Computer vision systems and methods for ground surface condition detection and extraction from digital images

1. A system for detecting and extracting a ground surface condition from an image comprising:

a memory; and

a processor in communication with the memory, the processor:

performing a high resolution scan of at least one input image;

generating an orthomosaic model and a digital surface model based on the high resolution scan of the at least one input image;

generating an image tile based on the generated orthomosaic model and the digital surface model;

determining a label for each pixel of the image tile, via a computer vision model, the label indicating a probability of a presence of a ground surface condition;

generating a label tensor for the at least one input image based on each label for each pixel of the generated image tiie;

extracting a two-dimensional geospatial representation of a detected ground surface condition based on the generated label tensor for the at least one input image; and

generating a report indicative of damage associated with the detected ground surface condition based on the extracted two-dimensional geospatial representation.

2. The system of claim 1 , wherein the processor:

receives a geospatial region of interest, the geospatial region of interest being a polygonal boundary indicative of latitudinal and longitudinal coordinates of a region; and

retrieves the at least one input image and the metadata of the at least one input image from the memory based on the received geospatial region of interest.

3. The system of claim 1 , wherein the processor:

receives a geospatial region of interest, the geospatial region of interest being a polygonal boundary indicative of latitudinal and longitudinal coordinates of a region; and

captures the at least one input image at a sub-inch ground sample distance based on the received geospatial region of interest.

4. The system of claim 1 , wherein the at least one input image is an aerial image, a satellite image, a ground-based image, a photograph and a scan.

5. The system of claim 1 , wherein the processor:

performs a high resolution scan of a plurality of images, the plurality of images being captured by a plurality of image capturing sources;

determines a spatial position and an orientation of each image capturing source among the plurality of image capturing sources relative to one another by selecting a matching key point in a determined image pair among the plurality of images;

determines at least one extrinsic parameter of the plurality of image capturing sources based on a transformation of the selected matching key point from one image of the determined image pair to another image of the determined image pair;

generates the orthomosaic model based on the determined at least one extrinsic parameter of the plurality of image capturing sources by stitching the plurality of images together to form a first image; and

generates the digital surface model by stitching the plurality of images together to form a second image.

6. The system of claim 1 , wherein the ground surface includes a parking lot, a roadway, and a driveway.

7. The system of claim 1 , wherein:

the generated image tile is an image tile tensor having a first shape including a first height, a first width and a first number of channels, and

the determined label is a score label tensor having a second shape including a second height and a second width corresponding to the first height and the first width of the image tile tensor and a second number of channels, the score label tensor being indicative of a probability of a presence of at least one type of ground surface condition including cracking, distortion, disintegration, polished aggregate, bleeding, flushing and utility cut depression.

8. The system of claim 7 , wherein the label is a Boolean tensor label, the Boolean tensor label being derived from the score label tensor and being indicative of one of an absence of a ground surface condition and the detection of the at least one type of ground surface condition based on a predetermined threshold.

9. The system of claim 1 , wherein the computer vision model is a fully convolutional network.

10. The system of claim 1 , wherein the processor generates the label tensor for the at least one input image based on each determined label for each pixel of the generated image tile by performing one of a cropping operation and a stitching operation on the determined labels.

11. The system of claim 1 , wherein the two-dimensional geospatial representation includes a polygon, a line segment, a point and a bounding box.

12. The system of claim 1 , wherein the processor extracts the two-dimensional geospatial representation of the detected ground surface condition based on the generated label tensor for the at least one input image by:

extracting vector data indicative of the detected ground surface condition in pixel space via one of a contour extraction algorithm and a bounding box finding algorithm;

projecting the extracted vector data to world geospatial coordinates using metadata of the at least one input image and a ground elevation of the generated digital surface model; and

exporting the projected extracted vector data.

13. A method for detecting and extracting a ground surface condition from an image comprising the steps of:

performing a high resolution scan of at least one input image;

generating an orthomosaic model and a digital surface model based on the performed high resolution scan of the at least one input image;

generating an image tile based on the generated orthomosaic model and the digital surface model;

determining a label for each pixel of the generated image tile via a computer vision model, the label being indicative of a probability of a presence of a ground surface condition;

generating a label tensor for the at least one input image based on each determined label for each pixel of the generated image tile;

extracting a two-dimensional geospatial representation of a detected ground surface condition based on the generated label tensor for the at least one input image; and

generating a report indicative of damage associated with the detected ground surface condition based on the two-dimensional geospatial representation.

14. The method of claim 13 , further comprising:

receiving a geospatial region of interest, the geospatial region of interest being a polygonal boundary indicative of latitudinal and longitudinal coordinates of a region; and

retrieving the at least one input image and the metadata of the at least one input image from a memory based on the received geospatial region of interest.

15. The method of claim 13 , further comprising:

receiving a geospatial region of interest, the geospatial region of interest being a polygonal boundary indicative of latitudinal and longitudinal coordinates of a region; and

capturing the at least one input image at a sub-inch ground sample distance based on the received geospatial region of interest.

16. The method of claim 13 , further comprising:

performing a high resolution scan of a plurality of images, the plurality of images being captured by a plurality of image capturing sources;

determining a spatial position and an orientation of each image capturing source among the plurality of image capturing sources relative to one another by selecting a matching key point in a determined image pair among the plurality of images;

determining at least one extrinsic parameter of the plurality of image capturing sources based on a transformation of the selected matching key point from one image of the determined image pair to another image of the determined image pair;

generating the orthomosaic model based on the determined at least one extrinsic parameter of the plurality of image capturing sources by stitching the plurality of images together to form a first image; and

generating the digital surface model by stitching the plurality of images together to form a second image.

17. The method of claim 13 , further comprising:

generating the label tensor for the at least one input image based on each determined label for each pixel of the generated image tile by performing one of a cropping operation and a stitching operation on the determined labels.

18. The method of claim 13 , wherein the step of extracting the two-dimensional geospatial representation of the detected ground surface condition based on the generated label tensor for the at least one input image comprises the steps of:

extracting vector data indicative of the detected ground surface condition in pixel space via one of a contour extraction algorithm and a bounding box finding algorithm;

projecting the extracted vector data to world geospatial coordinates using metadata of the at least one input image and a ground elevation of the generated digital surface model; and

exporting the projected extracted vector data.