Binocular system of active vision to object localisation on partially unknown environment

The global objective of this thesis is to design a flexible vision system to provide robots with self-localisation capacity and object localisation capacity, for them to act “wisely” in poorly structured environment. This Doctoral Thesis presents some work for the improvement of this field, following three paths: (1) Design and develop an opened hardware/software platform allowing the development and verification of different active vision strategies for different problems and applications. (2) Suggest new efficient and robust methods and algorithms for solving visual localisation problems. (3) Analyse the influence of uncertainties in the redundant visual information fusion for 3D reconstruction.; Concerning the first purpose, an active vision binocular system called SiviS has been designed. This vision system, when compared to the already existing systems, widens the field of possible applications allowing base line control among cameras, thus allowing high accuracy in a wide range of distances. SiviS is an adaptable system because of its modular software architecture that allows flexibly planning the system's motor and visual behaviour. In addition, one of the most important features in this system is that it offers a co-operation link between Marr's vision paradigm and qualitative vision paradigm.; In case of working on a partially unknown environment, the available information can be used for endowing localisation greater efficiency and robustness. This idea is introduced in the thesis and a effective and robust localisation technique called N-points Hough Transform is applied. This technique is a generalization of the well known Hough Transform, with accumulation phase time costs minimised.; Finally, in order to obtain robust algorithms in the case of completely unknown environments, time restrictions should be relaxed. In these cases, robustness can be obtained from redundant information fusion. In this sense and to allow a robust 3D reconstruction in an unknown environment, this thesis provides and applies uncertainty calculation for monocular and binocular techniques used in active vision binoculars. Furthermore, some image processing techniques able to filter noises produced by shadows as well as image reflections in natural lighting conditions have been designed and applied.