Trained machine learning model for estimating structure feature measurements
A computer system trains a machine learning model to estimate a real-world measurement of a feature of a structure. The machine learning model is trained using a plurality of digital image sets, wherein each image set depicts a particular structure, and a plurality of measurements, wherein each measurement is a measurement of a feature of a particular structure. After the machine learning model is trained, it is used to estimate a measurement of a feature of a particular structure depicted in a particular image set.
1 . A method comprising:
receiving a plurality of images captured via one or more image capture devices, the image capture devices including one or more sensors, and the plurality of images comprising:
a plurality of lateral images of an exterior of a real-world structure and no orthographic images of the exterior of the real-world structure, wherein a lateral image includes a ground-level view of the real-world structure, wherein an orthographic image includes a top-down view of the real-world structure, and wherein the plurality of lateral images depict a first real-world feature of the real-world structure; and
metadata indicating a measurement of a second real-world feature of the real-world structure,
wherein the metadata is derived from one or more orthographic images using a different image capture device than used for the plurality of lateral images,
and wherein the measurement of the second real-world feature is associated with features forming a perimeter associated with the real-world structure according to a particular scale; and
estimating a measurement associated with the first real-world feature of the real-world structure based on execution of a machine learning model and input comprising the plurality of lateral images and the metadata, the measurement being reflective of a real-world area associated with the real-world structure,
wherein training the machine learning model comprises scale normalizing a set of training orthographic images to establish the particular scale, such that pixels of the set of training orthographic images correspond to a same real-world metric distance, and training the machine learning model based on lateral training images and training measurements according to the particular scale,
wherein the first real-world feature and second real-world feature are separate features.
2 . The method of claim 1 , further comprising training the machine learning model before estimating, wherein training the machine learning model comprises providing to the machine learning model:
a plurality of image sets comprising training images of an exterior of a corresponding training real-world structure, and
a plurality of real-world measurements corresponding to training measurements, the plurality of real-world measurements comprising, for each image set, a real-world measurement of a feature of the exterior of the corresponding training real-world structure.
3 . The method of claim 2 , wherein training the machine learning model comprises:
training a first machine learning model to identify a type of the real-world structure; and
training a second machine learning model with the plurality of image sets and the plurality of real-world measurements.
4 . The method of claim 2 , wherein the plurality of image sets include only metadata describing one or more images and no photos or digital image files.
5 . The method of claim 1 , further comprising generating an indication of a confidence level associated with the estimated measurement associated with the first real-world feature of the real-world structure, wherein generating the indication of the confidence level comprises:
comparing the real-world structure, as depicted in one or more lateral images of the plurality of lateral images, and a training real-world structure, as depicted in training images used to train the machine learning model,
wherein the indication of the confidence level is based on a degree of differences between the real-world structure, as depicted in the one or more lateral images, and the training real-world structure, as depicted in the training images used to train the machine learning model.
6 . The method of claim 1 , wherein estimating comprises using the machine learning model to normalize the plurality of images.
7 . The method of claim 6 , wherein the plurality of images are further normalized by correlating azimuth information associated with the real-world structure depicted in the plurality of images.
8 . The method of claim 1 , wherein the real-world structure comprises a roof, wherein the first real-world feature of the real-world structure is indicative of a pitch of the roof.
9 . The method of claim 1 , wherein one or more lateral images of the plurality of lateral images comprises a ground level image.
10 . The method of claim 1 , wherein the metadata for the plurality of images comprises one or more of an indication of whether particular features are present in an image of the plurality of images and a structure type.
11 . The method of claim 2 , wherein an individual training image of the training images was normalized via rotating the individual training image such that a bottom of the corresponding training real-world structure was depicted as parallel to a bottom of the individual training image.
12 . The method of claim 1 , wherein the lateral images further depict a portion of the second real-world feature of the real-world structure.
13 . The method of claim 1 , further comprising:
normalizing the lateral images via rotating individual orientations associated with individual lateral images, the normalization being based on sensor information derived from the one or more sensors and the sensor information including azimuth information.
14 . A system comprising:
one or more processors;
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more processors, cause performance of:
receiving a plurality of images captured via one or more image capture devices, the image capture devices including one or more sensors, and the plurality of images comprising:
a plurality of lateral images of an exterior of a real-world structure and no orthographic images of the exterior of the real-world structure, wherein a lateral image includes a ground-level view of the real-world structure, wherein an orthographic image includes a top-down view of the real-world structure, and wherein the plurality of lateral images depict a first real-world feature of the real-world structure; and
metadata indicating a measurement of a second real-world feature of the real-world structure,
wherein the metadata is derived from one or more orthographic images using a different image capture device than used for the plurality of lateral images,
and wherein the measurement of the second real-world feature is associated with features forming a perimeter associated with the real-world structure according to a particular scale; and
estimating a measurement associated with the first real-world feature of the real-world structure based on execution of a machine learning model and input comprising the plurality of lateral images and the metadata, the measurement being reflective of a real-world area associated with the real-world structure,
wherein training the machine learning model comprises scale normalizing a set of training orthographic images to establish the particular scale, such that pixels of the set of training orthographic images correspond to a same real-world metric distance, and training the machine learning model based on lateral training images and training measurements according to the particular scale,
wherein the first real-world feature and second real-world feature are separate features.
15 . The system of claim 14 , wherein the instructions, when executed by the one or more processors, further cause performance of:
training the machine learning model before estimating, wherein training the machine learning model comprises providing to the machine learning model:
a plurality of image sets comprising training images of an exterior of a corresponding training real-world structure, and
a plurality of real-world measurements corresponding to training measurements, the plurality of real-world measurements comprising, for each image set, a real-world measurement of a feature of the exterior of the corresponding training real-world structure.
16 . The system of claim 15 , wherein training the machine learning model comprises:
training a first machine learning model to identify a type of the real-world structure; and
training a second machine learning model with the plurality of image sets and the plurality of real-world measurements.
17 . The system of claim 15 , wherein the plurality of image sets include only metadata describing one or more images and no photos or digital image files.
18 . The system of claim 14 , wherein the instructions further cause performance of generating an indication of a confidence level associated with the estimated measurement associated with the first real-world feature of the real-world structure, wherein generating the indication of the confidence level comprises:
comparing the real-world structure, as depicted in one or more lateral images of the plurality of lateral images, and a training real-world structure, as depicted in training images used to train the machine learning model,
wherein the confidence level is based on a degree of differences between the real-world structure, as depicted in the one or more lateral images, and the training real-world structure, as depicted in the training images used to train the machine learning model.
19 . The system of claim 14 , wherein the real-world structure comprises a roof, wherein the first real-world feature of the real-world structure is indicative of a pitch of the roof.
20 . The system of claim 14 , wherein one or more lateral images of the plurality of lateral images comprises a ground level image.
21 . The system of claim 14 , wherein the metadata for the plurality of images comprises one or more of an indication of whether particular features are present in an image of the plurality of images and a structure type.
22 . The system of claim 15 , wherein an individual training image of the training images was normalized via rotating the individual training image such that a bottom of the corresponding training real-world structure was depicted as parallel to a bottom of the individual training image.
23 . A non-transient computer-readable storage medium having instructions embodied thereon, the instructions being executable by one or more processors to perform a method for measuring an angle of a roof facet, the method comprising:
receiving a plurality of images captured via one or more image capture devices, the image capture devices including one or more sensors, and the plurality of images comprising:
a plurality of lateral images of an exterior of a real-world structure and no orthographic images of the exterior of the real-world structure, wherein a lateral image includes a ground-level view of the real-world structure, wherein an orthographic image includes a top-down view of the real-world structure, and wherein the plurality of lateral images depict a first real-world feature of the real-world structure; and
metadata indicating a measurement of a second real-world feature of the real-world structure,
wherein the metadata is derived from one or more orthographic images using a different image capture device than used for the plurality of lateral images,
and wherein the measurement of the second real-world feature is associated with features forming a perimeter associated with the real-world structure according to a particular scale; and
estimating a measurement associated with the first real-world feature of the real-world structure based on execution of a machine learning model and input comprising the plurality of lateral images and the metadata, the measurement being reflective of a real-world area associated with the real-world structure,
wherein training the machine learning model comprises scale normalizing a set of training orthographic images to establish the particular scale, such that pixels of the set of training orthographic images correspond to a same real-world metric distance, and training the machine learning model based on lateral training images and training measurements according to the particular scale,
wherein the first real-world feature and second real-world feature are separate features.
24 . The computer-readable storage medium of claim 23 , wherein the method further comprises:
training the machine learning model before estimating, wherein training the machine learning model comprises providing to the machine learning model:
a plurality of image sets, each image set comprising an individual 2-dimensional image of an exterior of a corresponding training real-world structure, and
a plurality of real-world measurements corresponding to training measurements, the plurality of real-world measurements comprising, for each image set, a real-world measurement of a feature of the exterior of the corresponding training real-world structure.
25 . The computer-readable storage medium of claim 24 , wherein the individual 2-dimensional image was normalized via rotating the individual 2-dimensional image such that a bottom of the training real-world structure was depicted as parallel to a bottom of the individual 2-dimensional image.