Automatic assessment of damage and repair costs in vehicles

1. A method, comprising:

causing a displaying of an outline of a selected damaged part of a damaged vehicle to be captured with a camera of a client device;

causing the camera of the client device to capture an image of the damaged vehicle based on the displaying of the outline of the selected damaged part;

receiving, at a server computing device over an electronic network and from the client device, the image of the damaged vehicle;

aligning the image to an undamaged version of the damaged vehicle;

segmenting the image into vehicle parts; and

detecting damage to a set of parts of the damaged vehicle by comparing portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle, wherein detecting damage to the set of parts includes:

comparing at least one of edge distribution, texture comparison, and spatial correlation of portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle;

determining whether at least one of the edge distribution, the texture comparison, and the spatial correlation exceeds a respective threshold difference value, wherein damage is detected in a portion of a vehicle part in the image if at least one of the edge distribution, the texture comparison, and the spatial correlation exceed the respective threshold difference value;

detecting a pose of the damaged vehicle in the image; and

determining which external vehicle parts are damaged in the image; and

calculating an estimated repair cost for the damaged vehicle based on which external vehicle parts are damaged based on accessing a parts database that includes repair costs.

2. The method of claim 1 , wherein the parts database that includes repair costs includes estimates for parts and labor for individual parts.

3. The method of claim 1 , further comprising removing artifacts from the image by:

removing background material from the image; and

removing specular reflection due to incident light on the damaged vehicle shown in the image.

4. The method of claim 1 , wherein damage is detected in a portion of a vehicle part in the image if at least two of the edge distribution, the texture comparison, and the spatial correlation exceed the respective threshold difference value.

5. The method of claim 1 , wherein the detecting damage to the set of parts includes comparing each of edge distribution, texture comparison, and spatial correlation of portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle.

6. The method of claim 1 , wherein the detecting the pose of the damaged vehicle in the image comprises:

training a first Convolutional Neural Networks (CNN) of a plurality of CNNs to detect the pose of a damaged vehicle in the image; and

training each of the plurality of CNNs to detect damage on a respective vehicle part of a plurality of external vehicle parts; and

executing the first CNN to detect the pose of the damaged vehicle in the image.

7. The method of claim 1 , wherein detecting damage to the set of parts further includes inferring damage to internal parts of the damaged vehicle from the determined damaged external vehicle parts; and

wherein calculating the estimated repair cost for the damaged vehicle is further based on which internal vehicle parts are inferred to be damaged.

8. The method of claim 7 , wherein inferring damage to internal parts of the damaged vehicle from the determined damaged external vehicle parts comprises executing a Markov Random Field (MRF) algorithm.

9. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause a computing device to perform operations of:

causing a displaying of an outline of a selected damaged part of a damaged vehicle to be captured with a camera of a client device;

causing the camera of the client device to capture an image of the damaged vehicle based on the displaying of the outline of the selected damaged part;

receiving, from the client device, the image of the damaged vehicle;

aligning the image to an undamaged version of the damaged vehicle;

segmenting the image into vehicle parts; and

detecting damage to a set of parts of the damaged vehicle by comparing portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle, wherein detecting damage to the set of parts includes:

comparing at least one of edge distribution, texture comparison, and spatial correlation of portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle;

determining whether at least one of the edge distribution, the texture comparison, and the spatial correlation exceeds a respective threshold difference value, wherein damage is detected in a portion of a vehicle part in the image if at least one of the edge distribution, the texture comparison, and the spatial correlation exceed the respective threshold difference value;

detecting a pose of the damaged vehicle in the image; and

determining which external vehicle parts are damaged in the image; and

calculating an estimated repair cost for the damaged vehicle based on which external vehicle parts are damaged based on accessing a parts database that includes repair costs.

10. The computer-readable medium of claim 9 , wherein the parts database that includes repair costs includes estimates for parts and labor for individual parts.

11. The computer-readable medium of claim 9 , further comprising removing artifacts from the image by:

removing background material from the image; and

removing specular reflection due to incident light on the damaged vehicle shown in the image.

12. The computer-readable medium of claim 9 , wherein damage is detected in a portion of a vehicle part in the image if at least two of the edge distribution, the texture comparison, and the spatial correlation exceed the respective threshold difference value.

13. The computer-readable medium of claim 9 , wherein the detecting damage to the set of parts includes comparing each of edge distribution, texture comparison, and spatial correlation of portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle.

14. The computer-readable medium of claim 9 , wherein the detecting the pose of the damaged vehicle in the image comprises:

training a first Convolutional Neural Networks (CNN) of a plurality of CNNs to detect the pose of a damaged vehicle in the image; and

training each of the plurality of CNNs to detect damage on a respective vehicle part of a plurality of external vehicle parts; and

executing the first CNN to detect the pose of the damaged vehicle in the image.

15. The computer-readable medium of claim 9 , wherein detecting damage to the set of parts further includes inferring damage to internal parts of the damaged vehicle from the determined damaged external vehicle parts; and wherein calculating the estimated repair cost for the damaged vehicle is further based on which internal vehicle parts are inferred to be damaged.

16. The computer-readable medium of claim 15 , wherein the inferring damage to internal parts of the damaged vehicle from the determined damaged external vehicle parts comprises executing a Markov Random Field (MRF) algorithm.

17. A computing device, comprising: a processor; and

a memory storing instructions that, when executed by the processor, cause the computing device to perform operations of:

causing a displaying of an outline of a selected damaged part of a damaged vehicle to be captured with a camera of a client device;

causing the camera of the client device to capture an image of the damaged vehicle based on the displaying of the outline of the selected damaged part;

receiving, from the client device, the image of the damaged vehicle;

aligning the image to an undamaged version of the damaged vehicle;

segmenting the image into vehicle parts; and

detecting damage to a set of parts of the damaged vehicle by comparing portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle, wherein detecting damage to the set of parts includes:

comparing at least one of edge distribution, texture comparison, and spatial correlation of portions of each vehicle part in the image to corresponding portions of each vehicle part in the undamaged version of the damaged vehicle;

determining whether at least one of the edge distribution, the texture comparison, and the spatial correlation exceeds a respective threshold difference value, wherein damage is detected in a portion of a vehicle part in the image if at least one of the edge distribution, the texture comparison, and the spatial correlation exceed the respective threshold difference value;

detecting a pose of the damaged vehicle in the image; and

determining which external vehicle parts are damaged in the image; and

calculating an estimated repair cost for the damaged vehicle based on which external vehicle parts are damaged based on accessing a parts database that includes repair costs.

18. The computing device of claim 17 , wherein the detecting the pose of the damaged vehicle in the image comprises:

training a first Convolutional Neural Networks (CNN) of a plurality of CNNs to detect the pose of a damaged vehicle in the image; and

training each of the plurality of CNNs to detect damage on a respective vehicle part of a plurality of external vehicle parts; and

executing the first CNN to detect the pose of the damaged vehicle in the image.

19. The computing device of claim 17 , wherein detecting damage to the set of parts further includes inferring damage to internal parts of the damaged vehicle from the determined damaged external vehicle parts; and

wherein calculating the estimated repair cost for the damaged vehicle is further based on which internal vehicle parts are inferred to be damaged.

20. The computing device of claim 19 , wherein the inferring damage to internal parts of the vehicle from the determined damaged external vehicle parts comprises executing a Markov Random Field (MRF) algorithm.