Systemes de stereovision passive dedies a un stimulateur intra-cortical visuel

The main objective of the present Master thesis is to design a passive stereovision system to be used in the cortical visual stimulators application. The system aims to provide blind patients with information on the spatial positioning of objects that are present in a given environment. Main requirements are real-time operability and high disparity estimation accuracy.;The second chapter presents two phase based approaches to solve the correspondence problem. The first system (Enhanced Phase Only Correlation, EPOC) is an enhancement of the 1D version of the Phase Only Correlation. The second system (Multiple Modes Phase Only Correlation, MMPOC) optimizes the 2D Phase Only Correlation method. The 2D Phase Only Correlation method is known to be more accurate than the 1D version, but takes a longer time to execute. A real-time implementation of the EPOC algorithm on a FPGA is described.;The third chapter discusses in more detail the EPOC system in a graphical framework. Key systems comparison is made in a Bayesian framework.;In the fourth chapter, we design two Hopfield based quantum associative memories that have the ability to infer depth from two images, the serial architecture and parallel architecture of Hopfield Quantum network. A comparison relates the drawbacks of the proposed systems with results of experimentations that were performed on the human brain.;The first chapter presents different approaches to solve the correspondence search problem, a main step to infer depth with passive stereovision systems. The first system is based on a Hidden Markov Model. We review the drawbacks of the proposed system and present a more stable approach that exists in the literature, Belief Propagation. We also present faster approaches to solve the correspondence problem, with a special attention on phase correlation.