François Cabestaing

A comparison of classification techniques for the P300 Speller.

This study assesses the relative performance characteristics of five established classification techniques on data collected using the P300 Speller paradigm, originally described by Farwell and Donchin (1988 Electroenceph. Clin. Neurophysiol. 70 510). Four linear methods: Pearsons correlation method (PCM), Fishers linear discriminant (FLD), stepwise linear discriminant analysis (SWLDA) and a linear support vector machine (LSVM); and one nonlinear method: Gaussian kernel support vector machine (GSVM), are compared for classifying offline data from eight users. The relative performance of the classifiers is evaluated, along with the practical concerns regarding the implementation of the respective methods. The results indicate that while all methods attained acceptable performance levels, SWLDA and FLD provide the best overall performance and implementation characteristics for practical classification of P300 Speller data.

Catching moving objects with snakes for motion tracking

We propose an efficient method for tracking several objects moving through a sequence of monocular images against a non-uniform background. Each object entering the scene is intercepted by an active contour model which locks on it as long as it moves in the scene. The procedure does not necessitate an interactive initialization. Some results are presented in case of real traffic scenes.

Automatic tracking and 3D localization of moving objects by active contour models

The matching problem is an important task in computer vision. It consists either to find the image of the same object in all the images of a sequence (temporal matching), or in two stereoscopic images (stereoscopic matching). In order to simplify the matching, only particular features of the images, called primitives, are used. Thus the matching is performed with primitives containing a low level of descriptive information which do not take into account the global structure of the objects. We propose to model moving objects by active contours, or snakes, to simplify both the temporal and the stereoscopic matching. When the stereoscopic matching is achieved, the distance between the object and the cameras is computed by triangulation. The mono vision algorithms used for the tracking and the stereo vision matching are described.

A People Counting System Based on Dense and Close Stereovision

We present in this paper a system for passengers counting in buses based on stereovision. The objective of this work is to provide a precise counting system well adapted to buses environment. The processing chain corresponding to this counting system involves several blocks dedicated to the detection, segmentation, tracking and counting. From original stereoscopic images, the system operates primarily on the information contained in disparity maps previously calculated with a novel algorithm. We show that one can obtain a counting accuracy of 99% on a large data set including specific scenarios played in laboratory and on some video sequences shot in a bus during exploitation period.

Multimodal data association based on the use of belief functions for multiple target tracking

In this paper we propose a method for solving the data association problem within the framework of multi-target tracking, given a set of environmental measurements obtained by complementary and redundant sensors. The proposed method exploits belief theory, which is a powerful tool for handling imperfect data. We applied the method to situations where colored moving targets emit an audio signal. The basic belief assignment is computed using a confidence measure between targets and incoming measurements based on multi-modal attributes. This allows the ambiguity in association between measurements and targets to be reduced especially for targets that come closely spaced. The proposed method has been tested using different sets of simulated data. The results obtained are very satisfactory and show that the method provides a useful mechanism for data association.

Narrow-band frequency analysis for laser-based glass thickness measurement

This paper presents a narrow-band frequency analysis approach for a new laser interferometric heterodyne system, which is used for noncontact glass bottle wall thickness measurement. The measurement signal consists of a number of spectral components, the strongest of which is a reliable representation of the above-mentioned thickness. A fast method for searching and locating this frequency is vital for real-time implementations. As the standard fast Fourier transform (FFT) proved to be ineffective for the given problem, we use a combination of a zero-padding discrete Fourier transform (DFT) with a coarse-to-fine technique to locate this frequency at a smaller processing cost. Considering also the Chirp-z transform, a comparison of the different methods under investigation demonstrates the effectiveness of the proposed approach for online thickness estimation.

Joint audio-video people tracking using belief theory

This paper is concerned with the use of belief theory to resolve the data association problem in the context of tracking and identifying people using audio and video data. In order to associate measurements with targets, the proposed method exploits different features such as color, position and acoustic parameters. This has the advantage of providing a robust solution to data association in challenging tracking scenarios.

A similarity-based adaptive neighborhood method for correlation-based stereo matching

In this paper, we present a new method for dense stereo matching. In area-based methods, the similarity between one pixel of the left image and one pixel of the right image is measured using a correlation index computed on neighborhoods of these pixels. In our method, the neighbor pixels not similar to the center one are excluded when computing the correlation index, which corresponds to adjusting the equivalent size and shape of the correlation neighborhood. Our algorithm yields a precise estimation of the disparity in nontextured areas, while avoiding undesired smoothing at discontinuities. This method is suitable for real-time implantation using specialized hardware. We demonstrate and discuss performances using synthetic stereo pairs.

An Effective Active Vision System for Gaze Control

This paper presents the performances of an active vision system that mimic the human gaze control. A human can shift his gaze either by quickly moving his fixation point or by keeping a moving target in the fovea (high resolution). These two visual phenomena are called saccadic and smooth pursuit eye movements respectively. In order to mimic this human behavior, we have developed a novel active vision system based on a particular stereo-vision setup. It is composed with one camera, one prism and a set of mirrors. To point the field of view of the sensor at a target, the prism is rotated about its axis by a motorized stage. The system is designed for fast and accurate dynamical adjustments of gaze. To study the mechanical performances of our active vision system we have used three different but classical input signals. A step signal that simulates a change of target (saccadic eye movement), a velocity ramp and a sinusoidal signal that simulate a moving target (smooth pursuit). Whatever the input signal, the objective is to maintain the target in the middle of the image. The experiments demonstrate the efficiency of our vision sensor, in term of dynamical properties and measurement accuracy.

A Colour Correlation-Based Stereo Matching Using 1D Windows

In this paper, we propose an original approach to colour correlation-based stereo matching with mono-dimensional windows. The result of the algorithm is a quasi-dense disparity map associated with its confidence map. For each pixel, correlation indices are computed for several widths of windows and several positions of the current pixel. Three criteria, extracted from each correlation curve, are combined by a fuzzy filter to define a confidence measure. A basic decision rule computes the disparity value and its associated confidence for most of the image pixels. A first study shows results obtained on grey level images with our 1D method and a classical 2D method. The method is applied to the RGB colour space: three disparity maps are computed and fused to compute the final disparity map. The method is validated on the Tsukuba image pair. On the first hand, we show that our method presents lower error rates with the RGB colour space than with the grey level image for identical density rates. On the other hand, our results are compared with those obtained using similar colour 2D methods (presented on the Middlebury Website). Our algorithm is ranked in the first places for each area of the image.