International Journal of Recent Technology and Engineering (IJRTE)

ISSN: 2277-3878, Volume-8 Issue-3S3, November 2019

Scrutiny of Methods for Image Detection and

Recognition of Different Species of Animals
Elham Mohammed Thabit A. ALSADI, Nidhal K. El Abbadi

Abstract: Animal detection-based study is useful in many It helps people to have a better understanding of living
real-life applications. Techniques involved in animal detection creatures on earth by monitoring animal movements,
are useful in observing the locomotive behavior of the engaged particularly how the animal interacts with its environment.
animal and in result it prevent harmful interruption of animals in The third element, the animal identification, is used to
residential areas. There are some branches of research in animal classify the animal that has been identified. It was used for
detection. Some of these branches will therefore be discussed in
domestic animals in the health monitoring program [22, 35].
this journal. Humans have developed many algorithms and
techniques to gain a better understanding of animal behaviour. Management of animal care becomes an important issue
Therefore, for early preventive measures, these technologies can because animals directly affect the psychological and
also serve as a warning system for humans from encroachment physical health of humans.
of dangerous wild animals. Such tasks can be reduced to three Although there are many remedies, animal abuse and animal
main branches, namely animal detection, tracking and health threats are on the rise. In order to better control
recognition. Through these papers, new approaches for study and animals in the retrieval of complex data, location tracking
a variety of technologies/algorithms implemented in the past are and RFID-based mobile monitoring system (RFID-MMS)
identified and appropriate ways for solving the research gaps are are designed to assist users over a wireless network [23].
suggested to fill the gap.
Animal identification has made it easier for people to track
Keywords: Deep Learning, Deep Neural Networks, Artificial
and control their pets. This paper will pay more attention to
Intelligence, Camera-Trap Images, Animal Detection &
Recognition.
animal detection methods and will study them. The analysis
also prohibited the use of digital images or digital video
I. INTRODUCTION techniques. In the next chapter, the review will be issued.
1.1. Animal Detection Methods in Field Of Image
Animal detection is a major and changing field due to a Processing
large number of real-life applications. Showing animal The field of Animal detection is a major area in study of
welfare on roads or in residential areas will use different various applications. Techniques of animal detection are
methods of detection for the animal and warning systems. useful in studying targeted animal's locomotive behavior
Requirements that are very relevant in real life are and also to prevent dangerous animal invasion in residential
prevention of animal vehicle collision on highways, areas. Early research on the identification of animals is to
prevention of dangerous animal invasion in residential areas, determine how easily human eyes can detect the presence of
awareness of targeted animal locomotive behaviour, etc. All animals in the natural world. Animal detection by human
of these technologies can be reduced to three fields, namely eyes was seen from the computational point of view as the
animal detection, tracking and identification. The very first most reliable method of detection [24]. This is due to the
area of animal detection was applied to different real-life ambiguity of the object form in natural objects. It is found in
applications areas. As an example, hundreds of camel- that a human observer can determine if a quickly flashed
vehicle accidents have been recorded each year resulting in animal scene includes an object as fast as 150ms. In the
numerous deaths and property damages involving millions results of median reaction time suggest a speed accuracy of
of Saudi Riyals. Using global positioning system 92 percent for 390ms reaction time and increase to 97
technology, a deployable and intelligent Camel Vehicle percent for 570ms correctness. Although human detection is
Accident Avoidance System (CVAAS) was designed to successful and achieves a satisfactory level, human eyes can
address this problem. Researchers in developed a Light quickly become tired, resulting in reduced efficacy. In fact,
Detection and Range Data Algorithm (LIDAR) to enable human eyes are unable to function 24 hours a day to identify
fishermen to find the right place for deep-sea fish [21]. animals. The use of computer vision in image processing
Scientists use micro-Doppler signals to ensure human health [18] for animal identification will counter these flaws.
and security by detecting potential dangerous animal The main task is ecology in its natural environment, that
intrusions into the residential area. follows wild animals. Speedy population growth and endless
The second branch focuses on monitoring animal growth were over-exploitation of natural resources and
locomotives and their contact with the animal tracking result in rapid, novel and significant changes to the country's
environment. One application is the development of new ecosystems. The through amount of land area, species
animal animal trace system geology applications, change patterns, environment and actions have been
monitoring and theft control and safety in zoos[36] for changed by human action. More importantly, various
sensor technology, radio frequency identification (RFID) species of animals are being thought to be extinct on earth
and global positioning system (GPS). and many are being introduced in new fields that can
damage both natural and human systems. Several
Revised Manuscript Received on November 11, 2019. contemporary techniques, including radio tracking, wireless
* Correspondence Author
Elham Mohammed Thabit A. ALSADI, College of Information
sensors and camera sensitive
Technology, University of Babylon, Iraq trap detection, have been
Nidhal K. El Abbadi, Education College, University of Kufa, Iraq produced for wild animally
monitoring [ 25 ]. Motion-led

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
151 & Sciences Publication
 Scrutiny of Methods for Image Detection and Recognition of Different Species of Animals

remote cameras or "camera traps," that have novel features emotions, along with happiness, tristy, incredulity, disgust
equipped for wide trade accessibility and ease of and surprise.
deployment and use, become a more and more common tool Despite these developments, automatic instruments to detect
for wildlife control. A standard coated camera model can facial expressions and evaluate emotional outcomes for non-
collect time, temperature and moon phase information, human primates have not been available for the tracking of
incorporated into the image data, to capture strong- animal models in mental health research. Given these
determination images both day and night [37 ]. advances, for non-human primates, manual equipment exist
In addition, generous and versatile camera setup allows for for identifying face expressions and assessing emotional
secret and consistent follow-up of animals. A camera can state, impeding animal models of mental health studies.
take thousands of images and provide a lot of information AFER decodes a range of predetermined stimulations such
when completely loaded. These standards make camera as happiness, liking, sorrow, anger, surprise, fear, and
traps a powerful tool for geologists to document all wildlife surprise. AFER decodes pre-determined emotions like joy,
aspects [27 ]. These standards are a powerful tool to sorrow, rage, disgust, confusion or anxiety for people.
environmentalists as camera traps can document all wildlife AFER schemes have variability among subjects and the
aspects. Camera trapping is being carried out quickly in the objective challenge for some adaptive states such as having
field of wildlife monitoring with digital technological to suffer from pain or depression [28] in finding correct
improvements, which produce more contemporary camera grounds for truth. AFER algorithms in humans are primarily
traps with system part automation, but reduce costs for based on models or model-free statistics. Model-based
procurement [26,38]. Although the human visual system can techniques naturally assume a pre-determined prototype of
quickly and easily process images, manually manipulating expressions and are directly linked by Action Units (AU) to
such a large amount of images is still very costly. Although decoding muscle activity blocks of facial expressions Each
the human visual system can easily and quickly process a AU has its own number of muscle movement and facial
large number of pictures manually, it is very wasteful. appearance qualities.
Furthermore, the large collections of images and ⮚ Any more organized decision-making method,
shortcomings in image quality, including experts, are an including fundamental affective recognition, could be used
important factor in the speed and precise human evaluation. by Aus . The Facial Action Coding System (FACS) is
Some photos in Snapshot Serengeti are classified as ' configured to objectively and deeply decode human facial
unidentifiable ' in particular, more than 9600 images were expression. Model-free techniques are based on the use of
marked as ' anyway ' or ' video problems, ' and thousands greatly training data sets, such as profound neural network
were marked inconsistently in the data set of Wildlife learning with pre-labeled voice inflections, with statistical
Spotter and Victoria South Central. In the past years, various machine learning instruments. The detection and recognition
governments and organizations have invested additional of animals is usually a challenging task and no special
funding in the protection of the habitats of various animal technique offers an efficient and robust solution for all
species which are at risk in their natural environment, situations. The algorithms for animal detection are
particularly by means of the construction of new facilities commonly done as a binary classification task in animal
[27 ]. When new infrastructure is growing, animal migration detection[30] . That is, the input image is divided into
corridors are challenging. This is the main reason for blocks and translated into an element in every block. A
implementing an embedded automatic system with artificial certain class is used to train features of the animal relating to
intelligence components to monitor animal activity and to a given class Model-free methods are based on the program,
provide data about animal migration in a specific area. This such as in-depth learning based on convolution neural
program will replace standard methods that could not span a network, of statistical tools with massive training sets with
period of time (direct observation, field routes, droppings, re-marked facial expressions. The detection / recognition of
etc). When planning and financing a new highway system, objects stays a difficult challenge and there is no specific
information on wildlife migration opportunities can be used approach that can address all problems in a stable and
to reliably model foraging corridors. The animal detection efficient way. Broadly speaking, animal detection
system in the desert of Mojave was used to defend algorithms is used as a binary classification task[17] for
endangered and endangered species from animals in animal detection. Then the classifier will decide if the
wildlife. The expressions of the eyes of internal emotional sample is the animal, when a new input image is given. For
states play an important role in human and other primates. the following basic applications the animal recognition
Additionally, continuous knowledge of primate facial system classification can be seen in Figure 1:
expressions may increase mental health and behavior. Facial
expressions are standard for human beings and share most
common characteristics in cultures Technologies for the
detection of human feelings are more and more automatic
and reliable due to increased computation capacity and
space availability. Non-human primates remain unable,
despite this development, to identify facial expressions and
evaluate emotional states by using automatic means which
prevent the development of animal models in mental health
researchNo instruments are available for detecting facial
expressions automatically and assessing emotional states in
semi-homan primates, in spite of such innovations, which
impede animal models for studies on psychological health.
The AFER decodes a range of predetermined people's

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
152 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-3S3, November 2019

is segmented and segmentation process cannot be

implemented without at least partial detection. The word
detection was used to refer to numerous distinct visual
capabilities which include the identification, discrimination
and categorization. In this paper we will discuss the various
techniques and methods adapted by the researchers in past
few years.

Figure 1 Classification of animal recognition approaches

[14].
⮚ Compares the animal and confirms or continues to Figure 2. Basic Objects Detection Model [13]
refuse the identity of the discovered animal (the one - to-one
matching). Even though verification and identification often II. RELATED WORK
share the same classification algorithms, the two modes are In numerous applications research has been pulled out into
intended for specific applications[17 ]. animals in image processing Many human algorithms and
⮚ Identification-compares the picture of the animal techniques have been established to better understand
with all other animals in the database and gives the matches evolutionary biology. In addition, these applications serve as
a categorized list (one to n matching). an early warning mechanism for people against harmful
⮚ To better comprehend the task and issues of animal intrusions to wild life[15].
identification and recognition, the following factors have to Boon Tatt Koik et al. (2012) Survey proposed to detect,
be considered because the outcomes of animal track and recognize animals in some of these branches.
detection and recognition system may be greatly impaired: Research based on animal detection is helpful for many real
⮚ Illumination and other image acquisition conditions life applications. Animal detection methods are useful in
-variables such as light differences in source distribution and researching the targeted animal's locomotive behavior and
intensity or camera characteristics such as sensor reaction also in preventing hazardous animal intrusion in residential
and lenses can be influenced by the animal picture input. areas. First is the issue of lighting, where a sudden shift of
⮚ Occlusions–Animal images can partially be lighting impact can influence the efficiency of identifying
overlooked by other objects and other animals the presence of animal intrusion mostly in indoor
applications. In addition, the detection may also be affected
Object detection (OD) is a technologically difficult and by the luminance issue with day-to-night modifications in
practically helpful computer vision issue and has seen the natural environment at the outdoor monitoring scheme.
important progress in recent years [1]. In computer vision Furthermore, moving background, such as such as wind
(CV) and picture processing, detection and classification of leaves could be considered as a foreground image and the
objects are significant tasks. Classification of objects could algorithms could erroneously interpret some inactive animal
be considered a normal problem of recognition of patterns. that remains static for a long time as a background image.
Unlike people who can easily recognize any object in the Visual descriptors were chosen to develop object
actual globe, it is a hard job to computerize object representation, and descriptors were classified using Support
recognition in picture. The Detection and classification of Vector Machine (SVM).
animals is a challenge for the scientist owing to numerous Ferryman, J., et al. (2013) addressed a technique of video
problems that decrease performance and effectiveness. surveillance that detects abandoned objects in surveillance
Effortlessly and instantly, humans conduct object scenes robustly and effectively. This technique used a fresh
recognition. Algorithmic description of this assignment was algorithm for threat assessment that incorporated ownership
very hard to implement on computers. notion with automatic knowledge of social relationships to
The basic model for object detection is shown in Figure 2. infer object abandonment. Implementation was
Basically an OD scheme can be readily defined by seeing accomplished by developing a logic-based Prolog-based
Figure 1, showing the fundamental phases engaged in the inference engine. The efficiency of threat detection was
object detection method. In the case of videos, the basic performed by testing against a range of data sets describing
input to the OD system may be an image or scene. The realistic scenarios and showing a decrease in the amount of
fundamental purpose of this scheme is to identify objects false alarms produced.
(animals) current in the picture or scene, or merely to Slavomir Matuska et al. (2014) A new automatic animal-
categorize the different objects (animals) into the recognition system (ASFAR) and grade system were
corresponding object classes. The issue of object detection described in this paper. AFSAR collects data from its wild
can be described. Due to an image comprising one or more nature watch tools. System detection track animals and then
interesting objects and a set of labels corresponding to a set send information to MCU. This
of models known to the system, the system is intended to information is reviewed by
assign accurate labels to the regions of the image. The MCU and animal migration
problem of object detection cannot be solved until the image corridors are built in the given

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
153 & Sciences Publication
 Scrutiny of Methods for Image Detection and Recognition of Different Species of Animals

country. In MCU, SVM classifier BOW software was observation, especially for wild kangaroos, is not as
deployed in c. This technique of classification has been effective and expensive as automated systems. A significant
tested in our experiments. This technique of classification step in experimentation with computer vision instruments is
has been tested in our studies. The research demonstrates to help study kangaroos in the wild. A kangaroo picture
that an algorithm in terms of dealing with the SURF dataset was developed to investigate feasibility from
detector, SIFT descriptor, Flann Based matcher, 18,000 gathered information from various domestic parks
clustering descriptors and 20,000 learning descriptors throughout Queensland. A Multi-position strategy was
achieved the highest grade score of 94 percent. studied, and a framework was suggested to obtain
Slavomir Matuska et al. (2014) in this paper proposed a reasonable detection precision by using the state-of - the-art
novel method based on hybrid descriptors is suggested for Deformable Part Model (DPM). Since DPM cannot detect
object recognition. This technique uses several methods kangaroos caused by various poses
(SIFT-transforming invariant-scale features, SURF-robust Kumar, S., & Singh, S. K. et al. (2016) Focused primarily
speed-up features) and consists of two components. Some on pet animal tracking in Smart City, a significant challenge
pictures from the dataset demonstrate the applicability of the for authorities. The required level of animal safety and
hybrid techniques submitted. Classes of data set include the management is not delivered for by classic processes of
wolf, fox, brown bear, wild boar and horse-drawn livestock animal identification and tracking. In this report, it
in the Slovak nation. The technique shown can also be used suggested the use of the key animal biometric identifier to
in other fields of classifying images and extracting feature. track individual animal animals (dogs). The recognition
The experimental findings indicate that local descriptor technique was used to match and classify the extracted
combinations have a positive impact on the detection of features of pet animals (dogs) with single-shoot
objects. It presents and tests two hybrid main points- impermissibility and distance metric approaches. A
detectors compared with other detectors. The BOW and prototype was developed to evaluate the accuracy of the
SVM classificator combinations have been tested. recognition system.
Experiments have indicated that an algorithm based on Tibor TRNOVSZKY et al. (2017) Proposed a device input
SISURF, Opponent SIFT Descriptor, the Brute Force for the Convolutionary Neural Network (CNN). This
matcher and 15000 descriptors in a clustering system has approach is contrasted with well-known image recognition
attained a highest grade achievement rate of 86 percent. In approaches such as Primary Binary Pattern (LBPH)
addition, in four other cycles, success rates of over 80% Histograms and System Valve (SVM) Linear Discriminate
were accomplished. The proposed SISURF hybrid key Analysis (LDA). The primary purpose is to equate the
detector produced promising outcomes similar to other key general recognition accuracy of PCA, LDA, LBPH and
point detectors. In addition, other conventional detectors SVM with the existing CNN technique. For the experiments,
were outperformed by SISURF.The proposed SISURF the wild animals database is created. This database contains
hybrid key point’s detector promising results that are 500 various subjects (5 classes/100 images per class). The
comparable to other key point detectors. wild animal database is created for the experiments. There
Peter Christianse et al. (2014) The automatic method for will be 500 subjects in this list (5 classes / 100 images per
the detection and validation of animals using thermal class). The overall performance was obtained by various
cameras in the UAV company is introduced by On the basis training pictures and test images. The experimental results
of a complex limit, hot points are identified and the indicate that the approach described has a beneficial impact
trademark extraction is carried out. The new technique of on animal recognition's overall performance and exceeds
extraction of thermal signature described in this article is the other techniques tested. The article presents an expected
extraction for each subject and a parameterisation based on CNN (PCA, LDA, SVM and LBPH) compared against well-
DCT. A grading technique is presented with measurements known algorithms, for the identification, extraction and
from single and multiple frames. With a balanced object classification of images. The proposed CNN was
classification precision of 93.5 percent in the altitude range tested in the generated animal database. For different
of 3–10 m and 77.7 percent in the 10–20 m altitude, the training images and test images the overall performance was
combination of multi-frame measurements achieves the obtained. Several learning objects and videos were used to
highest performance, thus showing a clear connection generate the general results. The results show that the LBPH
between investigator performance and altitude classification. algorithm produces the best results for large running groups
The proposed system for detecting and classification is than PCA, LDA and SVM. The requested system achieves a
based on wildlife top-view images as seen by a UAV. In precision of almost 82.5 percent for animal (cow) detection.
terms of performance of the whole algorithm the simulated The proposed system can be used with other current and
and limited dataset is favorable. The identification and active pedestrian and car detection technologies to prevent
classification model proposed is based on highest level-of - accidents and loss of human life on roads as a full solution
the-range images seen by a wildlife UAV. The automatic (Package)[19].
tracking and recognition of wildlife by UAV software is Jaskó, G., et al. (2017) provided a system capable of
primarily part of ongoing wildlife-friendly agriculture study. identifying from traffic scenes various enormous wild
In the paper, a concept is proposed for CNN (PCA, LDA, animals. Visual information from a monocular color vision
SVM, and LBPH) in contrast to the well-known frame camera was acquired. The goal was to evaluate the image of
recognition, extraction and object classification. The the traffic scene, identify the areas of concern and classify
suggested CNN was analyzed on the animal database them properly for the discovery of the animals that were on
created. the highway and could trigger
Zhang, T., et al. (2015) addressed the problem of current an accident. Using intensity,
research in kangaroo such as population tracking and color and orientation features, a
monitoring of activities. However, traditional manual saliency map was generated

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
154 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-3S3, November 2019

from the image of the traffic scene. It was presumed that the Gyanendra K. Verma et al. (2018) focus on tracking and
outstanding areas of this map were areas of concern. A analyzing animal wildlife through the identification of
database of a big amount of pictures comprising multiple natural scenes from camera trapping systems. The camera
wild animals with four legs was collected. A database was trap picture sequences consist of extremely encroached
compiled from a large number of images containing various pictures which prevent animal detection leading to low
four-legged wild animals. Relevant features were extracted detection rates and high false detection rates. We used a
from these and were utilized for training Support Vector camera trap database that has animal proposals candidates
Machine (SVM) classifiers. These classifiers provide an with a multilevel graph cut in the spatiotemporal domain to
accuracy of above 90% and are used to predict whether or deal with this issue. These proposals are used to produce a
not the selected regions of interest contain animals. check stage which will determine whether a particular patch
Hung Nguyen et al. (2017) proposed and demonstrated the is animal or background. In this study we have taken animal
feasibility of constructing a scalable automated wildlife movement and the spatial context into account in developing
surveillance system with a deep learning strategy. In order candidate animal areas, using the IEGC method for
to inform conservation and management decisions, efficient analyzing the camera-trap picture sequences. In addition, we
and reliable surveillance of wild animals in their natural have noted that the picture functions of the Deep
habitats is essential. Use the Wildlife Spotter dataset in this Convolutionary Neural Network (DCNN) function well and
paper, which includes a big amount of pictures taken by trap enhance classification efficiency using separate classifiers.
cameras in South Central Victoria, Australia. In recognizing We have intended a solid and reliable camera-trap picture
pictures with animals, this model accomplished more than model that is extremely dynamic and cluttered.
96 percent and nearly 90 percent in identifying three most Eric T. Psotaet al. (2019) introduced a fully convolution
common animals (bird, rat, and bandicoot). neural network used to identify the location and orientation
Rey N., et al. (2017) investigated a semi-automatic system of multiple group-housed pigs. The target output of the
capable of detecting large mammals in semi-arid Savanna network is goal is to display an image space of each part of
was explored. It uses machine learning to rely on an animal the pig along with a method for associating them to form
detection scheme, trained with crowd-sourced annotations complete instances. A new dataset with 2000 images and
supplied by volunteers who manually interpreted color 24,842 pig cases is used to train the network. The dataset is
pictures with sub-decimeter resolution. A high recall rate divided into a training set and a testing set, and the testing
has been accomplished for the scheme and a human operator set is subdivided into two sets: one with images depicting
can then with restricted effort remove false detections. The the same environments as the training set, and another with
system offered excellent prospects for developing data- images of new environments not represented in the training
driven wildlife conservation management methods. set. This dataset design allows the robustness of detection
Tarrit, K., et al. (2018) addressed the significance of visual algorithms to be tested against novel animal presentations
surveillance in public spaces and especially in train and and environments. The three key contributions of this work
subway platforms that are particularly vulnerable to many are:
kinds of crime from small-scale theft to terrorist activity. If (1) A fully convolution instance detection method,
image quality cannot be enhanced by means of high- (2) A public dataset for training and evaluation, and
resolution sensors, high-end lenses or IR illumination, the (3) The design and/or training methodology should be
visual surveillance system may need to increase the image revised to improve the robustness to environmental
resolution capacity by software to provide accurate outputs variability.
such as disappearing points (VPs) in our case. In this work, Regarding the third interpretation, it is not possible for
we provided a three-stage technique for accurately detect the spatial transformation to mimic changes in luminous
primary lines and vanishing points (VPs) in low-resolution conditions while significant increase has been applied to the
images obtained in indoor and outdoor railway platform input and output images during training. Therefore, a new
settings through visual surveillance systems. Initially, set of non-uniform color-space transformations may provide
several frames were used by a multi-frame picture enhancer a solution that improves the robustness of the trained
to increase the resolution capacity. Adaptive edge detection network.
was then performed and a new line clustering algorithm was 2.1. Animal Detection Systems and Algorithms
applied to determine the parameters of the lines that Comparison
converge at VPs; this used the detected lines and heuristics An in-depth literature analysis brought forward many
statistics about the type of scene. Finally, gaps were innovative animal detection systems / algorithms. Although
calculated through a voting system to optimize detection in the number of publications on this particular topic is smaller
an effort to omit spurious lines. The proposed approach is than that of vehicle detection (another important factor of
very robust as it is not affected by ever-changing road safety), we found a variety of animal detection
illumination and weather conditions of the scene, and it is algorithms. The figure below illustrates different techniques
resistant to vibrations. used so far to identify animals. This involves the
identification and tracking of endangered species.

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
155 & Sciences Publication
 Scrutiny of Methods for Image Detection and Recognition of Different Species of Animals

Figure 3 Different techniques used for detecting

Table 1 Comparison of different images and videos based object detection techniques

Ref No Title Attributes Merits Demerits

Animal Recognition and Robust, stable and Low accuracy

Identification with Deep appropriate for
Convolutional Neural Networks Bird, Rat, capturing wild images
for Automated Wildlife Bandicoot, Rabbit,
[1] Monitoring Mouse, Cat etc.

Detecting Kangaroos In The Kangaroo dataset Reasonable detection Low precision

Wild: The First Step Towards accuracy
[2] Automated Animal
Surveillance

Monitoring of pet animal in Pet animal (dogs) Improve the Does not used the real
smart cities using animal recognition accuracy data of pet animals
[3] biometrics

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
156 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-3S3, November 2019

Detecting animals in African Kuzikus dataset High recall rate Low precision value
Savanna with UAVs and the
[4] crowds

Only low-frequency
Vanishing point detection for 18 rail platform Very good results for information of the
visual surveillance systems in environments detecting vanishing images is used.
[5] railway platform environments points in Manhattan-
like scenes

Problem occur when the

same animal crosses
Detection of Cattle Using Cattle detection Improve the tracking
Drones and Convolutional accuracy several times on the path
[6] of the multicolor.
Neural Networks

Animal Intrusion Detection Null, Human being Improve the accuracy Low accuracy of animal
Based on Convolutional Neural and Animal of detection detection
[7] Network

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
157 & Sciences Publication
 Scrutiny of Methods for Image Detection and Recognition of Different Species of Animals

Classification of Wild Animals Wild boar, brown Promising outcomes Poor outcomes with a
Based on SVM and Local bear, wolf, fox and comparable to other classification success
Descriptors deer important point rate of only around 50%.
detectors
[8]

Towards Automatic Wild

Animal Monitoring:
[9] Identification of Animal Snapshot Serengeti High performance in
Species in Camera-trap Images dataset classification
using Very Deep Convolutional Poor illumination
Neural Networks

Table 2: Algorithms for detecting animals

Reference Algorithms and Techniques Implementation
[10] Foreground-background segmentation and object verification Detects animals in highly cluttered
videos

[11] Fusion of deep learning and HOG features Robust algorithm Detects animals in
highly cluttered natural images
[12] Deep CNN with machine learning Accuracy of 91% and works for both
daytime and nighttime

Table 3: Performance comparison of above mentioned algorithms

Study Year Method Precision F-score
Zhi Zhang et al. [10] 2015 Ensemble graph cuts for object 0.8293 0.8695
classifier as foreground-background
segmentation followed by object
verification
Zhi Zhang et al. [11] 2016 Combination of deep learning and 0.8209 0.8398
HOG features
Animal Detection 2017 Deep CNN features with machine 0.91625 0.9476
with deep CNN [12] learning

2.2. Research Gap ⮚ Animal detection is comparatively easier if the

The various practical solutions available for automatic device is looking for an animal (say cat) to be detected.
detection and recognition of animal for highways are going However, inherently acknowledging many animals needs
on. These are the gaps we found during automatic animal the ability to distinguish between one animal and the other,
detection: although they may not be of the same sort. Such a issue for
⮚ Animals in distinct forms, dimensions, stances, computers is very hard if they do not understand about
colors, and conduct are not completely predictable in animals' different opportunities.
comparison with walking or other objects, which are the ⮚ Given the complications apart from a moving or
biggest problem. stationary animal, along with the interest involved in
⮚ Although human beings have a beautiful basic identifying the animal on the
shape and size, animals do not share the average and roads, there are other objects
standard. such as vehicles, pedestrians,

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
158 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-3S3, November 2019

shade of trees and other objects that make the animal very [3] Kumar, S., & Singh, S.K. (2016). Monitoring of pet animal in smart
cities using animal biometrics. Future Generation Comp. Syst., 83,
difficult to detect.
553-563.
[4] Rey, Nicolas & Volpi, Michele & Joost, Stéphane &Tuia, Devis.
III. CONCLUSION (2017). Detecting animals in African Savanna with UAVs and the
crowds. Remote Sensing of Environment. 200. 341–351.
We reviewed the literature on the automatic detection and [5] Tarrit, Katy &Molleda, Julio & Atkinson, Gary & Smith, Melvyn &
classification of animal image. We faced many problems C. Wright, Glynn &Gaal, Peter. (2018). Vanishing point detection for
during the detection and image classification of animal such visual surveillance systems in railway platform environments.
Computers in Industry. 98. 153-164.
as large variation in shape and color appearances,
[6] Rivas, Alberto &Chamoso, Pablo & González Briones, Alfonso
light/illumination conditions problem. In recent years, with &Corchado Rodríguez, Juan. (2018), “Detection of Cattle Using
the increasing importance of animal identification and Drones and Convolutional Neural Networks,” Sensors, 18.
tracking, animal recognition algorithms have received more [7] W. Xue, T. Jiang and J. Shi, "Animal intrusion detection based on
and more attention, and the application of neural networks convolutional neural network," 2017 17th International Symposium
on Communications and Information Technologies (ISCIT), Cairns,
has opened up new and individuals. This paper uses this QLD, 2017, pp. 1-5.
principle to use convolutional neural networks for animal [8] SlavomirMatuska, Robert Hudec, PatrikKamencay, Miroslav Benco,
identification studies. The experimental results show that the Martina Zachariasova, Classification of Wild Animals based on SVM
optimal feature extraction algorithm and optimization and Local Descriptors,AASRI Procedia, Volume 9, 2014, Pages 25-
30, ISSN 2212-6716.
algorithm are obtained. It is hoped that later researchers will [9] Alexander Gomez Villa, Augusto Salazar, Francisco Vargas,Towards
be able to conduct more in-depth research on this basis to automatic wild animal monitoring: Identification of animal species in
further improve the performance of animal recognition camera-trap images using very deep convolutional neural networks,
algorithms based on neural networks for animal Ecological Informatics,Volume 41, 2017, Pages 24-32, ISSN 1574-
9541.
identification. Animal characteristics help to better describe
[10] Z. Zhang, Z. He, G. Cao and W. Cao, "Animal Detection from Highly
and identify different species. However, there still exist Cluttered Natural ScenesUsing Spatiotemporal Object Region
research gaps for the further improvement and innovation in Proposals and Patch Verification," in IEEE Transactions on
this area. Multimedia, vol. 18, no. 10, pp. 2079-2092, Oct. 2016.
[11] Z. Zhang, T. X. Han and Z. He, "Coupled ensemble graph cuts and
object verification for animal segmentation from highly cluttered
IV. FUTURE SCOPE videos," 2015 IEEE International Conference on Image Processing
(ICIP), Quebec City, QC, 2015, pp. 2830-2834.
Most animal detection systems are found to have their own
[12] Verma G.K., Gupta P. (2018) Wild Animal Detection Using Deep
advantages and disadvantages. Yet in a country like India, Convolutional Neural Network. In: Chaudhuri B., Kankanhalli M.,
given the growing number of accidents with stray animals, Raman B. (eds) Proceedings of 2nd International Conference on
there is no realistic implementation of such systems. Future Computer Vision & Image Processing. Advances in Intelligent
research needs to expand in order to develop a device Systems and Computing, vol 704. Springer, Singapore.
[13] Sharma, K.U. and Thakur, N.V., 2014. An Approach For Single
capable of detecting animals during the day, at night and Object Detection In Images. International Journal of Image
also in fog. In foggy weather, none of the existing systems Processing (IJIP), 8(5), p.278.
promised to detect animals. In fact, most AVCs (Animal- [14] Kumar, S., & Singh, S.K. (2016). Monitoring of pet animal in smart
Vehicle Collision) occur when dogs or cows strike the cities using animal biometrics. Future Generation Comp. Syst., 83,
vehicle in urban environments, so we need to concentrate on 553-563.
[15] Sharma S and Shah D, 2013, A brief overview on different animal
identifying animals like dogs and cows instead of moose or detection method, Signal and Image Processing: An International
giraffe. The effectiveness of a system depends on two Journal, vol. 4, pp: 77- 81.
factors: the reliability of detection and the signaling of the [16] Verma G.K., Gupta P. (2018) Wild Animal Detection Using Deep
animal to the driver. In terms of detection accuracy, the Convolutional Neural Network. In: Chaudhuri B., Kankanhalli M.,
Raman B. (eds) Proceedings of 2nd International Conference on
technologies / algorithms studied above indicate that a
Computer Vision & Image Processing. Advances in Intelligent
rained DCNN performed quite well by giving approximately Systems and Computing, vol 704. Springer, Singapore
91% precision, 91-93% true positives and 6-10% false [17] Trnovszky, Tibor &Kamencay, Patrik&Orješek, Richard &Benco,
positive rate[16]. On both day and night, it performed well. Miroslav &Sykora, Peter. (2017), Animal Recognition System Based
In contrast, RADAR is also capable of working in harsh on Convolutional Neural Network. Advances in Electrical and
Electronic Engineering. 15. 10.15598/aeee.v15i3.2202.
weather. Those two can be further expanded to make it work [18] Ran, Y., Weiss, I., Zheng, Q. and Davis, L.S., 2007. Pedestrian
on highways (background continues to change) and in real detection via periodic motion analysis. International Journal of
time. The most important fact to communicate with the Computer Vision, 71(2), pp.143-160.
driver is to tell well in advance so that the driver has ample [19] S. U. Sharma and D. J. Shah, "A Practical Animal Detection and
Collision Avoidance System Using Computer Vision Technique,"
time to take the action. All of these criteria have created a
in IEEE Access, vol. 5, pp. 347-358, 2017.
large area for further animal detection and alarm system [20] Manjot Kaur and Rajneesh Randhawa, (2019) Animal Detection:
work to be applied on a large scale. Techniques, Challenges and Future Scope, International Journal of
Innovative Technology and Exploring Engineering (IJITEE), vol.9.
REFERENCES [21] Sharma, S.U. and Shah, D., 2017. DESIGN AND DEVELOPMENT
OF ANIMAL DETECTION ALGORITHM USING IMAGE
[1] H. Nguyen et al., "Animal Recognition and Identification with Deep PROCESSING.
Convolutional Neural Networks for Automated Wildlife [22] Neethirajan, S., 2017. Recent advances in wearable sensors for animal
Monitoring," 2017 IEEE International Conference on Data Science health management. Sensing and Bio-Sensing Research, 12, pp.15-29.
and Advanced Analytics (DSAA), Tokyo, 2017, pp. 40-49. [23] Pathare, S.P., 2015. Detection of black-backed jackal in still images
[2] Zhang, T., Wiliem, A., Hemsony, G., & Lovell, B.C. (2015). (Doctoral dissertation, Stellenbosch: Stellenbosch University).
Detecting kangaroos in the wild: the first step towards automated
animal surveillance. 2015 IEEE International Conference on
Acoustics, Speech and Signal Processing (ICASSP), 1961-1965.

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
159 & Sciences Publication
 Scrutiny of Methods for Image Detection and Recognition of Different Species of Animals

[24] Koik, B.T. and Ibrahim, H., 2012. A literature survey on animal
detection methods in digital images. International Journal of Future
Computer and Communication, 1(1), p.24.
[25] Figueroa, K., Camarena-Ibarrola, A., García, J. and Villela, H.T.,
2014, November. Fast automatic detection of wildlife in images from
trap cameras. In Iberoamerican Congress on Pattern Recognition (pp.
940-947). Springer, Cham.
[26] G. C. White and R. A. Garrott, Analysis of wildlife radio-tracking
data. Elsevier, 2012.
[27] A. Swanson, M. Kosmala, C. Lintott, R. Simpson, A. Smith, and C.
Packer, “Snapshot Serengeti, high-frequency annotated camera trap
images of 40 mammalian species in an African savanna,” Scientific
Data, vol. 2, p. 150026, 2015.
[28] M. S. Bartlett, G. C. Littlewort, M. G. Frank, and K. Lee, “Automatic
decoding of facial movements reveals deceptive pain expressions,”
Curr. Biol., vol. 24, no. 7, pp. 738–743, 2014.
[29] Blumrosen, G., Hawellek, D. and Pesaran, B., 2017. Towards
automated recognition of facial expressions in animal models. In
Proceedings of the IEEE International Conference on Computer
Vision (pp. 2810-2819).
[30] Cai, C. and Li, J., 2013, October. Cattle face recognition using local
binary pattern descriptor. In 2013 Asia-Pacific Signal and Information
Processing Association Annual Summit and Conference (pp. 1-4).
IEEE.
[31] Koik, B.T. and Ibrahim, H., 2012. A literature survey on animal
detection methods in digital images. International Journal of Future
Computer and Communication, 1(1), p.24.
[32] Christiansen, P., Steen, K., Jørgensen, R. and Karstoft, H., 2014.
Automated detection and recognition of wildlife using thermal
cameras. Sensors, 14(8), pp.13778-13793.
[33] Nguyen, H., Maclagan, S.J., Nguyen, T.D., Nguyen, T., Flemons, P.,
Andrews, K., Ritchie, E.G. and 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.
[34] Psota, E.T., Mittek, M., Pérez, L.C., Schmidt, T. and Mote, B., 2019.
Multi-Pig Part Detection and Association with a Fully-Convolutional
Network. Sensors, 19(4), p.852.
[35] Jukan, A., Masip-Bruin, X. and Amla, N., 2017. Smart computing and
sensing technologies for animal welfare: a systematic review. ACM
Computing Surveys (CSUR), 50(1), p.10.
[36] Anu, V.M., Deepika, M.I. and Gladance, L.M., 2015, February.
Animal identification and data management using RFID technology.
In International Confernce on Innovation Information in Computing
Technologies (pp. 1-6). IEEE.
[37] Zhu, C., Li, T.H. and Li, G., 2017. Towards automatic wild animal
detection in low quality camera-trap images using two-channeled
perceiving residual pyramid networks. In Proceedings of the IEEE
International Conference on Computer Vision (pp. 2860-2864).
[38] Giraldo-Zuluaga, J.H., Salazar, A., Gomez, A. and Diaz-Pulido, A.,
2017, November. Recognition of Mammal Genera on Camera-Trap
Images Using Multi-layer Robust Principal Component Analysis and
Mixture Neural Networks. In 2017 IEEE 29th International
Conference on Tools with Artificial Intelligence (ICTAI) (pp. 53-60).
IEEE.

Retrieval Number: C10461183S319/2019©BEIESP Published By:

DOI: 10.35940/ijrte.C1046.1183S319 Blue Eyes Intelligence Engineering
160 & Sciences Publication