I am interested in various aspects of cognition and intelligence, especially as it relates to visual perception and understanding.
The underlying motivation is fourfold. Firstly, to gain a better understanding of myself and the world by
investigating what it fundamentally means to see, analyse, understand, conceive, and communicate.
Secondly, to share the insights with interested parties, in particular similarly motivated graduate students and research collaborators.
Thirdly, to master and be able to effectively teach fundamental and practical aspects of computer science and engineering.
Fourthly, to apply my creativity, knowledge, and know-how to help develop original solutions to complex real-world challenges.
My main field of research is cognitive
This work is in continuation with over thirty years of research and development on one
main theme that is, generic modelling and recognition of unexpected objects in images of complex scenes.
A number of efforts resulted in algorithmic methods and software systems for the
extraction, grouping, and interpretation of topologically-valid
constant-curvature contour primitives-based descriptions of multi-part
volumetric objects from static intensity images.
Various psychological and computational models were used as inspirations towards
building hierarchies of qualitative categorical representations of objects at multiple scales.
The expertise developed from my work in cognitive computer vision, coupled with the expertise developed in chairing computer engineering programs,
developing and teaching an advanced undergraduate course on the conception, analysis, and simulation of parallel real-time systems and various design
courses from basic methodology to final-year projects, includes, apart from general and specialized computer science and engineering elements,
a number of other elements that give me a unique ability to contribute to multi-disciplinary challenges with complexity, logistics, simulation,
visualisation, optimisation, and decision-making issues.
Presently active computer vision projects address the anytime classification of interactive actions from regular video,
the detection and characterization of objects carried by persons, from regular or regular/infrared video, the
classification of scenes from static images, and the precise segmentation of MRI images for custom-made prostheses.
Past cognitive computer vision projects include
MAGNO (Multi-level Access to Generic Notable Objects)
ConStruct (Object-level Sructured Contours),
two large computer vision systems developed at the contour and junction grouping
level. Their shared goal was to generically provide an object-level
structured description of the man-made objects in a static image.
The whole approach was integrated in the
PLASTIQUE (Part, Link, and ASsociated Template Image QUEry)
computer vision system, a successor of
PARVO (Primal Access Recognition of Visual Object),
and applied to automatic content-based
indexing and retrieval of images of objects from large databases.
More recent efforts aimed at applying the developed models to
the observation and communication of behaviors in
MONNET (Monitoring of Extended Premises: Tracking Pedestrians Using a Network of Loosely Coupled Cameras),
was a distributed system developed in the context of a large team project
in human motion capture and action modeling from image sequences.
A project issued from MONNET aimed at automatically
modelling pedestrians' gaits for the purpose of comparison. This was obtained despite
variations in walking directions and viewpoints. Possible applications are
in the surveillance and health care domains. Another project aimed at tracking
and modelling animals in challenging natural environments.
Past projects not part of my main cognitive computer vision theme include robot vision (2D space mapping by a mobile
robot, 3D pose estimation for prehension, arm control), range image analysis (multiview
registration, generalized cylinder extraction, multiview integration using geometry and
photometry, 2D-3D fusion, surface inspection), and applications (road sign extraction,
wireframe extraction using smart sensor).
All projects are made possible thanks to undergraduate and graduate students and external collaborators.
On a more fundamental level, I am also actively interested in research methodology, specifically in clarifying
the respective roles of formalism and subjectivity. This is a fundamental topic
in artificial intelligence and philosophy, having connections with fundamental
concepts such as knowledge, reality, interpretation, proof, and truth.
I have been busy developing and promoting an appropriate research methodology addressing
concerns related to the above motivations. The SAFE (Subjectivity And Formalism Explicitely) methodology
has been applied to the figure-ground segmentation problem which considers situations where
expectations about the contents of the observed scene are limited and generic.
methodology was developed in the context of a more global personal reflection on the role and impact
of communication and language in science and society.
My conclusions following that reflection were summarized in
DESHU (Discours épistémologique de la science humaine),
a text comprising 75 loosely-ordered but strongly-coupled propositions
addressing scientific research and models, from physics to biology, and
including computer vision up to generic object recognition.
Main themes comprise elementary and generic entities, properties, categories,
structure, space, time, infinity,
theory, experimentation, prediction, subjectivity, formalism,
existence, selection, state, measurement, perception,
recognition, interpretation, validation, and language.
This text has not been submitted for publication yet but it could be made available upon individual request
(it is written in French).
Following these works, IbTT (Image-based Turing Test) is proposed as an alternative
to the Turing Test for artificial intelligence. It may be viewed as
an experimental validation method for generic object recognition systems, developed
according to SAFE principles.
In systems modelling and simulation, I am interested to apply my model for Parallel Real-Time Systems to any human activity domain, including
science and technology, at any abstraction, detail, and complexity level.
I am a follower of Comprehensive Anticipatory Design Science as defined by R. Buckminster Fuller.
Cognitive Computer Vision, Generic Object Modelling, Detection and Recognition, Object Discovery,
Categorization, Parts, Shape, Contours, Junctions, Primitive Maps, Grouping, Segmentation,
Scale, Systems, Complexity, Parallelism, Simulation, Temporal Integration, Multi-criteria Optimisation, Datasets,
Research and Development Methodology, Video Analysis, Gait Modelling, Animal Tracking,
Activity Recognition, Scene Recognition, Carried Object Detection and Characterization, Medical Image Segmentation.