Extraction of 3D Structure using Texture Flow

Studenteropgave: Speciale (inkl. HD afgangsprojekt)

  • Ole Thomsen Buus
This report investigates the application of texture flow for direct estimation of local three-dimensional structure of a rigid textured surface moving relative to a fixed perspective camera. Texture flow is defined here as a spatiotemporal gradient, based on the estimation of local image texture densities and its derivatives. A major part of the work deals primarily with computational experiments. A simulation framework was developed for exact simulation of local surface and image texture density. An experimental setup was also built and experimentation with a digital camera and a movable textured surface was attempted. The conclusion is two-fold: (1) In the computational experiments, it was indeed found possible to directly estimate local surface orientation and depth using the estimated texture flow derivatives. Sometimes also for velocities (ego-motion). (2) The camera experiments need further work. The estimation of texture density in real digital images is difficult and the methods chosen for texel segmentation showed to be highly sensitive to the lighting conditions.
Antal sider140
Udgivende institutionAalborg Universitet Esbjerg
ID: 14458505