10 X October 2022

https://doi.org/10.22214/ijraset.2022.47012
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

Animal Detector System for Forest Monitoring

Using OpenCV and Raspberry-pi
Aniket Gat1, Hrishikesh Gaikwad2, Rahul Giri3, Dr. Mohini P Sardey4, Milind P Gajare5
1, 2,
3Student, Department of Electronics and Telecommunication, AISSMS IOIT, Maharashtra, India
4
Associate Professor, AISSMS IOIT, Maharashtra, India
5
Assistant Professor, AISSMS IOIT, Maharashtra, India

Abstract: Today, even without the help of technology, monitoring wildlife has become a challenging job. To address this
problem, we have developed a solution for animal classification cameras that can detect and store data of these animals in SD
card. The system can also be used to protect wildlife in zoos. The system is also able to identify new species. When a new species
is discovered, it stores its data in a separate database for further study. It is also possible to control the system remotely if there is
internet in the area.
The system uses Python-based code that includes pretrained classification models. We have used the pi camera to feed live
footage to the Raspberry Pi. It then uses the OpenCV module to segment the image into frames and compares the obtained
frames with the pretrained module and gives labels accordingly. The system and stored data can also be accessed through the
mobile application Rasp Controller and through computer using the software named VNC viewer. The system is also able to
move from one place to another and also as it is a part of whole system it is also remotely monitored.
Keywords: Raspberry-pi, OpenCV modules, Rasp Controller, VNC viewer

I. INTRODUCTION
Wildlife monitoring is very important these days as many animal species are threatened with extinction, some of which have already
disappeared. Therefore, it is necessary to keep records of wild animals. But keeping these records becomes a daunting task without
the help of technology. Traditional methods, such as setting up camera traps in the forest, manually clicking on photos, and then
classifying and tagging animals by looking at the images, require too much time and effort. Also, it is impossible for a person to
take pictures in the forest around the clock.
With modern techniques such as machine learning, we can classify animals based on images and videos taken by them. Information
can also be stored in the system.
Therefore, we developed an animal classification system in the project. Our devices are able to identify animals by capturing real-
time images of the animals and storing the data in a database.
We have used Python Idle of Raspbian OS, Raspbian OS Buster, OpenCV module, pretrained detection model and Raspberry Pi,
camera as hardware. The objectives of this work are aiming to classify animals using their images/ live video and store the
respective data to help forest monitoring process

The objectives of this work are:

1) Identify the animal and save the data.
2) To store the information of new creature/species in the system.
3) To remotely manipulate the system.
4) To move system from one place to another

The system consists of five main parts:

a) Connect the SD card to the Raspberry Pi.
b) Connect the camera module to the Raspberry Pi.
c) Take videos/pictures of animals and categorize them accordingly.
d) View the output on a computer or smartphone and save the information.
e) Remotely access the system and obtain information (depending on the network).

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 477
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

This type of system is useful for monitoring wildlife and also contribute to wildlife conservation in zoos. It is also useful to maintain
safety of animals in zoos as well as domestic animals also, there are so many threats for the domestic animals such as attack by other
wild animals, person trying to steal them etc.
This paper is organized as follows: The section 2 discusses the literature survey. In section 3, Machine learning is discussed
followed by OPEN CV in section 4. Raspberry Pi module description is provided in section 5. The proposed system is discussed in
section 6 followed by methodology in section 7. Results are discussed in section 8. The paper is concluded in section 9.

II. LITERATURE SURVEY

Slavomir Matuska et al.,[1] proposed a device referred to as ASFAR (automated system For Animal recognition) is primarily based
on disbursed so-called `looking tool' in targeted location and important computing unit (MCU) performing as server and system
manager. Looking devices are located in wild nature and their challenge is to discover animal and then ship facts to MCU to
assessment. The primary venture of whole machine is to decide migration corridors of untamed animals in specified area. To create
object representation, visible descriptors were chosen and assist Vector gadget (SVM) turned into used to categories descriptors.
Priya Sharma et al., [2] proposed the system which can protect the crop from animals those who can harm the crop. The proposed
system uses raspberry-pi and PIR sensor. The PIR sensor detects the motion of the animal and then camera starts to capture the
images. These images are sent to raspberry-pi for the classification of the animal. According to the detected animal the appropriate
action is taken.
Morrow-Tesch J et. al. [3] has developed a video data base system to kick off a system that uses a multi-object tracking and
reasoning system to extract animal motion information from an input animal activity video clip. The material will eventually be
examined and summarised using standard animal behaviour definitions.
DP Gibson et. al. [4] proposed a system, from a minimal set of monitored data, a system that recognises gait and quadruped
structure Because the motion information in this work is generated from dynamic wildlife film footage, it is incredibly complex and
loud. Gait analysis is performed on footage containing quadrupeds that typically walk in silhouette towards the camera, but this part
of the system is poseagnostic and often useful for gait detection. The main motion is considered to be produced by the background
and its relationship to the camera motion, so it can be removed as a first step. Together with the frequency analysis, eigenmodes are
used as templates to synchronize point clusters and establish the underlying spatiotemporal structure. Given this synchronized
structure, further tracking observations were used to deform the structure to better match the overall motion. They show that using
the eigengait model enables spatiotemporal localization of walking animals, helping to overcome difficulties caused by occlusions,
tracking errors, and noisy measurements
T Burghardt and J Calic [5] focused on algorithm detection system which uses Haar-like features and AdaBoost classifiers to mimic
a human face detection method. Face tracking is accomplished with the Kanade-Lucas-Tomasi tracker and a special interest model
applied to the detected face.
Matthias Zeppelzauer [6] proposed a system which is vigorous to impediments (e.g., by vegetation) and accurately handles camera
movement and different lighting conditions. There system show that both allover far-off elephants can be distinguished and
followed dependably. The proposed technique empowers scientists productive and direct admittance to their video assortments
which works with additional social and environmental examinations. The strategy doesn't make hard limitations on the types of
elephants themselves and is in this way effectively versatile to other creature species.
Prajna. P et. al. [7] have worked to design a system that will monitor the field. That is, it will first use a sensor to detect trespass
surrounding the field, then use a camera to capture the intruder's image and categorise them using image processing, and finally take
appropriate action based on the type of invader and at the last it will send notification to the farm owner.
Rasool and Moorthy [8] proposed a system which highlights the issue of safety and security of animals in the zoological parks. The
proposed system uses Raspberry-pi, pi-camera and OpenCV platform to classify the data. The camera takes the footage from the
cage or from a required area and forward it to the raspberry-pi and the data is classified using OpenCV. Deep Zhou, [9] has
proposed a system for real time animal detection with criteria such as detection accuracy, detection time, and system energy
consumption. They have used the LBP adopting AdaBoost algorithm in the first stage. Using Two HOGSVM based classifiers, the
second stage rejects the false positive ROIs.
Sk. Almas Tabassum et. al. [10] proposed the system which prevent the crops from animals. Sometimes animals like cow, buffalo,
elephants etc. enters in crop yield and by running here there or eating the crop keeps vandalizing it. So, this system detects the
animals and send the photo of detected animal to the farmer also it continuously keeps on making the buzzer sound until the animal
is in front of camera.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 478
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

In summary, many researchers have proposed their work for animal monitoring and detection. Most of the work carried out in
literature focuses on controlling in a specific geographical area within a certain range. Our proposed system can be controlled
remotely from any place of the world using pitunnel that allows raspberry-pi to be controlled from any host or network. Some of the
machine learning algorithms used cannot work properly on raspberry-pi as they require high computing power but they are accurate,
they might take time to classify but the time varies from model to model. There are other models that require very less time but are
not accurate. So, to meet these both requirements i.e., computing power and the time required we are using ssd-Mobilenet v3.

III. MACHINE LEARNING

Machine learning algorithms study the historical data to predict the new output. Machine learning algorithm uses such historic data
which can be termed as training data to build a model which is going to be really helpful in predictions and to take decisions
accordingly. There are three most important machine learning methods: supervised learning, unsupervised learning, and
reinforcement learning. Supervised learning: Supervised learning is a method of machine learning in which already labelled data is
provided as input. Based on these inputs the model predicts the output when an unlabeled data is given. From the total available data
20% data is used for testing purpose and 80% data is used for training the model. Unsupervised learning: In this method of machine
learning, no trained data or labelled data is provided as an input. Instead of labelled data unlabeled data is directly given to the
model. In this method the machine combines the unsorted information according to the similar patterns and any other similarities.
Here no trainer is provided like as in supervised learning. Reinforcement learning: Reinforcement learning has a learning agent.
This type of machine learning need feedback as input received from the learning agent. As the name suggests the learning agent gets
trained on its own and it learns from the environment. Some points are credited to this learning agent if the decision taken is correct.
More the points more is the accurate the model will be.
IV. OPENCV
OpenCV (Open-Source Computer Vision Library) is an open-source library for machine learning and computer vision. OpenCV has
more than 2500 algorithms which also includes machine learning and computer vision algorithms. These algorithms are able to
detect the objects and also detect and identify human faces, detects motion of a moving object and many other operations related to
computer vision and machine learning. That’s why OpenCV library plays an important role in our system.
OpenCV library supports languages like Python, Java and C++ and MATLAB interface. OpenCV can also be used on operating
systems like Linux, MAC OS, Windows and Android. OpenCV library is written natively in C++. OpenCV library can run on any
device that can execute C program. As it is written in C++ the execution of OpenCV is quite fast also it uses low RAM. OpenCV
library has more than 400 image-processing functions which are most useful for computer vision and machine learning applications.
OpenCV works on BGR (Blue Green Red) color format.

V. RASPBERRY-PI
The Raspberry Pi is a cheap, small-sized computer but have low performance compared to laptops and desktops. Raspbian OS is the
official OS which is installed in this small computer. This small sized computer is able to perform all the operations that a computer
performs in a day-to-day life. Also, this computer can be used for machine learning, computer vision, real time applications and
many more. As it is compact in size it is possible to reduce the hardware in the robotics applications.
Raspberry Pi comes with ram size ranging from 512mb up to 8 GB depending upon the model of Raspberry pi. Raspberry pi has a
clock frequency/CPU speed which can ranges from 700MHz up to 1.2GHz again depending on the model of Raspberry pi. The
figure 1 shows the Raspberry Pi module.

Fig.1 Raspberry-pi 4B board

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 479
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

VI. PROPOSED SYSTEM

The figure 2 shows the block diagram of proposed system. After setting up the raspberry-pi for wireless or wired manner, and after
coding for the required operations to be executed the raspberry-pi and the whole system is all set to execute the operations. The
system is initialized after booting up of raspberry-pi. Locate the code and run the program.
At very first step pi-camera will initialized and will start to stream the real time video. If there is nothing in front of the camera no
data will be stored and no images will be captured. Once the identical animal is in the frame of pi-camera the video will be
processed using OpenCV library and after processing the video the obtained output will be compared with the pretrained model’s
dataset and it will label the animal accordingly. The information i.e., the name of animal and the time and date of appearance will be
stored in the form of text in a new text file. Also, the images of animals will be captured and the images will be stored in a separate
folder. Along with the above process the information i.e., the name of animal and the real time video will be displayed on the screen
at the same time.
If the animal appeared in the camera frame is not identical i.e., it is not detected then using some program for motion detection its
information will also be stored in the separate folder named as ‘New Animals’. Motion detection program is used because if animal
is not detected by detecting any kind of motion, the system will capture images. Images will be captured only if the animal is not
detected and there is certain movement in the camera frame. These images can be studied later to get the more information about the
unidentified animals. This can be useful for the wildlife researchers if any new species or animal is found. This system can be
controlled remotely using the pitunnel (www.pitunnel.com). Pitunnel allows to control raspberry-pi remotely, hence it is not
necessary to have a common network connectivity between PC and raspberry-pi. The proposed system also consists of a moving
robot which allows whole system to travel. As it is a part of whole system it is also controlled remotely.

Fig. 2. Block Diagram

A. Algorithm
1) Taking Image/video as input from the pi-camera.
2) Converting input into necessary formats/frames which are required for OpenCV.
3) Comparing the obtained output after processing with the pretrained model.
4) Storing the information about the detected animal like time of occurrence, images etc in a separate folder.
5) Storing the information of undetected animals in a separate folder named as ‘New Animals’.
6) Displaying the obtained results after performing operations on the camera inputs using python code on the smartphone/ mobile
(using Rasp controller and VNC viewer).

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 480
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

B. Flowchart

VII. METHODOLOGY
A. Step 1: Set up Raspberry-pi
Set up the raspberry-pi using Raspberry pi imager software which is available on www.raspberrypi.com To set up Raspberry-pi a
SD-card of 16 GB or more than that storage capacity should be used for better performance. There are number of Raspberry-pi OS
are available install the required OS. The ‘Raspbian OS
Buster’ is used in our system.
Insert the micro-SD card in the pc (Format the SD card before using it for set up). Open Raspberry-pi imager and select the OS.
Select storage type as your inserted SD card. Click on the gear icon and make necessary settings there (if you are not using any
external monitor for set up). Click on the write button and OS will be installed on the SD card. Insert the SD card in Raspberry pi
and power on.
(Note: Need not to follow Step 2 if external monitor is connected to Raspberry-pi)

B. Step 2: Set up Raspberry-pi (headless set up using VNC viewer)

If an external monitor is not available, it is possible to control the Raspberry-pi wirelessly. After selecting the storage type click on
the gear icon and give the information about Wi-Fi available in the room, select the Wi-Fi country so that when the pi will boot up it
will connect automatically to that Wi-Fi. This is necessary to make a connection with raspberry-pi using our PC. The default host
name of Raspberry-pi is raspberrypi. local, it is possible to change the host name while setting up. Open the command shell as
admin in the PC and type command ‘ssh -p 22 pi@raspberrypi.local’, then it will ask to join type ‘Yes’ and it will ask for the
password the default password is ‘raspberrypi’, password can also be changed at set up. After hitting enter PC will log in the
raspberrypi command shell. Type ‘raspi-config’ and select ‘Interface options’ and enable the VNC option there. Reboot raspberry-
pi, open VNC viewer software installed in pc, click new connection and enter hostname it will again ask for password type the
password and the screen of raspberry-pi will be visible in the VNC.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 481
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

C. Step 3: Install OpenCV, pandas, NumPy, ssd-Mobilenet v3

Install the most important library OpenCV for the detection. Install other required libraries like pandas, NumPy. Most of the times
they are pre-installed, if they are not install using the command shell. Download the pretrained model ssd-Mobilenet v3 which has
data of no of objects and few animals.

D. Step 4: Coding
Prepare a code for animal detection in such a way that it will process the image/video sent by camera. After processing the
processed image should compared with the pretrained model and if animal is detected the data should be stored accordingly and if
not, then its data should be stored separately.

VIII. RESULTS AND DISCUSSION

The input given to the system is in the form of image/video obtained using pi-camera. After processing the video, the video is
further compared with the model and the name of animal is given as output. The data of the detected animals is also stored in text
format in a new text file. The text file contains information like time and date of the animal detected and also the name of that
animal. This kind of information is useful for further study and monitoring of the forest area. One of the most important features of
the system is that it can be remotely monitored.
As the system uses ssd-Mobilenet v3 the system is sometimes not accurate as expected but the system performs faster in the
detection process. The main reason to use ssd-Mobilenet v3 is that its lightweight and the computational power of raspberry-pi is
also low compared to the normal computer. Hence ssd-Mobilenet performs better on raspberry-pi than other detection models. ssd-
Mobilenet v3 takes less time for the detection purpose. As shown in the figure 3, it is displaying real time image of the animal
appeared in the camera frame and also the name of animal is displayed along with box. The box indicates that the animal is detected.
The same image is captured and stored in the folder of SD card.
Also, name of animal i.e., ‘Elephant’ here is stored in the text file along with the timestamp.

Fig.3. Output shown as Elephant detected

The figure 4 shows the movable robot attached along with the system. So that, it can travel a particular distance. The purpose of this
robot is to study the wild animals closely so that the researchers can extract more information about that animal. The robot consists
of 2 DC motor, L293d motor driver ic and a robot chassis.

Fig.4. System with a movable robot

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 482
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue X Oct 2022- Available at www.ijraset.com

IX. CONCLUSION AND FUTURE SCOPE

It is observed from the above experimentations, the system is capable of detecting animal and storing its data. Even if the animal is
not detected system can store its images for the further studies and research. The system can be controlled remotely using PC and
also smartphone.
The proposed system is prepared in a such a way that it will be very helpful for the wildlife researchers and the forest officers. They
will be able to keep records of the animals in the forest without any extra efforts. The traditional way of doing this kind of jobs is a
bit harder. The above discussed system will reduce their efforts. Along with the researchers and the forest officers the system can
also help farmers, those who allow their domestic animals to live in an open space having a large area and compound. The system
will monitor each and every action of animals in the area so that any outsider will not harm them or any animal will not escape from
that open space.
After doing some modifications in the system it can be used as an alert device for the farmers. The system will alert the farmers if
any animal enters their farm. As most of the times animals cause harm to the crop in the farms the system can alert the farmers.
In future the system can also used for identification of new species, new animals in more precise way so that it will be more
accurate.

REFERENCES
[1] S. Matuska, R. Hudec, M. Benco, P. Kamencay and M. Zachariasova, "A novel system for automatic detection and classification of animal," 2014
ELEKTRO, 2014, pp. 76-80, doi: 10.1109/ELEKTRO.2014.6847875.
[2] Priya Sharma, Sirisha C K, Soumya Gururaj, and Padmavathi C, “Neural Network Based Image Classification for Animal Intrusion Detection
System”, IJPRSE, vol. 1, no. 4, pp. 1–7, Jul. 2020.
[3] Morrow-Tesch J, Dailey JW, Jiang H. A video data base system for studying animal behavior. J Anim Sci. 1998 Oct;76(10):2605-8. doi:
10.2527/1998.76102605x. PMID: 9814900.
[4] D. P. Gibson, N. W. Campbell and B. T. Thomas, "Quadruped gait analysis using sparse motion information," Proceedings 2003 International Conference on
Image Processing (Cat. No.03CH37429), 2003, pp. III-333, doi: 10.1109/ICIP.2003.1247249.
[5] Burghardt, T. & Calic, Janko. (2006). “Analysing animal behaviour in wildlife videos using face detection and tracking”, Vision, Image and Signal
Processing, IEE Proceedings -. 153. 305 - 312. 10.1049/ip-vis:20050052.
[6] Zeppelzauer, M., “Automated detection of elephants in wildlife video”, J Image Video Proc 2013, 46 (2013). https://doi.org/10.1186/1687-5281-2013-46
[7] Prajna. P, Soujanya B.S, Mrs. Divya, “IoT-based Wild Animal Intrusion Detection System”, International Journal Of Engineering Research & Technology
(IJERT) ICRTT – 2018 (Volume 06 – Issue 15).
[8] Sk. Nayab Rasool and T. SR.CH.Murthy, “Wildlife Monitoring in Zoological Parks Using RASPBERRYPI and Machine Learning”, International Journal of
Recent Technology and Engineering (IJRTE), ISSN: 2277-3878, Volume-8, Issue-2S11, September 2019
[9] Depu Zhou, “Real-time Animal Detection System for Intelligent Vehicles”, Depu Zhou, Ottawa, Canada, 2014
[10] Sk. Almas Tabassum, B. Sri Vaishnavi, Dr. K. S. Sagar Reddy, “SMART CROP PROTECTION WITH IMAGE CAPTURE OVER IOT”, International
Journal of Research in Engineering, IT and Social Science, ISSN 2250-0588, Impact Factor: 6.565, Volume 09, Special Issue 2, May 2019.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 483