Computer Vision

(3 Credit Hours) · Undergraduate

The aim of Computer Vision is to give computers the ability to "understand" what they "see" in images and videos taken by one or more sensors (most often visible-light cameras). The goal of this course is to introduce and discuss methods for interpreting the visual information captured by machines to give them this ability. The course is divided into four parts. In the first part, we define the notion of computer vision, the progress made in this discipline in recent decades, current challenges, successful applications, and its limitations. We also discuss selected biological vision mechanisms as an inspiration to create better computer vision solutions. The second part explains the basics of signal processing from a computer vision perspective. This part includes image formation, image acquisition, understanding and effective use of color (and in general multi-wavelength) information, and image processing (filtering and segmentation). The third part focuses on the automatic recognition of patterns. It covers feature extraction and selection, texture descriptors, Bayesian inference, classification, and decision making. In this part, we will also discuss how these tasks can be solved using deep learning techniques, especially convolutional neural networks. Several meetings in this third part will be devoted to the reliability of modern deep learning-based, generative, and image-to-image translation models. Finally, the fourth part considers multiple-view and geometry topics in vision: motion analysis, including object tracking, projective geometry, camera geometric model, camera calibration, and 3D reconstruction. One meeting at the end of the semester will be devoted to the non-technical aspects of designing trustworthy and reliable computer vision (and AI in general) systems. After completing this course, students will be able to understand computer vision literature, recognize the frontiers of state-of-the-art computer vision systems, and select appropriate mathematical and software tools to develop algorithms solving the most important computer vision problems. Practical classes will utilize high-level programming languages (Python will be our main coding language) and popular computer vision tools and machine learning packages, such as OpenCV, Keras, Tensorflow, or Pytorch, to illustrate in practice selected topics discussed in class. The goal of the semester project is to exercise the entire computer vision pipeline on the selected vision problem. The most recent syllabus is available at https://adamczajka.com/teaching/computer-vision.