Tutorials and seminars

Title: Shape analysis techniques for exploring collections of 3D data
Speakers: S. Biasotti and M. Spagnuolo
Abstract: The ever-growing availability of data acquired by different devices has been inducing an exponential growth of the volume of the available data and has been deeply modifying the research approach in shape analysis and reasoning. These data are often stored using different data formats and have to be analyzed, interpreted and stored with significant computational efforts and experts' commitment. Moreover, data are often affected by noise, for instance when the acquisition conditions are not optimal (low resolution of instruments, partial acquisition, motion, etc.). Therefore it is everyday more urgent the definition of strategies for the description and representation of data in an efficient way, as well as the definition of tools that are able to perform computations over this kind of information even in presence of partial data or affected by noise. This course is meant as a practical guide for researchers who are willing to explore 3D shape analysis, and thus require to manage the rather complex mathematical tools which most methods rely on. The attendees will familiarize with basic concepts in differential geometry and topology, and proceed to notions of computational topology, always keeping an eye on the concepts effectively used for 3D shape analysis. Examples of applications to shape correspondence, symmetry detection, and shape retrieval will be shown to demonstrate how these mathematical notions can be transferred into practical solutions.

Title: A topological point of view
Speakers: Massimo Ferri and Claudia Landi
Abstract: Geometrical analysis of images and shapes is powerful and is used quite frequently in Pattern Recognition and Computer Vision. It suffers, however, of an intrinsic rigidity which is fine when dealing with mechanical pieces, buildings etc., but which becomes a drawback when natural shapes have to be worked on. Topology is - in a sense - an extension of geometry which allows for much wider equivalences. It is actually at the other end of the spectrum, so to say, in the sense that shapes which we would not want to consider similar are equivalent in topological terms. A recent development, called Persistent Topology, tries to cope with this problem. A first part is devoted to surveying some basic concepts of topology: homeomorphism, homology, Morse functions. A second part will concentrate on applying the theoretical notions introduced in the first part to the practical setting of Pattern Recognition. In particular, the Persistent Homology approach to shape description and shape comparison will be illustrated, mostly by means of examples on 2D and 3D shapes.

Title: Introduction to image-based modeling
Speaker: Andrea Fusiello
Abstract: This lecture will illustrate principles and technique of image-based modeling, i.e., the computational process that brings from pictures of a subject to its three-dimensional model. The topics that will be covered include structure-and-motion and multi-view stereo.

Title:
Speaker: Giulio Sandini and Giorgio Metta
Abstract:

Title: Deep representation hierarchies for interactive 3D vision
Speaker: Silvio Sabatini
Abstract: Enabling visually-guided behaviors in artificial agents implies picking-up and organizing appropriate information from the visual signal at multiple levels. The question arises how to carefully define which feature to extract, or, from a different perspective, which kind of representation to adopt for the visual signal itself. Although the two issues cross relate each other, there is a substantial distinction between them, which set the "feature detection hypothesis" against the "signal analysis hypothesis". The former relies on matched operators that extract the most informative - symbolic - elements of an image, the latter performs a mapping to a quasi holographic description, meaning that the visual signal is described in terms of more general structural properties of a local portion of the plenoptic space. In the first part of the talk, I will present how diversified and composite visual descriptors emerge from different hierarchical combinations of the harmonic content of the visual signal. The resulting deep hierarchy networks can specialize to solve different tasks and trigger different behaviors, without necessarily getting through an explicit measure of the reconstructive visual attributes of the observed scene. In the second part of the talk, a showcase of applications in active binocular systems will be presented. Specifically, the advantaged of not abandoning distributed representations of multiple solutions to prematurely construct integrated description of cognitive entities and commit the system to a particular behavior, will be discussed.

Title: Interactive 3D vision for ecological virtual reality
Speaker: Fabio Solari
Abstract: The recent diffusion of the stereoscopic 3D technologies has yielded the development of affordable devices for the visualization of such information. This has paved the way for powerful human and computer interaction systems. However such systems may produce visual fatigue and misperception of 3D scene structure, since the interaction between the user and the virtual reality environment is not taken correctly into account. To overcome these issues we have to endow the virtual reality systems with the capability of computing the dynamic 3D description of the user's pose and movement, and of exploiting such information in order to properly render the virtual world. In the last part of the talk, I will describe a prototype that uses a low-cost RGB-D camera (like the Kinect) and a modified stereoscopic virtual camera to improve the human machine interactions through natural interfaces and veridical visual stimuli. The use of off-the-shelf technologies could permit a large diffusion of such ecological virtual reality systems that, in perspective, could allow supporting solutions of everyday life problems of the elderly and impaired people.

Title: An introduction to 3-D human motion analysis from multi-camera sequences
Speaker: Emanuele Trucco
Abstract: The tutorial will introduce the topic of 3-D motion capture (mocap) and analysis from multi-camera sequences. The key topics covered will be multi-camera mocap techniques, articulated human motion representation, tracking, and action representation and classification. Challenges and techniques for each topic will be presented and discussed. The tutorial will touch upon related, relevant issues of wide applicability in computer vision, e.g., dimensionality reduction and optimization.

Title: About 3D faces...
Speaker:Alberto del Bimbo
Abstract: In this tutorial we will provide an overview of 3D face recognition. Automatic human target identification has been an active research area in pattern recognition and computer vision, since '90s with major emphasis on detection and matching of human faces in 2D still images and videos. 2D face recognition suffers anyway from several problems: illumination conditions; acquisition pose; facial expressions; occlusions; aging... But the increasing availability of 3D face scans has raised the interest in 3D face recognition to improve the effectiveness of face recognition systems. The full 3D geometry of the face can be used instead of just a 2D representation of how the light is reflected from the face surface. and discuss relationships with advanced surveillance applications.

Title: Iris Recognition and the NICE (Noisy Iris Challenge Evaluation) Contest
Speaker: Michele Nappi and Daniel Riccio
Abstract: The tutorial will discuss about the recognition of degraded iris images acquired in visible wavelengths. About this subject, in the last years the University of Beira Interior (Portugal) promoted two International evaluation initiatives, named Noisy Iris Challenge Evaluation (NICE) I and II. The main innovation of NICE framework is the use of heavily degraded data acquired in the visible wavelength and uncontrolled setups, with subjects moving and at widely varying distances. The NICE contest is composed of two separate modules: NICE (I) is focussed on iris segmentation and noise detection techniques, while NICE (II) evaluated encoding and matching strategies for biometric signatures.

Title: Tutorial on multi-sensor, multi-camera surveillance systems
Speaker: Andrea Prati
Abstract: This 3-hour tutorial will provide a broad overview of the different scientific topics addressed in the field of surveillance. Several new research topics will be tackled with a proper level of details and plenty of pointers for further reading. A not-comprehensive list of them includes: active and multi-camera systems, human action recognition, crowd analysis, people/object detection, trajectory analysis, multi-sensor integration for surveillance, mobile vision challenges.

Title: Performance Evaluation of Video Content Analysis products in the international market.
Speaker: Giovanni Garibotto
Abstract: A quantitative evaluation of Video Analytics technologies is a fundamental objective in the area of Physical Security. It is common opinion that Video Content Analysis (VCA) represents the most valuable component of the industrial offer of Video Surveillance solutions. Anyway, the possibility to measure and compare the achieved results of Video Analysis and Computer Vision is quite relevant also in the scientific and academic community, to measure the progress achieved by the most advanced research activities. System integrators and final users are also interested subjects since they need a quantitative support to beleive in the real potential of VCA technologies. The lecture is mainly addressed to industrial application requirements and specification analysis, including camera configuration, image quality issues, up to the definition of a metrics for Video Analysis performance evaluation. Market constraints are also esamined to classify most traditional solutions (rule-based approaches) and newly emerging solutions (abnormal behaviour analysis). Another classification key will be based on the use of calibrated systems against the uncalibrated approach. The adopted metrics for performance evaluation is based on a carefully selected mixture of video recorded sequences, including background "normal" scenes as well as "acted" scenarios where human actors will play some selected events, (in the category of alarms). The referred experimental results have been collected with two advanced well known technologies in the International market, quite representative of the two different approaches of rule-based and behavioral analytics. The present discussion is not intended to provide any judgement or classification. Rather, it is aimed to help in the definition of some effective tools to be used by technologists in the task of video analysis performance evaluation.

Title: Recent achievements on Multicamera People Surveillance: the Re-identification problem
Speaker: Roberto Vezzani
Abstract:

Title: Game Theory in Computer Vision and Pattern recognition
Speaker:Marcello Pelillo
Abstract: The development of game theory in the early 1940's by John von Neumann was a reaction against the then dominant view that problems in economic theory can be formulated using standard methods from optimization theory. Indeed, most real-world economic problems typically involve conflicting interactions among decision-making agents that cannot be adequately captured by a single (global) objective function, thereby requiring a different, more sophisticated treatment. Accordingly, the main point made by game theorists is to shift the emphasis from optimality criteria to equilibrium conditions. As it provides an abstract theoretically-founded framework to elegantly model complex scenarios, game theory has found a variety of applications not only in economics and, more generally, social sciences but also in different fields of engineering and information technologies. In particular, in the past there have been various attempts aimed at formulating problems in computer vision, pattern recognition and machine learning from a game-theoretic perspective and, with the recent development of algorithmic game theory, the interest in these communities around this topic is growing at a fast pace. The goal of this short course is to offer an introduction to the basic concepts of game theory and to provide an overview of recent work on the use of game-theoretic models in computer vision and pattern recognition problems. I shall assume no pre-existing knowledge of game theory by the audience, thereby making the course self-contained and understandable by a non-expert. [The course is based on tutorials given at CVPR 2011 and ICPR 2010.]