See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/381128194

A comparative study of animal detection and classification algorithms,

applications and challenges A Comparative Study of Animal Detection and
Classification Algorithms, Applicat...

Conference Paper · June 2024

DOI: 10.1063/5.0211343

CITATIONS READS

0 4

2 authors:

Divya Meena Sundaram Johnwesily Chappidi

VIT-AP University VIT University - AP
62 PUBLICATIONS 273 CITATIONS 3 PUBLICATIONS 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Johnwesily Chappidi on 04 June 2024.

The user has requested enhancement of the downloaded file.

RESEARCH ARTICLE | JUNE 03 2024

A comparative study of animal detection and classification

algorithms, applications and challenges 
Johnwesily Chappidi; Divya Meena Sundaram 

AIP Conf. Proc. 3112, 020020 (2024)

https://doi.org/10.1063/5.0211343


View Export
Online Citation

04 June 2024 10:17:30

 A Comparative Study of Animal Detection and
Classification Algorithms, Applications and Challenges
a)
Johnwesily Chappidi, Divya Meena Sundaram

SCOPE, VIT-AP University, Amaravathi, India

Corresponding Author:- a) divyameena.s@vitap.ac.in

Abstract—Animals suddenly obstructing roads and wild animals infiltrating human dwellings are involved
in many of the world's worst road accidents. Aside from property damage, fatal deaths and permanent
disabilities also occur. As a result, mitigation of such problems has become a prominent area of research,
leading to the development of a wide range of detection techniques. This paper analyses ongoing research
approaches to the prevention of animal-vehicle conflicts (AVC) and animal-human conflict (AHC). The
primary goal of this study is to investigate the various image and video-based object detection approaches
that are used to detect and solve image and video-based animal detection challenges. The various animal
monitoring system, as well as some of the methods used to detect animals and collect their data sets, will be
discussed in this paper. The challenges associated with animal identification are briefly discussed.

Keywords—Animal Detection; Animal classification; Animal monitoring system; Animal-Vehicle Collision;

Animal-HumanConflict.

INTRODUCTION

04 June 2024 10:17:30

Innovative technological advancements, industry expansions, changes in human lifestyle and Habits,
overpopulation, aspirations for economic development, and the over-exploitation of natural resources have all
contributed to the acceleration of deforestation and the depletion of the planet's ecosystems. As a consequence,
many animals are now migrating to human communities, leading to an increase in accidents[1]. Road accidents
involving fatalities (humans, animals) property damage, and sometimes fatalities, are caused by animals suddenly
blocking fast-moving vehicles and drivers not being able to spot them in the blink of an eye [2]. They fall into the
category of Animal-Vehicle Collisions [AVC] species population dwindles. Sometimes dangerous animals leave
the forest and infiltrate human habitable areas instigating rebellion. When this happens, humans kill animals or
animals kill humans. They fall into the category of Animal-Human Conflicts [AHC].A study of statistics [3]
illustrates the seriousness of the issue. Many destructions caused by animal involvement can be prevented by
detecting the movements of dangerous animals in advance. Animal detection is to identify the moments of the
animals to prevent wild animal intrusion in human places, conservation of endangered animals, AVC and on the
highways. Traditional methods, such as electric fence systems and warning boards, are ineffective and lead to
injuries and death. To reduce AVC on roads, prevent dangerous animal invasions, and establish alerting systems,
modern technologies can help identify and track animal activity. Deep learning and image processing techniques
developed as AHC, and AVC preventing algorithms to give alert messages to the human.

ANIMAL MONITORING SYSTEM

Applying Deep learning algorithms and image processing techniques for animal identification, recognition and
using image sensing devices and sensors to collect animal data, processing and storing a wide variety of animal
data to alert living beings from Predatory animals to prevent road accidents that happen suddenly. Animal Monitor
system [AMS] divides the collision prevention system into different categories to help to study and develop
warning systems for humans from the encroachment of animals. Fig. 1 depicts the AMS including animal
Detection, classification, and tracking, to assist humans in the research and development of warning systems when
detected Animal movements.

International Conference on Contemporary Challenges in Science, Engineering, and its Applications

AIP Conf. Proc. 3112, 020020-1–020020-7; https://doi.org/10.1063/5.0211343
Published under an exclusive license by AIP Publishing. 978-0-7354-4965-7/$30.00

020020-1
 Fig. 1Animal Monitoring System

The followingsub-sections discuss AMS and different sections involving to detecting animals to avoid AVCs.The
AMS has been divided into the following categories as shown in figure 1.
a) Animal Detection:This is detecting the type of animal and it is used in several applications. Light detection and
ranging (LIDAR), is the technology most fishermen use for fish detection to get fish locations in the deep sea.
Airborne remote sensing images are used to detect the motion of wild animals [4]. Another way, micro-Doppler
signals are used to detect harmful animal intrusion in residential places. convolutional neural network (CNN)
techniques are applied to detect animals by extracting features [5].
b) Animal Classification:It is sorting animals into classes based on their features like size, shape, colour, etc., to
identify animals. thermal infrared remote sensing Technologies and data sets used to study and recognize
specific animals. The spectral resolution, Panchromatic imagery records, Spectral resolution, multispectral
imagery that captures multiple bands, hyperspectral data collect different types of bands, and RGB colours

04 June 2024 10:17:30

visible light [6]. Digital Image Processing algorithms are used to classify and extract animal features from
images and videos taken from high-end cameras. Many remote sensing applications, such as farmland, earth
science, forest management, weather prediction, marine studies, archaeology research, and so on, require
landslide susceptibility maps of land cover/ soil types, that were created using image processing techniques [7].
c) Animal tracking:It is to monitor animal moments in identified areas. Using sensors, location tracking and
radio-frequency identification (RFID) technologies able to trace, to prevent Animal-human collisions, animal
trafficking, and species population dwindling. NGOs and Animal Protectors can use animal tracking data to
help them care for dwindling animal populations. Satellites, manned aircraft, and unmanned aircraft systems are
all examples of unmanned aircraft systems (UASs) Drones, unmanned aerial vehicles (UAVs), and advanced
DSLR cameras were utilized to track the animals and inform humans [8].
In AVC, the detection and alerting systems are usually split up into two subsystems to prevent conflict. Detection
and tracking subsystems and relevant alert subsystems are used in a range of applications. There were applications
that were road-based, animal-based, vehicle-based, and hybrid systems.

Fig.2 Various Animal Detection Mechanisms

020020-2
 Animal-Based Systems
The animal-based system works by attaching collars or necklaces to the animal's neck. These collars aid in
activating the animal detection subsystem. For this case, several methodologies would be used. Radio technology
and GPS are two examples. The following is a list of various collars used on animals.
 Glowing Collar: Glowing or indicative collars hooked up to the animal will assist drivers in sensing the animal
while driving. This system was put in place in Canada in 2006. Despite this, these techniques are ineffective at
preventing accidents. This is due to the fact that from a distance, drivers are unable to see the collars that are
attached to them. In addition, the collar's reflective materials degrade with time [9].
 Radio Transmitter Collar: Animals have been fitted with non-lethal short-range wireless radio transmitter
collars. These circuits are in-charge of transmitting waves to the road's receivers. If any signals are received, the
object recognition siren will be sounded to alert drivers. The problem with radio collars is that they consume a
lot of electricity. It means that the necessity to replace batteries on a regular basis puts financial hardship on the
animals' owners.
 GPS Collar: Install a GPS receiver circuit in the animal collar device to track satellite location data. As a result,
the animal's position will be included in the collar signal data. This will aid in the detection of animals as well
as, as a result, the generation of advanced warning alarms. These collars have the drawback of consuming a lot
of battery power and having low GPS accuracy, which affects the system's cost and accuracy. No animal-based
approach is suitable for preventing AHC in general. Because either the animals or the people who own them
will lose control.

Vehicle-based system
The circuit has been equipped with sensors in this vehicle-based system. These systems will be categorized as
follows.
 Alerting Alarms-Based Systems: In this system, the vehicle emits warning whistles or ultrasonic frequencies
which are noisy signals. These noisy signals are sound waves that have the ability to frighten animals and keep
them away.These systems are not always advantageous to many animals, such as camels1. These techniques are

04 June 2024 10:17:30

ineffective in preventing camel-related accidents [10].
 Sensor-Based Systems: The described technology involves installing infrared sensors and thermal cameras on
the front of a vehicle, serving two main functions. The infrared sensor detects animals, while the thermal
imaging camera and IR sensor measure the distance between the vehicle and other objects.“Gazal" vehicle has
been specifically designed with precise dimensions and magnitude to match the physical proportions of a camel,
providing King Saud University with an exceptional driving experience. In this context, sensor-based systems
have proven to be more effective than other types of systems. However, implementing and monitoring an
animal-based system can be challenging. [11].

Road-based system
In this concept, an animal detection system is installed on the roads or polls nearby roads. The most effective animal
detection has different models. So, the need to focus on accuracy should be increasing and reliability for this type of
road-based system model.
 Traditional Methods:Many traditional methods, such as wayside fencing, warning signs, and streetlights, could
be used to detect species on roads. These are not cheap, but rather expensive techniques that require a huge
amount of maintenance. These technologies, however, are ineffective at preventing AVC [12].
 Wireless Sensor-Based Methods:Sensor technologies used in wireless sensor-based systems include laser
sensors, electromagnetic radar sensors, and infrared sensors. Methodologies for detecting and alarming based on
infrared sensors 50% of detections produced false alarms. For the false output reasons may be either Presence of
a strong breeze or storms may misalign the sensors or maybea power issue due to solar panels covered by sand
or snow [13].

LITERATURE REVIEW
The accuracy of the various methods utilized in animal monitoring systemsare described in this section, as well as
the limits. The following three tables present image processing approaches, classical machine learning techniques,
and contemporary deep learning techniques with limitations. While there are several mechanisms used here to
support image processing approaches and research is ongoing. Table 1 includes the contemporary deep learning
methods, their accuracy, limitations, and references.

020020-3
 TABLE 1: COMPARISON OF VARIOUS ANIMAL MONITORING SYSTEMS

Ref. Attributes (species) Merits and Demerits

No.
1 The tested number is 1 Merits: Classifier Results high accurate
to 10 based on the Demerits: region delimiter box is a static size
SVM classifier
2 zebras and antelope Merits: Global pixel motion differs between the foregroundandthe
background.
Demerits: Accuracy Low
3 Rat, cat, rabbit, bird, Vigorous, solid, and suitable for image data set collected from the
mouse, bandicoot etc. wild. Demerits: Accuracy Low
4. Whale Fluke, Merits: This is a better approach for ecological information to
Reticulated Giraffe, conservationists.
Grevy’s Zebra, Sea Demerits:Accuracy Low
Turtle, Masai, Giraffe,
5. Foxes, wolves, brown Merits: When compared to other point detectors, this one get better
bears, wild boars, and results. Demerits: The classification rate of success is only 50%, so
deer are among some the results are poor.
of the animals that can
be found in the area.
6. Ahumanbeings,Anima Demerits: Animal detection accuracy is low
ls and Null
7. VWM dataset Merits: Quality Improves with the

04 June 2024 10:17:30

of wild animal detection and more robustness
Demerits:F-measure value is Low
8. Kangaroo dataset Accuracy is reasonable for animal detection.
Demerits:precision is Low
9. Snapshot Serengeti Merits: Performance is High in classification.
dataset Demerits:Illumination is Poor
10. Buffalo, Sheep, goat, Merits: Rapid and simple method.
cattle, deer, pig yak, Demerits:Accuracy needs to improve
and camel
11. Dog Merits: Recognition accuracy improved
Demerits:Not using the real dataset
12. Mice Merits: Detected mice effectively with a high precision rate.
Demerits: This approach is not suitable for high-speed applications
13. SegTrack v2dataset Merits: Objects should be segmented when the lighting and texture
of the environment drastically change.
Demerits:Algorithm runtime not satisfactory
14. Downloaded nature Merits: Contourlet transform is good
images and captured Demerits: Performance is low for the segmentation process
them with a camera
and mobile phone.
15. Porcine, Bovine Merits: Obtained accountable and acceptable technique
performance
Demerits:Accuracy not determined
16. animal activity, Merits: Better prediction of the relative activity of the group.
Human activity dataset Demerits: For feeding activities intense is low
17. SUBITO and Merits: Tracking accuracy improved
PETS2006 datasets Demerits:recall value and Precision is low
18. railway station Merits: Detecting vanishing points in Manhattan-style scenes
platform environments yielded better results.

020020-4
 Demerits: Accuracy needs to improve
19. Kuzikus dataset Merits: The recall rate is high.
Demerits:Precision value is low
20. Caltech101 dataset Merits: Complexity is low and the true positivity rate is high.
Demerits:Precision value is low

CHALLENGES
Animal detection algorithms may fail to detect animals at night due to limited visibility caused by low natural light
or artificial light pollution, the elusive behaviour of nocturnal animals, adaptive colouration and camouflage,
technological limitations of traditional cameras, weather conditions, and human error. These factors can make it
challenging to detect animals accurately and consistently at night.Object detection accuracy is hampered in
nighttime images captured by image-sensing devices due to several atmospheric conditions, such as fog, rain, or
dust, and the low contrast of salient objects. Consequently, the intensity, colour, polarization, and coherence of the
scene undergo changes, resulting in night images with subjective lighting and weather patterns that present a
daunting challenge for computer vision analysis. [54].
This section describes many problematic scenarios that cause problems in identifying animals with atypical
characteristics that make detection difficult. The main issue, and the first stage in tracking, is to reliably detect the
object in various settings. However, tracking an object became challenging due to complex backdrops, cast
shadows, weather conditions, and occlusions. This section will focus on the challenges of animal detection and
tracking.

04 June 2024 10:17:30

Fig. 3 Challenges for a AMS. a. Night vision, b. Natural Camouflage, c. Artificial Camouflage, d. Sea animal
Camouflage, e. Background clutter, f. Disruptive Coloration, g. colour countershading, h. countershading, i.
Cluttered environmental setting, j. object occlusion problem.

RESEARCH GAPS
Although there are numerous feasible solutions for automatic lane and pedestrian detection on highways, it is
essential to persist with research on completely automated detection and tracking on roadways. The following are
some of the unresolved challenges:
a) In the past, attempts were made to detect animals in forests or videos, as well as in controlled areas on the
seabed, but the task of detecting species on roadways, which are uncontrolled areas, presents significantly
greater challenges. This is due to the need for both the species and the camera-equipped vehicle to navigate

020020-5
 around other moving or stationary obstacles on the road. Unlike in wildlife films, where speed and distance
recognition are not problematic, these factors are essential and highly critical for animal detection on
highways.
b) Animal detection is much more challenging than detecting pedestrians or other objects since species come in a
variety of sizes, shapes, poses, and colours, and their activity is unpredictable. Although a human's basic
shape and size are quite average and standard, this is not the case for species.
c) Despite the various measures and strategies that have been implemented and continue to be developed for
detecting, addressing, and minimizing AVCs, the absence of feasible systems specifically designed for AVCs
on roadways has impeded notable advancements in this field.

CONCLUSION
Many problems should be considered while developing animal detection algorithms. The lighting problem is the
primary thing to consider, which is that an unexpected change of lighting condition in the indoor environment
affects the productivity in detecting the moments of animal intrusion. The next issue is brightness, which arises
when the natural habitat shifts from day to night. This might be difficult when it comes to animal detection while
using an exterior surveillance system. Furthermore, non-constant backgrounds, such as leaves blowing in the wind,
may be misidentified as foregrounds, and some motionless species may be misinterpreted as background images by
detection algorithms.

REFERENCES
[1] Alsadi, E. M. T. A., & El Abbadi, N. K. (2019). Scrutiny of methods for image detection and recognition of
different species of animals. International Journal of Recent Technology and Engineering, 8(3), 151-160.
[2] Bagge, L. E., Osborn, K. J., & Johnsen, S. (2016). Nanostructures and monolayers of spheres reduce
surface reflections in hyperiid amphipods. Current Biology, 26(22), 3071-3076.
[3] Banupriya, N., Saranya, S., Swaminathan, R., Harikumar, S., & Palanisamy, S. (2020). Animal detection

04 June 2024 10:17:30

using deep learning algorithm. Journal of Critical Reviews, 7(1), 434-439.
[4] Besteiro, R., Rodríguez, M. R., Fernández, M. D., Ortega, J. A., & Velo, R. (2018). Agreement between
passive infrared detector measurements and human observations of animal activity. Livestock Science, 214, 219-
224.
[5] Bíl, M., Andrášik, R., Duľa, M., & Sedoník, J. (2019). On reliable identification of factors influencing
wildlife-vehicle collisions along roads. Journal of environmental management, 237, 297-304.
[6] Bravo, M. J., & Farid, H. (2006). Object recognition in dense clutter. Perception & psychophysics, 68(6),
911-918.
[7] Camouflaged Object Sensing. Sensors, 21(16), 5455.Joseph, J., & Vineetha, S. Review on Various Object
Detection Methods in Camouflage Images.
[8] Cao, Z., Principe, J. C., Ouyang, B., Dalgleish, F., & Vuorenkoski, A. (2015, October). Marine animal
classification using combined CNN and hand-designed image features. In OCEANS 2015-MTS/IEEE Washington
(pp. 1-6). IEEE.
[9] Chabot, D., & Francis, C. M. (2016). Computer‐automated bird detection and counts in high‐resolution
aerial images: a review. Journal of Field Ornithology, 87(4), 343-359.
[10] Chandel, H., & Vatta, S. (2015). Occlusion detection and handling: a review. International Journal of
Computer Applications, 120(10).
[11] Chapelle, O., Haffner, P., & Vapnik, V. N. (1999). Support vector machines for histogram-based image
classification. IEEE transactions on Neural Networks, 10(5), 1055-1064.
[12] Chrétien, L. P., Théau, J., & Ménard, P. (2016). Visible and thermal infrared remote sensing for the
detection of white‐tailed deer using an unmanned aerial system. Wildlife Society Bulletin, 40(1), 181-191.
[13] Clevenger, A. P., & Wierzchowski, J. (2006). Maintaining and restoring connectivity in landscapes
fragmented by roads. Conservation biology series-cambridge-, 14, 502.
[14] Dollar, P., Wojek, C., Schiele, B., & Perona, P. (2011). Pedestrian detection: An evaluation of the state of
the art. IEEE transactions on pattern analysis and machine intelligence, 34(4), 743-761.
[15] Donohue, C. G., Hemmi, J. M., & Kelley, J. L. (2020). Countershading enhances camouflage by reducing
prey contrast. Proceedings of the Royal Society B, 287(1927), 20200477.
[16] Duran, D., Peng, D., Sharif, H., Chen, B., &Armstrong, D. (2007, September). Hierarchical character-
oriented wildlife species recognition through heterogeneous wireless sensor networks. In 2007 IEEE 18th

020020-6
 International Symposium on Personal, Indoor and Mobile Radio Communications (pp. 1-5). IEEE.
[17] Fang, Y., Du, S., Abdoola, R., Djouani, K., & Richards, C. (2016). Motion based animal detection in aerial
videos. Procedia Computer Science, 92, 13-17.
[18] Ferryman, J., Hogg, D., Sochman, J., Behera, A., Rodriguez-Serrano, J. A., Worgan, S., ... & Dose, M.
(2013). Robust abandoned object detection integrating wide area visual surveillance and social context. Pattern
Recognition Letters, 34(7), 789-798.
[19] Geronimo, D., Lopez, A. M., Sappa, A. D., & Graf, T. (2009). Survey of pedestrian detection for advanced
driver assistance systems. IEEE transactions on pattern analysis and machine intelligence, 32(7), 1239-1258.
[20] Gupta, P., & Verma, G. K. (2017, May). Wild animal detection using discriminative feature-oriented
dictionary learning. In 2017 International Conference on Computing, Communication and Automation (ICCCA)
(pp. 104-109). IEEE.
[21] guyen, H., Maclagan, S. J., Nguyen, T. D., Nguyen, T., Flemons, P., Andrews, K., ... & Phung, D. (2017,
October). Animal recognition and identification with deep convolutional neural networks for automated wildlife
monitoring. In 2017 IEEE international conference on data science and advanced Analytics (DSAA) (pp. 40-49).
IEEE.
[22] Haering, N., Qian, R. J., & Sezan, M. I. (2000). A semantic event-detection approach and its application to
detecting hunts in wildlife video. IEEE Transactions on Circuits and Systems for Video technology, 10(6), 857-
868.
[23] Hannuna, S. L., Campbell, N. W., & Gibson, D. P. (2005, September). Identifying quadruped gait in
wildlife video. In IEEE International Conference on Image Processing 2005 (Vol. 1, pp. I-713). IEEE.
[24] Hollings, T., Burgman, M., van Andel, M., Gilbert, M., Robinson, T., & Robinson, A. (2018). How do you
find the green sheep? A critical review of the use of remotely sensed imagery to detect and count animals. Methods
in Ecology and Evolution, 9(4), 881-892.
[25] Huerta, I., Rowe, D., Mozerov, M., & Gonzàlez, J. (2007, June). Improving background subtraction based
on a casuistry of colour-motion segmentation problems. In Iberian Conference on Pattern Recognition and Image
Analysis (pp. 475-482). Springer, Berlin, Heidelberg.

04 June 2024 10:17:30

020020-7
View publication stats