Underwater machine vision : recovering orientation and motion of Lambertian planar surfaces in light attenuating media

Abstract

Although much progress has been made for land-based systems, the developed techniques cannot be directly applied in environments where attenuation of light intensity is significant, and artificial lighting is the only source of illumination, as in deep ocean environments. To achieve this, various illumination effects including those caused by an artificial light source with limited power and light intensity attenuation due to absorption and scattering in the medium should be taken into account. In this dissertation, equations for scene radiance and image irradiance are developed which model the attenuation phenomenon in an absorbing and scattering medium such as sea water. Based on these models, the reflectance map for a Lambertian surface illuminated by a point light source is derived. The dissertation then shows how the orientation of planar patches with Lambertian reflectance properties can be recovered from their image shading. Closed-form solutions are given for cases when either parallel or perspective projection is assumed. In either case, more robust solutions can be obtained when other information such as distance to the surface is known. This dissertation also investigates effects due to motion of the camera and light source relative to the scene. Based on a general brightness change equation, pure translational motion and pure rotational motion are discussed. A least square formula is derived to recover general motion of an object provided it has three or more planar patches which can be identified. Computer simulation results are presented to show the sensitivity of these solutions to noise in the data.