Homepage of Vincent Barra

Modular Ensemble Tracking

vbarra — 2011-07-22T10:06:38Z

Numerous works identify object tracking as a critical issue in many applications such as surveillance and anomaly detection. Among all definitions, the simplest but clearest one defines tracking as the estimation from a video sequence of the trajectory of moving object in the image plane. The tracking by itself is composed of two steps: an object detection step, where potential candidates are identified in each frame of the sequence, and a tracking step, where a specific candidate is tracked all along the frames. Depending on the constraints imposed on these two steps, several methods and algorithms are available.

We herein impose four constraints: we want the tracker to be (i) robust (ii) real-time (iii) able to track pedestrians and (iv) usable from mobile cameras acquisitions. Considering these four points, numerous tracking methods are available, that can be separated into four main categories. Background subtraction, silhouette tracking, points tracking and supervised learning methods. Due to constraints (iii), the -first two ones are inappropriate and we focus on the remaining two other categories.

We propose to have a double point of view, by combining Point tracking and supervised learning methods into a modular version of the Ensemble Tracking Algorithm (MET). Classifiers are trained with Adaboost on homogeneous feature spaces, and the classification decisions are used by a particle filter specially designed for the application.

The MET algorithm is based on two ideas. The main drawback of the classical Ensemble Tracking algorithm relies on the definition of its strong classifier, since each weak classifier works on an heterogeneous feature space. We thus decided to split this feature space into several homogeneous subspaces (called modules) where a strong classifier lives and takes a decision on the position of the object to tracking is computed. Splitting the feature space strongly modifies the objective of the tracking process: a tracking algorithm now has to estimate a hidden state composed on the one hand of the position and the dimensions of the object, and on the other hand of the linear weights of the decisions, leading to the most discriminant observation. The second modification we propose, resolving this second objective, is based on the construction of a specific particle filter jointly managing both the positions and dimensions of the object and the weights of the modules.

Results

Video: comparison Ensemble Tracking / Modular Ensemble Tracking with MCMC

IMG/flv/ET_MC2MET_comparison.flv

Tracking on a mobile camera

Scaling on a camera with mobile optical center.

Associated references

PENNE T, TILMANT C, CHATEAU T, BARRA V, Modular Ensemble Tracking, Proceedings of IEEE International Conference in Image Processing, Theory and Applications 2010, Paris France, 7-10 July 2010

PENNE T, TILMANT C, CHATEAU T, BARRA V, Monte Carlo Markov Chain Modular Ensemble Tracking, Proceedings of VISAPP'12, Roma, Italy, February 24-26 2012

PENNE T, BARRA V, TILMANT C, CHATEAU T., Ensemble tracking modulaire, Actes conférence Reconnaissance des Formes et Intelligence Arti-cielle, RFIA'10, 19-22 janvier 2010, Caen, France.

PENNE T, BARRA V, TILMANT C, CHATEAU T, Une version modifiée de l'Ensemble Tracking, ORASIS'09, Tregastel, juin 2009

Transcranial Magnetic Stimulation

vbarra — 2011-06-29T08:08:04Z

Today, using TMS is still empirical. The clinician who wants to stimulate a specific cortex area given an expected result (phosphene, modification of vision, motor effect) will perform several stimulations, scanning over the patient's scalp to find the most important response. We thus propose to develop a transcranial magnetic stimulation simulator : according to position and orientation of the coil and to the parameters of the stimulation, the simulator will give a modelling of the stimulation effects. Result will be mapped on MRI image of the patient.

The TMS simulator is composed of three main tools:

a numerical tool, to compute magnetic field generated by a given coil at any point of the empty-space. The aim is first to assess the magnetic field induced by the coil, then to provide a finite-element scheme boundaries conditions and finally to finite-element computation in free space.
a design tool, able to take into account the specificity of each winding copper shape. For a given contour, the user has only to define a 3D-parametrical curve and the expression of its derivative
An interface tool, providing some results such as the maps of magnetic and potential vector field, mapped on a 3D brain of the patient, and some interactions ( position/orientation of the coil w.r.t. the patient...)

Associated references

INNOCENTI E, LUQUET S, HILL D, BARRA V, Model driven Engineering for a Transcranial Magnetic Stimulation simulation : Application to software versioning, Proceedings of ESM'10, European Simulation and Modelling Conference, Hasselt University, Hasselt, Belgium, 25-27 Octobre , 2010

LUQUET S., BARRA V, HAUG C., LEMAIRE JJ, Development of a software for transcranial magnetic stimulation : potential magnetic field mapping on human brain, Proceedings of the International Brain Mapping and Intraoperative Surgical Planning Symposium, Los Angeles, 2005.

LUQUET S., BARRA V, LEMAIRE JJ., Transcranial Magnetic Stimulation : Magnetic Field Computation in empty free space, Proceedings of the IEEE Engineering in Medicine and Biology Conference, Shangai, septembre 2005.

LUQUET S., BARRA V, HAUG C., LEMAIRE JJ., Transcranial magnetic stimulation : magnetic field computation using a parametrical coil model, Proceedings of the First Open International Conference on Modelling and Simulation, June 2005.

Topology correction of brain surfaces

vbarra — 2011-06-29T08:00:02Z

Brain surfaces provide a reliable representation for cortical mapping. The construction of correct surfaces from magnetic resonance images (MRI) segmentation is a challenging task, especially when genus zero surfaces are required for further processing such as parameterization, partial inflation and registration. The generation of such surfaces has been approached either by correcting a binary image as part of the segmentation pipeline or by modifying the mesh representing the surface. During this task, the preservation of the structure may be compromised because of the convoluted nature of the brain and noisy/imperfect segmentations. In this paper, we propose a combined, voxel and surface based, topology correction method which preserves the structure of the brain while yielding genus zero surfaces. The topology of the binary segmentation is first corrected using a set of topology preserving operators applied sequentially. This results in a white matter/gray matter binary set with correct sulci delineation, homotopic to a filled sphere. Using the corrected segmentation, a marching cubes mesh is then generated and the tunnels and handles resulting from the meshing are finally removed with an algorithm based on the detection of nonseparating loops. The approach was validated using 20 young individuals MRI from the OASIS database, acquired at two different timepoints. Reproducibility and robustness were evaluated using global and local criteria such as surface area, curvature and point to point distance. Results demonstrated the method capability to produce genus zero meshes while preserving geometry, two fundamental properties for reliable and accurate cortical mapping and further clinical studies.

Associated references

ACOSTA O., FRIPP J., BOURGEAT P., FAVREAU J.M., RUEDA A., BONNER E., CANTOR D., XIAO D., RANIGA P.,BARRA V., SALVADO O., Cortical surface mapping using topology correction, partial flattening and 3D shape context-based non-rigid registration for inter-individual statistical analysis, Human Brain Mapping

GRIS F, FAVREAU JM, ACOSTA O, BARRA V, SALVADO O, A combined voxel and surface based method for topology correction of brain surfaces , spie medical imaging, 13 - 18 February 2010 San Diego, California, USA

GRIS F, FAVREAU JM, ACOSTA O, BARRA V, SALVADO O, A combined voxel and surface based method for topology correction of brain surfaces, ICTCC'09, CSIRO ICT Centre Conference 2009

Cutting Organic Surfaces

vbarra — 2011-06-29T07:20:00Z

Mapping a given surface onto a piece of the plane is an usual problem, and conformal mappings with border free approaches are classical as non-distorting methods. However, algorithms of this category are only able to produce a mapping from a surface topo- logically equivalent to a disc. In the general case, a topological cutting is thus required. The approach we are reporting here is a method that does not only take into account the topologi- cal properties by cutting the original surface into a disc, but also the the geometrical properties by globally selecting the local extrema of original arm- like surfaces, we called organic surfaces.

Associated references

FAVREAU JM, BARRA V, Cutting an organic surface, European Workshop on Computational Geometry, Brussels, 2009

Topographic Analysis of Electrode Contacts in Human Cortical Stimulation

vbarra — 2011-06-29T07:12:59Z

Electric chronic stimulation of the human motor cortex (ECSM) has been reported to alleviate chronic severe pain. However the mechanism of action of ECSM is still hypothetical. This is due mainly to the poor knowledge of, 1) the electric diffusion through the multiple structures beneath the epidural contacts (i.e. dura matter, cerebrospinal fluid space, arachnoid membrane, grey and white matter layers, pie mere and vascular tree), 2) the absence of consensus concerning the stimulation parameters (mono versus bipolar stimulation, cathodic or anodic current) and 3) the detailed cortical topography of the contacts. In this study we focused on the precise identification of the cortical areas covered by the electric contacts in a series of twelve patients operated on for ECSM. We proposed a new automatic tool for topographic analysis able to compute 2D maps from the 3D anatomic MRI with bijective transformation (point-to- point correspondance). Anatomical regions of interest (AROIs) were visually identified, manually outlined and extracted (Iplan, BrainLab, Germany) for further analysis: 1) for the anatomic structures, on pre operative T1-weigthed magnetic resonance imaging (MRI), the frontal (superior or F1, intermediate or F2 and inferior or F3), the pre central and the post central gyrus; 2) for the electrode contacts (Resume, Medtronic, USA), on post operative computerized tomography (CT). After getting white and gray matter membership maps by automatic segmentation, we produced a cortical mask to build a triangular mesh. We defined a homeomorphism between the 3D mesh and a subset of R2 and could apply in consequence the circle packing algorithm. We built depth maps (distance to the skull), distance to- contact maps (distance to a given electrode contact) and anatomic structure maps. Results showed that it was easier to accurately define the location of the contact projection on the cortex allowing physicians to correlate the benefit with the topography. In particular, because of the unfolding, it was easier to integrate the cytoarchitectonics (i.e. the manually identified AROIs) knowledge in the analysis. Beyond the better understanding of ECSM and indirectly of the pathophysiologic process of chronic pain, this new tool might be used in the future for image guided electrode positioning.

Associated references

FAVREAU JM, HEMM S, NUTI C, COSTE J, BARRA V, LEMAIRE JJ, A Tool for Topographic Analysis of Electrode Contacts in Human Cortical Stimulation, ICCV 2007

Low-resolution Surface Mapping

vbarra — 2011-06-29T07:07:04Z

There are several ways to produce 3D acquisitions from a real object (volumic acquisitions, point clouds generated from 3D scanners or cameras) Both of the data structures are usually turned into surfaces described by triangles, using isosurface reconstruction methods . Due to the precision and resolution, the resulting virtual surface may not capture the full topological and/or geometrical details of the original surface, and are prone to artifacts that misrepresent the data.

Wa propose here to tackle the problem of mapping a surface acquired from a real object onto a piece of the plane, taking into account the topological and geometrical properties of the surface, as well as the specificity of low-resolution acquisitions. We introduce a cutting process used to manage the topological constraint of the one-to-one mapping, with some speedup improvements. Specific geometrical constraints linked to the low resolution context can be introduced

In medical imaging, an unfolding approach has been described in a previous work, to produce a flattened map of the region of interest on the cortical surface, using T1-weigthed Magnetic Resonance Imaging scans. After some segmentation steps and a surface reconstruction, a mesh is obtained, modeling the region of interest on the cortical surface. The cutting method used here did not take into account geometry, producing maps with overlaps for high genus original surfaces. The approach described in here was applied to the meshes stemming from the original data. The resulting surface was then unfolded using classical unfolding tools.

(a): A neighbourhood of a central sulcus extracted from the original volumic data. (b), (c), (d): Cortical maps unfolded by Circle Packing after cutting steps (na-ve, topological, and patching + topological). Dark grey shows regions of overlappings.

Associated references

FAVREAU JM, BARRA V, Low-resolution Surface Mapping : a Topological and Geometrical Approach, European Workshop on Computational Geometry, Brussels, 2009

Hidden Markov models and temporal Image Processing

vbarra — 2011-06-22T12:16:55Z

Remote sensing is an important technique for observing Earth surface applied to different areas such as, land use, urban planning, remote monitoring, real time deformation of the soil that can be associated with earthquakes or landslides, the variations in thickness of the glaciers, the measurement of volume changes in the case of volcanic eruptions, deforestation, etc .... To follow the evolution of these phenomena and to predict their future states, many approches have been proposed. However, these approches do not respond completely to the specialists who process yet more commonly the data extracted from the images in their studies to predict the future. In the context of a cotutelle with Tunisia, we propose an innovative methodology based on hidden Markov models.

Associated references

ESSID H, TOUJANI A, BEN ABBES A, FARAH R, BARRA V, Spatio temporal modelling using HMM for satellite imaging, Proceedings of 6th IEEE SETIT, October 26-29, Sousse, 2011

ESSID H, FARAH IR, BARRA V , BENGHZALA H, Analyse de la variation spatio temporelle des objets dans des images satellitaires à base de modèles de Markov cachés couple, Actes de la conférence Extraction et Gestion des Connaissances, Hammamet, Tunisie, 26-29 janvier 2010

ESSID H, FARAH IR, BENGHZALA H, BARRA V , Système d'interprétation spatio-temporelle d'images satellitaires à base de modèle de Markov caché hiérarchique, Actes Spatial Analysis and GEOmatics - SAGEO'09, 25-27 Novembre 2009, Paris, France.

ESSID H, FARAH IR, BARRA V, BENGHZALA H, Approche hybride pour l'analyse de la dynamique d'objets de scènes d'images satellitaires à base de réseaux Bayésiens dynamiques, Actes COSI'2009, 25-27 Mai 2009, Annaba, Algérie.

Elements of machine Learning

vbarra — 2011-01-12T07:55:51Z

Agenda

vbarra — 2010-10-14T15:52:23Z

Slides & homework

vbarra — 2010-02-12T16:17:37Z

the Slides
homework : implementation of Metropolis Algorithm