Democritus University of Thrace
Loading the player ...
- Offer Profile
- The main scope of the Robotics
and Cognitive Systems Group is to perform and promote research in
application problems that rise in the area of robotics, computer vision,
multimodal integration, haptics, image analysis and understanding, quality
control, visual surveillance, intelligent sensory networks. The tools that
the group uses to expand the front of the science and the corresponding
research areas of interest are:
- Artificial Vision (including Machine Vision, Cognitive Vision and
- Intelligent Systems (such as Fuzzy Systems and Artificial Neural
- Sensor Data Fusion
- Pattern Recognition
- Cognitive science is the interdisciplinary study of mind
and intelligence, embracing philosophy, psychology, artificial intelligence,
neuroscience, linguistics, and anthropology. Attempts to understand the mind
and its operation go back at least to the Ancient Greeks, when philosophers
such as Plato and Aristotle tried to explain the nature of human knowledge.
The study of mind remained the province of philosophy until the nineteenth
century, when experimental psychology developed.
Cognitive science intellectual origins are in the mid-1950s when researchers
in several fields began to develop theories of mind based on complex
representations and computational procedures. Its organizational origins are
in the mid-1970s when the Cognitive Science Society was formed and the
journal Cognitive Science began. Since then, more than sixty universities in
North America, Europe, Asia, and Australia have established cognitive
science programs, and many others have instituted courses in cognitive
Visual and other kinds of images play an important role in human thinking.
Pictorial representations capture visual and spatial information in a much
more usable form than lengthy verbal descriptions. Computational procedures
well suited to visual representations include inspecting, finding, zooming,
rotating, and transforming. Such operations can be very useful for
generating plans and explanations in domains to which pictorial
In the Department of Production Management and Engineering (PME) of the
Democritus University of Thrace (DUTH) in Greece, assiduous research in
cognitive vision has been made. Results of this research are the
construction of disparity, saliency and depth maps as also the generation of
algorithms responsible for the extraction of optic flow in complex
backgrounds and the estimation of motion.
- Digital image stabilization is the process that
compensates the undesired fluctuations of a frame’s position in an image
sequence. The techniques for image stabilization are consisted by two
successive units. The first one is the motion estimation unit and the next
one is the motion compensation or correction unit. During the motion
estimation phase, the global motion vector is extracted, which is composed
by two principal components, the indented camera movement and the unwanted
one. The accuracy of this estimation is important due to the fact that the
compensation unit corrects the estimated vector, which means that any
possible mistake will affect the final output.
Digital stabilization preserves the intentional camera movements, while
smoothens the video output from the unwanted oscillations. Almost any
acquired image sequence is affected by noise and undesired camera jitters.
Depending on the application those unwanted fluctuations are caused by a
rough terrain, the shaking of a hand etc. Image stabilization is a
necessity, as vision plays a key role to many applications including
automatic localization, mapping, and navigation. Therefore, the output of
the image sequence should be free from noise, and should be smooth enough in
order for useful results to be extracted. Image stabilization is application
depended. In the case of a camera mounted on an active servo mechanism, the
undesired oscillations are mostly the rotational ones and the stabilization
is implemented by servo motors, which compensate the pan and the tilt camera
movement, respectively. This technique is known as the optical
stabilization. When electronic hardware is utilized the stabilization is
referred as electronic stabilization. Finally, when only pure image
processing techniques are adopted the stabilization is called digital image
stabilization (DIS). This is the process of preserving the intended camera
motion, while removing the unwanted noise and motion effects by means of
digital image processing. DIS is performed in many ways, either real-time or
non real-time, and as pre-process or as post-process.
- The issue of stereo correspondence is of great importance
in the field of Machine Vision. It concerns the matching of points, or any
other primitive, between a pair of pictures of the same scene. Assuming a
calibrated stereo setup, matching points reside on corresponding horizontal
lines. The disparity is calculated as the distance of these points when one
of the two images is projected onto the other. The disparity values for all
the image points comprise the disparity map. Once the stereo correspondence
problem is solved the depth of the scenery can be estimated.
This issue is of interest in the contexts of 3D reconstruction, virtual
reality, robot navigation, Simultaneous Localization and Mapping (SLAM) and
many other aspects of production, security, defense, exploration and
The problem is usually addressed using software implemented
hardware. On the other hand, many tasks require real-time performance
without the use of a PC. As a result there are hardware implemented and
optimized algorithms. The evolution of FPGAs has made them an appealing
choice towards this direction.
- In the last decade, pattern recognition tasks have
flourished and become one of the most popular tasks in computer vision. A
wealth of research focused on building vision systems capable of recognizing
objects in cluttered environments. Generally, recognizing objects in a scene
is one of the oldest tasks in computer vision field and still constitutes
one of the most challenging. Every pattern recognition technique is directly
related with the decryption of information contained in the natural
environment. During the past few years, remarkable efforts were made to
build new vision systems capable of recognizing objects in cluttered
Moreover, emphasis was given to recognition systems based on appearance
features with local estate. Local neighborhood data are discerned and
organized using efficient detectors and descriptors respectively. The main
idea behind interest location detectors is the pursuit of points or regions
with unique information in a scene. These spots or areas contain data that
distinguish them from others in their local neighborhood. It is apparent
that, detector’s efficiency relies on its ability to locate, as many
distinguishable areas as possible, in an iterative process.
In turn, a descriptor organizes the information collected from the detector
in a discriminating manner. Thus, locally sampled feature descriptions are
transformed into high dimensional feature vectors. In other words, parts of
an object located in a scene are represented by descriptors. Putting these
descriptors in logical coherence fulfills the final object representation.
Finally, during the last decade, several important techniques were
presented, such as SIFT (Scale Invariant Feature Transform) and SURF
(Speeded-Up Robust Features).
Research within Funded Projects:
- The fundamental objective of the INFRA project is to
research and develop novel technologies for personal digital support
systems, as part of an integral and secure emergency management system to
support First Responders in crises occurring in Critical Infrastructures
under all circumstances.
The specific objectives of the project fall under the following categories:
Communications objectives, which involve the research and development of an
integral and interoperable wireless communications system that will allow
First Responders to have reliable means of communications as they enter
subway tunnels and buildings with thick concrete walls.
First Responders objectives, which entail the research and development of a
robust indoor site navigation system based on three location sensors (an
inertial sensor, a wireless sensor and a video sensor), a video annotation
system for First Responder PDAs, sensors for real time identification of
radiation exposure and hazardous materials, and applications for gas leakage
and hidden fire detection.
Standardization objectives, which includes R&D of a European level proposal
for the standardization of the framework of communications and applications
as proposed by INFRA.
Demonstration objectives, which consist on the demonstration of the validity
of INFRA’s standards, communications and First Responder applications being
DUTh is responsible for the following key tasks:
- Implementation of a reliable real-time indoor mapping based on
- Implementation of a reliable real-time indoor mapping based on
existing 802.11 Wi-Fi networks
- Visual-inertial Data fusion for indoor mapping
- The project aims to develop a radically new robot
locomotion technology that can effectively be used in a home and/or in a
workplace environment for manipulating small object autonomously or in close
cooperation with humans. Further more the robot could assist human occupants
of the room by following spoken directions, or by offering assistance with
f0 movements or exercises. This new type of mobile robot will be designed to
move fast and in any direction in an indoor environment.
The whole system is divided into several sub-systems: 1. The moving platform
depends on the anchor points-units placed in a raster fixed to the ceiling
of the room, 2. The pendulum-like structure corresponds to the swinging unit
(SU) that hangs on a wire, 3. The necessary vertical movements are provided
by a winding mechanism (WM), 4. Place on the climber unit (CU), 5. The
vision system (VS) comprises of four cameras installed in the four corners
of the room and one mounted on the CU.
DUTh is responsible for the vision
system VS of the ACROBOTER which, in turn, must provide vital visual
- the position of the platform in the 3D working space,
- the topology of possible objects/obstacles in the platform's
The overall goal is to adequately accomplish demanding manipulation
tasks. Furthermore, the VS is responsible for three tasks that affect
directly the overall efficiency of the project:
- estimate the SU's pose in the room,
- the reconstruction of the 3D working environment of the platform,
- the recognition of objects found in the scene.