Patent Application Titled “Vehicle Accident Image Processing Method And Apparatus” Published Online (USPTO 20190251395)

2019 AUG 29 (NewsRx) -- By a News Reporter-Staff News Editor at Internet Daily News -- According to news reporting originating from Washington, D.C., by NewsRx journalists, a patent application by the inventors Zhang, Haitao (Hangzhou, China); Liu, Yongchao (Hangzhou, China), filed on February 13, 2019, was made available online on August 15, 2019.

The assignee for this patent application is Alibaba Group Holding Limited (George Town, Cayman Islands).

Reporters obtained the following quote from the background information supplied by the inventors: “As cars gradually become commonly used travel tools, the vehicle insurance market also rapidly develops, and vehicle insurance services are growing fast too. It is a very important step to fast and accurately assess a vehicle accident loss in the vehicle insurance service. For assessing the vehicle accident loss, a group of photos taken at a vehicle accident scene are needed. The group of photos can be taken by the insurance inspector at the scene, or can be independently taken by the insured and sent to the insurance inspector. In some cases, photos that are taken at a scene other than the vehicle accident scene can be mixed in the vehicle accident photos on purpose, and the photos can be of a similar vehicle but not the exact vehicle, are taken at a different place, or are taken at a later time. Some photographing devices store metadata information such as the photographing time, the photographing device, and the photographing location in photos, and the metadata information can be used to determine whether the photos are taken at the same vehicle accident scene. However, not all photos include metadata such as the time and the location, and such metadata is very easy to modify. In addition, Manual photo examination is labor-consuming, and examination quality cannot be ensured. Therefore, a more effective vehicle accident photo processing solution is needed.”

In addition to obtaining background information on this patent application, NewsRx editors also obtained the inventors’ summary information for this patent application: “Implementations of the present specification are intended to provide a more effective vehicle accident photo processing solution, to reduce disadvantages in the existing technology.

“To achieve the objective, an aspect of the present specification provides a method for training a convolutional neural network for processing a vehicle accident image, including the following: obtaining at least one positive sample pair, where each positive sample pair includes a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and training the convolutional neural network by using the at least one positive sample pair, to decrease the sum of at least one first distance respectively corresponding to the positive sample pair, where the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network.

“Another aspect of the present specification provides a method for training a convolutional neural network for processing a vehicle accident image, including the following: obtaining at least one negative sample pair, where each negative sample pair includes a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one negative sample pair, to increase the sum of at least one second distance respectively corresponding to the negative sample pair, where the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“Another aspect of the present specification provides a method for training a convolutional neural network for processing a vehicle accident image, including the following: obtaining at least one positive sample pair and at least one negative sample pair, where each positive sample pair includes a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair includes a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting the sum of at least one second distance corresponding to the negative sample pair from the sum of at least one first distance corresponding to the positive sample pair, where the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“In an implementation, in the method for training a convolutional neural network, the distance is a Euclidean distance.

“Another aspect of the present specification provides an apparatus for training a convolutional neural network for processing a vehicle accident image, including the following: an acquisition unit, configured to obtain at least one positive sample pair, where each positive sample pair includes a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and a training unit, configured to train the convolutional neural network by using the at least one positive sample pair, to decrease the sum of at least one first distance respectively corresponding to the positive sample pair, where the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network.

“Another aspect of the present specification provides an apparatus for training a convolutional neural network for processing a vehicle accident image, including the following: an acquisition unit, configured to obtain at least one negative sample pair, where each negative sample pair includes a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and a training unit, configured to train the convolutional neural network by using the at least one negative sample pair, to increase the sum of at least one second distance respectively corresponding to the negative sample pair, where the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“Another aspect of the present specification provides an apparatus for training a convolutional neural network for processing a vehicle accident image, including the following: an acquisition unit, configured to obtain at least one positive sample pair and at least one negative sample pair, where each positive sample pair includes a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair includes a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and a training unit, configured to train the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting the sum of at least one second distance corresponding to the negative sample pair from the sum of at least one first distance corresponding to the positive sample pair, where the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“Another aspect of the present specification provides a vehicle accident image processing method, including the following: obtaining N vehicle accident images, where N is a natural number greater than or equal to 2; obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images to a convolutional neural network obtained by using the previous training method; calculating a distance between any two of the feature vectors; and determining that two vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

“In an implementation, the vehicle accident image processing method further includes the following: determining the number B of distances greater than the first predetermined threshold after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a first probability P1=B/N; and determining that the N vehicle accident images are generally abnormal when the first probability is greater than a second predetermined threshold.

“In an implementation, the vehicle accident image processing method further includes the following: determining the number M of abnormal vehicle accident images in the N vehicle accident images after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a second probability P2=M/N; and determining that the N vehicle accident images are generally abnormal when the second probability is greater than a third predetermined threshold.

“Another aspect of the present specification provides a vehicle accident image processing apparatus, including the following: an acquisition unit, configured to obtain N vehicle accident images, where N is a natural number greater than or equal to 2; a feature acquisition unit, configured to input the vehicle accident images to a convolutional neural network obtained by using the previous training method, to obtain N feature vectors respectively corresponding to the vehicle accident images; a first calculation unit, configured to calculate a distance between any two of the feature vectors; and a first determining unit, configured to determine that two vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

“In an implementation, the vehicle accident image processing apparatus further includes the following: a second determining unit, configured to determine the number B of distances greater than the first predetermined threshold after it is determined that the two vehicle accident images corresponding to the distance are abnormal; a second calculation unit, configured to calculate a first probability P1=B/N; and a third determining unit, configured to determine that the N vehicle accident images are generally abnormal when the first probability is greater than a second predetermined threshold.

“In an implementation, the vehicle accident image processing apparatus further includes the following: a fourth determining unit, configured to determine the number M of abnormal vehicle accident images in the N vehicle accident images after it is determined that the two vehicle accident images corresponding to the distance are abnormal; a third calculation unit, configured to calculate a second probability P2=M/N; and a fifth determining unit, configured to determine that the N vehicle accident images are generally abnormal when the second probability is greater than a third predetermined threshold.

“Another aspect of the present specification provides a computer readable storage medium. The computer readable storage medium stores instruction code, and when the instruction code is executed in a computer, the computer performs the method for training a convolutional neural network and/or the vehicle accident image processing method.

“Another aspect of the present specification provides a computing device including a memory and a processor. The memory stores executable code, and when executing the executable code, the processor implements the method for training a convolutional neural network and/or the vehicle accident image processing method.

“Based on the method and apparatus for training a convolutional neural network for processing a vehicle accident image, and the vehicle accident image processing method and apparatus according to the implementations of the present specification, vehicle accident images can be quickly processed with high-accuracy and a high recall rate. As such, the vehicle accident images are fast and automatically processed.”

The claims supplied by the inventors are:

“1. A computer-implemented method for vehicle accident image processing, comprising: obtaining N vehicle accident images, wherein N is a natural number greater than or equal to 2; obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; calculating a distance between any two feature vectors of the N feature vectors; and determining that two vehicle accident images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

“2. The computer-implemented method of claim 1, further comprising: determining a number B of distances greater than the first predetermined threshold after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a first probability based on B and N; and determining that the N vehicle accident images are generally abnormal when the first probability is greater than a second predetermined threshold.

“3. The computer-implemented method of claim 1, further comprising: determining a number M of abnormal vehicle accident images in the N vehicle accident images after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a second probability P2 based on M and N; and determining that the N vehicle accident images are generally abnormal when the second probability is greater than a third predetermined threshold.

“4. The computer-implemented method of claim 1, wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises: obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and training the convolutional neural network by using the at least one positive sample pair, to decrease the sum of at least one first distance respectively corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network.

“5. The computer-implemented method of claim 1, wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises: obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one negative sample pair, to increase the sum of at least one second distance respectively corresponding to the negative sample pair, wherein the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“6. The computer-implemented method of claim 1, wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises: obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting the sum of at least one second distance corresponding to the negative sample pair from the sum of at least one first distance corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“7. The computer-implemented method of claim 1, wherein the distance is a Euclidean distance.

“8. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising: obtaining N vehicle accident images, wherein N is a natural number greater than or equal to 2; obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; calculating a distance between any two feature vectors of the N feature vectors; and determining that two vehicle accident images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

“9. The non-transitory, computer-readable medium of claim 8, further comprising: determining a number B of distances greater than the first predetermined threshold after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a first probability based on B and N; and determining that the N vehicle accident images are generally abnormal when the first probability is greater than a second predetermined threshold.

“10. The non-transitory, computer-readable medium of claim 8, further comprising: determining a number M of abnormal vehicle accident images in the N vehicle accident images after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a second probability P2 based on M and N; and determining that the N vehicle accident images are generally abnormal when the second probability is greater than a third predetermined threshold.

“11. The non-transitory, computer-readable medium of claim 8, wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises: obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and training the convolutional neural network by using the at least one positive sample pair, to decrease the sum of at least one first distance respectively corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network.

“12. The non-transitory, computer-readable medium of claim 8, wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises: obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one negative sample pair, to increase the sum of at least one second distance respectively corresponding to the negative sample pair, wherein the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“13. The non-transitory, computer-readable medium of claim 8, wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises: obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting the sum of at least one second distance corresponding to the negative sample pair from the sum of at least one first distance corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“14. The non-transitory, computer-readable medium of claim 8, wherein the distance is a Euclidean distance.

“15. A computer-implemented system, comprising: one or more computers; and one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising: obtaining N vehicle accident images, wherein N is a natural number greater than or equal to 2; obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; calculating a distance between any two feature vectors of the N feature vectors; and determining that two vehicle accident images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

“16. The computer-implemented system of claim 15, further comprising: determining a number B of distances greater than the first predetermined threshold after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a first probability based on B and N; and determining that the N vehicle accident images are generally abnormal when the first probability is greater than a second predetermined threshold.

“17. The computer-implemented system of claim 15, further comprising: determining a number M of abnormal vehicle accident images in the N vehicle accident images after determining that the two vehicle accident images corresponding to the distance are abnormal; calculating a second probability P2 based on M and N; and determining that the N vehicle accident images are generally abnormal when the second probability is greater than a third predetermined threshold.

“18. The computer-implemented system of claim 15, wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises: obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and training the convolutional neural network by using the at least one positive sample pair, to decrease the sum of at least one first distance respectively corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network.

“19. The computer-implemented system of claim 15, wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises: obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one negative sample pair, to increase the sum of at least one second distance respectively corresponding to the negative sample pair, wherein the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.

“20. The computer-implemented system of claim 15, wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises: obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting the sum of at least one second distance corresponding to the negative sample pair from the sum of at least one first distance corresponding to the positive sample pair, wherein the first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network.”

For more information, see this patent application: Zhang, Haitao; Liu, Yongchao. Vehicle Accident Image Processing Method And Apparatus. Filed February 13, 2019 and posted August 15, 2019. Patent URL: http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PG01&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.html&r=1&f=G&l=50&s1=%2220190251395%22.PGNR.&OS=DN/20190251395&RS=DN/20190251395

(Our reports deliver fact-based news of research and discoveries from around the world.)