Deep learning image processing method for determining vehicle damage

Patent metadata

Field	Value
Publication number	US-2023222179-A1
Application number	US-202318110510-A
Country	US
Kind code	A1
Filing date	Feb 16, 2023
Priority date	Jun 29, 2017
Publication date	Jul 13, 2023
Grant date	—

Abstract

In a computer-implemented method and associated tangible non-transitory computer-readable medium, an image of a damaged vehicle may be analyzed to generate a repair estimate. A dataset populated with digital images of damaged vehicles and associated claim data may be used to train a deep learning neural network to learn damaged vehicle image characteristics that are predictive of claim data characteristics, and a predictive similarity model may be generated. Using the predictive similarity model, one or more similarity scores may be generated for a digital image of a newly damaged vehicle, indicating its similarity to one or more digital images of damaged vehicles with known damage level, repair time, and/or repair cost. A repair estimate may be generated for the newly damaged vehicle based on the claim data associated with images that are most similar to the image of the newly damaged vehicle.

First claim

What is claimed:

1 . A computer-implemented method, comprising:
determining, using a predictive similarity model, a similarity score for a given two digital images of damaged vehicles, wherein the predictive similarity model is trained using a historical training dataset populated with digital images of damaged vehicles and historical claim data related to the damaged vehicles, to identify damaged vehicle image characteristics that are predictive of one or more of damage level, repair time, or repair cost; and based on the similarity score for the given two digital images, one or more of:
(i) displaying the two digital images; and
(ii) predicting a damage level, repair time, or repair cost for a first damaged vehicle depicted in a first digital image of the given two digital images based on a damage level, repair time, or repair cost for a second damaged vehicle depicted in a second digital image of the given two digital images.

2 . The computer-implemented method of claim 1 , wherein determining the similarity score for the given two digital images of damaged vehicles further comprises weighting the damaged vehicle image characteristics based on a degree to which each characteristic is predictive of at least one of damage level, repair time, or repair cost.

3 . The computer-implemented method of claim 1 , wherein determining the similarity score for the given two digital images of damaged vehicles includes:
determining a plurality of similarity scores each corresponding to a digital image of a newly damaged vehicle and a respective one of a plurality of digital images in a given dataset of damaged vehicles; and the method further comprising finding, using an image search engine, a set of one or more digital images of damaged vehicles from the given dataset for which the similarity score is above a threshold similarity score.

4 . The computer-implemented method of claim 3 , wherein the given dataset of damaged vehicles is populated with digital images of at least one different damaged vehicle than the historical training dataset.

5 . The computer-implemented method of claim 3 , further comprising displaying the set of digital images found using the image search engine.

6 . The computer-implemented method of claim 5 , further comprising displaying one or more of a damage level, a repair time, or a repair cost associated with the one or more digital images of damaged vehicles found using the search engine.

7 . The computer-implemented method of claim 3 , further comprising:
generating predictive statistics for a damage level, repair time, or repair cost for the newly damaged vehicle based on a damage level, repair time, or repair cost for damaged vehicles depicted in the set of one or more digital images of damaged vehicles found using the image search engine; and displaying the predictive statistics.

8 . A tangible, non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to:
determine, using a predictive similarity model, a similarity score for a given two digital images of damaged vehicles, wherein the predictive similarity model is trained using a historical training dataset populated with digital images of damaged vehicles and historical claim data related to the damaged vehicles, to identify damaged vehicle image characteristics that are predictive of one or more of damage level, repair time, or repair cost; and based on the similarity score for the given two digital images, one or more of the following:
(i) display the given two digital images; and
(ii) predict a damage level, repair time, or repair cost for a first damaged vehicle depicted in a first digital image of the given two digital images based on a damage level, repair time, or repair cost for a second damaged vehicle depicted in a second digital image of the given two digital images.

9 . The tangible, non-transitory computer-readable medium of claim 8 , wherein the instructions to determine the similarity score for the given two digital images of damaged vehicles further include instructions to weight the damaged vehicle image characteristics based on a degree to which each characteristic is predictive of at least one of damage level, repair time, or repair cost.

10 . The tangible, non-transitory computer-readable medium of claim 8 , wherein:
the instructions to determine the similarity score for the given two digital images of damaged vehicles further include instructions to determine a plurality of similarity scores each corresponding to a digital image of a newly damaged vehicle and a respective one of a plurality of digital images in a given dataset of damaged vehicles; and wherein the instructions further cause the one or more processors to find, using an image search engine, a set of one or more digital images of damaged vehicles from the given dataset for which the similarity score is above a threshold similarity score.

11 . The tangible, non-transitory computer-readable medium of claim 10 , wherein the given dataset of damaged vehicles is populated with digital images of at least one different damaged vehicle than the historical training dataset.

12 . The tangible, non-transitory computer-readable medium of claim 10 , wherein the instructions further include instructions to display the set of digital images found using the image search engine.

13 . The tangible, non-transitory computer-readable medium of claim 12 , further including instructions to display one or more of a damage level, a repair time, or a repair cost associated with the one or more digital images of damaged vehicles found using the search engine.

14 . The tangible, non-transitory computer-readable medium of claim 10 , further including instructions to:
generate predictive statistics for a damage level, repair time, or repair cost for the newly damaged vehicle based on a damage level, repair time, or repair cost for damaged vehicles depicted in the set of one or more digital images of damaged vehicles found using the image search engine; and display the predictive statistics.

15 . A computer system comprising:
one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to:
determine, using a predictive similarity model, a similarity score for a given two digital images of damaged vehicles, wherein the predictive similarity model is trained using a historical training dataset populated with digital images of damaged vehicles and historical claim data related to the damaged vehicles, to identify damaged vehicle image characteristics that are predictive of one or more of damage level, repair time, or repair cost; and
based on the similarity score for the given two digital images, one or more of the following:
(i) display the two digital images; and
(ii) predict a damage level, repair time, or repair cost for a first damaged vehicle depicted in a first digital image of the given two digital images based on a damage level, repair time, or repair cost for a second damaged vehicle depicted in a second digital image of the given two digital images.

16 . The computer system of claim 15 , wherein the instructions to determine the similarity score for the given two digital images of damaged vehicles further include instructions to weight the damaged vehicle image characteristics based on a degree to which each characteristic is predictive of at least one of damage level, repair time, or repair cost.

17 . The computer system of c

Assignees

• State Farm Mutual Automobile Insurance Co

Inventors

• Yang He
• Sliz Bradley A
• Clymer Carlee A
• Andrus Jennifer Malia

Classifications

• G06N3/08

  Learning methods · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06Q40/08 Primary

  Insurance · CPC title

• G06F16/583

  using metadata automatically derived from the content · CPC title