Dipartimento di Sistemi e Informatica
Visual Information Processing Laboratory
Image and Video
Content Based Retrieval
Visual information retrieval has become a major area of research due to the ever increasing rate at which images are being generated in many application fields. Given a collection of images or videos, the purpose of VIR is to retrieve those images or videos which are relevant to a visual query.
The PICASSO system developed at the Lab supports retrieval of images by directly addressing image visual features such as color, shape and spatial relationships. The system supports querying by visual examples. This paradigm requires that the user provides a prototype image which is used as a reference example so as to express perceptual aspects of low/intermediate features of visual content.
Prototype systems have been recently developed, supporting retrieval by content for a number of video categories, such as news, movies, and commercials. Video content representation involves the following steps: video sequence segmentation into coherent shots through the extraction of video editing effects (cuts, dissolves, etc.); description of the visual content each shot (based on colors, shapes, camera motions, number of independently moving objects, etc.). Video querying is made by real examples, by virtual sequences generation, by gestures.
Real-Time Computer Vision
The research on Computer Vision at the Lab focuses on devising lightweight algorithms and strategies for the processing and interpretation of image sequences that allow real-time applications such as gesture-based human-computer interaction and robot control.
Recent technological advances allow experimentation and development of new human-machine interfaces. Visual interfaces permit visualization and interaction with information in a more natural way with respect to common 2D graphics interfaces. On the one hand, dynamic tracking of several parts of the human body (head, eyes, hands) as well as the use of input devices with several degrees of freedom, support multimodal communication from user to machine. On the other hand, 3D computer graphics support representation of real world scenes and efficient display of multidimensional data. In order to improve acceptance and effectiveness of these systems, usability engineering also plays a central role in the lifecycle of interactive systems. At the Lab, 3D graphics have been successfully employed in different contexts. Recent projects include the development of 3D graphics interfaces to access image databases, as well as virtual environments to support training of construction workers. Lab as also carried out the usability evaluation of some non-immersive virtual reality systems.
Original algorithms have been developed for:
(1) 2D and 3D parameter
estimation (optical flow, time to collision),
(2) motion based image segmentation,
(3) extraction of object pose and structure,
(4) visual tracking (corners, active contours),
(5) design of new space-variant sensors.
Prototype systems recently developed include:
(1) Image-based visual servoing for robot navigation and manipulation,
(2) Collision monitoring for obstacle avoidance in autonomous vehicles,
(3) Moving object identification and tracking for visual surveillance,
(4) Eye, head and hand tracking for advanced human-machine interaction based on pointing and clicking actions.
web page: http://viplab.dsi.unifi.it
Research | Teaching
to reach DSI
Services (DSI) only | Webadmin