Medição da torção ocular através de análise de fluxo óptico direto

• Autor(a) principal: Stahl, Bruno Konzen
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da PUC_RS
• Texto Completo: http://tede2.pucrs.br/tede2/handle/tede/7144
• topic: PROCESSAMENTO DE IMAGENS; OLHOS - MOVIMENTOS; COMPUTAÇÃO GRÁFICA; ENGENHARIA BIOMÉDICA; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Eye movement measurement is done through exams called videonystagmography. Equipment currently available captures horizontal and vertical displacement movements called nystagmus. This equipment, however, do not capture ocular torsion movements, which are also important because they provide indicators to diagnose diseases such as balance disorders, motion sickness, seasickness, dizziness, and even space sickness. Conventional methods used for these measurements capture videos of the eyes and determine the torsion angles in the eye based on the contrast of the iris characteristics or terminations of conjunctival blood vessels. In these methods, the circular image of the iris is converted into a rectangular image through a polar transformation. This process allows converting a twisting motion into a horizontal displacement, facilitating the administration of screening techniques such as Optical Flow and Template Matching (TM). These methods operate satisfactorily when, between two frames of a video, there is no displacement of the center of the eye. However, this displacement is quite common and reduces measurement accuracy, once displacements as small as one pixel between images of the torsional axes can lead to errors in the range of 1.0°. To solve this problem, this paper proposes a new method which minimizes these variations, enabling the measurement of ocular torsion angles more accurately than conventional methods. In this study, the images of the ocular rotational movements were captured in six subjects. For evaluation of the method, the results were compared with data obtained by assisted visual measurement. In the results, the proposed method was capable of measuring ocular torsional motions with an average error smaller or equal to 0.15 degrees at a processing speed near 30 FPS, even with reflexes and variation in pupil diameter. In this study, the proposed method has a maximum error of 0.69 degrees. These results indicate that the proposed method has higher accuracy and performance equivalent to conventional methods.