Data processing: measuring – calibrating – or testing – Measurement system – Orientation or position
Reexamination Certificate
2002-03-05
2002-10-15
Hilten, John S. (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Orientation or position
C345S418000, C345S419000, C345S420000, C382S100000, C382S152000, C382S154000, C702S152000, C702S155000
Reexamination Certificate
active
06466892
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a technique for composing three-dimensional multi-viewpoints data obtained by measuring a three-dimensional object from a plurality of directions and/or distances.
2. Description of the Background Art
There are two methods of composing three-dimensional multi-viewpoints data obtained by measuring a three-dimensional object from a plurality of directions, i.e., “polygon-based method” and “volumetric method”. Both these methods are used for composing a plurality of original data obtained by measuring an object from a plurality of viewpoints (i.e, viewing points), and the original data are referred to as “polygon mesh data”.
Prior to discussion on these methods, polygon mesh data will be discussed. A camera capable of measuring a three-dimensional shape of an object is sequentially set at a plurality of positions of viewpoints. Data obtained by the camera at the viewpoints are converted into range images representing a range from the camera to parts of the object with reference to camera parameter values.
The range image is a set of latticed discrete points, and three-dimensional coordinate value data are defined at each lattice point. In the range image having this data structure, each point is connected to its adjacent four points in four directions by segments, to form a square polygon or a triangular polygon. In each polygon, a direction towards a viewpoint in obtaining a range image is defined as an upward direction. A set of polygons thus obtained is a polygon mesh data. Since the polygon mesh data are obtained for each viewpoint, the number of polygon mesh data is equal to the number of viewpoints.
The polygon-based method will be discussed below. In this method, two of polygon mesh data obtained from a number of viewpoints are sewn together to compose one polygon mesh data. By sewing each pair of adjacent polygon mesh data, all the polygon mesh data are composed, to eventually obtain one composed polygon mesh data.
Two polygon mesh data are composed as below. When polygons in the respective meshes defined by two polygon mesh data are overlapped, one of these polygons is deleted. After that, one of these meshes is clipped on the other.
FIGS. 12
to
14
are illustrations showing a clipping of meshes. In these figures, for explanatory convenience, meshes
31
and
32
are assumed on the same plane. Referring to
FIGS. 12
to
14
, the clipping of meshes will be discussed below.
First, as shown in
FIG. 12
, intersections
33
between respective boundaries of the mesh
31
and edges of the mesh
32
are created. With the intersections
33
as boundaries, an unnecessary disposal region
34
as shown in
FIG. 13
is deleted, and a set of intersections
33
are connected to vertexes on the boundaries of the mesh
32
, to form triangular polygons shared by the meshes
31
and
32
. As a result, as shown in
FIG. 14
, a composed mesh
35
is obtained by composing the meshes
31
and
32
.
Thus, clipping of meshes is performed as the boundary of two meshes is zippered to form a new polygon, filling a clearance.
The polygon-based method is disclosed, in detail, for example, in Greg Turk and Mark Levoy, “Zippered Polygon Meshes From Range Images”.
Proceedings of ACMSIG GRAPH '
94, pp. 311 to 318. ACM Press, 1994.
Next, the volumetric method will be discussed. In the volumetric method, three-dimensional space is divided into grids of a small unit in x, y and z directions and a small cube of unit grid, i.e., voxel, is used. For example, a rectangular pyramid
21
shown in
FIG. 15A
is represented by using a number of voxels
22
as shown in FIG.
15
B.
With respect to each of the above voxels, a potential value corresponding to a range from a polygon defined by the polygon mesh data to the voxel, is determined.
For example, when the voxel
22
, a viewpoint
23
and a mesh
24
are positioned as shown in
FIG. 16
, a potential value is determined on the basis of a distance dx from the voxel
22
on a line of sight
25
to a point x in a polygon
26
of the mesh
24
. In such a case, in general, when the voxel
22
is positioned on a side of the viewpoint
23
with respect to the polygon
26
, the potential value is negatively signed and when positioned on an opposite side of the viewpoint
23
, the value is positively signed. Therefore, in the example of
FIG. 16
, the potential value is negatively signed.
Since the above calculation of the potential value is executed with respect to a plurality of polygon mesh data obtained from a number of viewpoints, usually, a plurality of potential values are obtained for each voxel. The sum of those potential values is an added potential value for each voxel.
Then, polygon mesh data are generated again by using Marching Cubes algorithm and the like on the basis of the added potential value for each voxel.
The volumetric method is disclosed, in detail, for example, in Brian Curless and Marc Levoy, “A Volumetric Method For Building Complex Models From Range Images”
Computer Graphics
, Annual Conference Series, pp. 303 to 312, 1996 (Proceedings Siggraph '96).
As discussed above, composition of a plurality of polygon mesh data, i.e., original data obtained from a number of viewpoints is conventionally performed by using the polygon-based method or the volumetric method.
Since in the polygon-based method, as discussed above, two of the original polygon mesh data are sequentially composed to eventually obtain one composed polygon mesh data, if any of the original polygon mesh data includes some inaccuracies such as errors, the effect directly appears in the composed polygon mesh data. It is impossible to get rid of these inaccuracies from the composed polygon mesh data.
Further, the composed polygon mesh data are sensitive to errors in the original polygon mesh data and those generated in registration of a plurality of original polygon mesh data on a coordinate system. Especially, such a shape as a thin stick, sensitive to even a small error, cannot! be restructured with the composed polygon mesh data.
Furthermore, since it is necessary to reconstruct a segment connecting two points, i.e., a phase every time when two original polygon mesh data are composed, as the multiplicity of the original polygon mesh data, which is in proportion to the number of polygon elements constituting a mesh, becomes larger, a processing efficiency becomes lower.
On the other hand, in the volumetric method, since the sum of a plurality of potential values is the added potential value for each voxel, an error of one original polygon mesh data and that of other data offset each other in many cases, and therefore some inaccuracies of the original polygon mesh data have few effects. As a result, even higher multiplicity of the original mesh data hardly reduces the processing efficiency. In other words, the volumetric method can resolve all the problems of the polygon-based method.
Since the size of voxel is uniform, however, in the volumetric method, the resolution of the eventually-obtained composed polygon mesh data becomes uniform on the whole. Therefore, when the size of voxel is made small in accordance with a portion of high resolution among a plurality of polygon mesh data obtained from a number of viewpoints, the composed polygon mesh data have high redundancy to deteriorate efficiency of calculation. When in accordance with a portion of low resolution, high-resolution data of the original data are lost.
Further, when the potential value is calculated for each voxel, the polygon mesh data are resampled, and when faces of each voxel, i.e., polygons are extracted on the basis of the added potential value, interpolated vertexes are calculated. These two calculations inevitably cause deterioration in accuracy. To maintain the accuracy in the volumetric method, it is necessary to, set the size of voxel at a small value, and as mentioned above, that gives the composed polygon mesh data too high redundancy and requires a larger amount of all and is impractical.
SUMMARY OF
Fujii Eiro
Shiono Koichi
Hilten John S.
Le John
Minolta Co. , Ltd.
Sidley Austin Brown & Wood LLP
LandOfFree
Method of composing three-dimensional multi-viewpoints data does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of composing three-dimensional multi-viewpoints data, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of composing three-dimensional multi-viewpoints data will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2939074