And communications of the republic of uzbekistan tashkent information technologies university named after mukhammad al-kharazmi

MINISTRY OF DEVELOPMENT OF INFORMATION TECHNOLOGIES
AND COMMUNICATIONS OF THE REPUBLIC OF UZBEKISTAN
TASHKENT INFORMATION TECHNOLOGIES UNIVERSITY NAMED AFTER
MUKHAMMAD AL-KHARAZMI
Theme: Development of an efficient object tracking method for real-time applications
Graduate: Azamov Ajam
Supervisor: Atoyev Suxrob
Tashkent – 2021
Main parts of the presentation
Goals and objectives of the work
Use cases in the industry and other domains
Analyzing algorithms
What did we do?
Conclusion

Goals and objectives of the work

Creating an efficient method that can track an object on a real-time
Objectives :
Learning to work with video frames
Choosing algorithms for a tracker
Implementing and adding new features to algorithms
Creating graphic user interface application

Use cases in the industry and other domains

Analyzing traffic jam

People counting

Analyzing algorithms

We try to implement our method on tree different object tracking algorithms
KCF – Kernelized Correlation Filters
MIL – Multiple Instance Learning
CSR – Channel and Spatial Reliability

Analyzing algorithms

Kernelized Correlation Filters

Analyzing algorithms

Multiple Instance Learning

Analyzing algorithms

Channel and Spatial Reliability
What did we do?
x
y
(n – 1) th frame
n th frame
(n+1) th frame
a
b
Vectors: a, b

Graphic interface of the application

Next, next, finish
Graphic interface of the application
Main window
of the application
Graphic interface of the application
KFC algorithm
Our method is added
Graphic interface of the application
MIL algorithm
Our method is added
Graphic interface of the application
CSR algorithm
CSR algorithm

Conclusion

During the graduation work, how to deal with video frames and analyzing
them are learned
Object tracking problems and algorithms are briefly investigated
Implemented three different object tracking algorithms and created
real-time single object tracker application
Applying our method with KCF algorithm showed great result and with CSR algorithms
almost nothing changed, but with MIL algorithm slowed down
Thank you for your attention