Module I: Advanced Computer Vision Methods

Subject description

We'll begin with the basics of the human visual system and how it differs from computer vision, establishing a base for understanding the contrast between natural and artificial visual processing. Next, we'll examine image datasets, emphasizing their role in advancing computer vision through the development and evaluation of algorithms.

We'll explore feature point detection and description techniques, such as SIFT, HOG, and MSER, focusing on multiresolution and scale space to understand image interpretation at various details.

The course will then cover visual object detection and tracking, highlighting methods like Bayesian filtering and the Kalman filter, to show how computer vision tracks objects in diverse settings.

Finally, we'll study learnable image descriptors and convolutional neural networks, diving into their application in deep learning for automation and robotics, showcasing how these technologies enable the creation of smart systems for visual world interaction.

The subject is taught in programs

Objectives and competences

The aims of this course are to cover selected existing and emerging topics in advanced computer vision, and to prepare students for teamwork, as well as independent work in research and development.

Teaching and learning methods

The lectures provide a theoretical background on particular subjects together with practical examples in Matlab and Python.

Practical work is being performed as the part of laboratory exercises, and is accomplished in the form of multiple assignments, acquainting students with computer vision algorithms. Students work in groups, consisting of two or three students, and the results are in the form of Matlab and Python code. After completing each part, students present their results to the assistant.

Homework project addresses the particular problem from the machine vision domain, either robot vision or industrial machine vision applications.

Expected study results

After successful completion of the course, the students should be able to:

Explain the differences between human visual system and computer vision methods
Decide, which of the advanced computer vision methods is appropriate for the problem at hand
Implement modern computer vision algorithms
Develop complex solutions for computer vision problems, made up of multiple advanced methods
Judge the performance of computer vision methods on the available data
Propose solutions to computer vision problems from the field of automation and robotics that require the use of the advanced and complex algorithms, including neural networks and other learnable algorithms.

Basic sources and literature

Forsyth, David A., in Jean Ponce. Computer Vision: A Modern Approach. 2nd ed., Pearson, cop. 2012. Boston [etc.]. ISBN 978-0-13-608592-8, 0-13-608592-X. COBISS.SI-ID: 10640724. Dostopno v knjižnici UL FE.
Szeliski, Richard. Computer Vision: Algorithms and Applications. Springer, cop. 2011. London [etc.]. ISBN 978-1-84882-934-3, 1-84882-934-5, 1-84882-935-3. Dostopno na SpringerLink za uporabnike z dostopom do univerzitetne računalniške mreže ali preko storitve Oddaljeni dostop do informacijskih virov UL. COBISS.SI-ID: 9030996.
Temeljni članki, objavljeni v znanstvenih revijah (Basic scientific papers, published in scientific journals)