Michel Dufaut

An Optimized Segmentation Method for a 2D Laser-Scanner Applied to Mobile Robot Navigation

Abstract This article presents a system for segmentation the data of a laser scanner on board a mobile robot. Three algorithms were tested in three different environments. They were compared in terms of their robust segmentation for indoor and outdoor environments, good representation of the environment and their handling for the map building. For environment modelling a polygonal representation was chosen. Furthermore, it should have good real-time capability to be able to integrate the information of the laser-scanner into the navigation algorithms of the mobile robot.

Capture of homotopy classes with probabilistic road map

Feasibility tests in virtual reality for nuclear power plant maintenance or dismantling operations are a source of problems for motion planning because finding a way in a cluttered environment is not easy for the bulky loads, mobile devices and robots used in such operations. Standard probabilistic roadmap methods (PRM) have been successfully used to answer such feasibility problems. These methods provide, at the most a single solution but do not provide a complete overview of the possible motions which have to be evaluated in a complete engineering task. We focus here on the open question of building probabilistic roadmaps which can provide an exhaustive list of all the solutions which can not be distorted from one to another while staying collision free. We call such roadmaps homotopy preserving probabilistic roadmap (HPPR). We propose a new algorithm for creating HPPR.

Vehicle detection at night using image processing and pattern recognition

We present a methodology to detect vehicles at night using image processing and pattern recognition. The using of mathematical morphology, the techniques of pattern recognition and the studying of perspective influences are the major innovations of our method. First, we present an initialisation phase that involves a road modelling. Afterwards, we present methodologies to detect vehicles on highways. We also discuss the results and give perspectives and conclusions.<<ETX>>

A method of mobile robot localisation by fusion of odometric and magnetometric data

This paper presents the dead reckoning localisation technique of the robot ROMANE (“Robot Mobile Autonome de Navigation en Extérieur”) which is conceived to move outdoors on level roads without traffic. Vehicle localisation with incremental shaft encoders on the wheels is improved by integration of a magnetic compass into the robot system. Fusion of odometric and magnetometric data by Kalman filtering increases the accuracy of the estimated robot heading. Consequently the uncertainty of the vehicle position is reduced. To evaluate our localisation method we present some experimental results of trajectory estimations.

Connectivity estimation of three parametric methods on simulated electroencephalogram signals

The global framework of this paper is the connectivity estimation in multichannel electroencephalogram (EEG) recordings, modeled as multidimensional autoregressive (AR) processes. The coherence, directed transfer function and partial directed coherence functions are evaluated on two simulated EEG signals for their later application on real EEG recordings. The results were evaluated computing the relative error and a second proposed performance criterion (η) based on the entropy of the estimated connectivity matrix.

Paradigms and experimental set-up for the determination of the acceptable delay in Telesurgery

The aim of this work was to develop an experimental set-up and realistic paradigms to study the effect of delay on video flux transmission and surgical performance. Four exercises were performed by 15 surgeons with 5 different simulated transmission delays. Large standard deviation of the duration of an exercise was found. Even with a short transmission delay of 150 ms, some surgeon found that the surgical procedure was not possible. Further work has to be done to have a better evaluation of the surgical precision.

Analysis and inspection of road traffic using image processing

A methodology using image processing to detect vehicles on highways is presented in this paper. Vehicle detection is done in two stages. In stage one, the dimensional and morphological study of the analyzed highway is carried out in order to model and calibrate the vision system. In stage two, vehicles are detected owing to the method of matching clear and dark blobs. These are extracted by mathematical morphology techniques and bright headlight images by using pattern recognition techniques. The position of the vehicle allows access to spatial-temporal information which characterize traffic flow and permit automatic traffic control.

Sensor Cooperation and World Perception of the Autonomous Mobile Robot Romane

Abstract This paper presents a map building method of an autonomous mobile robots environment. It is based on an original « focus of attention » cooperation system using a laser rangefinder as the main source of information and a video camera mounted on a pantilt unit to perform close inspection of selected objects in the scene; selection of such objects being performed by analysing data from the laser rangefinder. This cooperation process provides a sufficiently rich description of the robots environment to call for more complex tasks such as localisation.

Systeme de vision utilisant le flou optique: application a la recherche automatique de defauts d'aspects

The possibilities of a hybrid image processor, entitled Net Flou, used for the automatic detection of defects are presented. In this paper, the new idea is to optimize the Net-Flou system to detect defects with particular spatial frequency characteristics. An application example in detecting visual defects of mechanical objects (connecting rods) will be illustrated.