Vehicle scratch similarity recognition method and system

Abstract

The application discloses a vehicle scratch similarity recognition method and system, and belongs to the technical field of computer vision. The vehicle scratch similarity recognition method comprises the following steps: determining the number of vehicle scratches in each of a first image and a second image to be recognized, and the type and position of each scratch; extracting and classifying sub-images in the first image and the second image according to the type and position of the scratches to obtain the image categories of the sub-images; marking feature points of each sub-image according to the image categories and a preset feature recognition model, wherein the preset feature recognition model is used to determine the coordinate vectors of the feature points of the sub-images; and comparing the coordinate vectors of the feature points of each sub-image in the first image with the coordinate vectors of the feature points of each sub-image in the second image to determine the similarity between the vehicle scratches in the first image and the vehicle scratches in the second image. The problems of low artificial inspection efficiency and low picture recognition accuracy in the related art are solved.

Description

Technical Field

[0001] This application belongs to the field of computer vision technology, specifically relating to a method and system for vehicle scratch similarity recognition. Background Technology

[0002] In the auto insurance claims process, photos of the damaged parts of the vehicle are taken during the inspection and damage assessment stages. By comparing the inspection photos and the damage assessment photos, it is sometimes found that the extent of the damage to the vehicle during the inspection is inconsistent with that during the damage assessment. In order to avoid such situations, relevant technologies usually use manual spot checks and image similarity recognition for verification.

[0003] Due to the massive scale of the business and the large number of photos, manual sampling can only select a small percentage, resulting in insufficient case coverage and an inability to effectively prevent such situations from occurring. Image similarity recognition aligns the photos from the inspection and damage assessment through stretching, rotation, scaling, and deformation, then compares the similarity between the aligned images. This method is computationally complex, time-consuming, and highly dependent on factors such as the photo's shooting angle, ambient lighting, and image sharpness, resulting in low accuracy and poor recognition performance. Summary of the Invention

[0004] This application provides a method and system for vehicle scratch similarity recognition, which can solve the problems of low efficiency of manual spot checks and low accuracy of image recognition in related technologies.

[0005] In a first aspect, embodiments of this application provide a method for vehicle scratch similarity recognition, including:

[0006] Determine the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage;

[0007] Based on the type and location of the injury, sub-images are extracted from the first image and the second image and classified to obtain the image category of the sub-image. The sub-image is the image of the location of the injury in the first image and the second image.

[0008] According to the image category and the preset feature recognition model, feature points are labeled for each of the sub-images. The preset feature recognition model is used to determine the coordinate vector of the feature points of the sub-image.

[0009] The coordinate vectors of feature points in each sub-image of the first image are compared with the coordinate vectors of feature points in each sub-image of the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image.

[0010] Secondly, embodiments of this application provide a vehicle scratch similarity recognition system, including:

[0011] The first determining module is used to determine the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage.

[0012] The extraction module is used to extract and classify sub-images from the first image and the second image according to the type and location of the injury, and obtain the image category of the sub-image, wherein the sub-image is the image of the location of the injury in the first image and the second image;

[0013] The annotation module is used to annotate feature points of each of the sub-images according to the image category and a preset feature recognition model, wherein the preset feature recognition model is used to determine the coordinate vector of the feature points of the sub-images;

[0014] The second determining module is used to compare the coordinate vectors of feature points of each sub-image in the first image with the coordinate vectors of feature points of each sub-image in the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image.

[0015] Thirdly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0016] In this embodiment, the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage, are first determined. Then, based on the type and location of the damages, sub-images are extracted from the first and second images and classified to obtain image categories. Each sub-image represents the location of the damage in the first and second images. Feature points are labeled for each sub-image according to the image category and a preset feature recognition model. The preset feature recognition model is used to determine the coordinate vectors of the feature points in the sub-images. Finally, the coordinate vectors of the feature points in each sub-image in the first image are compared with those in each sub-image in the second image to determine the similarity between the vehicle damages in the first and second images. This embodiment, by labeling each damage in the two images to be identified with feature points, determines the coordinate vectors of multiple feature points for each damage, and then compares them to determine the similarity between the two images. This can automatically identify whether the degree of damage to the vehicle in the first image (inspection image) is consistent with the degree of damage to the vehicle in the second image (damage assessment image), improving the accuracy and efficiency of identification. Attached Figure Description

[0017] Figure 1 This is a flowchart of a vehicle scratch similarity recognition method provided in one embodiment of this application;

[0018] Figure 2 This is a structural block diagram of a vehicle scratch similarity recognition system provided in one embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0021] The following description, in conjunction with the accompanying drawings, details a vehicle scratch similarity recognition method and system provided in this application through specific embodiments and application scenarios.

[0022] like Figure 1 As shown in the embodiments of this application, a method for vehicle scratch similarity recognition is also provided, such as... Figure 1 As shown, the vehicle damage similarity recognition method may include the contents shown in S101 to S104.

[0023] In S101, the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage, are determined.

[0024] The first image can be an image of the damaged part of the vehicle collected during the inspection stage, and the second image can be an image of the damaged part of the vehicle collected during the damage assessment stage, or an image collected under other circumstances, depending on the actual application. This embodiment does not make any specific limitations.

[0025] It is worth noting that when a vehicle is damaged, there may be damage to multiple parts, and each part may have multiple scratches.

[0026] In S102, sub-images are extracted from the first and second images and classified according to the type and location of the scars to obtain the image category of the sub-images.

[0027] In this context, a sub-image is an image showing the location of the injury in the first and second images. In other words, a sub-image is a local image within the overall captured image, which can more clearly show the details of the injury at that location, making subsequent comparisons more convenient. To clearly show the injury, each sub-image may include only one injury.

[0028] Image category refers to the type of damage in the image, such as scratches, deformation, cracks, etc. The specific type depends on the actual application and is not limited in this embodiment.

[0029] In S103, feature points are labeled for each sub-image according to the image category and the preset feature recognition model.

[0030] The preset feature recognition model is used to determine the coordinate vectors of feature points in the sub-image.

[0031] In other words, different feature points can be used to annotate images according to their different categories. For example, scratch images can be annotated using feature points such as scratch initiation point, scratch end point, scratch center point, the positive diameter point of the scratch center, and the widest point of the scratch; deformation images can be annotated using feature points such as deformation edge points and deformation center points; and fracture images can be annotated using feature points such as fracture initiation point, fracture branch end point, fracture center point, and fracture branch point. Each type of annotation can include multiple points; for example, scratch initiation point can include scratch initiation point 1, scratch initiation point 2, scratch initiation point 3, and so on.

[0032] In S104, the coordinate vectors of the feature points of each sub-image in the first image are compared with the coordinate vectors of the feature points of each sub-image in the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image.

[0033] In this embodiment, the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage, are first determined. Then, based on the type and location of the damages, sub-images are extracted from the first and second images and classified to obtain image categories. Each sub-image represents the location of the damage in the first and second images. Feature points are labeled for each sub-image according to the image category and a preset feature recognition model. The preset feature recognition model is used to determine the coordinate vectors of the feature points in the sub-images. Finally, the coordinate vectors of the feature points in each sub-image in the first image are compared with those in each sub-image in the second image to determine the similarity between the vehicle damages in the first and second images. This embodiment, by labeling each damage in the two images to be identified with feature points, determines the coordinate vectors of multiple feature points for each damage, and then compares them to determine the similarity between the two images. This can automatically identify whether the degree of damage to the vehicle in the first image (inspection image) is consistent with the degree of damage to the vehicle in the second image (damage assessment image), improving the accuracy and efficiency of identification.

[0034] In one possible implementation of this application, determining the number of vehicle damages in the first image and the second image to be identified, as well as the type and location of each damage, may include: determining a first number of vehicle damages in the first image to be identified, as well as the type and location of each damage, based on a preset detection model; and determining a second number of vehicle damages in the second image to be identified, as well as the type and location of each damage, based on a preset detection model.

[0035] This embodiment uses a preset detection model to determine the number of vehicle damages in the image, the type of each damage (e.g., scratches, deformation, cracks), and the location (coordinates) of each scratch. By classifying and locating vehicle damages, each damage can be obtained more clearly for subsequent comparison, improving the accuracy of recognition.

[0036] It is worth noting that if the damage to the vehicle in the first image is the same as the damage to the vehicle in the second image, the first number and the second number will be the same. However, due to differences in the shooting time, angle, environment, and other factors of the first and second images, the first number and the second number may differ.

[0037] In one possible implementation of this application, the training method for the preset detection model may include the contents shown in steps one to three.

[0038] Step 1: Obtain images of different types of vehicle damage.

[0039] In this embodiment, different types of vehicle damage images can be extracted from the database. These different types of vehicle damage images can include images from different shooting angles such as front, side, top, bottom, long distance, and close-up; images under different lighting conditions such as natural light, nighttime lights, and workshop lights; images of different vehicle models; images of different damage types such as scratches, deformation, and cracks; and other types of vehicle damage images. The specifics depend on the actual application, and this embodiment does not impose any limitations.

[0040] Step 2: Label each scratch in the scratch images of each vehicle to obtain the scratch type and location coordinates of each scratch.

[0041] In other words, each scratch in the different types of vehicle scratch images obtained above is labeled. Each scratch can be labeled by selecting a box in the image sample using an image labeling tool.

[0042] The annotation can include the damage-class of each damage, such as scratch, deformation, or crack, and the position coordinates (x, y, width, height) of each damage. For example, x and y are the coordinates of the upper left corner of the damage in the image, width is the width of the damage, and height is the height of the damage. Other methods can also be used to represent the position coordinates of the damage, depending on the actual application. This embodiment does not impose any specific limitations.

[0043] Step 3: Input the vehicle damage image and its corresponding damage type and location coordinates into the initial detection model for training to obtain the preset detection model.

[0044] In this embodiment, the vehicle damage images collected above and their corresponding annotations, namely the damage type and location coordinates, are input into the initial detection model for training to obtain the trained preset detection model.

[0045] The initial detection model can be any model applicable to the fields of machine learning and deep learning, and no specific limitation is made in this embodiment.

[0046] In this embodiment, a preset detection model can be obtained through model training. This preset detection model can automatically detect the type and location coordinates of the damage in the vehicle damage image, so as to improve the efficiency of subsequent image comparison.

[0047] In one possible implementation of this application, sub-images are extracted from the first image and the second image according to the type and location of the scar, and classified to obtain the image category of the sub-image. This may include: extracting sub-images from the first image and the second image according to the location of the scar; and classifying each extracted sub-image according to the type of scar to obtain the image category of the sub-image.

[0048] The image may contain 0 to n scars.

[0049] In this embodiment, each scratch is extracted from the original image according to its location coordinates to obtain a sub-image. These sub-images are then classified according to the type of scratch, resulting in multiple image category sample sets. For example, scratch types include scratches, deformations, and cracks. The sub-image category sample sets obtained according to these types include sample-scratches {sample-images}. 划痕}、sample-images 变形}、sample-images 破裂 A set of samples for three image categories.

[0050] This application describes the training method of a preset feature recognition model for a certain image category as an example. The training methods for other image categories are the same.

[0051] In one possible implementation of this application, the training method for the preset feature recognition model includes the steps one through three.

[0052] Step 1: Obtain multiple sub-images of the first image category.

[0053] Obtain multiple sub-images of the first image category from the multiple image category sample set obtained above. For example, obtain the scratch sample set {sample-images}. 划痕 Multiple sub-images in}

[0054] Step 2: Label the feature points for each sub-image to obtain the coordinate vector of each feature point.

[0055] Among them, the feature points include the starting point, the center point, the ending point, the orthogonal diameter point passing through the center point, the widest point, the edge point, and the branch point.

[0056] It is worth noting that the types of scars are different in each image category, so the feature point annotations for scars in each image category are also different.

[0057] For example, the feature points of a scratch may include scratch start point 1, scratch start point 2, scratch start point 3, scratch end point 1, scratch end point 2, scratch end point 3, scratch center point 1, scratch center point 2, scratch center point 3, scratch center point 4, scratch center point 5, orthogonal diameter 1 passing through the scratch center (point 1, point 2, point 3, point 4, point 5), orthogonal diameter 2 passing through the scratch center (point 1, point 2, point 3, point 4, point 5), and the widest point of the scratch (point 1, point 2, point 3). Up to 24 feature points can be marked, but the number of feature points can be appropriately reduced according to the size of the scratch. The specific number depends on the actual application, and this embodiment does not impose any limitations.

[0058] The deformable feature points may include: deformation edge point 1, deformation edge point 2, deformation edge point 3, deformation edge point 4, deformation edge point 5, deformation edge point 6, deformation edge point 7, deformation edge point 8, deformation edge point 9, deformation edge point 10, deformation center point 1, deformation center point 2, deformation center point 3, deformation center point 4, and deformation center point 5. Up to 15 feature points can be marked, or the number of feature points can be appropriately reduced according to the size of the deformation. The specific number depends on the actual application, and this embodiment does not limit it.

[0059] The feature points of the fracture may include: fracture initiation point 1, fracture initiation point 2, fracture initiation point 3, fracture branch 1 endpoint 1, fracture branch 1 endpoint 2, fracture branch 1 endpoint 3, fracture branch 2 endpoint 1, fracture branch 2 endpoint 2, fracture branch 2 endpoint 3, fracture center point 1, fracture center point 2, fracture center point 3, fracture center point 4, fracture center point 5, fracture branch point 1, fracture branch point 2, fracture branch point 3, fracture branch point 4, and fracture branch point 5. Up to 19 feature points can be marked, or the number of feature points can be appropriately reduced according to the degree of fracture. The specific number depends on the actual application, and this embodiment does not limit it.

[0060] Step 3: Input multiple sub-images of the first image category, and the coordinate vectors of multiple feature points corresponding to each sub-image, into the initial feature recognition model for training, to obtain the preset feature recognition model corresponding to the first image category.

[0061] In this embodiment, the sub-images classified above and the corresponding annotations of each sub-image, i.e. the feature points in each sub-image, are input into the initial feature recognition model for training to obtain the trained preset feature recognition model.

[0062] The initial feature recognition model can be a residual network deep learning algorithm, or any model applicable to machine learning and deep learning. No specific limitation is made in this embodiment.

[0063] In this embodiment, a preset feature recognition model can be obtained through model training. This preset feature recognition model can mark the location of the scars in each sub-image so as to improve the accuracy of subsequent image comparison.

[0064] It is worth noting that multiple models can be obtained by training the above model, for example, the scratch feature recognition model M-recognize. 划痕 M-recognize deformation recognition model 变形 M-recognize fracture recognition model 破裂 The aforementioned model can annotate feature points for the corresponding scar type and return a vector containing the coordinates of all feature points in the image. If there is no corresponding scar type in the image, it returns an empty coordinate vector with a length equal to the number of feature points for the scar type.

[0065] In one possible implementation of this application, comparing the coordinate vectors of feature points in each sub-image of the first image with the coordinate vectors of feature points in each sub-image of the second image to determine the similarity between vehicle damage in the first image and vehicle damage in the second image may include: converting the coordinate vectors of feature points in each sub-image of the first image category in the first image into 128-dimensional value vectors through a dimensionality-up operation to obtain a first vector; converting the coordinate vectors of feature points in each sub-image of the first image category in the second image into 128-dimensional value vectors through a dimensionality-up operation to obtain a second vector; and determining the Euclidean distance between the first vector and the second vector.

[0066] The Euclidean distance ranges from [0,1]. The closer the value is to 1, the more obvious the difference. The closer the value is to 0, the more similar the scars are.

[0067] In this embodiment, before comparing the coordinate vectors in the first image with the coordinate vectors in the second image, the coordinate vectors can be upgraded to obtain more details of the scars, making the comparison more accurate.

[0068] In one possible embodiment of this application, the vehicle damage similarity recognition method may further include: determining that the vehicle damage in the first image and the vehicle damage in the second image are the same damage when the Euclidean distance is less than a distance threshold; and determining that the vehicle damage in the first image and the vehicle damage in the second image are different damages when the Euclidean distance is greater than a distance threshold.

[0069] In other words, a distance threshold can be set. If the determined Euclidean distance is less than the distance threshold, the scar can be determined to be a uniform scar; otherwise, it is not the same scar.

[0070] It should be noted that the vehicle damage similarity recognition method provided in this application embodiment can be executed by a vehicle damage similarity recognition system, or a control module within that system for executing the vehicle damage similarity recognition method. This application embodiment uses the execution of the vehicle damage similarity recognition method by a vehicle damage similarity recognition system as an example to illustrate the vehicle damage similarity recognition system provided in this application embodiment.

[0071] like Figure 2 As shown in the figure, this application embodiment also provides a vehicle scratch similarity recognition system. The vehicle scratch similarity recognition system may include: a first determining module 201, an extraction module 202, an annotation module 203, and a second determining module 204.

[0072] The first determining module 201 is used to determine the number of vehicle damages in the first image and the second image to be identified, as well as the type and location of each damage; the extraction module 202 is used to extract and classify sub-images in the first image and the second image according to the type and location of the damages, and obtain the image category of the sub-images, wherein the sub-images are images of the locations of the damages in the first image and the second image; the annotation module 203 is used to annotate feature points of each sub-image according to the image category and a preset feature recognition model, wherein the preset feature recognition model is used to determine the coordinate vectors of the feature points of the sub-images; the second determining module 204 is used to compare the coordinate vectors of the feature points of each sub-image in the first image with the coordinate vectors of the feature points of each sub-image in the second image, and determine the similarity between the vehicle damages in the first image and the vehicle damages in the second image.

[0073] In this embodiment, the first determining module 201 first determines the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage. Then, the extraction module 202 extracts and classifies sub-images from the first and second images according to the type and location of the damages, obtaining the image category of the sub-images. The sub-images are images of the locations of the damages in the first and second images. The annotation module 203 annotates feature points of each sub-image according to the image category and a preset feature recognition model. The preset feature recognition model is used to determine the coordinate vectors of the feature points of the sub-images. Finally, the second determining module 204 compares the coordinate vectors of the feature points of each sub-image in the first image with the coordinate vectors of the feature points of each sub-image in the second image to determine the similarity between the vehicle damages in the first image and the vehicle damages in the second image. This embodiment, by annotating feature points of each damage in the two images to be identified, determines the coordinate vectors of multiple feature points of each damage, and then compares them to determine the similarity between the two images, can automatically identify whether the degree of damage of the vehicle damage in the first image (inspection image) is consistent with the degree of damage of the vehicle damage in the second image (damage assessment image), improving the accuracy and efficiency of identification.

[0074] In one possible implementation of this application, the first determining module 201 can be used to: determine a first number of vehicle scratches in a first image to be identified, as well as the type and location of each scratch, based on a preset detection model; and determine a second number of vehicle scratches in a second image to be identified, as well as the type and location of each scratch, based on a preset detection model.

[0075] In one possible implementation of this application, the first determining module 201 can be used to: acquire vehicle scratch images of different types; annotate each scratch in each vehicle scratch image to obtain the scratch type and location coordinates of each scratch; input the vehicle scratch images and their corresponding scratch types and location coordinates into an initial detection model for training to obtain a preset detection model.

[0076] In one possible implementation of this application, the extraction module 202 can be used to: extract sub-images from the first image and the second image according to the location of the scar; and classify each extracted sub-image according to the type of scar to obtain the image category of the sub-image.

[0077] In one possible implementation of this application, the annotation module 203 can be used to: acquire multiple sub-images of a first image category; annotate each sub-image with feature points to obtain the coordinate vector of each feature point, wherein the feature points include the start point, center point, end point, orthogonal diameter point passing through the center point, widest point, edge point, and branch point; and input the multiple sub-images of the first image category and the coordinate vectors of the multiple feature points corresponding to each sub-image into the initial feature recognition model for training to obtain the preset feature recognition model corresponding to the first image category.

[0078] In one possible implementation of this application, the second determining module 204 can be used to: convert the coordinate vectors of feature points of each sub-image of the first image category in the first image into 128-dimensional value vectors through a dimensionality-up operation to obtain a first vector; convert the coordinate vectors of feature points of each sub-image of the first image category in the second image into 128-dimensional value vectors through a dimensionality-up operation to obtain a second vector; and determine the Euclidean distance between the first vector and the second vector.

[0079] In one possible implementation of this application, the second determining module 204 can be used to: determine that the vehicle damage in the first image and the vehicle damage in the second image are the same damage when the Euclidean distance is less than a distance threshold; and determine that the vehicle damage in the first image and the vehicle damage in the second image are different damages when the Euclidean distance is greater than a distance threshold.

[0080] The vehicle scratch similarity recognition system in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0081] The vehicle damage similarity recognition system in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0082] The vehicle scratch similarity recognition system provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method embodiments achieve the same technical effect, and will not be described again here to avoid repetition.

[0083] This application also provides a readable storage medium storing a program or instructions. When executed by a processor, the program or instructions implement the various processes of the vehicle scratch similarity recognition method embodiments provided in any of the above embodiments. Furthermore, it achieves the same technical effects, and to avoid repetition, it will not be described again here.

[0084] Among them, readable storage media include computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, etc.

[0085] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0087] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for vehicle scratch similarity recognition, characterized in that, include: Determine the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage; Based on the type and location of the injury, sub-images are extracted from the first image and the second image and classified to obtain the image category of the sub-image. The sub-image is the image of the location of the injury in the first image and the second image. According to the image category and the preset feature recognition model, feature points are labeled for each of the sub-images. The preset feature recognition model is used to determine the coordinate vector of the feature points of the sub-image. Each image category corresponds to a preset feature recognition model. The coordinate vectors of feature points in each sub-image of the first image are compared with the coordinate vectors of feature points in each sub-image of the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image. The training method for the preset feature recognition model includes: Obtain multiple sub-images of the first image category; Each sub-image is labeled with feature points to obtain the coordinate vector of each feature point. The feature points include the starting point, center point, ending point, orthogonal diameter point passing through the center point, widest point, edge point, and branch point. Multiple sub-images of the first image category, and the coordinate vectors of multiple feature points corresponding to each sub-image, are input into the initial feature recognition model for training, thereby obtaining the preset feature recognition model corresponding to the first image category.

2. The vehicle scratch similarity recognition method according to claim 1, characterized in that, Determining the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage, includes: Based on a preset detection model, determine the first number of vehicle scratches in the first image to be identified, as well as the type and location of each scratch; Based on the preset detection model, the second number of vehicle damages in the second image to be identified, as well as the type and location of each damage, are determined.

3. The vehicle scratch similarity recognition method according to claim 2, characterized in that, The training method for the preset detection model includes: Acquire images of different types of vehicle damage; Each scratch in the vehicle scratch images is labeled to obtain the scratch type and location coordinates of each scratch; The vehicle damage image and its corresponding damage type and location coordinates are input into the initial detection model for training to obtain the preset detection model.

4. The vehicle scratch similarity recognition method according to claim 1, characterized in that, The step involves extracting and classifying sub-images from the first and second images based on the type and location of the injury, thereby obtaining the image category of the sub-images, including: Based on the location of the injury, extract sub-images from the first image and the second image; According to the type of injury, each extracted sub-image is classified to obtain the image category of the sub-image.

5. The vehicle scratch similarity recognition method according to claim 1, characterized in that, The step of comparing the coordinate vectors of feature points in each sub-image of the first image with the coordinate vectors of feature points in each sub-image of the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image includes: The coordinate vectors of feature points of each sub-image of the first image category in the first image are converted into 128-dimensional value vectors through a dimensionality-up operation to obtain the first vector; The coordinate vectors of feature points of each sub-image of the first image category in the second image are converted into 128-dimensional value vectors through a dimensionality-up operation to obtain the second vector; Determine the Euclidean distance between the first vector and the second vector.

6. The vehicle scratch similarity recognition method according to claim 5, characterized in that, The method further includes: If the Euclidean distance is less than the distance threshold, the vehicle damage in the first image and the vehicle damage in the second image are determined to be the same damage. If the Euclidean distance is greater than the distance threshold, the vehicle damage in the first image and the vehicle damage in the second image are determined to be different damages.

7. A vehicle scratch similarity recognition system, characterized in that, include: The first determining module is used to determine the number of vehicle damages in the first and second images to be identified, as well as the type and location of each damage. The extraction module is used to extract and classify sub-images from the first image and the second image according to the type and location of the injury, and obtain the image category of the sub-image, wherein the sub-image is the image of the location of the injury in the first image and the second image; The annotation module is used to annotate feature points of each of the sub-images according to the image category and the preset feature recognition model. The preset feature recognition model is used to determine the coordinate vector of the feature points of the sub-image. Each image category corresponds to one preset feature recognition model. The second determining module is used to compare the coordinate vectors of feature points of each sub-image in the first image with the coordinate vectors of feature points of each sub-image in the second image to determine the similarity between the vehicle damage in the first image and the vehicle damage in the second image. The annotation module is used to acquire multiple sub-images of a first image category; to annotate each sub-image with feature points to obtain the coordinate vectors of each feature point, wherein the feature points include a start point, a center point, an end point, an orthogonal diameter point passing through the center point, the widest point, an edge point, and a branch point; and to input the multiple sub-images of the first image category and the coordinate vectors of the multiple feature points corresponding to each sub-image into an initial feature recognition model for training to obtain a preset feature recognition model corresponding to the first image category.

8. The vehicle scratch similarity recognition system according to claim 7, characterized in that, The extraction module is used for: Based on the location of the injury, extract sub-images from the first image and the second image; According to the type of injury, each extracted sub-image is classified to obtain the image category of the sub-image.

9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the vehicle scratch similarity recognition method as described in any one of claims 1-6.

Images