IP Library Granted Patent US 12,651,297
Granted Patent B2
US 12,651,297 · App. 19/006,744 · Granted Jun 9, 2026

Systems and methods for automated data processing using machine learning for vehicle loss detection

Inventors: Jean-Christophe Bouëtté (Montreal, CA); Jimmy Lévesque (Blainville, CA); Marc Poulin (Saint-Lambert, CA); Satya Krishna Gorti (Toronto, CA); Keyu Long (Toronto, CA); Nicolas Gervais (St-Hubert, CA); Jennifer Bouchard (Montreal, CA)
Assignee: The Toronto-Dominion Bank
G06Q40/08222
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,651,297
App. No.
19/006,744
Granted
Jun 9, 2026
Kind
B2
Abstract

A data processing system comprising: inputting a tiled image of a vehicle including four different angle views of the vehicle combined into a single image to a first machine learning model (e.g. CNN), the model trained based on historical image data to predict a first likelihood of total loss vehicle; inputting a multi-fusion of images each into a second set of machine learning models; the multi-fusion of images including a set of separate and distinct images for each of the views input separately into the second set of machine learning models, and extracting features to predict a second likelihood of total loss vehicle; inputting tabular data relating to the vehicle into a third machine learning model to predict a third likelihood of total loss vehicle for the vehicle; and aggregating the first, second and third likelihood of total loss vehicle to determine the overall likelihood of total loss.

Claims (44)

1 . A computer system for processing a plurality of digital images, the computer system comprising:

a processor;

a non-transient computer-readable medium comprising instructions that, when executed by the processor, cause the processor to:

determine a location of a vehicle and define a bounding box that surrounds the location based on an object detection machine learning model being applied to distinct images of the vehicle;

crop the distinct images to display only the vehicle and rotate the distinct images to a defined orientation;

generate a tiled image of the vehicle the distinct images being combined into a single image of different angle views in equal portions of the tiled images;

process the generated tiled image via a first convolutional neural network configured to process images and trained based on historical tiled image data to extract a first set of image features from tiled images to predict a first likelihood of total loss for the vehicle; and

process via a second set of distinct and separate convolutional neural networks a multi-fusion set of images that comprise the set of distinct images, each of the second set of convolutional neural networks trained for a different non-overlapping view of the vehicle, with historical multi-fusion images, to extract a second set of image features from multi-fusion images to predict a second likelihood of total loss for the vehicle.

2 . The computer system of claim 1 , wherein the instructions, when executed, further cause the processor to rotate each cropped image to the defined orientation.

3 . The computer system of claim 1 , wherein the tiled image includes at least two cropped images combined into the single image.

4 . The computer system of claim 1 , wherein each of the images in the set of cropped images show the vehicle from a different view.

5 . The computer system of claim 4 , wherein the first convolutional neural network is applied to extract the first set of image features and the second set of distinct convolutional neural networks each trained for a particular view of the vehicle is applied as the respective machine learning model to extract the second set of image features.

6 . The computer system of claim 1 , wherein the different angle views are pre-defined.

7 . The computer system of claim 6 , wherein the instructions, when executed, further cause the processor to apply an image recognition machine learning model to classify cropped images as belonging to respective pre-defined views.

8 . The computer system of claim 1 , wherein the respective machine learning models process images with non-overlapping views of the vehicle.

9 . The computer system of claim 1 , wherein, to predict the likelihood of total loss, the instructions cause the processor to:

predict the first likelihood of total loss based on the first set of extracted features;

predict the second likelihood of total loss based on the second set of extracted features; and

predict the likelihood of total loss based on the first likelihood of total loss and the second likelihood of total loss.

10 . The computer system of claim 9 , wherein, to determine the second likelihood of total loss, the instructions cause the processor to fuse together the second set of extracted features.

11 . A computer implemented method for processing a plurality of digital images, the computer implemented method comprising:

determining a location of a vehicle and defining a bounding box surrounding the location based on applying an object detection machine learning model to distinct images of the vehicle;

cropping the distinct images to display only the vehicle and rotating the distinct images to a defined orientation;

generating a tiled image of the vehicle by combining the distinct images into a single image of different angle views in equal portions of the tiled image;

processing the generated tiled image via a first convolutional neural network configured for image processing and trained based on historical tiled image data to extract a first set of image features from tiled images to predict a first likelihood of total loss for the vehicle; and

processing via a second set of distinct and separate convolutional neural networks a multi-fusion set of images comprising the set of distinct images, each of the second set of convolutional neural networks trained for a different non-overlapping view of the vehicle, using historical multi-fusion images, to extract a second set of image features from multi-fusion images to predict a second likelihood of total loss for the vehicle.

12 . The computer implemented method of claim 11 , further comprising rotating each cropped image to the defined orientation.

13 . The computer implemented method of claim 11 , wherein the tiled image includes at least two cropped images combined into the single image.

14 . The computer implemented method of claim 11 , wherein each of the images in the set of cropped images show the vehicle from a different view.

15 . The computer implemented method of claim 14 , wherein the first convolutional neural network is applied to extract the first set of image features and the second set of distinct convolutional neural networks each trained for a particular view of the vehicle is applied as the respective machine learning model to extract the second set of image features.

16 . The computer implemented method of claim 11 , wherein the different angle views are pre-defined.

17 . The computer implemented method of claim 16 , further comprising applying an image recognition machine learning model to classify cropped images as belonging to respective pre-defined views.

18 . The computer implemented method of claim 11 , wherein the respective machine learning models process images with non-overlapping views of the vehicle.

19 . The computer implemented method of claim 11 , wherein predicting the likelihood of total loss further comprises:

predicting the first likelihood of total loss based on the first set of extracted features;

predicting the second likelihood of total loss based on the second set of extracted features; and

predicting the likelihood of total loss based on the first likelihood of total loss and the second likelihood of total loss.

20 . The computer implemented method of claim 19 , wherein, determining the second likelihood of total loss further comprises fusing together the second set of extracted features.

21 . A non-transitory computer-readable medium containing computer program code for processing a plurality of digital images, the computer program code being executable by a processor for the processor to perform a method, the method comprising:

determining a location of a vehicle and defining a bounding box surrounding the location based on applying an object detection machine learning model to distinct images of the vehicle;

cropping the distinct images to display only the vehicle and rotating the distinct images to a defined orientation;

generating a tiled image of the vehicle by combining the distinct images into a single image of different angle views in equal portions of the tiled image;

processing the generated tiled image via a first convolutional neural network configured for image processing and trained based on historical tiled image data to extract a first set of image features from tiled images to predict a first likelihood of total loss for the vehicle; and

processing via a second set of distinct and separate convolutional neural networks a multi-fusion set of images comprising the set of distinct images, each of the second set of convolutional neural networks trained for a different non-overlapping view of the vehicle, using historical multi-fusion images, to extract a second set of image features from multi-fusion images to predict a second likelihood of total loss for the vehicle.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 31, 2024
From: BOUËTTÉ, JEAN-CHRISTOPHE; LÉVESQUE, JIMMY; POULIN, MARC; GORTI, SATYA KRISHNA; LONG, KEYU; GERVAIS, NICOLAS; BOUCHARD, JENNIFER
To: THE TORONTO-DOMINION BANK
Reel/Frame 069708/0633 →
Continuity (2)
Continuation 17747819 · May 18, 2022
Related Publication 20250139708A1 · May 1, 2025
References Cited (20)
US 10657647B1 · Chen · 2020 [cited by examiner]
US 10692050B2 · Taliwal et al. · 2020 [cited by applicant]
US 11836684B2 · Gulati · 2023 [cited by examiner]
US 20170286763A1 · Fukuda · 2017 [cited by examiner]
US 20180260793A1 · Li · 2018 [cited by examiner]
US 20180300576A1 · Dalyac et al. · 2018 [cited by applicant]
US 20180365772A1 · Thompson · 2018 [cited by examiner]
US 20210081698A1 · Lindeman et al. · 2021 [cited by applicant]
US 20210358047A1 · Hantehzadeh · 2021 [cited by applicant]
US 20210358048A1 · Hantehzadeh · 2021 [cited by applicant]
US 20210358105A1 · Hantehzadeh · 2021 [cited by applicant]
US 20230067026A1 · Huts · 2023 [cited by examiner]
US 20230334438A1 · Horstmann · 2023 [cited by examiner]
US 20240086734A1 · Singh · 2024 [cited by examiner]
CA 3124237A1 · 2022 [cited by examiner]
CN 112862702A · 2021 [cited by applicant]
CN 114493896A · 2022 [cited by examiner]
WO 2021217852A1 · 2021 [cited by applicant]
“Reddy et al., Automatic Vehicle Damage Detection Classification framework using Fast and Mask Deep learning, Sep. 8, 2022, IEEE, entire document” (Year: 2022). [cited by examiner]
“Mallios et al., Vehicle Damage Severity Estimation for Insurance Operations Using In-The-Wild Mobile Images, Jul. 26, 2023, IEEE, entire document” (Year: 2023). [cited by examiner]