Statistical point pattern matching technique
A statistical point pattern matching technique is used to match corresponding points selected from two or more views of a roof of a building. The technique entails statistically selecting points from each of orthogonal and oblique aerial views of a roof, generating radial point patterns for each aerial view, calculating the origin of each point pattern, representing the shape of the point pattern as a radial function, and Fourier-transforming the radial function to produce a feature space plot. A feature profile correlation function can then be computed to relate the point match sets. From the correlation results, a vote occupancy table can be generated to help evaluate the variance of the point match sets, indicating, with high probability, which sets of points are most likely to match one another.
1. A microprocessor based method comprising:
receiving two or more different views of a structure, the views being in the form of digital images;
assigning, by the microprocessor, points from each view that represent a shape of the structure;
associating, by the microprocessor, corresponding points from the different views;
deriving height data from the points by performing a transform that produces radial point patterns for the assigned points from each view; and
generating, by the microprocessor, a three-dimensional model of the structure from the points and the height data.
2. The method of claim 1 wherein the receiving two or more different views includes receiving a top-down plan (orthogonal) view.
3. The method of claim 2 wherein at least one of the two or more different views is an aerial view.
4. The method of claim 2 wherein at least one of the two or more views is a satellite view.
5. The method of claim 2 wherein the structure is one or more of a building, a vehicle, a bridge, a roadway, an element of transportation infrastructure, a golf course, or a garden.
6. The method of claim 1 wherein the number of points on a first view differs from the number of points on a second view.
7. The method of claim 1 wherein the associating includes performing a statistical method of point pattern matching.
8. The method of claim 7 wherein the statistical method of point pattern matching executes in a point pattern matching computation machine.
9. The method of claim 7 wherein the statistical method of point pattern matching includes a variational analysis to determine a probability that each one of a pair of matched points represents substantially the same location on the structure.
10. A microprocessor based method comprising:
receiving two or more different views of a structure, the views being in the form of digital images;
assigning, by the microprocessor, points from each view that represent a shape of the structure;
associating, by the microprocessor, corresponding points from the different views;
deriving height data from the points by performing a transform; and
generating, by the microprocessor, a three-dimensional model of the structure from the points and the height data comprises:
selecting a first subset of points identified from a first view of the structure;
calculating an origin for a first point pattern that corresponds to the first subset of points;
calculating an origin for a second point pattern that corresponds to a set of points identified from a second view of the structure;
geometrically transforming the first point pattern to a second view image plane, taking into account one or more camera orientations; and
repeating the acts of calculating an origin for the first point pattern and geometrically transforming, on a trial basis to generate point matching statistical data in an iterative fashion, in which each trial includes a different subset of points identified from the first view.
11. The method according to claim 10 , wherein associating corresponding points from the different views further comprises evaluating the point matching statistical data by:
plotting angular variational data from each of the point patterns;
Fourier transforming the angular variational data to create Fourier space data;
correlating the Fourier space data to produce a feature space correlation function relating the two point patterns;
performing a Gaussian distribution-based clustering analysis that takes into account a weighted proximity of nearest neighbor points;
tabulating statistics for point matches from the different trials; and
determining a point match probability based on the tabulated statistics.
12. The method of claim 11 wherein the first view of the structure is an orthogonal view and wherein the second view of the structure is an oblique view of the structure.
13. The method of claim 11 further comprising generating a report that contains a three-dimensional model of the structure based at least on the point match probability.
14. The method of claim 13 wherein the structure is a roof.
15. The method of claim 14 wherein the generated report is a roof report that includes linear dimensions of roof features.
16. The method of claim 14 wherein the points from each view that represent a shape of the roof include points where roof lines of the structure converge.
17. The method of claim 11 , wherein the structure is the exterior of a building.
18. The method of claim 17 , wherein the points from each view that represent the shape of the exterior of the building include points where the exterior edges of the building converge.
19. The method of claim 17 , wherein the points from each view that represent the shape of the exterior of the building include points identifying where edges of one or more of exterior windows, exterior doors or balconies converge.
20. A system comprising:
a networked microprocessor;
a non-transitory computer-readable memory coupled to the microprocessor, the memory including stored microprocessor-executable instructions that, when executed by the microprocessor, cause the microprocessor to:
receive two or more different views of a structure, the views being in the form of digital images;
assign points from each view that represent a shape of the structure;
associate corresponding, points from the different views;
derive height data from the points by performing a transform;
generate, by the microprocessor, a three-dimensional model of the structure from the points and the height data; wherein associate corresponding points from the different views comprises:
select a first subset of points identified from a first view of the structure;
calculate an origin for a first point pattern that corresponds to the first subset of points;
calculate an origin for a second point pattern that corresponds to a set of points identified from a second view of the structure;
geometrically transform the first point pattern to a second view image plane, taking into account one or more camera orientations; and
repeat the acts of calculate origins for the point patterns and geometrically transform, on a trial basis to generate point matching statistical data in an iterative fashion, in which each trial includes a different subset of points identified from the first view.
21. The system of claim 20 , further comprising:
plot angular variational data from each of the point patterns;
Fourier transform the angular variational data to create Fourier space data;
correlate the Fourier space data to produce a feature space correlation function relating the two point patterns;
perform a Gaussian distribution-based clustering analysis that takes into account a weighted proximity of nearest neighbor points;
tabulate statistics for point matches from the different trials; and
determine a point match probability based on the tabulated statistics.