Publications

  • Framework interoperability through a standardized approach based on model-driven engineering

Abstract : Software applications need to be more flexible and communicative across their environment. Nevertheless, it is not always obvious as they face incalculable heterogeneity and constantly evolving technologies. Interoperability today imposes a great decoupling effort. This will promote the reusability of the application potential and above all a facility for interoperability management. In this study, a conceptual interoperability framework namely Software Framework Interop (SFI) is proposed. It is based on a standardized architecture while characterized by a model-driven approach. The aim of SFI is to decouple the software bricks while ensuring their interoperability.

23rd IEEE International  conference on engineering, technology and innovation, june 27-29 2017, Madeira Island, Portugal, http://www.ice-conference.org/

  • Detection of urban trees in multiple-source aerial data

Abstract : Standard Remote Sensing analysis uses machine learning methods such as SVMs with HOG or SIFT descriptors, but in recent years neural networks are emerging as a key tool regarding the detection of objects. Due to the heterogeneity of remote sensing information (optical, infrared, DSM) the combination of multi-source data is still an open issue. In this paper, we focused on localization of urban trees, and we evaluate the performances of CNNs compared to standard classification methods that employ descriptor-based representation.

42nd IEEE International conference on acoustics, speech and signal processing, march 5-9 2017, New Orleans, United-States

  • Evaluation of deep learning methodology for localization of urban trees in multi-source aerial data

Abstract : Object detection is a fundamental problem in aerial remote sensing images. Most of standard Remote Sensing analysis uses machine learning methods based on image descriptions as HOG or SIFT and a classifier as SVM. But in recent years neural networks have emerged as a key tool regarding the detection of objects. Due to the heterogeneity of remote sensing information (optical, infrared, LiDAR) the combination of multi-source data is still an open issue. In this paper, we focus on localization of urban trees in multi-source (optical, infrared, DSM) aerial images, and we evaluate the performances of CNNs compared to more standard classification methods based on a descriptor with a classifier scheme.

IEEE Geoscience and remote sensing letters, http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?reload=true&punumber=8859

  • Speeding-up a convolutional neural network by connecting an SVM network

Abstract : Deep neural networks yield positive object detection results in aerial imaging. However, real-time detection remains unfeasible. To deal with the massive computational time required, we propose to connect an SVM Network to the different feature maps of a CNN. After the training of this SVM Network, we use an activation path to cross the network in a predefined order. We stop the crossing as quickly as possible. This early exit from the CNN allows us to reduce the computational burden. Experimental results are obtained for an industrial application in urban object detection. The computational cost is reduced by 98%. Additionally, performance is slightly improved; for example, for a 55% recall, precision increases by 5%.

IEEE International conference on image processing, September 25-28 2016, Phoenix, United-States, http://2016.ieeeicip.org/default.asp

  • La véritable route de la ” Smart City ” à la française

Mustapha Derras répond à une tribune publiée dans Les Echos « La Smart City à la française fait fausse route » …

https://www.lesechos.fr/idees-debats/cercle/cercle-165550-la-veritable-route-de-la-smart-city-a-la-francaise-2062193.php

  • Doit-on parler de smart city “inclusive” ?

La ville, qui est par construction un lieu de convergence réunissant de nombreuses populations variées et diversifiées, n’a pas attendue d’être smart pour mettre en œuvre des actions visant à améliorer la qualité de vie …

http://www.smartcitymag.fr/article/152/doit-on-parler-de-smart-city-inclusive

  • Axes technologique, juridique et éthique du Big data par Berger-Levrault

Quels sont les réalisations concrètes du Big data à l’ouverture massive des données publiques dans l’Union européenne ? Pourquoi les nouvelles réglementations vont-elles impacter les éditeurs de logiciels en France ? Comment réintégrer la notion de confiance sur le Big data dans l’industrie du logiciel pour rester différenciant, et donc compétitif, face au GAFA ?

https://www.appvizer.fr/magazine/experts/axes-technologique-juridique-et-ethique-du-big-data-par-berger-levrault-1466078998

  • DECO: a Design Space for Device Composition

Abstract : Numerous interaction devices are designed through device composition. However, there is no conceptual support for this process and designers are left out to explore the space of combinations in an ad-hoc manner. In this paper we propose a design space for device composition, DECO. This design space is built around two main axes, namely Physical arrangement, which describes how elements are physically combined, and Physical manipulation, which describes how users manipulate each element. We first classify existing devices using our design space and then compare four of them to illustrate their similarities and differences. Using DECO, we design a new compound device: RPM2. This device is based on the combination of a regular mouse with the Roly-Poly Mouse. We describe in detail the user-centered iterative design process that leads to the final prototype. Finally, based on our experience designing RPM2, we propose a set of design guidelines to apply DECO to device composition.

Designing Interactive Systems, June 4 – 8 2016, Brisbane, Australia, http://www.dis2016.org/

  • Optimizing color information processing Inside an SVM network

Abstract : Today, with the higher computing power of CPUs and GPUs, many different neural network architectures have been proposed for object detection in images. However, these networks are often not optimized to process color information. In this paper, we propose a new method based on an SVM network, that efficiently extracts this color information. We describe different network architectures and compare them with several color models (CIELAB, HSV, RGB…). The results obtained on real data show that our network is more efficient and robust than a single SVM network, with an average precision gain ranging from 1.5% to 6% with respect to the complexity of the test image database. We have optimized the network architecture in order to gain information from color data, thus increasing the average precision by up to 10%.

Electronic Imaging, February 14-18 2016, San Francisco, USA, http://www.imaging.org/ist/conferences/ei2016/index.cfm

  • Class Model Normalization : Outperforming formal concept analysis approaches with AOC-posets

Abstract : Designing or reengineering class models in the domain of programming or modeling involves capturing technical and domain concepts, finding the right abstractions and avoiding duplications. Making this last task in a systematic way corresponds to a kind of model normalization. Several approaches have been proposed, that all converge towards the use of Formal Concept Analysis (FCA). An extension of FCA to linked data, Relational Concept Analysis (RCA) helped to mine better reusable abstractions. But RCA relies on iteratively building concept lattices, which may cause a combinatorial explosion in the number of the built artifacts. In this paper, we investigate the use of an alternative RCA process, relying on a specic sub-order of the concept lattice (AOC-poset) which preserves the most relevant part of the normal form. We measure, on case studies from Java models extracted from Java code and from UML models, the practical reduction that AOC-posets bring to the normal form of the class model.

Concept Lattices and their Applications, October 13-16th 2015, Clermont-Ferrand, France, http://cla2015.isima.fr/

  • An efficient multi-resolution SVM network approach for object detection in aerial images

Abstract : In this paper, we deal with the problem of object detection in aerial images. A lot of efficient approaches are based on the use of a cascade of classifiers processing vectors of descriptive features such as HOG. In order to take into account the variability in dimension of the objects, feature vectors at different resolutions are often concatenated in a large descriptor. This prevents from taking into account explicitly the different resolutions and results in losing valuable information. To overcome this problem, we propose to use a new method based on an SVM network. Each resolution is processed, regardless to the other ones, at the input layer level by a dedicated SVM. The main drawback of using a network is that the computational complexity for the classification phase drastically increases. We propose then to foster an incomplete exploration of the network by defining an activation path. This activation path determines an order to activate the network neurons, one after the other and introduces a rejection rule which allows the process to end before crossing the whole network. Experimental results are obtained and assessed in an industrial application of urban object detection. We can observe an average gain of 17% in precision and a computational cost divided by more than 5 with respect to a standard method.

Machine Learning for Signal Processing, september 17-20 2015, Boston, USA, http://mlsp2015.conwiz.dk/home.htm

  • The Roly-Poly Mouse: Designing a Rolling Input Device Unifying 2D and 3D Interaction

Abstract : We present the design and evaluation of the Roly-Poly Mouse (RPM), a rolling input device that combines the advantages of the mouse (position displacement) and of 3D devices (roll and rotation) to unify 2D and 3D interaction. Our first study explores RPM gesture amplitude and stability for different upper shapes (Hemispherical, Convex) and hand postures. 8 roll directions can be performed precisely and their amplitude is larger on Hemispherical RPM. As minor rolls affect translation, we propose a roll correction algorithm to support stable 2D pointing with RPM. We propose the use of compound gestures for 3D pointing and docking, and evaluate them against a commercial 3D device, the SpaceMouse. Our studies reveal that RPM performs 31% faster than the SpaceMouse for 3D pointing and equivalently for 3D rotation. Finally, we present a proof-of-concept integrated RPM prototype along with discussion on the various technical challenges to overcome to build a final integrated version of RPM.

Human Computer Interaction, 18 – 23 April 2015, Seoul, Korea, http://chi2015.acm.org/

  • Detection of manhole covers in high-resolution aerial images of urban areas by combining two methods

Abstract : The detection of small objects from aerial images is a difficult signal processing task. To localise small objects in an image, low-complexity geometry-based approaches can be used, but their efficiency is often low. Another option is to use appearance-based approaches that give better results but require a costly learning step. In this paper, we treat the specific case of manhole covers. Currently many manholes are not listed or are badly positioned on maps. We implement two conventional previously published methods to detect manhole covers in images. The first one searches for circular patterns in the image while the second uses machine learning to build a model of manhole covers. The results show non optimal performances for each method. The two approaches are combined to overcome this limit, thus increasing the overall performance by about forty percent.

Joint Urban Remote Sensing Event, 30 March – 1 April 2015, Lausanne, Switzerland, http://www.jurse2015.org/

  • Comparaison de la segmentation pixel et segmentation objet pour la détection d’objets multiples et variables dans des images

Résumé : Cet article étudie et compare deux méthodes de segmentation. La première est la segmentation par objet [VJ01] où l’on cherche à détecter des fenêtres à partir d’un modèle. La seconde est la segmentation pixel [ARTLdM10, FVS08], où l’on cherche à déterminer à quelle classe appartient chaque pixel. De plus, nous proposons une extension au modèle classique de détection par cascade de HOG [ZYCA06] en utilisant les sacs de mots visuels. Des expérimentations sur des jeux de données réelles permettront la comparaison et mettront en avant un gain de performance non négligeable de notre méthode.

COmpression et REprésentation des Signaux Audiovisuels, 25-28 novembre 2014, Reims, France, http://reimsimage2014.univ-reims.fr/coresa-2014/

  • Conception d’un Dispositif pour Interagir avec des Données Multidimensionnelles : Disco

Résumé : Dans le contexte de l’accroissement du nombre, de l’accessibilité et de l’usage des données numériques dans différents contextes, des situations interactives de plus en plus longues et complexes sont effectuées. En s’inspirant de travaux fondateurs proposant des souris à multiples degrés de liberté, cet article présente la conception d’un nouveau dispositif d’interaction basé sur le principe du culbuto : Disco. Nous présentons divers scénarii d’usage et explorons la manipulation de ce dispositif au travers de deux études. D’abord nous observons la prise en main de trois versions de Disco avec différents facteurs de forme. Ensuite nous étudions les capacités et limites liées à la translation, rotation ou inclinaison physique de deux versions de Disco selon trois prises en main différentes. A partir des résultats nous proposons des guides de conception afin de créer des techniques d’interaction qui exploitent au mieux les différents degrés de liberté du dispositif pour interagir avec des données multidimensionnelles.

Interaction Homme-Machine, 28-31 octobre 2014, Lille, France, http://ihm14.lille.inria.fr/

  • Automatic localization of tombs in aerial imagery : application to the digital archiving of cemetery heritage (2013 – Full Paper)

Abstract : This paper deals with digital archiving of cemetery heritage. A built cemetery is a tangible evidence of historical and cultural periods through the style and the shape of tombs. It gives quantitative information on the local population, about its history (by reading birth and death dates), its culture (by analysing name typology) and its temporal evolution (by using the family names written on the tombs). There is thus a crucial need to archive cemetery data for heritage purposes. The first step for digital archiving is to locate the tombs. A practical way is to use aerial images. We propose to automate this process by using image processing algorithms. This is a challenging problem, as in aerial images, tombs have very variable appearance, size and disposition, and many artefacts can occur such as occluding vegetation, shadows or walking people. We focused our study specifically on French cemeteries in Haute-Marne department, all located in villages close to the Langres city. We compare three automated localization methods. All the preliminary results are commented and we discuss other image-processing applications which could be used to enrich cemetery archiving such as writing recognition on headstones.

International Congress, 28 october – 01 november 2013, Marseille, France, http://www.digitalheritage2013.org/

  • Ouvrir le champs des possibles

http://fierdetredeveloppeur.org/?p=221

  • Berger-Levrault : « Que l’imagination reprenne ses droits »

http://www.lejournaldesentreprises.com/editions/31/actualite/entreprise-du-mois/berger-levrault-la-piste-du-demenagement-se-precise-07-12-2012-169471.php

  • Berger-Levrault gagnant de la catégorie « SAAS : Enterprise Solution » des European Software Excellence Awards 2012
Berger-Levrault, expert de la réglementation française, a été primé lors des European Software Excellence Awards 2012, dans la catégorie « SAAS : Enterprise Solution », pour sa solution logicielle e.Enfance
  • Berger-Levrault et Microsoft innovent dans la gestion des établissements de santé (Full Paper)

Microsoft et Berger-Levrault, éditeur de solutions d’avenir pour la fonction publique et la santé s’associent pour proposer de nouvelles solutions de gestion adaptables et performantes aux professionnels de la santé. Grâce aux technologies Microsoft Visual Studio 2010 et Silverlight[1], la plateforme Santé de Berger-Levrault apporte des solutions logicielles évolutives, centrées sur le métier et les usages. Finances, Ressources Humaines, Patients/Résidents, les modules Berger-Levrault permettent une gestion optimale des établissements de santé et se révèlent de puissants outils d’aide à la décision.

http://www.microsoft.com/france/hub-presse/communiques-de-presse/fiche-communique.aspx?EID=c9a095ec-e636-41ec-a39d-c0757d083862

Early 2007, we decided to use a GWT frontend backed by a Hibernate/Spring server stack to build our new Financial and HR software offer. As far as Berger-Levrault is concerned, this represented a deep change as we historically used proprietary environments (Microsoft® mainly) to code our products. Since then, the architectural model we worked on became the firm most advanced and promising technical stack and is now the environment of choice for most of our new projects. Moreover, we are working on a SaaS offer for our GWT products as the firm is undergoing a deep evolution towards open source environments.

http://googlewebtoolkit.blogspot.com/2011/07/gwt-spotlight-berger-levrault.html

  • Les solutions de Berger-Levrault embarquent des processus d’intégration Talend (Full Paper)

Talend, l’un des leaders mondiaux des logiciels open source, annonce aujourd’hui la conclusion d’un partenariat avec Berger-Levrault, éditeur de solutions d’avenir pour la fonction publique et les collectivités locales. A travers ce partenariat technologique, Berger-Levrault intègre dans ses solutions logicielles de gestion des processus de traitements de données développés avec Talend Open Studio, afin d’accélérer les processus de traitements de données réalisés dans le cadre de leur exploitation.

http://fr.talend.com/press/Les-solutions-de-Berger-Levrault-embarquent-des-processus-d-integration-Talend-pour-aider-les-etablissements-du-secteur-public-a-accelerer-leurs-processus-de-traitement-de-donnees.php

  • La conduite de l’innovation : Enjeux et bonnes pratiques en matière de R&D Logicielle (Full Paper)

Abstract : “Le métier d’éditeur de logiciels se distingue des métiers de services informatiques par une composante fondamentale, l’investissement. En particulier, éditer un logiciel, c’est prendre le pari de construire, avant même de l’avoir vendue, une solution qui sera compétitive au moment de sa mise sur le marché et qui pourra être rentabilisée par le nombre de clients qui l’achèteront. De fait, gérer correctement l’effort de recherche et développement, aussi bien dans sa composante de coût que de valeur, est le nerf de la guerre du métier d’éditeur de logiciels. On comprend alors aisément qu’il ne s’agit pas simplement d’apporter une réponse technique à un instant t à un besoin identifié et bordé, mais également à anticiper l’avenir, voire à le dessiner. Pour cette raison, la dimension de « Recherche et Développement » chez un éditeur de logiciels doit procéder d’une réelle gestion de l’innovation. Cette gestion comporte de nombreux enjeux, avec bien évidemment la maîtrise de la technologie et de ses changements rapides, mais également la capacité à connecter cette technologie aux besoins du marché, en privilégiant les axes qui créent une valeur maximale, pour un coût maîtrisable. Dans un contexte de plus en plus mondialisé, la lutte pour disposer des meilleures compétences constitue l’un des leviers majeurs, mettant la problématique des ressources humaines au cœur des enjeux des départements de Recherche et Développement. Rédigé sur la base de deux ans de réunions de la commission R&D de l’AFDEL, créée début 2008, ce livre blanc a pour objectif de donner une première vue synthétique du métier de la R&D logicielle et de ses différentes dimensions. Le lecteur y trouvera les bonnes pratiques partagées par les experts du domaine et l’ensemble des acteurs de la profession du marché.

Co-Auteur, Livre blanc AFDEL, 2010, http://www.afdel.fr/livrables-afdel-newsletter-livre-blanc-indice-position.asp

  • Use of VPM Calibration : A VCC Software Performance Estimation Methodology for Design Space Exploration

Abstract: “Virtual Component Co-Design (VCC) is a Cadence software tool for hardware/software co-design at the system level that separates the behavior (application) of an embedded system from its architecture (implementation). Such a tool allows the system designer to experiment with various implementations (hardware or software resource) of an application in many possible candidate architectures. This Design Space Exploration (DSE) is necessary to assess the impact of decisions early in the design process with respect to performance and cost constraints. In that context, the execution time of an application running on a software resource needs to be quickly evaluated. The success of software estimation techniques depends on abstracting away implementation details while accurately modeling the performance effects. One of the modeling techniques is based on the definition of a Virtual Processor Model (VPM), which is formed by a Virtual Machine Instruction (VMI) set. This model represents the performance aspect of a target processor’s instruction set architecture. A VMI is a generic low-level instruction that exists on all architectures. This paper presents a tool flow and a methodology developed to work in conjunction with VCC: the VPM-Calibration.

Derras Mustapha, Canut Frederic, The Regatta, Cadence R&D Newsletter, volume 13, issue 3, October 2002

  • Segmentation and Real Time Control

Abstract:An original region segmentation based on a Markovian modelling of a set of sites (representing a 16 by 16pixels elementary region) is presented. Each site is characterized by two texture parameters, deduced from co-occurrence matrices and two luminance parameters obtained from normalized local histograms. The application performed demonstrates the strength of a segmentation algorithm using texture analysis of mowed/unmowed natural zones. An optimum splitting limit has been obtained which is going to become the basic primitive in order to hook up a mower robot. In order to obtain a real time application (200 ms computing time per image), a simple parallelization of the algorithm and a control unit (control servoing) are realized. Lots of results which are layed out represent samples of a large land campaign measures under various meteorological conditions, with varied grass covers, image sensors, etc…

Pierre Bonton, Mustapha Derras, Christophe Debain, Michel Berducat, Microscopy Microanalysis Microstructure Vol. 7, N°5/6  345-353, 1996, (http://cat.inist.fr/?aModele=afficheN&cpsidt=2739316)

  • Asservissement Visuel d’Engins Agricoles par Commande Référencée Vision (Full paper)

Abstract : “L’objectif de cet article est de montrer les techniques que nous avons mises en œuvre afin d’asservir un engin mobile sur une interface quelconque (par exemple une ligne blanche ou la limite entre une zone fauchée et une zone non-fauchée d’un couvert végétal). La détection de cette interface est assurée par un algorithme de segmentation en région traitant des informations issues d’une caméra CCD. Cet algorithme est basé sur la théorie des champs de Markov appliquée à des paramètres de texture et de luminance déduits respectivement de matrices de coocurrences et d’histogrammes locaux normalisés. A partir des informations visuelles extraites, nous avons élaboré une loi de commande permettant d’asservir notre machine. L’originalité de notre loi de commande réside dans le fait qu’elle utilise des données de l’espace du capteur, elle s’affranchit donc de toute reconstruction de la scène. Nous avons effectué différents essais sur simulateur et en conditions réelles permettant de montrer la robustesse de nos algorithmes.

Christophe Debain, Mustapha Derras, Djamel Khadraoui, Philippe Martinet, Pierre Bonton, Proceedings of the International Conference on Quality Control by Artifical Vision, QCAV’95, pp 250-260, Le Ceusot, France, 17-19 May 1995, (http://cemadoc.cemagref.fr/cemoa/PUB00000431)

  • Development of a visual guidance system for an upkeep robot of natural environments

Abstract: “The aim of this project consists in the design of an help guidance system for an auto-propelled mower. We want to illustrate the fact that in a natural environment we need to use lots of different algorithms to reach our goal: control the upkeep machine. We briefly present here two methods used to detect a reap limit. The first one is a good segmentation process which also demands a powerful hardware because of its computing time costs. The second one is a more simple threshold technique less efficient but less expensive in computing time. We introduce discussion elements concerning the possibility of algorithms merging and, overall, date merging. It’s also necessary to use methods that overlap their different areas of disfunctionning.

C. Debain, M. Derras, M. Berducat, P. Bonton and J. Gallice, in Proceedings of the International Symposium on Signal Processing, Robotics an Neural Networks, IMACS, Lille, France, pages 121-124, 1995

  • Image processing and algorithms merging : Real time control of an upkeep machine for natural spaces

Abstract: “The aim of this project consists in the design of an help guidance system for an auto-propelled mower. We want to illustrate the fact that in a natural environment we need to use lots of different algorithms to reach our goal: control the upkeep machine. We briefly present here two methods used to detect a reap limit. The first one is a good segmentation process which also demands a powerful hardware because of its computing time costs. The second one is a more simple threshold technique less efficient but less expensive in computing time. We introduce discussion elements concerning the possibility of algorithms merging and, overall, date merging. It’s also necessary to use methods that overlap their different areas of disfunctionning.

M. Derras, L. Verrat, M. Berducat and P. Bonton, in Proceedings of the Fourth Workshop on Robotics in Agriculture and the Food Industry, Toulouse, France, pages 93-98, 1995, (http://cemadoc.cemagref.fr/cemoa/PUB00000433)

  • Segmentation texturale originale. Analyse et parallélisation temps réel

Abstract : “Dans cet article, nous présentons une segmentation non supervisée originale en régions basée sur une modélisation markovienne d’un ensemble de sites. Chaque site (représentant une région élémentaire de 16.6 pixels) est caractérisé par deux paramètres de texture, déduits de matrices de cooccurrences, et deux paramètres de luminance déduits d’histogrammes locaux normalisés. L’application réalisée montre la robustesse d’un algorithme de segmentation par l’analyse de la texture fauchée non fauchée d’espaces naturels. Nous obtenons ainsi une limite optimum de séparation des deux zones qui sera la primitive de base permettant d’assurer l’asservissement visuel d’un robot de fauche. De nombreux résultats sont fournis, ils représentent des échantillons d’une large campagne de mesures réalisées sur le terrain sous diverses conditions météorologiques, pour divers couverts végétaux, capteurs d’images etc… Nous présentons également une parallélisation simple de l’algorithme qui débouche sur une application temps réel (200 ms de temps de traitement par image).

M. Derras, M. Berducat, P. Bonton, J. Gallice et S. Nauded, 9ème Congrès RFIA (Reconnaissance des Formes et Intelligence Artificielle) AFCET, Paris, 11-14 janvier 1994, (http://cemadoc.cemagref.fr/cemoa/PUB00001299)

  • Segmentation texturale en régions – paramètre fractal – modélisation markovienne

Abstract: “In this article, we are presenting an original segmentation into regions based on an unsupervized markovian modelling of a set of sites. Each site (representing an elementary region of 16.16 pixels) is characterized by two texture parameters, deduced from co-occurrence matrixes, two luminance parameters deduced from standardized local histograms and one parameter related to the fractal dimension of each site. The application shows the robustness of the segmentation algorithm by analysis of the mowed and unmowed texture of natural environments. We obtain an optimum separation limit of the two areas which will be the basic pattern permitting the visual automatic control of a mowing robot. The results show samples of large field measurement campaigns for various groundcovers, image sensors, weather conditions, etc…”

M. Derras, M. Berducat, P. Bonton, International society for stereology 17ème réunion annuelle de la section française, Paris, 3 février 1994, (http://cemadoc.cemagref.fr/cemoa/PUB00001363)

  • Unsupervised Regions Segmentation: Real Time Control of an Upkeep Machine of Natural Spaces (Full paper)

Abstract: “An original image segmentation based on a Markovian modeling of a set of four parameters is presented. The application performed demonstrates the strength of an algorithm using texture analysis of natural scenes. A splitting limit is obtained which is going to become the basic primitive in order to hook up a mower robot. To obtain a real time application we also present a simple parallelization of the algorithm and a control servoing.

Mustapha Derras, Christophe Debain, Michel Berducat, Pierre Bonton, Jean Gallice, ECCV’94, Third European Conference on Computer Vision, Stockholm, Sweden, 2-6 May 1994, pages 207-212, (http://www.springerlink.com/content/d758073t4w322754/)

  • Unsupervised regions segmentation : a self-guidance system for agricultural equipments (Full paper)

Abstract:An original image segmentation based on a Markovian modeling of a set of four parameters is presented. The application performed demonstrates the strength of an algorithm using texture analysis of natural scenes. A splitting limit is extracted which is going to become the basic primitive in order to control a mower robot. To obtain a real time application we also present the algorithm parallelization and the architecture embedded on the machine.

Mustapha Derras, Christophe Debain, Michel Berducat, Pierre Bonton, Jean Gallice, XII CIGR World Congress, AgEng’94, Milano, 29-1 September 1994, volume 2, pages 1192-1199

  • La vision au service du guidage d’un engin d’entretien des espaces naturels

Abstract : “L’objectif du projet présenté consiste à démontrer la faisabilité d’un système de guidage par traitement d’image capable d’assurer l’asservissement d’un engin d’entretien sur une limite de fauche. Ce type de guidage repose uniquement sur la qualité de détection de la limite de travail séparant les parties fauchée et non fauchée. Modélisée sous la forme d’un segment de droite, cette primitive constituera l’information de base servant à l’asservissement. Une phase essentielle de ce travail a donc été le développement d’une méthode d’extraction de l’interface suffisamment robuste pour s’adapter aux conditions de travail très variées rencontrées en milieu naturel (nature et hauteur des couverts végétaux, conditions d’ensoleillement…) Cet algorithme est une segmentation non supervisée reposant sur une modélisation markovienne d’un ensemble de sites décrits chacun par quatre paramètres.

M. Berducat, M. Derras, P. Bonton, MECAFUTUR, L’application de l’analyse d’images dans l’agro-industrie : de la production à la transformation des produits agricoles, Montpellier, 18 septembre 1994, pages 37-42, (http://cemadoc.cemagref.fr/cemoa/PUB00001362)

  • Segmentation Texturale Originale Appliquée au Guidage Visuel d’un Robot d’Entretien d’Espaces Naturels (Full paper)

Abstract: “Dans cet article nous présentons une segmentation originale en régions basée sur une modélisation markovienne d’un ensemble de sites. Chaque site (représentant une région élémentaire de 16.16 pixels) est caractérisé par deux paramètres de texture, déduits de matrices de cooccurrences, et deux paramètres de luminance déduits d’histogrammes locaux normalisés. L’application réalisée montre la robustesse d’un algorithme de segmentation par l’analyse de la texture fauchée/non fauchée d’espaces naturels. Nous obtenons ainsi une limite optimum de séparation des deux zones qui sera la primitive de base permettant d’assurer l’asservissement visuel d’un robot de fauche. De nombreux résultats sont fournis, ils représentent des échantillons d’une large campagne de mesures réalisées sur le terrain sous diverses conditions météorologiques, pour divers couverts végétaux, capteurs d’images, etc… ”

Mustapha Derras, Michel Berducat, Pierre Bonton, Jean Gallice, 14th Collocium on the « Signal and Image Processing », GRETSI, Juan-Les-Pins, 13-16 September 1993, pages 771-774, (http://documents.irevues.inist.fr/handle/2042/11943)

  • Segmentation non supervisée d’images texturées par champs de Markov : Application à l’automatisation de l’entretien des espaces naturels (Full document)

Abstract: “This doctoral thesis demonstrate the feasibility of a guidance system using image processing to ensure the control of an upkeep machine of natural spaces and, generally, of agricultural equipments. The research of the tools which are able to automatically indicate in the image the reap limit position is the main problem of this approach.

The first step of our work is dedicated to an outdoor measures campaign which allows us to characterize our work environment. We have realized an image data bank grouping lots of different mowing configurations (camera view orientation, sunshine variations, kind of grass cover, ..). The first conclusions, coming from a first image analysis method, lead us to the use of a more sophisticated technique. We re-evaluate our image perception by integrating texture information to obtain an homogeneous regions splitting of the images. Then we use an adapted unsupervised segmentation tool which is a markovian process and, especially, an ICM (Iterated Conditional Mode) algorithm. This implemented approach giving satisfaction, the segmentation result is then entrusted to an analysis step which aim is to obtain the reap limit position in the image.

The project imperative is to realize real condition tests of the entire process on the machine. For that purpose we have conceived a parallel computer which able us to implant the algorithm. This parallel architecture is constituted by three MOTOROLA microprocessors (two 68040 and one 68030) which have respectively in charge the data texture extraction, the image segmentation and the control of the mower.

We have adopted a general processes which had lead us to real tests of the guidance system. Demonstrate the possibility of the realization of such a vision system permit us to introduce a branch of activity which group all the problems encountered for moving robots and vision systems.

Mustapha Derras, Thèse de Doctorat d’Université, D.U. 562, 10 décembre 1993, 163 pages

  • Vision Guided Mower for the Upkeep of Natural Environment (Full paper)

Abstract: “The present project aims at developing an artificial vision help guidance system on a self propelled radio-commanded machine. To achieve this goal a conceivable solution consists in a positioning system relative to the natural reference line formed by the reap limit. Then the first step of our project has consisted in determining methods to detect this limit.

Mustapha Derras, Michel Berducat, Pierre Bonton, International Seminar of the 3rd and 6th Technical Sections of the CIGR, Use of On Machine Vision Systems for The Agricultural and Bio-Industries, Montpellier, 1991, September 3-6, pages 37-46, (http://cemadoc.cemagref.fr/cemoa/PUB00005711)

Advertisements