Image processing system using recurrent neural networks

1. A method, comprising:

obtaining, by each instance of a Recurrent Neural Network (RNN) included in a plurality of instances of the RNN and having a one-to-one correspondence with a respective Convolutional Neural Network (CNN) included in a plurality of CNNs, a respective set of CNN-specific features extracted from a set of images of a damaged vehicle by the respective CNN, the respective CNNs included in the plurality of CNNs having different architectures, and the plurality of CNNs and each instance of the RNN having the one-to-one correspondence with the respective CNN included in the plurality of CNNs trained using historical vehicle total loss and repair (TLR) data;

operating on, by the each instance of the RNN, the respective set of CNN-specific features, thereby generating a respective RNN output indicative of a Total Loss or Repair (TLR) decision, the TLR decision being one of: the damaged vehicle corresponds to a total loss or the damaged vehicle corresponds to being repaired;

determining a majority of the respective RNN outputs of the respective instances of the RNN, wherein the majority of the RNN outputs indicates a final Total Loss or Repair (TLR) decision indicative of one of: (i) the damaged vehicle is a total loss, or (ii) the damaged vehicle is to be repaired; and

providing an indication of the final Total Loss or Repair (TLR) decision to at least one of a user interface or an application.

2. The method of claim 1 , wherein:

the respective set of CNN-specific features extracted from the set of images of the damaged vehicle by the respective CNN includes a respective subset of CNN-specific features extracted from each image included in the set of images of the damaged vehicle; and

operating on, by the each instance of the RNN, the respective set of CNN-specific features comprises sequentially operating on, by the each instance of the RNN, the respective subsets of CNN-specific features.

3. The method of claim 1 , further comprising selecting the set of images from a plurality of images of the damaged vehicle based on at least one of: image quality, image content, or a total number of images that are to be selected.

4. The method of claim 1 , wherein:

the method further comprises obtaining the one or more types of other data corresponding to the damaged vehicle, the one or more other types of data including at least one of telematics data, drivability status data, point of impact data, accident report data, or video data corresponding to the damaged vehicle; and

operating on, by the each instance of the RNN, the respective set of CNN-specific features thereby generating the respective RNN output comprises operating on, by the each instance of the RNN, the one or more other types of data corresponding to the damaged vehicle in conjunction with the respective set of CNN-specific features thereby generating the respective RNN output.

5. The method of claim 1 , wherein at least one of: the damaged vehicle is a vehicle that is damaged via multiple points of impact, or the set of images includes only a single image.

6. The method of claim 1 , wherein the set of images of the damaged vehicle includes multiple images of the damaged vehicle.

7. An image processing system, comprising:

a processor and a memory storing processor-executable instructions for:

an input interface including processor-executable instructions to obtain a set of images of a damaged vehicle by the image processing system;

a plurality of instances of a Recurrent Neural Network (RNN), each RNN instance coupled to a respective Convolutional Neural Network (CNN) of a plurality of CNNs having different architectures, each instance of the RNN having the one-to-one correspondence with the respective CNN included in the plurality of CNNs and the plurality of CNNs trained using historical vehicle total loss and repair data, and each RNN instance operating on a respective set of CNN-specific features extracted from a set of images of a damaged vehicle by the respective CNN, thereby generating a respective RNN output indicative of whether the damaged vehicle corresponds to a total loss or the damaged vehicle corresponds to being repaired;

a determination module including processor-executable instructions to determine a majority of the respective RNN outputs of the respective RNN instances, wherein the majority of the RNN outputs indicates a final Total Loss or Repair (TLR) decision indicative of one of: (i) the damaged vehicle is a total loss, or (ii) the damaged vehicle is to be repaired; and

an output interface via which an indication of the final Total Loss or Repair (TLR) decision is provided to at least one of a user interface or an application.

8. The image processing system of claim 7 , wherein:

the respective set of CNN-specific features extracted from the set of images of the damaged vehicle by the respective CNN includes a respective subset of CNN-specific features extracted from each image included in the set of images of the damaged vehicle; and

each instance of the RNN sequentially operates on the respective subsets of CNN-specific features.

9. The image processing system of claim 7 , further comprising an image selection module that operates on a plurality of images of the damaged vehicle to select the set of images of the damaged vehicle, the selection based on at least one of: an image quality, an image content, or a total number of images that are to be selected.

10. The image processing system of claim 7 , wherein each CNN of the plurality of CNNs is selected for inclusion in the image processing system based on respective accuracies of Total Loss or Repair (TLR) decisions generated based on different combinations of different types of CNNs, the each CNN included in the different types of CNNs.

11. The image processing system of claim 7 , wherein:

one or more other types of data corresponding to the damaged vehicle is obtained via the input interface, the one or more other types of data including at least one of: telematics data, drivability status data, point of impact data, accident report data, or video data; and

the each RNN instance operates on the respective one or more CNN-specific features extracted by the respective CNN from the each image in conjunction with the one or more other types of data corresponding to the damaged vehicle to thereby generate the respective RNN output.

12. The image processing system of claim 7 , wherein the damaged vehicle includes multiple points of impact.

13. The image processing system of claim 7 , where the set of images of a damaged vehicle includes multiple images of the damaged vehicle.

14. An image processing system, comprising:

one or more processors; and

one or more tangible, non-transitory computer-readable media storing computer-executable instructions that, when executed by the one or more processors, cause the image processing system to:

obtain, by each instance of a Recurrent Neural Network (RNN) included in a plurality of instances of the RNN and having a one-to-one correspondence with a respective Convolutional Neural Network (CNN) included in a plurality of CNNs, a respective set of CNN-specific features extracted from a set of images of a damaged vehicle by the respective CNN, the respective CNNs included in the plurality of CNNs having different architectures, and the plurality of CNNs and each instance of the RNN having the one-to-one correspondence with the respective CNN included in the plurality of CNNs trained using historical vehicle total loss and repair (TLR) data;

operate on, by the each instance of the RNN, the respective set of CNN-specific features, thereby generating a respective RNN output indicative of a Total Loss or Repair (TLR) decision, the TLR decision being one of: the damaged vehicle corresponds to a total loss or the damaged vehicle corresponds to being repaired;

determining a majority of the respective RNN outputs of the respective instances of the RNN, wherein the majority of the RNN outputs indicates a final Total Loss or Repair (TLR) decision indicative of one of: (i) the damaged vehicle is a total loss, or (ii) the damaged vehicle is to be repaired; and

provide an indication of the final Total Loss or Repair (TLR) decision to at least one of a user interface or an application.

15. The image processing system of claim 14 , wherein the computer-executable instructions are executable to cause the image processing system further to:

select the set of images from a plurality of images of the damaged vehicle based on at least one of: image quality, image content, or a total number of images that are to be selected.

16. The image processing system of claim 14 , wherein:

the respective set of CNN-specific features extracted from the set of images of the damaged vehicle by the respective CNN includes a respective subset of CNN-specific features extracted from each image included in the set of images of the damaged vehicle; and

each instance of the RNN sequentially operates on the respective subsets of CNN-specific features.

17. The image processing system of claim 14 , wherein each CNN of the plurality of CNNs is selected for inclusion in the image processing system based on respective accuracies of Total Loss or Repair (TLR) decisions generated based on different combinations of different types of CNNs, the each CNN included in the different types of CNNs.

18. The image processing system of claim 14 , wherein:

one or more other types of data corresponding to the damaged vehicle is obtained via the input interface, the one or more other types of data including at least one of: telematics data, drivability status data, point of impact data, accident report data, or video data; and

the each instance of the RNN operates on the respective set of CNN-specific features extracted by the respective CNN from the each image in conjunction with the one or more other types of data corresponding to the damaged vehicle to thereby generate the respective RNN output.

19. The image processing system of claim 14 , wherein the set of images of the damaged vehicle includes a plurality of images of the damaged vehicle.

20. The image processing system of claim 14 , wherein the set of images of the damaged vehicle includes only a single image of the damaged vehicle.

21. The image processing system of claim 14 , wherein the damaged vehicle includes multiple points of impact.