See discussions, stats, and author profiles for this publication at: [Link]

net/publication/365091324

Fast and Economical Object Tracking using Raspberry Pi 3.0

Article in International Journal of Engineering and Advanced Technology · September 2019

DOI: 10.35940/ijeat.F1070.0886S19

CITATIONS READS

6 265

1 author:

Nagalakshmi Tj
Saveetha University
135 PUBLICATIONS 342 CITATIONS

SEE PROFILE

All content following this page was uploaded by Nagalakshmi Tj on 09 November 2023.

The user has requested enhancement of the downloaded file.

 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249-8958 (Online), Volume-8 Issue-6S, August 2019

Fast and Economical Object Tracking using

Raspberry Pi 3.0
Purnima Prakas, T.J Nagalakshmi
ABSTRACT--- This paper proposes a way to construct a information for tracking. Re- assignment and back tracking
financially cheap and fast object tracking using Raspberry Pi3. is based on the measurement of the predicted object and the
Multiple object detection is an important step in any computer last appearance position. OpenCV is used to manipulate and
vision application. Since the number of cameras included is more
acquire input runtime. We used OpenCV python to reduced
these gadgets are compelled by expense per hub, control
utilization and handling power. We propose a tracking system complexity and increase robustness of the algorithm In the
with low power consumption. The framework is completely further sections we will discuss about the hardware set up
designed with python and OpenCV. The tracking quality and and tracking efficiency.
accuracy is measured using publicly available datasets.
II. FUNCTIONAL DESCRIPTION
Keywords: Cameras, Computer vision, Python, Raspberry Pi,
Pi camera, Tracking. The Functional description of various components are
given below:
I. INTRODUCTION A) Raspberry Pi :
Multiple Object detection and recognition is an important The Raspberry Pi [3] is a low cost, credit card sized single
topic when it comes to computer vision. It is widely used in board with a considerable computing power. It provides
applications like automatic traffic control systems and for various features like Ethernet port, GPIO pins for
driver assistance system [1]. On the other hand, these connecting sensors and various devices, Camera connector
systems are suffering badly with power consumption and (Raspberry pi camera). USB ports to connect various
cost. The cost of these devices increase with every node. devices like Keyboard, mouse, Monitor etc. The raspberry pi
Visual Object tracking is a system which faces various does not have any internal mass storage unit or built in
challenges like illumination and pose. At many cases the operating system. It requires a version of an SD card which
number of objects in the field of detection may vary will serve as mass storage and holds Linux OS. The system
overtime. To solve these problems there are several methods proposed here uses Raspberry Pi3 as shown in the figure 1.
and algorithms are developed [1] [2]. However, these This model board is a micro controller kit with ARM11
systems are designed to achieve the projected accuracy and processor provided with internet / Ethernet connectivity, 512
processing speed on a personal computer. These personal MB memory, dual USB connectivity and supports Pidora,
computers has required computing capacity to drive the Raspbmc, Raspbian
algorithm designed but when these algorithms are deployed
in to an embedded system their accuracy and computing
power does not meet the expectations. We used Raspberry
pi3 which has a quad core processor. It’s has all functions of
a basic computer and operates on a Linux OS. Raspberry pi
operates as a heart of the proposed system as it holds the
tracking algorithm and the camera input.
The Kernalized correlation Filters (KCF) tracker is used
in the proposed system. KCF tracker is designed using The raspberry pi operates as the heart of the proposed
OpenCV python. This tracker is designed for individual system as it holds the Raspberry Pi camera and the proposed
objects firstly and at the end all the individual trackers are algorithm. It operates at 5V 2 A. It supports Wi-Fi
combined with frame to frame detection. Traditional KCF connectivity and hence this device can be controlled
trackers suffered tracking errors when the size of the object anywhere from the world via internet.
gradually changed. Hence we took advantage over size of
B) Raspberry Pi Camera:
The Raspberry Pi camera is a 8 mega pixel camera which
Manuscript published on 30 August 2019. uses Sony IMX219 sensor designed specifically to fit an on
* Correspondence Author (s) board device like raspberry pi. It can capture 3280x2464
Purnima Prakas, Department of ECE, Saveetha School of Engineering,
Chennai, Tamil Nadu, India. static images and also extends its support to 640x480p90,
T.J Nagalakshmi, Department of ECE, Saveetha School of Engineering, 1080p, 720p60 video. It is attached to the raspberry pi
Chennai, Tamil Nadu, India. through a dedicated CSI interface. The
© The Authors. Published by Blue Eyes Intelligence Engineering and
Sciences Publication (BEIESP). This is an open access article under the
CC-BY-NC-ND license [Link]

Retrieval Number: F10700886S19/2019©BEIESP Published By:

DOI:10.35940/ijeat.F1070.0886S19 Blue Eyes Intelligence Engineering
Journal Website: [Link] and Sciences Publication (BEIESP)
336 © Copyright: All rights reserved.
 FAST AND ECONOMICAL OBJECT TRACKING USING RASPBERRY PI 3.0

dimension of camera is 25mm x 23mm x 9mm. Since the

size and weight of the camera is less, it is flexible to be used
in various applications.

The KCF tracker can be dissected in to pieces which can

handle training, detection, model update, appearance
representation.

Raspberry pi camera should be connected to the camera

port of [Link] camera should be configured by opening
the preferences option from the [Link] requires a reboot to
There are various versions of KCF available, they are
have the system settings saved.
CTF [5], SCT [5], DPT [5], SMACF [5], MAD [5], sKCF
C) KCF Tracker: [4], KCF2 [4], SME [4], LDP [4].
Kernel Correlation Filter (KCF) is a version of correlation
filter. In a correlation filter two data points are taken and if III. IMPLEMENTATION
the correlation between these two points are high then the KCF tracker algorithm is implemented using OpenCV
correlation value is high. The same principle is adopted in version 4.0 which extends its support for deep learning and
KCF. The correlation between the previous patch in the various computer vision algorithms. Since Raspberry pi 3
frame and the future frame is compared. In a traditional supports Python for control of its ports [6] and GPIO its
correlation filter, the tracker performance degrades when the very flexible and robust to go with python. When the
object size changes. In a KCF tracker the object being program is executed the user has to use there mouse and
tracked is updated using a linear regression model. Kernel select the object to be tracked. We have not implemented
functions like Gaussian functions are used in calculation dot any object detection block since it limits the object to be
product of high dimensional spaces without calculating tracked. After selecting the region of interest (ROI) to be
higher dimensional [Link] regression weights are tracked the algorithm tracks the object seamlessly.
represented by

Where Kaa is a kernal function calculated on auto-

correlation of training samples a. The regression function
has to be calculated on several patches. The candid patches
forms circular matrix. Training and testing stages
completely rely on computation of kernal function . For
Gaussian systems the kernal can be calculated as

The regression weights given by equation 1 is updated by

linear interpolation.

Retrieval Number: F10700886S19/2019©BEIESP Published By:

DOI:10.35940/ijeat.F1070.0886S19 Blue Eyes Intelligence Engineering
Journal Website: [Link] and Sciences Publication (BEIESP)
337 © Copyright: All rights reserved.
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249-8958 (Online), Volume-8 Issue-6S, August 2019

V. ACKNOWLEDGEMENT
The authors would like to thanks [Link]. Nagalakshmi
for her support and guidance through the development of
this project and for her advice in making it possible

REFERENCES
1. N. Majer, K. Schindler, and B. Schiele, “New features
and insights for pedestrian detection,” in Computer
vision and pattern recognition ( CVPR), 2010 IEEE
conference on. IEEE, 2010, pp. 1030–1037.
2. Possegger, T. Mauthner, P. Roth, and H.
Bischof,“Occlusion geodesics for online multi- object
tracking,”in Proceedings of the IEEE Conference on
When the object is moved away from the line of sight of Computer Vision and Pattern Recognition, 2014, pp.
the camera , it remembers the object and when it appears 1306–1313.
again it detects the object. The frames per second of the 3. Matt Richardson and Shawn Wallace, Getting Started
video is kept at a minimum level to boost the accuracy. with Raspberry Pi. United States of America: O’Reilly
Media, 2013. [4]AM. Kristan et. al. (2015) “The Visual
Since Raspberry is an IOT (Internet Of Things) device , live
Object Tracking VOT2015 challenge results.” IEEE
streaming can be done via connecting it to a server. International Conference on Computer Vision
Workshop(ICCVW) 2015.
4. Kristan et. al. (2016) “The Visual Object Tracking
VOT2016 challenge results.” European Conference on
Computer Vision Workshop(ECCV).
5. Raspberry Pi Foundation, [Online], [Link]
berrypi. org/ help/what- is-a- raspberry-p

IV. RESULT
An efficient, robust and low cost tracking method is
proposed. Since this system occupies less space, it can be
implemented in robots, drones, Traffic signals etc. This
system can track both video and live feed input. We have
optimised the Frames per second to boost the tracker
efficiency.
Retrieval Number: F10700886S19/2019©BEIESP Published By:
DOI:10.35940/ijeat.F1070.0886S19 Blue Eyes Intelligence Engineering
Journal Website: [Link] and Sciences Publication (BEIESP)
338 © Copyright: All rights reserved.
View publication stats