Advanced topics in computer vision

2019/2020
Programme:
Computer Science and Mathematics, Second Cycle
Year:
1 in 2 year
Semester:
second
Kind:
optional
ECTS:
6
Language:
slovenian, english
Course director:

Matej Kristan

Lecturer (contact person):

Matej Kristan

Hours per week – 2. semester:
Lectures
3
Seminar
0.67
Tutorial
1.33
Lab
0
Prerequisites

There are no prerequisites.

Content (Syllabus outline)

The course will include selected advanced topics in motion perception using computer vision. Concrete topics will change each year according to trends in this fast developing field.
in computer science and industry. Potential topics will include:
Overview of the field motion estimation and applications.
Optical flow estimation using least-squares.
Variational optical flow estimation.
Parametric template tracking using Lucas-Kanade.
Histogram-based tracking using Mean Shift
Tracking as stochastic optimization using cross entropy.
Recursive Bayes filter for online state estimation.
Tracking by Kalman filter.
Tracking by particle filters.
Tracking deformable objects by constellation models.
Methodologies of tracker comparison.
Tracking by classification.
Long-term tracking by detection.

Readings

Simon J. D. Prince, Computer Vision: Models, Learning, and Inference, Cambridge University Press, 2012
Richard Szeliski, Computer Vision: Algorithms and Applications, Springer, 2010
David Barber, Bayesian Reasoning and Machine Learning, Cambridge University Press, 2012

Objectives and competences

The primary objective is obtaining an overview of scientifically challenging topics of computer vision and broader artificial intelligence. In this sense, the course is logical continuation of basic first-level courses in artificial intelligence, specifically, computer vision, multimedia and machine learning. The secondary objective is practical application of analytical and numerical methods that students learn at basic courses, but seldom use in practice. At the end of this course the students will be skilled in modern approaches for motion estimation and tracking using computer vision approaches. The studens will obtain practical experience with these approaches.

Intended learning outcomes

After completing this course a students will be able to:

  • know major methods for motion estimation and localization of moving objects,
  • understand the concept of optical flow estimation and be able to implement basic approaches,
  • understand mathematical background of template alignment using gradient descent,
  • understand mathematical background of probabilistic Bayesian models for target position estimation in images and be able to implement the basic algorithms from the family of these approaches,
  • understand the approaches for tracker evaluation and be able to critically analyze the algorithms,
  • understand the basics of long-term trackers and know the major representatives from this field,
  • implement applications for image-based object tracking,
  • understand modern algorithms in the field of object tracking.
Learning and teaching methods

Lectures, laboratory exercises, homeworks and project work. Special emphasis will be given on individual work.

Assessment

Continuing (lab exercises, homework, project)
Final (written exam)
Final (oral exam)
grading: 5 (fail), 6-10 (pass) (according to the Statute of UL)

Lecturer's references

LUKEŽIČ, Alan, ČEHOVIN ZAJC, Luka, KRISTAN, Matej. Deformable parts correlation filters for robust visual tracking. IEEE transactions on cybernetics, ISSN 2168-2267, 2017, vol. , no. , str. 1-13, [COBISS-SI-ID 1537625283],
KRISTAN, Matej, SULIĆ KENK, Vildana, KOVAČIČ, Stanislav, PERŠ, Janez. Fast image-based obstacle detection from unmanned surface vehicles. IEEE transactions on cybernetics, ISSN 2168-2267 , 2016, vol. 46, no. 3, str. 641-654, [COBISS-SI-ID 1536310979],
KRISTAN, Matej, MATAS, Jiří, LEONARDIS, Aleš, VOJÍŘ, Tomáš, PFLUGFELDER, Roman, FERNÁNDEZ, Gustavo, NEBEHAY, Georg, PORIKLI, Fatih, ČEHOVIN ZAJC, Luka. A novel performance evaluation methodology for single-target trackers. IEEE transactions on pattern analysis and machine intelligence, ISSN 0162-8828. [Print ed.], Nov. 2016, vol. 38, no. 11, str. 2137-2155, [COBISS-SI-ID 1536872643]
ČEHOVIN ZAJC, Luka, LEONARDIS, Aleš, KRISTAN, Matej. Visual object tracking performance measures revisited. IEEE transactions on image processing, ISSN 1057-7149, 2016, vol. 25, no. 3, str. 1261-1274, [COBISS-SI-ID 1536812739]
ČEHOVIN, Luka, KRISTAN, Matej, LEONARDIS, Aleš. Robust visual tracking using an adaptive coupled-layer visual model. IEEE trans. pattern anal. mach. intell.. [Print ed.], 2012, str. [1-14], [COBISS-SI-ID 9431124]

Celotna bibliografija je dostopna na SICRISu:
http://sicris.izum.si/search/rsr.aspx?lang=slv&,id=32801.