1. Inicio
  2. Atrás
Título: Towards robust voxel-coloring : handling camera calibration errors and partial emptiness of surface voxels
Autores: Anwar, Zeeshan.
Fecha: 2005
Publicador: McGill University - MCGILL
Fuente:
Tipo: Electronic Thesis or Dissertation
Tema: Engineering, Electronics and Electrical.
Descripción: In this thesis, we present two new methods to overcome the effects of both camera calibration errors and partial emptiness of surface voxels in voxel-coloring. Voxel-coloring is a relatively new volumetric method for 3D scene reconstruction from multiple calibrated views. The quality of reconstruction is affected by the presence of errors in the estimated calibration parameters. Furthermore, a voxel forming a scene surface may be partially empty as there is no prior knowledge about the scene surface. Both of these sources of error result in outlier pixels in voxel projections in the input images. These outlier pixels affect the photo-consistency test of the voxel and tend to result in over-carving of the reconstructed 3D scene. The existing methods to handle these errors are either insufficient or too complex. We propose a method to handle the effect of calibration errors and call it Adaptive Gaussian Averaging. It makes use of the information of the error probability distribution of projected pixel coordinates due to camera parameter errors. We propose another method to reduce the effects of partial emptiness of surface voxels and we call it Area Weighting. In this method we use the pixel count in voxel projections to weight the projections in voxel's color statistics calculations. Our proposed methods are simple and can be incorporated into the existing voxel-coloring algorithms easily. We also conduct experiments on our own calibrated data sets to verify the effectiveness of the proposed methods. The experimental results show that both of our proposed methods have the ability to improve the results of existing algorithms. We also compare the results of our proposed methods with the results of an existing method that handle these errors too, the r-Consistency. We find that our proposed methods have the ability to adapt to the level of errors present in the system, and perform better than r-Consistency when the effect of these errors is higher on voxel-coloring.
Idioma: en