Navigation d'un véhicule intelligent à l'aide d'un capteur de vision en lumière structurée et codée

Abstract

The purpose of the work presented in this thesis is the application of structured light vision (a<br />sensor composed by a CCD camera and a light source) to the navigation of mobile robots. This<br />led us to study various techniques and approaches of computer vision and image processing. First<br />of all, we reviewed the principal types of codification for structured light and its main applications<br />in robotics, medical imagery and metrology. Besides, we propose a method of image processing<br />for structured light with the aim to extract the segments of the image and to decode the pattern.<br />Then, we detail a method of three-dimensional reconstruction from an uncalibrated sensor. The<br />projection of a light pattern onto the environment imposes constraints to self-calibration methods.<br />It arises that the reconstruction has to be carried out in two steps, with a unique image capture and<br />a unique pattern projection. We specify the method of projective reconstruction used for our<br />experiments and we give a method which permits to pass from a projective to a Euclidean<br />reconstruction. By using the geometrical relations generated by the projection of the light pattern,<br />we show that it is possible to find Euclidean constraints between the points of the scene,<br />independent of the objects of the scene. We also propose a technique of quantitative detection of<br />obstacles, allowing to estimate the map of free space observed by the robot. Finally, we make a<br />complete study of the sensor in motion : it leads to an algorithm that allows to estimate the<br />displacement of the robot from planes correspondences.