Course information | Artificial Vision

Provide a robust understanding of some significant problems in computer vision and solutions, in particular for (autonomous or quasi-autonomous) cyberphysical systems

Introduction to computer vision for (autonomous) cyberphysical systems.

Detailed program

Geometry of image formation (cameras are bearing-only sensors)
Calibration of the projection parameters
Technological aspects of image formation
Model-based vision (monocular object recognition and localization)
3D reconstruction of points by leveraging more than one image at the same time (stereoscopy)
3D reconstruction of points by leveraging more than one image at different time instants (optical flow, feature tracking, structure from motion)
Sensors capable to natively give out depth information: LiDARs and 3D cameras, both structured light and ToF
Depth from monocular DNN-based techniques
Bayesian discrete time filtering (Kalman, EKF, UKF, gaussian mixtures, IF, histogram, particle)
Panoptic (semantic multi-instance) segmentation of images
Observer localization problem, classical techniques and DNN-based techniques
SLAM (Simultaneous Localization and Mapping) problem and its sub-problems (incremental SLAM, loop detection, relocation, loop closure), classical techniques and DNN-based techniques

Prerequisites

Linear 3D geometry (lines, planes).
Linear algebra.
Digital image processing.
Deep Neural Networks.

Teaching form

Classes and practices, both programming and hands-on.
The scheduled activities are: 32 hours of lessons in dispensing mode and/or in interactive mode, 24 hours of laboratory in interactive mode.

Textbook and teaching resource

Selected parts from well-known textbooks like, e.g.,
David A. Forsyth and Jean Ponce, "Computer Vision: A Modern Approach" 2nd edition, Pearson, 2012
Andrea Fusiello, "Computer Vision: Three-dimensional Reconstruction Techniques", Springer 2024
Emanuele Trucco, Alessandro Verri, "Introductory techniques for 3D Computer Vision", Prentice Hall, 1998
Other material, like e.g., tutorials, review papers, etc. for the less consolidated parts.