Home People Research Projects Teaching Theses Downloads Links   Contacts English Italiano

Research activities

2D and 3D medical imaging has become an extremely effective tool for early detection, diagnosis, follow up and assisted therapy in several diseases, mainly but not only in solid tumours. This has become possible through the improvement of technologies, instrument resolutions and image analysis techniques. Effective results in this field, yet more than in other ones, come from multi-disciplinary researches. Our group is active to cooperate with researchers in medicine, biology and nuclear physics, to develop new instruments and technologies, providing our specific expertise in image analysis, reconstruction and measurements.
On the other side, image analysis constitutes the enabling technology in several fields, ranging from automatic cell analysis to cosmetics, even when the detecting device is based on a non-optical sensor. Besides, part of our research activity is also spent in application-oriented projects in collaboration with researchers in medicine and biology.
- CT perfusion (PERFECT)
- Microscopy imaging (STAMINAL, DYNAMO)
- Digital mammography
- 3D Positron Emission Tomography (3D PET)
- Combined tomographic and emission methods (CT, SPECT)
- Characterization of the skin cells and wrinkles by using capacitive image analysis

Aerial and satellite imaging represent an increasing research area that is spreading its application domain, from environmental monitoring, to 3D aerial view reconstruction and panorama, to self-localisation to determine attitude and position (i.e., the pose) of flying vehicles or orbiting satellites exploiting image-based information only. However, merging this information with data coming from the status vector of the flying object works to strengthen measurement computations and improve the attitude accuracy, accordingly.
The CVG's quantitative imaging techniques have been developed and successfully applied to reconstruct the trajectory of an Earth LEO satellite, as well as to help determining the libration of the Mercury planet. The same technologies allow reconstructing a 3D aerial view acquired using a commercial low-cost drone.

Vision-based methods are more and more entering into diverse application fields due to three main advantages that the technology has brought in. First, vision-based systems can mimic the simplest human brain functions to perform automatically tasks like target's counting and classification, or even defect analysis. Second, a digital image yields photometric as well as geometric information and a video sequence offers information regarding the evolution of parameters of the objects being studied. This means that one cheap CCD sensor could replace an ensemble of traditional sensors, thus making a system far simpler and, possibly, cheaper. Finally, vision-based systems of course provide a visual feedback on what is being measured and, in general, on what happens. This makes these systems the best likely candidate to help, or even replace, human beings to automatize measuring tasks and to achieve reproducible and repeatable measurements.
- Outlines of image-based measurements and 3D stereoscopy
- Measurements of vehicle trajectory using 3D stereoscopy
- Measurements of geometric properties of a vehicle beam's light profile
- Measurements of angular relative motion between camera and shot object

Most of the information useful for scene understanding stem from motion detection. However, changes in scenes may also be originated by either acquisition sensor noise or more simply by changes in the environmental illumination. The first step of motion analysis addresses automatic methods to discriminate between uninteresting scene changes and real motion. Nevertheless, moving objects (people included) are often detected together with their cast shadow, which must be removed before analysing each object and extracting its properties.
After that, the objects detected as being moving are followed in time, by automatically identifying and tracking their features that keep invariant. For instance, this permits to count and classify the moving objects. Finally, their trajectory is built in real time and analysed so as to discover events or infer interesting behaviours.
Applications of our algorithms have been found in video surveillance, detection of events in traffic monitoring and behaviour analysis of people walking in museums, fairs, etc.
- Motion detection
- Removal of moving shadows
- People/Object tracking and event detection

Image registration is concerned with the extraction of coherent and coordinate information from collection of images of the same subject matter, coming from either video streaming or individual images.
An image mosaic is a single image that is generated by transforming and merging the set of partially overlapping smaller images. Transformations can be both geometric and photometric ones. An accurate determination of the inter-frame spatial and photometric relationships (registration) leads to the creation of one image with a greater spatial extent (mosaic) and/or augmented details (super-resolution).
- Automatic real time image mosaicing
- Joint spatial and tonal registration
- Motion Detection with Pan-Tilt-Zoom Cameras using Background Mosaics
Copyright 2008, A.G. - All Rights Reserved