We developed a method to estimated 3D shape of dynamic object. Seven
lights with different wavelengths is illuminated to the target object,
and multispectral camera with seven channel captured the image.
Smoothness of albedo, smoothness of surface normal, and the surface
normal at occluding boundary are used as constraints. After applying
bilateral filter to the average of input iamges, the resultant image is
used for the initinal value of albedo. The approximate shape calculated
from the object silhouette is used for the initial value of surface
normal. Using these initial values, shape is obtained with iterative
process. |