Tesis doctoral de Agustín Alfonso Navarro Toro
Monocular cues are spatial sensory inputs which are picked up exclusively from one eye. They are in majority static features that provide depth information and are extensively used in graphic art to create realistic representations of a scene. Since the spatial information contained in these cues is picked up from the retinal image, the existence of a link between it and the theory of direct perception can be conveniently assumed. According to this theory, spatial information of an environment is directly contained in the optic array. Thus, this assumption makes possible the modeling of visual perception processes through computational approaches. in this thesis, angular variation is considered as a monocular cue, and the concept of direct perception is adopted by a computer vision approach that considers it as a suitable principle from which innovative techniques to calculate spatial information can be developed. the expected spatial information to be obtained from this monocular cue is the position and orientation of an object with respect to the observer, which in computer vision is a well known field of research called 2d-3d pose estimation. In this thesis, the attempt to establish the angular variation as a monocular cue and thus the achievement of a computational approach to direct perception is carried out by the development of a set of pose estimation methods. Parting from conventional strategies to solve the pose estimation problem, a first approach imposes constraint equations to relate object and image features. In this sense, two algorithms based on a simple line rotation motion analysis were developed. These algorithms successfully provide pose information; however, they depend strongly on scene data conditions. To overcome this limitation, a second approach inspired in the biological processes performed by the human visual system was developed. It is based in the proper content of the image and defines a computational approach to direct perception. the set of developed algorithms analyzes the visual properties provided by angular variations. The aim is to gather valuable data from which spatial information can be obtained and used to emulate a visual perception process by establishing a 2d-3d metric relation. Since it is considered fundamental in the visual-motor coordination and consequently essential to interact with the environment, a significant cognitive effect is produced by the application of the developed computational approach in environments mediated by technology. In this work, this cognitive effect is demonstrated by an experimental study where a number of participants were asked to complete an action-perception task. The main purpose of the study was to analyze the visual guided behavior in teleoperation and the cognitive effect caused by the addition of 3d information. The results presented a significant influence of the 3d aid in the skill improvement, which showed an enhancement of the sense of presence.
Datos académicos de la tesis doctoral «Angular variation as a monocular cue for spatial percepcion«
- Título de la tesis: Angular variation as a monocular cue for spatial percepcion
- Autor: Agustín Alfonso Navarro Toro
- Universidad: Politécnica de catalunya
- Fecha de lectura de la tesis: 18/06/2009
Dirección y tribunal
- Director de la tesis
- Juan Aranda López
- Tribunal
- Presidente del tribunal: alícia Casals gelpí
- xavier Lladó bardera (vocal)
- víctor Fernando Muñoz martínez (vocal)
- raul Marín prades (vocal)