Pierre Miche

A cooperative approach to vision-based vehicle detection

In this paper two different vision based systems for vehicle detection are described and their integration discussed. The first approach is based on the use of a specific model for vehicles and mostly relies on monocular vision. Conversely, the second system is based on the use of stereo vision and allows to refine the coarse results obtained by the former. A preliminary integration of the two systems has been tested on the ARGO experimental vehicle and some remarks about reliability and robustness are also included.

Fast and automatic stereo vision matching algorithm based on dynamic programming method

We consider that the matching problem in a stereo vision process can be treated as the problem of finding an optimal path on a two-dimensional (2D) search plane. To obtain this path, we consider a non-linear gain function which varies as a function of threshold values calculated during the preliminary statistical analysis of the right and left images.

Fast and self-adaptive image segmentation using extended declivity

In this paper we present a new fast self-adaptive image segmentation operator for real-time vision applications. It is based on the extended declivity which is obtained by extending the basic declivity — whose concept is recalled — using a similar principle to region growing. This non-linear operator is able to detect low contrasted region boundaries without any pre-processing such as low-pass filtering. Experimental results are produced with different types of image and compared with those obtained with the basic declivity operator and with the Deriche’s operator as a reference.RésuméDans cet article un nouvel opérateur rapide et autoadaptatif de segmentation d’images pour des systèmes de vision en temps réel est présenté. Il est basé sur la déclivité étendue qui est obtenue par extension de la déclivité de base — dont le concept est rappelé — selon un principe semblable à celui de la croissance de régions. Cet opérateur non linéaire permet de détecter des limites de régions faiblement contrastées sans aucun pré-traitement tel qu un filtrage passe-bas. Des résultats expérimentaux portant sur différents types d’images sont présentés et comparés avec ceux obtenus avec l’opérateur de déclivité de base ainsi qu’avec l’opérateur de Deriche en tant que référence.

Analogue circuits of a learning spiking neuron model

Biological neurons communicate via sequences of calibrated pulses or spikes. The behaviour of spiking neurons is the following: input spikes from pre-synaptic neurons are weighted and summed up yielding a value called membrane potential. The membrane potential is time dependent and decays when no spikes are received by the neuron. If however spikes excite the membrane potential sufficiently so that it exceeds a certain threshold, a spike is emitted and transmitted through its axon via synapses to other neurons. After the emission of a spike the neuron is unable to spike again for a certain period called refractory period. Recently, a new theoretical formulation has been proposed by Gerstner (1999). The computational power of neural networks based on temporal coding by spikes, rather than on the traditional interpretation of analogue variables, has been investigated by Maass (1999). It is shown that simple operations on phase-differences between spike-trains provide a powerful computational tool.

Passive 3-D shape recovery of unknown objects using cooperative polarimetric and radiometric stereo vision processes

The polarimetric characterization of a beam reflected onto a dielectric surface provides information about the orientation of this surface. From this principle, we present an original method for passive 3-D reconstruction of unknown objects. A specific vision system has been designed that enables polarization image acquisition and processing. The 3-D reconstruction is achieved by an integration process whose initialization data are provided in an original way by a cooperative edge-based stereo vision algorithm. Results obtained from real scenes are presented, which show off the reliability of our method applied to objects without any available a priori information as well as for flat or curved surfaces.

Implementation of a 3D Vision System on DSPs TMS320C31

In this paper, the implementation of a new stereo vision process on a specialized architecture which comprises of three DSPs TMS320C31 is described. The first step of our stereo vision system is a self-adaptive image segmentation algorithm based on a new concept that we call declivity. The second step is a new and fast stereo matching algorithm based on dynamic programming and using self-adaptive decision parameters. The goal of our work is to develop a stereo vision system that achieves an acceptable level of performance using a modest amount of hardware. This implementation is organized as follows: declivity extraction from the two stereo images is performed in parallel on two DSPs, one for the right image and the other for the left one. Then, the last DSP computes the declivity matching based on our dynamic programming method as well as the 3D maps calculation. Finally, experimental results obtained using real pairs of stereo images on a VME 150/40 Imaging Technology Vision System are presented. They show the feasibility and the effectiveness of our system. These results can surely be improved by using a new generation of DSP in order to consider real-time applications.

The study of the radiance variation between two pixels for their stereo vision matching

In the stereo vision system, the key problem is a search problem which finds the corresponding pixels in the two stereoscopic images. These pixels must satisfy a very strict photometric constraint. In our approach, we use a special configuration of the two cameras in order to speed up the stereo vision process. In this paper, a theoretical study of the radiance variation between two stereo vision matching pixels is presented. This study shows the efficiency and the limitation of the used photometric constraint. Furthermore, we show the influence of the maximum value of the disparity and the one of the focal length of the two cameras on the stereo vision process. Some experimental results obtained on real images are given in the end of this paper.