1. Vision in Man and Machine: Computational aspects.
2. Image formation.
3. Monocular vision: views of planar surfaces.
4. Image registration.
5. Cameras: models and calibration.
6. 3D Structure from dense 2D motion: video sequences.
7. Stereo vision.
8. 3D reconstruction from single images and image collections.
9. Applications: Cultural heritage, Augmented reality, Human-machine interaction, Advanced robotics, multimedia, Information forensics, etc.
1. Hartley and Zisserman, MULTIPLE VIEW GEOMETRY IN COMPUTER VISION.
Cambridge University Press, 2003 (2nd edition).
2. Visione Computazionale - Tecniche di ricostruzione tridimensionale. Franco Angeli 2013.
Teaching notes and recent papers on selected topics will be made available to students.
Learning Objectives
The course aims at providing students with theoretical and practical tools for the visual analysis of images and videos. Such knowledge will be used to develop modern 2D and 3D computer vision systems, with applications in several fields, such as autonomous robotics and AGVs, fruition and valorization of cultural heritage, auxilia for disabled people, industrial automation, human-computer interaction based on movements and gestures, interactive 2D/3D comouter graphics, information forensics.
Prerequisites
Basic knowledge on image representation and processing, algebra and geometry.
Teaching Methods
Classroom teaching with blackboard and videoprojector. Classroom experiments with laptops and wi-fi.
Oral test (for both 6 and 9 CFU). The student will be asked to outline in some detail the theory of a selected topic, from which the discussion will move on to practical-implementation aspects, and exercises aimed at assessing the student's ability to use theory as a tool for solving concrete, real problems.
Homework assignment (only for 9 CFU students), which will consist in the writing of either a computer program (in MATLAB, C++ o Python) or a bibliographical essay. The topic will be either suggested by the teacher, or proposed by the student. The computer program will be accompanied by a document detailing the theoretical and implementation aspects of the work. The bibliographical essay will either review a specific vision topic (e.g. computation of stereo disparity) as addressed by the recent literature, or discuss in detail a selected application scenario of computer vision (e.g., vision algorithms for the movie industry).
Course program
1. VISION IN HUMAN AND MACHINES: COMPUTATIONAL ASPECTS. General introduction. Ambiguity of vision. Visual illusions.
The role of semantics in perception. Visual attention. Basic 3D cues.
2. IMAGE FORMATION. The image as the combination of light, material, geometry and the observer. Tipi di superfici. BRDF. Albedo. Diffuse and specular components. Image acquisition devices. Lenses.
3. MONOCULAR VISION: PLANAR SURFACES. Homographies and their anatomy.
Image rectification based on circular points.
4. IMAGE REGISTRATION
Robust homography estimation: RANSAC etc. Mosaicing.
Mosaicing in the presence of parallax. Image-based
rendering. Super-resolution.
5. CAMERA MODELS AND CALIBRATION
Camera pinhole, natural and affine camera. Radial distortion. Photogrammetric calibration. Self-calibration.
6. VIDEO ANALYSIS
Optical flow vs motion field. Structure from motion (continuous case).
Time to collision. SLAM (simultaneous localization and mapping).
7. STEREOPSIS
Two-view geometry. Fundamental and Essential matrices. Parallax. Stereo rectification. Projective and metric reconstruction. Disparity. Dense stereo algorithms.
8. 3D RECONSTRUCTION FROM SINGLE VIEWS AND IMAGE COLLECTIONS
3D reconstruction pipeline. Bundle adjustment. Single-view reconstruction: scene constraints (planes, surfaces of revolution, etc.). Single view metrology.
9. APPLICATIONS
Natural human-machine interfaces, advanced robot exterocepsis, video post-production, smart videoprojectors, computational photography, 3D television, information forensics, etc.