COMPUTER VISION AND 3D GRAPHICS - VISIONE COMPUTAZIONALE E GRAFICA 3D
INO2043950, A.A. 2016/17
Principali informazioni sull'insegnamento
Dettaglio crediti formativi
||Ingegneria delle telecomunicazioni
Modalità di erogazione
|Periodo di erogazione
|Anno di corso
|Modalità di erogazione
Organizzazione della didattica
|Inizio attività didattiche
|Fine attività didattiche
|3 A.A. 2016/2017
||Elaborazione Numerica dei Segnali e Analisi delle immagini e video sono consigliati ma non strettamente necessari.
Conoscenze e abilita' da acquisire:
||The course offers a guided tour of the computer vision and computer graphics topics needed for current virtual and augmented reality applications.
The course rationale is the introduction of the notions and techniques to go
a) from 3D scenes to images by way of real imaging systems;
b) from images to 3D scene models;
c) and from 3D models to images by way of virtual cameras.
Part a) has the objective of explaining the operation and the mathematical models of current imaging systems (e.g., video-cameras, Time of Flight systems, kinect) in the language of computational photography. The objective of Part b) is the coverage of two topics: 3D recontruction from images (with special focus on stereo and active stereo systems) and the 3D modeling pipeline (i.e., the procedures to obtain full 3D models from depth maps). The objective of Part (c) is to introduce the rendering methods as approximate solution of the rendering equation.
The course is structured in order to give the students a clear sense for the deep interconnections between the notions of computer vision and computer graphics encountered in virtual and augmented reality applications, interconnections due to the fact that that 3D reconstruction (computer vision) can be interpreted as the inverse problem of rendering (computer graphics).
Within its time limits, the course also aims to introduce the students to current computer vision and computer graphics tools such as OpenCV, Point CloudLibrary and OpenGL.
Modalita' di esame:
||Prova scritta + report
Criteri di valutazione:
||The Computer Vision and 3DGraphics class covers a wide range of topics mainly across Computer Vision and Computer Graphics since a wide panorama is a good asset to face the fast evolution of these fields.
Nevertheless the student evaluation will be focused on the concepts necessary for building and visualizing 3d models within typical augmented and virtual reality contexts.
Such topics will be clearly indicated during the course and in the course material.
Every efforts on the student part revealing personal involvement and special care will be recognized in terms of scores.
||a) From 3D scene to images via real imaging systems
1 Image formation and camera model
Simplified camera model
General camera model and its properties
2 Camera calibration
Computation of the homography (DLT)
b) From images to 3D scene model (or 3D reconstruction in Computer Vision)
3 Stereopsys: geometry
Essential matrix, factorizzation and computation
Motion and structure from calibrated homography
4 Salient points extraction
Harris e Stephens method
Salient points correspondence
Scale Invariant Feature Transform (SIFT)
5 Stereopsys: Corrispondence
Local correspondence methods
Window correspondence methods
Accuracy-reliability trade-off Indicatori di affidabilita’
Other local methods
Global correspondence methods
6 Mosaics and image synthesis
7 Non-calibrated reconstruction
Fundamental matrix and its computation
Projective reconstruction from 2 and N views
Methd of Mendonca e Cipolla
8 Optical flow
Motion field: computation of motion and structure
Optical flow: Lucas-Kanade method
9 3D Modeling Pipeline
Absolute orientation by Horn and Arun methods
Rotation interpolation by SLERP method
Iterated Closest Point (ICP)
Global Registration: Lu & Milios method
Surface integration: non-volumetric (Turk & Levoy method) and volumetric methods (Wheeler, Ikehuchi & Sato method)
c) From scene 3D models to images by way of virtual 3D video-cameras (Computer Graphics)
Proiective projection convention in graphics
BRDF and fundamental BRDF models (Lambertian, specular, glossy)
Lambertian surfaces: Relationship between radiosity and irradiance
The radiance (or rendering) equation and its solution
11 Illumination models
The radiance solution by local metohds: Phong and ModelloCook-Torrance models
The radiance solution by grobal metohds
Ray tracing: Whitted method
Radiosity: radiosity equation in continuous and discrete form
Hidden surface removal
Shading: Flat, Phong e Gouraud methods
The OpenGL pipeline
13 Mapping methods
Texture mapping in 1 e 2 passes
Foreword and backword mapping.
Aliasing: minification and maxification.
Attivita' di apprendimento previste e metodologie di insegnamento:
||The course offers a guided tour of the computer vision and computer graphics concepts needed for current virtual and augmented reality applications. The main topics are:
• Image formation: mathematical models of cameras and Time of Flight systems.
• Camera calibration: procedures for metrical measurements from images
• Computational stereopsis: 3D scene structure derived from 2 or more images obtained from calibrated cameras
• Structure from motion: 3D scene structure derived from 1 or more calibrated moving cameras
• Un-calibrated 3D reconstruction: 3D scene structure derived from un-calibrated cameras.
• 3D registration: pairwise and global registration (or SLAM) of depth-maps into a point cloud
• 3D data integration and geometrical simplification: integration of overlapping point clouds into tessellated surfaces and their simplification
• Rendering methods: ray casting, ray tracing, radiosity and rasterization
The topics are treated by means of frontal lectures with computational examples based on MATLAL, Open CV and Open GL.
The appraisal is stimulated by homeworks confronting the student with practical situations due to the concepts seen in class.
Eventuali indicazioni sui materiali di studio:
||The course purposely hybridates two disciplines, computer vision and computer topics, in order to focus on 3D model construction and visualization and there is no textbook on such a specific topic.
The study material is given by the class-notes made available before every class meeting.
The notes distill and condense various research papers and content coming from several textbooks, among which:
Richard Szeliski, Computer Vision: Algorithms and Applications, Springer, New York, 2010
A. Fusiello, "Visione Computazionale", F. Angeli, Milano, 2013
C. M. Bishop, "Pattern recognition and machine learning", Springer, New York, 2006
Testi di riferimento:
Fusiello, Andrea, Visione computazionaletecniche di ricostruzione tridimensionale. Milano: Angeli, 2013.
Szeliski, Richard, Computer visionalgorithms and applicationsRichard Szeliski. New York: Springer, 2011.
Bishop, Christopher M., Pattern recognition and machine learning. New York: Springer, --.
Forsyth, David; Ponce, Jean, Computer VisionA Modern ApproachDavid A. Forsyth, Jean Ponce. Boston: ©Pearson, 2012.