Building material classifications from imagery
Imagery is used to identify architectural elements that have known architectural patterns. Feature sets associated with a surface and architectural elements in a building model image are compared with known architectural standards of materials to determine the surface building materials and architectural details of a textured building model. In addition, specific texture patterns can assist final material selections for a repair/replacement.
1. A method of classifying building materials in an image comprising:
retrieving an image of a building object;
identifying contiguous pixels;
comparing the contiguous pixels against a known building material feature set;
calculating, based on at least the comparing, a probability that the contiguous pixels represent one or more of a plurality of known building materials; and
determining one of the plurality of known building materials for the identified contiguous pixels of the retrieved image by selecting a highest probability.
2. The method of claim 1 further comprising ranking a result of the calculated probability for the plurality of known building materials.
3. The method of claim 2 further comprising estimating a cost of replacement or repair of the building materials determined by the highest ranked result.
4. The method of claim 1 , wherein the plurality of known building materials include any of: siding, brick, stone, paint, glass, wood, shingles or stucco.
5. The method of claim 1 , wherein the image is used for building and texturing any of a two dimensional (2D) building model or three dimensional (3D) building model of the building object.
6. The method of claim 1 , wherein the image includes any of or a combination of: orthogonal, oblique, aerial, 2D model, 3D model or ground-level.
7. The method of claim 1 further comprising statistics for a plurality of known images.
8. The method of claim 7 , wherein the statistics for a plurality of known images are aggregated into buckets to create ranges of expected responses for the known building materials.
9. The method of claim 8 , wherein the ranges of expected responses for the known building materials are stored in a library of known materials.
10. The method of claim 1 further comprising repeating the retrieving, identifying, comparing and calculating steps as part of a training routine to a build a library of the known building materials based on the known building material feature set.
11. The method of claim 10 wherein the training routine is performed by machine learning.
12. A system for classifying building materials in an image comprising:
an image database configured to retrieve an image corresponding to a building object; and
a processor communicatively coupled to the image database and configured to:
retrieve an image of the building object;
identify contiguous pixels;
compare the contiguous pixels against a known building material feature set;
calculate, based on at least the compare, a probability that the contiguous pixels represent one or more of a plurality of known building materials; and
determine one of the plurality of known building materials for the identified contiguous pixels of the retrieved image by selecting a highest probability.
13. The system of claim 12 further comprising a module communicatively coupled to the processor and operable to generate statistics for a plurality of known images.
14. The system of claim 13 , wherein the statistics for a plurality of known images are aggregated in buckets to create ranges of expected frequency responses for known materials.
15. The system of claim 12 , wherein the processor is further configured to rank the probability for a plurality of the one or more of a plurality of known building materials.
16. The system of claim 15 , wherein the processor is further configured to estimate a cost of replacement or repair of the building materials determined by a highest probability for the identified contiguous pixels.
17. The system of claim 12 , wherein the plurality of known building materials include one or more of: siding, brick, stone, paint, glass, shingles or stucco.
18. The system of claim 12 , wherein the image is used for building and texturing any of a two dimensional (2D) building model or three dimensional (3D) building model of the building object and includes images including any of or a combination of: orthogonal, oblique, aerial or ground-level.
19. The system of claim 12 further comprising repeating the retrieving, identifying, comparing and calculating steps as part of a machine learning training routine to a build a library of the known building materials based on the known building material feature set.
20. A method of classifying building materials in an image comprising:
retrieving an image of a building object;
converting contiguous pixels of the retrieved image to frequency domain;
measuring frequency responses of one or more sections of the image;
comparing the measured frequency responses against a library of known material frequency responses;
calculating, based on at least the comparing, a probability that the contiguous pixels represent one of a plurality of known building materials based on the comparing; and
determining one of the plurality of known building materials for the one or more sections of the retrieved image by selecting a highest probability.