Collisions," 15 May 2020. [Online].
Detection Architectures," 4 August 2022. [Online]. Available: https://digitalcommons.usu.edu/hwi/vol14
Available: https://www.mdpi.com/2076-2615/12/ 15/ /iss1/13/.
1976. [14]. Karunarathna and S. Ranwala, "Impact of vehicular
[3]. J. Staver and E. Jackson, "Why Do More Car Crashes traffic on animal mortality in Horton Plains and Yala
Happen at Night?," [Online]. Available: National Parks of Sri Lanka," December 2015.
https://www.chicagolawyer.com/blog/why-do-more- [Online]. Available: https://www.researchgate.net/
car-crashes-happen-at-night/#:~:text=Reaction publication/301593977_Impact_of_vehicular_traffic
%20times%20have%20been%20found,to%20reduce _on_animal_mortality_in_Horton_Plains_and_Yala_
%20car%20crash%20fatalities.. National_Parks_of_Sri_Lanka.
[4]. F. M. Shilling and D. P. Waetjen, "Wildlife-vehicle [15]. K. Jiang, T. Xie, R. Yan and X. Wen, "An Attention
collision hotspots at US highway extents: scale and Mechanism-Improved YOLOv7 Object Detection
data source effects," 28 July 2015. [Online]. Algorithm for Hemp Duck Count Estimation," 10
Available: October 2022. [Online]. Available:
https://natureconservation.pensoft.net/article/4438/. https://www.mdpi.com/2077-0472/12/10/1659.
[5]. T. G. S. L. Prakash, A. W. Wijeratne and P. [16]. M. S. &. T. S. Hilaire, "Wildlife-Vehicle Collisions
Fernando, "Human-Elephant Conflict in Sri Lanka: Are a Big and Costly Problem and Congress Can
Patterns and Extent," 2020. [Online]. Available: Help," 10 May 2021. [Online]. Available:
https://www.asesg.org/PDFfiles/2020/51-16- https://www.pewtrusts.org/en/research-and-analysis/
Prakash.pdf. articles/2021/05/10/wildlife-vehicle-collisions-are-a-
[6]. S. Plainis, I. J. Murray and I. G. Pallikaris, "Road big-and-costly-problem-and-congress-can-help.
traffic casualties: understanding the night‐time death [17]. W. Burrard-Lucas, "Turn your DSLR Camera into a
toll," 12 April 2006. [Online]. Available: Wildlife Camera Trap," [Online]. Available:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC256 https://digital-photography-school.com/turn-dslr-
4438/. camera-wildlife-camera-trap/.
[7]. R. Pagany, "ScienceDirect," November 2020. [18]. Benten, P. Annighöfer and T. Vor, "Wildlife
[Online]. Available: https://www.sciencedirect.com/ Warning Reflectors' Potential to Mitigate Wildlife-
science/article/pii/S0006320720308168. Vehicle Collisions—A Review on the Evaluation
[8]. S. Mayadunnage, S. Colombage, D. Weerakoon and Methods," April 2018. [Online]. Available:
W. Mampearachchi, "Animal collision induced road https://www.researchgate.net/publication/324480597
accidents in the Southern Expressway," [Online]. _Wildlife_Warning_Reflectors%27_Potential_to_Mi
Available: http://dl.lib.uom.lk/handle/123/18179. tigate_Wildlife-Vehicle_Collisions-
[9]. J. A. Litvaitis and J. P. Tash, "An approach toward A_Review_on_the_Evaluation_Methods.
understanding wildlife-vehicle collisions," 22 April [19]. Abra, B. M. Granziera and M. P. Huijser, "Pay or
2008. [Online]. Available: prevent? Human safety, costs to society and legal
https://pubmed.ncbi.nlm.nih.gov/18427884/. perspectives on animal-vehicle collisions in São
Paulo state, Brazil," 11 April 2019. [Online].
Available: https://journals.plos.org/plosone/article?id
=10.1371/journal.pone.0215152.

IJISRT23OCT1283 www.ijisrt.com 1211

Volume 8, Issue 10, October – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[20]. "Wildlife-Vehicle Collision Reduction Study," 2008
August. [Online]. Available:
https://www.fhwa.dot.gov/publications/research/safet
y/08034/.
[21]. "Wildlife‐Vehicle Collision Mitigation," Arizona
Department of TransportaƟon, 2012.
[22]. LibreTextPhysics:https://phys.libretexts.org/Bookshe
lves/University_Physics/Book%3A_University_Phys
ics_(OpenStax)/Book%3A_University_Physics_I_-
_Mechanics_Sound_Oscillations_and_Waves_(Open
Stax)/17%3A_Sound/17.08%3A_The_Doppler_
Effect#:~:text=The%20Doppler%20effect%20is%20
an,stationary%20source%20and%20moving%20obse
rver.https://phys.libretexts.org/Bookshelves/Universit
y_Physics/Book%3A_University_Physics_(OpenSta
x)/Book%3A_University_Physics_I_-
_Mechanics_Sound_Oscillations_and_Waves_(Open
Stax)/17%3A_Sound/17.08%3A_The_Doppler_
Effect#:~:text=The%20Doppler%20effect%20is%20
an,stationary%20source%20and%20moving%20obse
rver.

IJISRT23OCT1283 www.ijisrt.com 1212