Granted Patent B2

US 12,190,491 · App. 18/522,655 · Granted Jan 7, 2025

Inventors: Abhilash Nvs (Bhubaneswar, IN); Ankit Sati (New Delhi, IN); Payanshi Jain (Jaipur, IN); Koundinya K. Nvss (Tirupati, IN); Rajat Katiyar (Lucknow, IN); Mohiuddin Khan (Bangalore, IN); Chirag Jain (Bangalore, IN); Sreekanth Menon (Bangalore, IN)

Assignee: Genpact USA, Inc.

G06T7/0002G06F18/22G06T7/11G06T7/62G06T2207/20081G06T2207/20221G06T2207/30252

Abstract

Claims (57)

1. A method for dimension estimation based on duplicate instance identification, the method comprising:

receiving a set of images of an object;

detecting, from each image in the set of images, a respective image segmentation representing a damage of the object;

determining a respective dimension for the damage represented by each of the image segmentations;

determining whether two or more of the image segmentations represent a same damage of the object; and

responsive to two or more of the image segmentations representing a same damage of the object, combining the respective dimensions determined for the damage represented by the two or more image segmentations to obtain a final dimension for the same damage.

2. The method of claim 1 , further comprising:

responsive to no two or more of the image segmentations representing a same damage, using the dimension respectively determined for the damage of the object represented by each image segmentation as a respective final dimension for the damage.

3. The method of claim 1 , further comprising:

training a first machine learning system to detect the respective image segmentation; and

training a second machine learning system to determine the respective dimension.

4. The method of claim 3 , wherein determining the respective dimension for the damage represented by a particular image segmentation comprises:

identifying a reference object with a known physical dimension from the image from which the particular image segmentation was detected;

determining a pixel dimension of the reference object based on a pose and visibility of the reference object on the image from which the particular image segmentation was detected; and

determining the respective dimension based on a pixel dimension of the particular image segmentation on the image from which the particular image segmentation was detected.

5. The method of claim 4 , wherein the second machine learning system is trained to classify image segmentations including the reference object.

6. The method of claim 4 , wherein the object is a vehicle, and the reference object is a wheel of the vehicle.

7. The method of claim 1 , further comprising:

tagging the respective image segmentation to one or more anchor parts on the image from which the respective image segmentation was detected; and

generating a similarity index between image segmentations based on the tagging of the respective image segmentation to the one or more anchor parts.

8. The method of claim 7 , wherein generating the similarity index comprises performing a four-point homographic transformation.

9. The method of claim 7 , further comprising:

comparing the similarity index to a threshold,

wherein determining whether two or more of the image segmentations represent the same damage is based on whether the similarity index exceeds the threshold.

10. The method of claim 1 , wherein combining the respective dimensions comprises generating a weighted average of the respective dimensions and a confidence score corresponding to the respective dimensions.

11. The method of claim 1 , wherein, prior to receiving the set of images of the object, the method further comprises:

receiving a plurality of images captured in an uncontrolled environment; and

filtering out one or more irrelevant images from the plurality of images to obtain the set of images of the object.

12. The method of claim 1 , further comprising performing part detection to identify different parts of the object from the set of images.

13. The method of claim 12 , further comprising:

performing damage-to-part association by associating the respective image segmentation to the identified parts of the object,

wherein determining the respective dimension is based on associating the respective image segmentation to the identified parts of the object.

14. A system for dimension estimation based on duplicate instance identification, the system comprising:

a processor; and

a memory in communication with the processor and comprising instructions which, when executed by the processor, program the processor to:

receive a set of images of an object;

detect, from each image in the set of images, a respective image segmentation representing a damage of the object;

determine a respective dimension for the damage represented by each of the image segmentations;

determine whether two or more of the image segmentations represent a same damage of the object; and

responsive to two or more of the segmentations representing a same damage of the object, combine the respective dimensions determined for the damage represented by the two or more image segmentations to obtain a final dimension for the same damage.

15. The system of claim 14 , wherein the instructions further program the processor to:

responsive to no two or more of the image segmentations representing a same damage, use the dimension respectively determined for the damage of the object represented by each image segmentation as a respective final dimension for the damage.

16. The system of claim 14 , wherein the instructions further program the processor to:

train a first machine learning system to detect the respective image segmentation; and

train a second machine learning system to determine the respective dimension.

17. The system of claim 16 , wherein to determine the respective dimension for the damage represented by a particular image segmentation, the instructions further program the processor to:

identify a reference object with a known physical dimension from the image from which the particular image segmentation was detected;

determine a pixel dimension of the reference object based on a pose and visibility of the reference object on the image from which the particular image segmentation was detected; and

determine the respective dimension based on a pixel dimension of the particular image segmentation on the image from which the particular image segmentation was detected.

18. The system of claim 17 , wherein the instructions further program the processor to train the second machine learning system to classify image segmentations including the reference object.

19. The system of claim 17 , wherein the object is a vehicle, and the reference object is a wheel of the vehicle.

20. A computer program product for dimension estimation based on duplicate instance identification, the computer program product comprising a non-transitory computer-readable medium having computer readable program code stored thereon, the computer readable program code configured to:

receive a set of images of an object;

detect, from each image in the set of images, a respective image segmentation representing a damage of the object;

determine a respective dimension for the damage represented by each of the image segmentations;

determine whether two or more of the image segmentations represent a same damage of the object; and

responsive to two or more of the image segmentations representing a same damage of the object, combine the respective dimensions determined for the damage represented by the two or more segmentations to obtain a final dimension for the same damage.

Assignments (3)

Continuity (2)

References Cited (42)