You are on page 1of 9

A

SYNOPSIS ON
Object Tracker Attachment for drone flight controller.
A Project Submitted in Partial Fulfillment of the Requirements for the Degree of Bachelor of
Technology
in
ELECTRONICS AND COMMUNICATION ENGINEERING

SUBMITTED BY

Shashwat Pandey – 1902220310038


Siddhi Singh – 1902220310038
Deepak Sharma - 1902220310012
UNDER THE SUPERVISION OF

Mr. Agha Asim Husain, Asst. Professor.

SUBMITTED TO
DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING
I.T.S ENGINEERING COLLEGE, GREATER NOIDA
(Affiliated to Dr. APJ Abdul Kalam Technical University, Lucknow)
INTRODUCTION

The idea is to develop a tracking system attachment for the drone


flight controller to tack the direction of a moving object from a
camera feed of drone. The said system is a processing device
combined with software to perform the stated tasks. The
mentioned system is implemented using, Object Detection, Object
Tracking and GPS mapping to autonomously and continuously
track an object till it exists in the feed. The device allows the
drone to maneuver in the direction of the moving object
autonomously, by sending control signals to the flight controller
present onboard. The requirements of the device is a 5W power
supply and stable internet connectivity, that allows the device to
take commands from the ground station. The device is supposed
to be actually present on the drone and connected to it’s flight
controller. The benefit of such system is to track a desired object
and collect positional information of the object being tracked in
terms of latitude, longitude, and directions. The device will start
tracking the desired object once it gets the command from a
‘ground station’. The mentioned ‘ground station’ can be any
computer running any operating system of type LINUX or
Windows, this computer will be running a command exchange
software, this command exchange software will be responsible
for notifications from drone and sending commands the device
present on drone . The system when implemented on a drone can
help in autonomously.
METHODOLOGY
The mentioned device is a device, capable of tracking moving
objects in the live video feed. The device is a combination of a Single
Board Computer (SBC), and a software. The SBC used in the
construction is a Raspberry pi4 Model B. The software mentioned
has the following features.
• Object Detection.
• Object Classification.
• Object Tracking.
• SORT.
• DeepSORT.
• Collecting Positional Information.
• Communicating with the on board flight controller.
• Communicating with the ground station.

1. Object Detection – Object detection is an image processing


technique that is used to find objects present in an image. The
technology uses image manipulation techniques of Computer
Vision, that can find the presence of any object in an image
regardless of its nature or type. Classical Object detection
techniques are inefficient and are prone to noise and lighting
variations.
2. Object Classification – Object Classification is a more detailed
and complex methodology used to find objects in an image, the
method does not only finds the object in any image but it can also
classify the type of object detected. Object Classification is
achieved using techniques of Deep Learning and Machine
Learning. It requires training of neural networks on image
datasets. The trained neural network is capable of classifying all
types objects that were present in the dataset.
The neural network used for classification are of Convolutional
Neural Network (CNN) architecture. Here we use one such
pretrained neural network namely YOLOV3 trained on the COCO-
80 dataset, the COCO dataset has 80 Classes.
3. Object Tracking – Object Tracking is another Computer Vision
technique that keeps the track of an object when it moves across
frames in an incoming video feed. The objective is to track the
position or path of the moving object. There are basically two
methods used for tracking. SORT (Simple Online Realtime
Tracking) and another more refined and accurate method called
DeepSORT.
4. SORT – Simple Online Realtime Tracking is a tracking method,
that is fairly simple yet effective, it uses a classical approach to
detect object that is purely algorithmic, and assign IDs to the
detected object, and then track the object using an estimation
model called Kalman Filter that predicts the position of a given
object in the next frame, based on its position and path in earlier
frames. This estimate or probability distribution is then used by
SORT to look for the same object in the probable position the
Kalman Filter suggested. And as soon as the object is lost in the
frame, say due to occlusion the ID is deleted. And the object is
lost.
5. DeepSORT – DeepSORT is an extended approach to SORT it uses
Object Classification as the technique for object detection and
localization. DeepSORT also solves the problem of occlusion as
when the object is lost in one frame and re appears in next the ID
is not deleted, hence we can still track the object.
6. Collecting Positional information – The device can also keep
track of the positional information and log it back to the ground
station, in terms of GPS coordinates, and directions of movement
by corelating the information with a map.
7. Communicating with the flight controller – The device
seamlessly communicates with the flight controller, in order to
update it for the change in position by providing it the right
signal, this way we capacitate the drone with autonomous
tracking technology.
8. Communicating with the ground station – The device also
communicates with the ground station in real-time updating
details of the object being tracked and it’s positional information
and listens for any command update from the ground station.
PROJECT TIMELINE

The project will progress by first creating all the core functionalities
of the device in terms of tracking and object detection, later we will
move towards implementation of a ground station software, that
will be responsible for control and command for the device, we will
then move to the implementation of the signaling function that will
be responsible for keeping the drone in motion as the object moves
through the scene. And after finalizing these functionalities we will
implement the code on the SBC.
References

• https://www.mathworks.com/discovery/object-
detection.html#:~:text=Object%20detection%20is
%20a%20computer,learning%20to%20produce%2
0meaningful%20results.
• https://github.com/theartificialguy/Yolo-v4-
DeepSort
• https://paperswithcode.com/task/object-tracking
• https://www.linkedin.com/pulse/object-tracking-
sort-deepsort-daniel-pleus/?trk=pulse-article_more-
articles_related-content-card
• https://www.raspberrypi.org/
• https://machinelearningmastery.com/object-
recognition-with-deep-learning/

You might also like