The examining board awarded Jonathan Felipe the highest grade of outstanding cum laude for his doctoral thesis, a work focused on the development of artificial vision techniques for the interpretation of 3D scenes.
On May 24 was presented at the School of Engineering and Technology of the University of La Laguna the doctoral thesis entitled 3D Reconstruction based on U-V Disparity techniques in stereo vision systems. Robustness analysis and practical applications, carried out by our colleague Jonatán Felipe García, coordinator of the supercomputing department, and directed by Dr. Leopoldo Acosta and Dr. Marta Sigut of the University of La Laguna.
The work presented originated within the development of the Verdino prototype, a self-guided vehicle designed to circulate in a gated community of 25 bioclimatic dwellings at the Instituto Tecnológico y de Energías Renovables (ITER). The prototype was developed between 2002 and 2022 in several research projects funded by the National R&D Plan, addressing aspects from the automation of the vehicle to the development of advanced algorithms for its control. In particular, a set of tasks dealt with the design and implementation of the sensory system of the prototype, fitting the content of this research within this line of work.
This thesis adopts an innovative approach using stereovision techniques based on U-V Disparity to retrieve 3D information, differentiating and representing elements of a scene by planes whose orientation is proportional to their location in the real world. The study includes a sensitivity analysis to evaluate how errors in the pitch angle of the vision system affect the accuracy of the reconstruction.
In addition, the research explores practical applications derived from the proposed reconstruction system. A methodology was developed to calibrate the pitch angle of stereoscopic systems, allowing the estimation of possible errors so that later stages of processing can benefit from their correction. A second application is also proposed that simplifies the reconstruction methodology to be used for obstacle detection. This adaptation allowed to carry out an implementation capable of detecting and characterizing in real time the obstacles that may appear in a self-guided vehicle such as the Verdino prototype.
