Computer vision systems and methods for geospatial property feature detection and extraction from digital images
Systems and methods for property feature detection and extraction using digital images. The image sources could include aerial imagery, satellite imagery, ground-based imagery, imagery taken from unmanned aerial vehicles (UAVs), mobile device imagery, etc. The detected geometric property features could include tree canopy, pools and other bodies of water, concrete flatwork, landscaping classifications (gravel, grass, concrete, asphalt, etc.), trampolines, property structural features (structures, buildings, pergolas, gazebos, terraces, retaining walls, and fences), and sports courts. The system can automatically extract these features from images and can then project them into world coordinates relative to a known surface in world coordinates (e.g., from a digital terrain model).
1. A computer vision system for detecting and extracting geometric features from digital images, comprising:
at least one computer system; and
computer vision system code executed by the at least one computer system, the computer vision system code causing the computer system to:
select a geospatial region of interest;
select and retrieve at least one digital image associated with the geospatial region of interest;
generate a plurality of image tiles from the at least one digital image;
generate a plurality of label tiles from the plurality of image tiles, each of the plurality of label tiles comprising a tensor having a height dimension, a width dimension, and a channel dimension;
combine the plurality of label tiles into a single labeling for the at least one digital image; and
extract two dimensional (“2D”) representations of geometric features from the at least one digital image using the single labeling.
2. The computer vision system of claim 1 , wherein the computer vision system code further causes the computer system to retrieve metadata associated with the at least one digital image, and extract the 2D representations of the geometric features using the metadata.
3. The computer vision system of claim 2 , wherein the metadata comprises intrinsic parameters and extrinsic parameters of a camera used to capture the at least one digital image.
4. The computer vision system of claim 1 , wherein the computer system uses a fully convolutional network to generate the plurality of label tiles.
5. The computer vision system of claim 1 , wherein the computer vision system code further causes the computer system to combine the plurality of label tiles into the single labeling using one or more operations including, cropping or stitching the plurality of label tiles, deriving Boolean labels, and unscaling label tensors.
6. The computer vision system of claim 1 , wherein the geometric features correspond to property features.
7. A method for detecting and extracting geometric features from digital images, comprising the steps of:
selecting a geospatial region of interest;
selecting and retrieving at least one digital image associated with the geospatial region of interest;
generating a plurality of image tiles from the at least one digital image;
generating a plurality of label tiles from the plurality of image tiles, each of the plurality of label tiles comprising a tensor having a height dimension, a width dimension, and channel dimension;
combining the plurality of label tiles into a single labeling for the at least one digital image; and
extracting two dimensional (“2D”) representations of geometric features from the at least one digital image using the single labeling.
8. The method of claim 7 , further comprising retrieving metadata associated with the at least one digital image and extracting the 2D geospatial representations of the geometric features using the metadata.
9. The method of claim 8 , wherein the metadata comprises intrinsic parameters and extrinsic parameters of a camera used to capture the at least one digital image.
10. The method of claim 7 , wherein the computer system uses a fully convolutional network to generate the plurality of label tiles.
11. The method of claim 7 , wherein combining the plurality of label tiles into the single labeling using one or more operations includes, cropping or stitching the plurality of label tiles, deriving Boolean labels, and unscaling label tensors.
12. The method of claim 7 , wherein the geometric features correspond to property features.
13. A computer vision system for detecting and extracting non-geometric features from digital images, comprising:
at least one computer system; and
computer vision system code executed by the at least one computer system, the computer vision system code causing the computer system to:
select a geospatial region of interest;
select and retrieve at least one digital image associated with the geospatial region of interest;
generate a plurality of image tiles from the at least one digital image;
generate a plurality of label tiles from the plurality of image tiles, each of the plurality of label tiles comprising a tensor having a height dimension, a width dimension, and channel dimension;
combine the plurality of label tiles into a single labeling for the at least one digital image; and
generate annotated models corresponding to the non-geometric features using the single labeling.
14. The computer vision system of claim 13 , wherein the computer vision system code further causes the computer system to retrieve metadata associated with the at least one digital image, and generate the annotated models corresponding to the non-geometric features using the metadata.
15. The computer vision system of claim 14 , wherein the metadata comprises intrinsic parameters and extrinsic parameters of a camera used to capture the at least one digital image.
16. The computer vision system of claim 13 , wherein the computer system uses a fully convolutional network to generate the plurality of label tiles.
17. The computer vision system of claim 13 , wherein the computer vision system code further causes the computer system to combine the plurality of label tiles into the single labeling using one or more operations which include, cropping or stitching the plurality of label tiles, deriving Boolean labels, and unscaling label tensors.
18. The computer vision system of claim 13 , wherein the annotated model comprises a three-dimensional (“3D”) model.
19. A method for detecting and extracting non-geometric features from digital images, comprising the steps of:
selecting a geospatial region of interest;
selecting and retrieving at least one digital image associated with the geospatial region of interest;
generating a plurality of image tiles from the at least one digital image;
generating a plurality of label tiles from the plurality of image tiles, each of the plurality of label tiles comprising a tensor having a height dimension, a width dimension, and channel dimension;
combining the plurality of label tiles into a single labeling for the at least one digital image; and
generating annotated models corresponding to the non-geometric features using the single labeling.
20. The method of claim 19 , further comprising retrieving metadata associated with the at least one digital image and generating the annotated models corresponding to the non-geometric features using the metadata.
21. The method of claim 20 , wherein the metadata comprises intrinsic parameters and extrinsic parameters of a camera used to capture the at least one digital image.
22. The method of claim 19 , wherein the computer system uses a fully convolutional network to generate the plurality of label tiles.
23. The method of claim 19 , wherein combining the plurality of label tiles into the single labeling using one or more operations includes, cropping or stitching the plurality of label tiles, deriving Boolean labels, and unscaling label tensors.
24. The method of claim 19 , wherein the annotated model comprises a three-dimensional (“3D”) model.