Academic and industrial research Projects
(the most recent ones)
Unmanned Aerial Vehicles (UAV
or, more informally, drones
) are nowadays widespread in several application fields.
Their success is mostly due to the possibility to achieve cheap and high-resolution aerial views of every place, since they
are relatively small and agile (the last ultra-compact models can even fly indoors). Nevertheless, they are not exploited at their best yet, since most of these applications just
collect high-resolution sequences or perform image analysis (e.g., object detection or planar automatic environmental
measurements) on classical 2D images
. In fact, having multiple data of same objects shot from different points of views
is known to allow 3D object reconstruction
, for instance through SLAM
(Simultaneous Localization And Mapping)
Nevertheless, their numerical instability and need of relavant computational resources has brought these algorithms
to work accurately only in well-controlled conditions, using expensive equipment.
The 3-DARE Project
aims at developing a software system to perform an accurate 3D virtual reconstruction
natural or urban environment
, just exploiting the visual information
delivered by a drone, hence suitable to
all UAV devices and for any application. This is achieved using a quad-core processor, a GPU, and optimization in hardware.
Additional (inertial, magnetic, etc.) sensors available on-board are exploited either
to perform in real-time or to improve the accuracy of the final 3D reconstruction. Then, our image analysis
by exploiting full 3D
This library exploits the experience
we gathered in satellite imaging
, since the first
SLAM algorithm we started developing back in 2007
in the STARS
and subsequently refined in the challenging ASIMO
The Project is partly financed by an Italian private company
3DVirBone - 3D Virtual model of a human Bone tissue generated
through a portable real-time laser scanning system
Precision surgery is one of the main goal of the technologic innovation in orthopedy, this including custom implants, whose
three dimensional prostheses are designed arising from the diagnostic images of the patient and based on the specific
osteo-cartilaginoid defects. Nevertheless, still there is a high number of failures, mainly due to the geometric mismatch between
the implanted prosthesis and the specific articulation, yet more in case of wrist, elbow, shoulder and ankle.
Meanwhile, laser-scanning technologies have taken several steps forward and several devices exist capable to perform a rapid
3D scanning of the most common objects. However, the high diffusion shown by translucent objects, such as bones, jeopardizes
a high-accuracy 3D object reconstruction.
The goal of the 3DVirBone Project is realizing an innovative and highly accurate system, based on an industrial camera and a
low-power laser, to scan a small bone cage to provide its virtual 3D CAD model in real time. This will constitute
the reference model to be manufactured by a robot working in a clean room.
This result can be achieved thanks to the advanced local-based signal denoising techniques developed by CVG in the course
of years, that allow extracting information even when they are subtle, such as in case of low light (e.g., night vision) or
in the presence of reflection and diffusion phaenomena (e.g., non-controlled industrial environment).
This sub-system is a part of a wider system to be realized in the Project "Custom Implants", involving four more
public and private partners.
This project is financed with a public grant of the European Regional Development Fund.
PERFECT - Automatic analysis of hepatic and lung PERFusion through the usE of
CT-4D image reconstruction
The Computed Tomography (CT) has been developed to combine together the diagnostic capabilities of Radiology and Digital Processing,
the latter typically provided by computers. The innovative 256-Slice CT scanner offers the possibility of investigating greater volumes in a shorter amount of time,
thus improving performances in terms of image resolution and definition, in particular for vascular structures, and reducing the radiation dose for the patient.
The perfusion analysis represents a valid support to assess the effectiveness of the therapies aiming at stopping the vascularization typical of tumour lesions.
This could permit to evaluate the possible reduction of the tumour neoangiogenesis even at the early stage, before observing the modifications in size,
which occur just at a subsequent stage.
The PERFECT Project aims at developing new methodologies based on semi-automatic analyses of a temporal sequence of 3D-TC images,
capable to enable quantitative assessments of the effectiveness of treatment for lung and hepatic lesions, through
analysing the temporal evolution of some parameters (radiometric, geometrical, statistical, etc.), extracted from
region of interests inside the lesions, acquired in a short time and manually selected by an expert radiologist.
Tracking these image-based parameters over all the image sequences acquired throughout the Dynamic Contrast Enhanced TC examination, allows measuring the perfusion parameters
needed to quantify the neoangiogenesis degree, and the effectiveness of therapy, accordingly. All the
algorithms are conceived for high performance distributed SIMD/MIMD architectures, so to allow even real-time
4D image analysis.
The project has been partly funded by the IRCCS - Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori - IRST
(The Cancer Institute of Romagna).
STAMINAL - Characterization of
STem cells by means of AutoMatIc
aNALysis of microscopic images in pre-clinic therapy
Nowadays, acute and chronic respiratory diseases (such as, pulmonary emphysema and fibrosis) are more and more widespread,
due to life style (e.g. cigarette smoke) and the increasing urban pollution, and these represent the third cause of death
after tumours and cardio-circulatory diseases, thus inducing relevant sociomedical problems.
Normal pulmonary stem cells offer a novel opportunity in regenerative medicine for clinical applications in
the treatment of chronic degenerative diseases, including Idiopathic Pulmonary Fibrosis (IPF).
The STAMINAL Project supports the on-going studies for biomolecular characterization of proliferation and differentiation of
tumour's and IPF's pulmonary stem cells. In particular, the Project aims at developing a tool for automatic culture analysis
of normal or aberrant (tumours' or IPF's) pulmonary stem cells. The possibility to monitor and to assess cell cultures enriched
with normal pulmonary stem cells through common microscope inspections without any intervention by operators would allow the
stem cells to be employed for grafting in the keeping of the Good Manufacturing Practice (GMP) rules.
The project has been partly funded by the Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori - IRST
(The Cancer Institute of Romagna).
advanced Analysis of the Sequence of Images acquired by the Mercury Orbiter
The ASIMO Project
is carried out in the framework of the BepiColombo
mission of the European Space Agency
(ESA), in cooperation with Japan, aiming at exploring the planet Mercury, and it is part of
(Mercury Orbiter Radioscience Experiment), one of the two experiments
funded by the Italian Space Agency (ASI). Among the goals of MORE
it will help to determine the size and physical state of Mercury's core, based on the analysis of its rotational state.
The rotational state of a celestial body is characterized by its obliquity and librations
. The obliquity is
the angular amplitude between the normal to the orbital plane of the planet and its spin axis direction. The physical
librations are characterized, in the case of Mercury, by a longitudinal motion (which takes the form of a
sinusoidal wave) superimposed to the nominal planet spin motion. This behaviour is due to the Sun gravity-gradient
effect inducing a forced libration whose maximum amplitude is expected to be about 400 m at the equator.
to estimate the rotational state of any solar system body, it is possible, in principle, to use complex dynamical
models where the fitting functions are very general, so that all possible dynamical states are considered. However,
in this case it is not accurate enough.
As a spinoff of the MORE
Project, the goal of the ASIMO Project
is to compute the libration of the planet. In particular, for the study of the Mercury
's rotational state we exploit the particular geometry of the problem
that permits the implementation of a more meaningful method, where the estimation problem is performed in two successive steps.
First, the estimation of the obliquity is performed and then, using a provisional reference frame, the residual information
is used to estimate the amplitude of the forced libration. The observations are computed by finding out the parameters of
the geometric transformation by correlating pairs of optical images of the planet surface, both of them partly looking
over the same region, at different times, under different scale and illumination condition. For each pair, from the shared
region we can extract the invariants that permit to determine the parameters of the geometric transformation between the
couple of views and, subsequently, to fully determine the three components of the libration, with sub-pixel
As a spinoff of the MORE Project, the ASIMO Project is funded by the Italian Space Agency (ASI)
This project tackles the problem of high dynamic range image analysis and
metrology. The objective of the AUTOBEAM Project
is to characterize, both from a geometric
(the profile) and a photometric point of view, the light distribution emitted by different automotive
headlamp beams projected on a panel, within the range of accuracy required by the European regulations.
The high accuracy in the vehicle's alignment is provided by recovering in real time
the 3D instant trajectory of a vehicle while approaching the measurement system, using natural
markers and 3D sparse feature matching with stereoscopy. After alignment, a commercial low cost CCD camera has been used.
Exploiting the camera's response function (recovered experimentally) an adaptive segmentation
algorithm has been developed that permits to perform accurate measurements
of the luminous beam profile as perceived by human eyes
It is worth noting that this is the first automatic prototype using a low cost CCD
sensor to assess the compliance of vehicle's headlight beams with the European regulations.
The project has been partly funded by industry
STARS - Standalone
spacecraft oRientation Sensors
Current attitude determination systems on-board Earth orbiting, Nadir pointing spacecraft rely on Earth crossing
sensors which allow the detection of the Earth/atmosphere horizon and, hence, the Earth centre direction. This
information enables the computation of two attitude angles, while the angle about the sensor line-of sight remains
unknown, unless another source of attitude information is available.
The goal of STARS
, carried out in collaboration with
the Aerospace Group of DIEM
is the development of novel standalone spacecraft attitude sensors, capable of estimating
the full three-axis orientation of an Earth-orbiting satellite. The novelty of the project consists in developing
an approach that only exploits the real time analysis of image sequences of Earth surface captured from space,
without using any information from other sensors.
The challenging task is to design an algorithm for pose estimation relying on image registration and able to measure
the attitude Euler angles whitin the strict accuracy required (1 arcsec).
The project has been funded by the University of Bologna as a Strategic Project
The increasing processing power available on PC and embedded architectures has opened the doors to real time
3D analysis of crowded scenes, where the environment under surveillance is completely covered by a loosely
synchronized camera network.
aims to exploit the experience gathered by CVG
in 3D people tracking and analysis
carried out with couples of cameras and
3D stereoscopy to develop a tracking people system for wide areas. The network camera system is self calibrating and the core of
the application is the people tracking technology we have developed. On top of the outcome of our group, it will be
developed a behaviour analysis system and an effective decision-making system, by two more academic units,
respectively. A Small Enterprise will develop the commercial product and a Large Enterprise is in charge of
bringing the system to the market.
The automated analysis of images and sequences plays a crucial role in most of the computer vision systems for
medical images. The assessment of pathologies or treatments is often entirely still left to human (visual) evaluation,
because of the objective difficulty to develop automated (or computer aided) assessment systems, also due to the human
health being a critical issue.
DERMOCAL aims at developing a portable and low-cost system able to achieve automatically, in vivo and routinely,
measurements of the skin surface to quantify the skin changes in dermo-cosmetic treatments. The system is based on a
2 cm2 capacitive device.
Several features have been extracted from the 3D skin topographic structure by using advanced image processing
and analysis methods.
The accuracy achieved in measuring depth and width of the skin's microrelief makes our system apt to quantify changes of
the skin surface and to be used in comparative follow-up studies in dermatology and dermocosmetics, accordingly.
The project has been partly funded by the University of Bologna.