Merging of Range Images

Cultural Heritage Preservation Project at the University of Tokyo

Digitally archive Japanese cultural heritage objects for...
  • Preservation
  • Restoration
  • Exhibition
The issues for modeling cultural heritages are...
  • Large size
  • Outdoor objects
How do we acquire the shape and color of objects?
How do we handle large data to represent their intricate shape?

Parallel Processing of Range Data Merging

We originally proposed a method to integrate range images into a geometric model, called Consensus Surface Algorithm (ppt). We have to handle large data set of range images to model cultural heritage objects. To accomplish this, we parallelize the merging algorithm. Our parallel merging algorithm consists of following two methods:
  • Distributed allocation of range images
  • Parallel traversal of octree

Consensus Surface with Photometric Attributes

We accomplish the photometrically rigid 3D model construction in our range image integration framework. As examples of photometric attributes attached to range images, we consider two different attributes: laser reflectance strength and intensity/color.

By taking consensus of photometric attributes of range images, we extract Lambertian refrectance parameter with discarding outliers such as specular refrection.

An example of merging with color:
Input images of a toy house (some images include specular reflection.)
Merged result (specular reflection is removed.)

Integrating in Adaptive Resolution

The original consensus surface algorithm efficiently computes signed distances by utilizing an octree. However, it generated a mesh model in finest resolution everywhere. To reduce the amount of data to represent the object and to use computational resources efficiently, we propose a method which generates a mesh model in adaptive resolution; with appropriate resolution according to the geometric and photometric characteristics of the observed object.

Our method splits an octree adaptively according to the following criteria:

  • Geometric Attribute: curvature of surface
  • Photometric Attribute: color variation