Vehicle appearance detection and parameter prediction model training method, device and equipment

Abstract

The present application relates to the field of image processing technology, and in particular to a method, device and equipment for training a vehicle appearance detection and parameter prediction model, which is used to improve the accuracy of vehicle appearance detection results. The method includes: obtaining an image to be detected and initial environmental parameters corresponding to the vehicle to be detected, a standard appearance image of the vehicle to be detected and an appearance image to be detected; inputting the image to be detected and the initial environmental parameters into a trained parameter prediction model to obtain target environmental parameters predicted by the parameter prediction model, which represent the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected; adjusting the initial acquisition device parameters according to the target environmental parameters to obtain target acquisition device parameters; collecting the target appearance image of the vehicle to be detected according to the target acquisition device parameters, and performing vehicle appearance detection based on the target appearance image, which is determined based on the adjusted acquisition device parameters, and based on this, a more accurate detection result can be obtained.

Description

Background Art

[0002] With the development of social economy, the popularity of automobiles is increasing, and the total number of vehicles is also increasing. In order to eliminate potential safety hazards of vehicles and ensure traffic safety, each new vehicle needs to be inspected by the motor vehicle manufacturer before leaving the factory. In addition, vehicles that have obtained official license plates and driving licenses must undergo annual vehicle inspections to urge and strengthen vehicle maintenance and reduce the occurrence of traffic accidents.

[0003] Vehicle appearance inspection requires that the photographed appearance should be consistent with the standard photo and the actual color of the vehicle body. However, when actually collecting vehicle appearance photos, due to the influence of many factors, the quality of the collected appearance photos is often poor.

[0004] To solve the above problems, the relevant technologies mainly adopt manual judgment and algorithm judgment to perform vehicle appearance inspection. Among them, manual judgment means that the vehicle inspection personnel manually judge the body color of the vehicle appearance picture and compare it with the recorded legal body color. This method cannot be adapted to all models, cannot fundamentally solve the problem, and is time-consuming, affecting the production and business efficiency of automobile companies.

[0005] Algorithm judgment, such as using a semantic segmentation model to identify the vehicle appearance image, labeling multiple labels based on the recognition results, and then comparing the probability of the labels with the vehicle standard color to detect the appearance color. This method cannot be widely used for the composite colors of many current car companies, because the probability of multiple labels after recognition is easily affected by external factors, resulting in a situation where the color is completely inconsistent with the actual body color, and thus cannot be used effectively.

[0006] In summary, how to reduce the interference of external factors and ensure that the results of vehicle appearance inspection can accurately reflect the consistency with the standard photo and physical body color has become a technical problem that needs to be solved urgently. Summary of the invention

[0007] The embodiments of the present application provide a method, device and equipment for training a vehicle appearance detection and parameter prediction model to improve the accuracy of vehicle appearance detection results.

[0008] A vehicle appearance detection method provided in an embodiment of the present application includes:

[0009] Acquire the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected; wherein the image to be detected includes: the standard appearance image of the vehicle to be detected and the appearance image to be detected, and the initial environmental parameters represent: the environmental conditions when the appearance image to be detected is collected;

[0010] Input the image to be detected and the initial environmental parameters into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model; wherein the target environmental parameters represent: the environmental conditions expected when the appearance image of the vehicle to be detected is currently collected; the parameter prediction model is trained based on the sample image and sample environmental parameters corresponding to the sample vehicle;

[0011] The initial acquisition equipment parameters are adjusted according to the target environment parameters to obtain the target acquisition equipment parameters for the vehicle to be detected; wherein the initial acquisition equipment parameters include: various index parameters in the acquisition equipment when acquiring the appearance image to be detected;

[0012] The target appearance image of the vehicle to be detected is acquired according to the target acquisition device parameters, and the vehicle appearance detection is performed based on the target appearance image to obtain a corresponding detection result.

[0013] An embodiment of the present application provides a vehicle appearance detection device, comprising:

[0014] A data acquisition unit, used to acquire an image to be detected and initial environmental parameters corresponding to the vehicle to be detected; wherein the image to be detected includes: a standard appearance image of the vehicle to be detected and an appearance image to be detected, and the initial environmental parameters represent: environmental conditions when the appearance image to be detected is collected;

[0015] An environmental parameter prediction unit is used to input the image to be detected and the initial environmental parameters into a trained parameter prediction model to obtain target environmental parameters predicted by the parameter prediction model; wherein the target environmental parameters represent: the environmental conditions expected when the appearance image of the vehicle to be detected is currently collected; the parameter prediction model is trained based on the sample image and sample environmental parameters corresponding to the sample vehicle;

[0016] The device parameter adjustment unit is used to adjust the initial acquisition device parameters according to the target environment parameters to obtain the target acquisition device parameters for the vehicle to be detected; wherein the initial acquisition device parameters include: various index parameters in the acquisition device when acquiring the appearance image to be detected;

[0017] The appearance detection unit is used to collect the target appearance image of the vehicle to be detected according to the target acquisition device parameters, and perform vehicle appearance detection based on the target appearance image to obtain corresponding detection results.

[0018] Optionally, the appearance detection unit is specifically used for:

[0019] Comparing the target appearance image with the standard appearance image to obtain the detection result;

[0020] If the detection result indicates that the target appearance image is inconsistent with the standard appearance image, the appearance detection unit is further configured to:

[0021] The current target appearance image and target environment parameters are used as the new appearance image to be detected and initial environment parameters corresponding to the vehicle to be detected, and the step of inputting the image to be detected and the initial environment parameters into the trained parameter prediction model to obtain the target environment parameters predicted by the parameter prediction model is returned, until the latest detection result indicates that the latest target appearance image is consistent with the standard appearance image.

[0022] Optionally, the parameter prediction model includes an image-generated video sub-model and a fixed parameter sub-model; and the environmental parameter prediction unit is specifically used for:

[0023] Inputting the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and obtaining a vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model;

[0024] Inputting the vehicle appearance video and the standard appearance image into the fixed parameter sub-model, segmenting and decoding the vehicle appearance video through the fixed parameter sub-model, and obtaining each frame of the vehicle appearance video reference appearance image and corresponding reference environment parameters;

[0025] Each frame of the reference appearance image is compared with the standard appearance image, and the target environment parameters are determined from the obtained reference environment parameters according to the comparison result.

[0026] Optionally, the environmental parameter prediction unit is specifically used for:

[0027] Inputting the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and extracting feature vectors corresponding to the appearance image to be detected and the initial environmental parameters respectively through the image-generated video sub-model;

[0028] The extracted feature vectors are fused to obtain a fused vector;

[0029] Generating a vehicle appearance video corresponding to the vehicle to be detected based on the fusion;

[0030] and, comparing each frame of the reference appearance image with the standard appearance image to obtain an image similarity corresponding to each frame of the reference appearance image;

[0031] The reference environment parameters corresponding to the reference appearance image with the highest image similarity are used as the target environment parameters.

[0032] A training method for a parameter prediction model provided in an embodiment of the present application includes:

[0033] Acquire a training sample set, wherein each training sample in the training sample set includes: a sample image corresponding to a sample vehicle and a sample environmental parameter; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: an environmental condition when the sample appearance image is collected;

[0034] For each training sample, perform the following operations:

[0035] Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle;

[0036] The vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained; based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image, and the parameters of the image-generated video sub-model are adjusted according to the comparison result; wherein the trained parameter prediction model is used to predict the first target environmental parameters for the sample vehicle, the first target environmental parameters representing the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the first target environmental parameters and perform vehicle appearance detection.

[0037] The present application provides a parameter prediction model training device, comprising:

[0038] A sample acquisition unit is used to acquire a training sample set, each training sample in the training sample set includes: a sample image corresponding to a sample vehicle and a sample environmental parameter; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: an environmental condition when the sample appearance image is collected;

[0039] The training unit is used to perform the following operations on each training sample:

[0040] Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle;

[0041] The vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained; based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image, and the parameters of the image-generated video sub-model are adjusted according to the comparison result; wherein the trained parameter prediction model is used to predict the first target environmental parameters for the sample vehicle, the first target environmental parameters representing the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the first target environmental parameters and perform vehicle appearance detection.

[0042] Optionally, the training unit is specifically used for:

[0043] Based on the fixed parameter sub-model, the vehicle appearance video is segmented and decoded to obtain a reference appearance image of each frame in the vehicle appearance video and corresponding reference environment parameters;

[0044] Comparing each frame of the reference appearance image with the corresponding standard appearance image to obtain the image similarity corresponding to each frame of the reference appearance image;

[0045] If the obtained similarities of each image are not higher than a preset threshold, the parameters of the image-generated video sub-model are adjusted according to the difference between the reference environment parameters and the corresponding sample environment parameters.

[0046] Optionally, the training unit is specifically used for:

[0047] Parameters of a low-rank matrix corresponding to the image-generated video sub-model are adjusted, wherein the low-rank matrix is ​​obtained by performing low-rank decomposition on a weight matrix to be adjusted in the image-generated video sub-model.

[0048] Optionally, the training device of the parameter prediction model is also used for:

[0049] If the similarities of the obtained images are not higher than the preset threshold, the reference appearance image and the corresponding reference environment parameters with the highest image similarity are used as the new sample appearance image and sample environment parameters corresponding to the sample vehicle, and the step of inputting the corresponding sample image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained to obtain the vehicle appearance video corresponding to the corresponding sample vehicle is returned, until the image similarities corresponding to each frame of the reference appearance image in the latest vehicle appearance video have an image similarity higher than the preset threshold;

[0050] If there is an image similarity higher than the preset threshold among the obtained image similarities, the reference environmental parameters corresponding to the reference appearance image with the highest image similarity are used as the second target environmental parameters corresponding to the sample vehicle. The second target environmental parameters represent the expected environmental conditions when the appearance image of the sample vehicle is currently being collected.

[0051] An electronic device provided in an embodiment of the present application includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the above-mentioned vehicle appearance detection methods or parameter prediction model training methods.

[0052] An embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any one of the above-mentioned vehicle appearance detection methods or parameter prediction model training methods.

[0053] An embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device performs the steps of any one of the above-mentioned vehicle appearance detection methods or parameter prediction model training methods.

[0054] The beneficial effects of this application are as follows:

[0055] The embodiments of the present application provide a vehicle appearance detection method, device and equipment. Specifically, the present application first obtains the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected, and by setting the initial environmental parameters, ensures that when the appearance image to be detected is collected, the environmental conditions at that time can be recorded, so that subsequent image processing and analysis can be adjusted based on the specific environmental conditions.

[0056] Secondly, the image to be detected and the initial environmental parameters are input into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model. Since the parameter prediction model based on deep learning is adopted and the image to be detected and the initial environmental parameters are input, the model can effectively predict the target environmental parameters required to collect high-quality appearance images in the current environment. By predicting the target environmental parameters, the system can automatically adjust the device settings so that the collected images are closer to the effects under ideal conditions, reducing the impact of environmental changes on image quality.

[0057] Then, the initial acquisition equipment parameters are adjusted according to the target environment parameters to obtain the target acquisition equipment parameters for the vehicle to be detected. Since a mechanism for automatically adjusting the equipment parameters is set, the system can optimize the various indicators of the acquisition equipment according to the predicted target environment parameters. This dynamic adjustment can keep the acquisition equipment in the best state under different environmental conditions, automatically optimize the acquisition equipment parameters, and ensure that the quality of the collected image meets the standard, thereby further improving the accuracy of the detection. On this basis, the target appearance image of the vehicle to be detected is collected according to the target acquisition equipment parameters, and the vehicle appearance detection is performed based on the target appearance image to obtain the corresponding detection results. Since the vehicle appearance detection is based on the high-quality target appearance image, the result is closer to the actual state of the vehicle, reducing the interference of external factors on the detection results, improving the applicability of the appearance image of the vehicle under different environmental conditions, reducing the impact of the surrounding environment on the photo effect, and ensuring that the detection results can accurately reflect the consistency of the vehicle appearance with the standard photo and the physical body color.

[0058] The embodiment of the present application also provides a training method, device and equipment for a parameter prediction model. Specifically, a training sample set is first obtained, and each training sample in the training sample set contains a sample image and sample environment parameters corresponding to the sample vehicle. Such a design enables the model to learn the diversity of vehicle appearance under different environmental conditions, improves the generalization ability of the model, and ensures that the model can perform well in different practical application scenarios.

[0059] Secondly, for each training sample, the corresponding sample appearance image and sample environmental parameters are input into the image-generated video sub-model in the parameter prediction model to be trained to obtain the vehicle appearance video corresponding to the corresponding sample vehicle. By utilizing deep learning technology, a realistic vehicle appearance video is generated, so that the model can simulate the changes in vehicle appearance under different environmental conditions, providing high-quality data input for subsequent parameter-determining sub-model processing.

[0060] Subsequently, the vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained. Based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image. By comparing the image similarity, the model can identify the difference between the video frame and the standard image, thereby judging whether the current video generation is accurate. Based on these differences, the system can further adjust the parameters of the image-generated video sub-model, gradually optimize the generation capability of the image-generated video sub-model, ensure that the generated video is as close as possible to the standard appearance image, and enable the trained parameter prediction model to more accurately predict the target environment parameters. In the model application stage, this capability can be used to optimize the collected appearance images and improve the detection accuracy.

[0061] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0063] Figure 1 A schematic diagram of an application scenario of an embodiment of the present application;

[0064] Figure 2 A flowchart of an implementation of a model training method provided in an embodiment of the present application;

[0065] Figure 3 A processing logic diagram of a parameter prediction model provided in an embodiment of the present application;

[0066] Figure 4 A flowchart of a parameter adjustment process of a parameter prediction model provided in an embodiment of the present application;

[0067] Figure 5 A schematic diagram of image similarity comparison between a set of standard photos and corresponding reference appearance photos provided in an embodiment of the present application;

[0068] Figure 6 A flowchart of a vehicle appearance detection method provided in an embodiment of the present application;

[0069] Figure 7 An implementation diagram of a vehicle appearance detection method provided in an embodiment of the present application;

[0070] Figure 8 A schematic diagram of the structure of a vehicle appearance detection device in an embodiment of the present application;

[0071] Fig. 9 A schematic diagram of the structure of a training device for a parameter prediction model in an embodiment of the present application;

[0072] Fig.10 A schematic diagram of a hardware structure of an electronic device to which an embodiment of the present application is applied;

[0073] Fig.11 The present invention is a schematic diagram of the hardware structure of another electronic device to which the embodiments of the present application are applied. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical solution of the present application, rather than all of the embodiments. Based on the embodiments recorded in the application documents, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the technical solution of the present application.

[0075] The following is an introduction to some concepts involved in the embodiments of the present application.

[0076] 1. Appearance image: refers to an image describing the appearance of a vehicle, which is used to analyze the appearance features of the vehicle, including but not limited to the color, shape and details of the vehicle body, and can also be called an appearance image. The appearance images in the embodiments of the present application are divided into two categories: standard appearance images and real appearance images of actual vehicles. In order to distinguish between the model training and application stages, the real appearance images are further divided into appearance images to be detected in the model application stage and sample appearance images in the model training stage. In addition, in the above two stages, the reference appearance images output by the model in the middle and the target appearance images finally used for detection are also included.

[0077] Among them, the standard appearance image refers to the vehicle image taken under standardized conditions and used as a reference image. Vehicles with the same appearance configuration (regardless of whether they are the same model) will share one or a group of standard appearance images, also called standard appearance pictures or standard photos. The "configuration" here refers to the combination of parameters that affect the judgment of the vehicle's appearance, such as body color, exterior decoration, wheel style, window film, etc. If vehicles of different models are exactly the same in these appearance configurations, they can share the same set of standard appearance images; the appearance image to be detected refers to the actual appearance image of the vehicle to be detected, which is compared with the standard appearance image to detect the vehicle appearance; the reference appearance image refers to each frame of the vehicle appearance video generated during the prediction process of the image-generated video sub-model in the parameter prediction model, which is used for comparison and reference with the standard appearance image during training or application; the target appearance image refers to the vehicle appearance image obtained by optimizing the configuration parameters of the acquisition device under the target environmental parameters, which is used for the final appearance detection. Here, "configuration" refers to a series of parameters in the acquisition device settings, which are used to adjust the performance of the acquisition device to adapt to specific acquisition conditions; the sample appearance image refers to the vehicle appearance image used in the model training process. These images are used together with the sample environmental parameters to train the parameter prediction model, thereby helping the model learn and generate high-quality prediction results.

[0078] 2. Environmental parameters: refers to the environmental conditions when the vehicle appearance image is collected, including but not limited to factors that affect the image quality, such as lighting, weather, and camera angle. The environmental parameters in the embodiments of the present application are divided into two categories: environmental parameters in the model application phase and environmental parameters in the model training phase, where the environmental parameters in the model application phase include initial environmental parameters and target environmental parameters (also known as first target environmental parameters), and the environmental parameters in the model training phase include sample environmental parameters and second target environmental parameters. In addition, in the above two stages, reference environmental parameters of the intermediate output of the model are also included.

[0079] Among them, the initial environmental parameters represent the environmental conditions when the appearance image to be detected is collected. These conditions reflect the actual collection environment at that time and are used for subsequent parameter prediction and adjustment; the reference environmental parameters refer to the environmental parameters that accompany the generation of each frame of the reference appearance image in the video frame generated by the image-generated video sub-model. The reference environmental parameters are used for comparison and reference during model training or application to ensure that the quality of the generated image meets the requirements; the target environmental parameters refer to the environmental conditions generated by the parameter prediction model during the model application process, which represent the ideal appearance image collection environment under the current collection environment, and are used to adjust the collection equipment to ensure that the collected target appearance image meets the expected detection standard; the sample environmental parameters refer to the environmental conditions used together with the sample appearance image during the model training process to help the model learn how to generate or optimize appearance images in different environments. The first target environmental parameters, like the target environmental parameters in the model application process, are generated by the parameter prediction model, which represent the ideal appearance image collection environment under the current collection environment, and are used to adjust the collection equipment to ensure that the collected target appearance image meets the expected detection standard. The second target environmental parameters refer to the reference environmental parameters when the image similarity is high by comparing the reference appearance image in the video frame generated by the image-generated video sub-model with the standard appearance image during the model training process. The reason for distinguishing between the "first target environment parameters" and the "second target environment parameters" here is to clarify the difference between the final target environment parameters (i.e., the first target environment parameters) and the intermediate target environment parameters (i.e., the second target environment parameters) generated during the parameter adjustment process in the model training phase. Standard environment parameters refer to the environmental variables recorded when taking standard appearance images.

[0080] 3. Parameter prediction model: The core model used to predict the ideal environmental conditions when acquiring vehicle appearance images. The model generates target environmental parameters by analyzing the input image to be detected and the initial environmental parameters to optimize the settings of the acquisition equipment and obtain high-quality target appearance images. The parameter prediction model can be divided into two parts: the image-generated video sub-model and the fixed parameter sub-model. The image-generated video sub-model is responsible for generating appearance video data containing the vehicle appearance, while the fixed parameter sub-model compares the reference appearance image in the generated video data with the standard appearance image to accurately obtain the target environmental parameters.

[0081] 4. Acquisition equipment parameters: refers to the various equipment settings that determine the image quality and appearance detail capture during the vehicle appearance image acquisition process. Among them, the acquisition equipment is the equipment used to acquire vehicle appearance images, which can be a camera, a high-definition camera, a dedicated image acquisition device, or an image sensor integrated in the intelligent transportation system. These devices are responsible for capturing clear vehicle images and providing basic data for subsequent image analysis and processing.

[0082] 5. Pre-inspection: refers to a preliminary inspection conducted by the vehicle manufacturer before the vehicle leaves the factory, which aims to ensure that the vehicle's appearance meets relevant standards and rule out possible illegal modifications.

[0083] 6. Parameter update record table: This record table records in detail the environmental parameters of each test process at different dates and times. By querying the parameter update record table, you can obtain past parameter settings and use them directly in similar environments, thereby improving detection efficiency and reducing unnecessary repeated calculations.

[0084] The following is a brief description of the design concept of this application:

[0085] With the development of social economy, the popularity of automobiles is increasing, and the total number of vehicles is also increasing. In order to eliminate potential safety hazards of vehicles and ensure traffic safety, each new vehicle needs to be inspected by the motor vehicle manufacturer before leaving the factory. In addition, vehicles that have obtained official license plates and driving licenses must undergo annual vehicle inspections to urge and strengthen vehicle maintenance and reduce the occurrence of traffic accidents.

[0086] Vehicle appearance inspection requires that the photographed appearance should be consistent with the standard photo and the actual color of the vehicle body. However, when actually collecting vehicle appearance photos, due to the influence of many factors, the quality of the collected appearance photos is often poor.

[0087] For example, the vehicle appearance inspection requires that the photographed appearance should be consistent with the standard photo and the actual color of the vehicle body. Before leaving the factory, since all vehicles must take appearance photos for inspection, they are usually collected on the production line or at the end of the line. However, due to the influence of factors such as the surrounding lighting, light, reflection of the production line chain, and curtain color, it is difficult to ensure that the body brightness in the collected appearance photos is uniform, without obvious reflections, and in natural and real tones. It is even affected by the surrounding ambient light, which often leads to the collected appearance photos being unable to distinguish some structures in the vehicle appearance. At the same time, after the shooting plans of different factories are adjusted, it is difficult to be universally applied in other factories.

[0088] In addition, during the annual vehicle inspection, it is also necessary to inspect the body color of the vehicle to be inspected, and then determine whether the vehicle to be inspected has been illegally modified. If the photos collected at the pre-inspection collection node cannot identify the vehicle's appearance due to environmental factors, it will greatly hinder the circulation of vehicles.

[0089] The embodiments of the present application provide a method, device and equipment for detecting the appearance of a vehicle. Since the present application first obtains the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected, by setting the initial environmental parameters, it is ensured that when the appearance image to be detected is collected, the environmental conditions at the time can be recorded, so that the subsequent image processing and analysis can be adjusted based on the specific environmental conditions. Secondly, the image to be detected and the initial environmental parameters are input into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model. Since a parameter prediction model based on deep learning is adopted and the image to be detected and the initial environmental parameters are input, the model can effectively predict the target environmental parameters required to collect high-quality appearance images in the current environment. By predicting the target environmental parameters, the system can automatically adjust the device settings so that the collected image is closer to the effect under ideal conditions, reducing the impact of environmental changes on image quality.

[0090] Then, the initial acquisition equipment parameters are adjusted according to the target environment parameters to obtain the target acquisition equipment parameters for the vehicle to be detected. Since the mechanism of automatically adjusting the equipment parameters is set, the system can optimize the various indicators of the acquisition equipment according to the predicted target environment parameters. This dynamic adjustment can keep the acquisition equipment in the best state under different environmental conditions, automatically optimize the acquisition equipment parameters, and ensure that the quality of the collected image meets the standard, thereby further improving the accuracy of the detection. Finally, the target appearance image of the vehicle to be detected is collected according to the target acquisition equipment parameters, and the vehicle appearance detection is performed based on the target appearance image to obtain the corresponding detection results. Since the vehicle appearance detection is based on the high-quality target appearance image, the result is closer to the actual state of the vehicle, reducing the interference of external factors on the detection results, improving the applicability of the appearance image of the vehicle under different environmental conditions, and reducing the influence of the surrounding environment on the photo effect, ensuring that the detection results can accurately reflect the consistency of the vehicle appearance with the standard photo and the physical body color.

[0091] The embodiments of the present application also provide a training method, device and equipment for a parameter prediction model. Since a training sample set is first obtained, each training sample in the training sample set contains a sample image and sample environment parameters corresponding to the sample vehicle. Such a design enables the model to learn the diversity of vehicle appearance under different environmental conditions, improves the generalization ability of the model, and ensures that the model can perform well in different practical application scenarios.

[0092] Secondly, for each training sample, the corresponding sample appearance image and sample environmental parameters are input into the image-generated video sub-model in the parameter prediction model to be trained, and the vehicle appearance video corresponding to the corresponding sample vehicle is obtained. By using deep learning technology, a realistic vehicle appearance video is generated, so that the model can simulate the changes in vehicle appearance under different environmental conditions, providing high-quality data input for subsequent fixed-parameter sub-model processing. Subsequently, the vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained. Based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image. By comparing the image similarity, the model can identify the difference between the video frame and the standard image, thereby judging whether the current video generation is accurate. Based on these differences, the system can further adjust the parameters of the image-generated video sub-model and gradually optimize the generation capability of the model.

[0093] Finally, the parameters of the image-generated video sub-model are adjusted based on the comparison results, so that the model can be continuously optimized to ensure that the generated video is as close to the standard appearance image as possible. Through this cyclic adjustment mechanism, the prediction ability of the model has been significantly improved, so that the trained parameter prediction model can more accurately predict the target environment parameters.

[0094] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0095] like Figure 1 As shown, it is a schematic diagram of an application scenario of an embodiment of the present application. The application scenario diagram includes a terminal device 110 and a server 120.

[0096] In the embodiment of the present application, the terminal device 110 includes but is not limited to mobile phones, tablet computers, laptop computers, desktop computers, e-book readers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals and other devices; the terminal device may be installed with a client related to vehicle appearance detection, which may be software (such as a browser, vehicle appearance detection software, etc.), or a web page, a small program, etc. The server 120 is a background server corresponding to the software or web page, small program, etc., or a server specifically used for vehicle appearance detection, which is not specifically limited in the present application. The server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0097] It should be noted that the vehicle appearance detection method and the parameter prediction model training method in each embodiment of the present application can be executed by an electronic device, which may be a terminal device 110 or a server 120, that is, the method can be executed by the terminal device 110 or the server 120 alone, or can be executed jointly by the terminal device 110 and the server 120.

[0098] For example, when the terminal device 110 and the server 120 jointly execute, the server 120 may pre-deploy a trained parameter prediction model, and the user uploads the image to be detected and the initial environmental parameters of the vehicle to be detected on the terminal device 110, and sends them to the server 120, which analyzes them to obtain more accurate target environmental parameters, and determines the appropriate target acquisition device parameters based on them, and returns them to the terminal device 110, and the terminal device 110 automatically adjusts the acquisition device according to the received target acquisition device parameters, re-acquires and generates a more accurate vehicle appearance image. The specific processing process is as follows:

[0099] Acquire the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected; input the image to be detected and the initial environmental parameters into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model; adjust the initial acquisition device parameters according to the target environmental parameters to obtain the target acquisition device parameters for the vehicle to be detected; acquire the target appearance image of the vehicle to be detected according to the target acquisition device parameters, and perform vehicle appearance detection based on the target appearance image to obtain a more accurate vehicle appearance detection result.

[0100] In addition, the trained parameter prediction model deployed by the server 120 can be trained in the following manner:

[0101] A training sample set is obtained, wherein each training sample in the training sample set includes: a sample image and sample environmental parameters corresponding to the sample vehicle; for each training sample, the corresponding sample appearance image and sample environmental parameters are input into a graph-generated video sub-model in a parameter prediction model to be trained, and a vehicle appearance video corresponding to the corresponding sample vehicle is obtained; the vehicle appearance video and the corresponding standard appearance image are input into a fixed-parameter sub-model in the parameter prediction model to be trained, and based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image, and parameters of the graph-generated video sub-model are adjusted according to the comparison result.

[0102] In an optional implementation, the terminal device 110 and the server 120 may communicate with each other via a communication network.

[0103] In an optional implementation, the communication network is a wired network or a wireless network.

[0104] It should be noted that Figure 1 The figure is only an example. In fact, the number of terminal devices and servers is not limited and is not specifically limited in the embodiments of the present application.

[0105] In an embodiment of the present application, when there are multiple servers, the multiple servers can form a blockchain, and the servers are nodes on the blockchain; such as the vehicle appearance detection method disclosed in the embodiment of the present application, the various appearance images, environmental parameters, detection results and other data of the vehicle involved can be saved on the blockchain.

[0106] In addition, the embodiments of the present application can be applied to various scenarios, including not only traditional vehicle detection scenarios, but also including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving and other scenarios.

[0107] For example, in smart transportation, the vehicle appearance detection method can be used to monitor traffic flow in real time and identify specific vehicles based on appearance features; in assisted driving or autonomous driving scenarios, the method can be used to detect changes in vehicle appearance to predict possible risk situations, etc. By being applied to these different scenarios, the method can provide strong support for intelligent and automated traffic management systems.

[0108] The following describes the parameter prediction model training method and vehicle appearance detection method provided by the exemplary embodiment of the present application in combination with the application scenario described above and with reference to the accompanying drawings. It should be noted that the above application scenario is only shown to facilitate understanding of the spirit and principles of the present application, and the implementation of the present application is not limited in this respect.

[0109] The solution provided by the embodiment of the present application mainly involves the application of artificial intelligence in vehicle appearance detection. When performing vehicle appearance detection, the present embodiment adopts an artificial neural network model based on deep learning, that is, the embodiment of the present application combines machine learning and methods related to the vehicle appearance detection architecture to obtain a parameter prediction model. The image-generated video sub-model in the parameter prediction model is used to generate a vehicle appearance video, and combined with the fixed parameter sub-model, the reference appearance image in the vehicle appearance video is compared and analyzed with the standard image, so as to obtain the desired target environment parameters, and then in the actual scene, the desired target environment parameters can be predicted by the parameter prediction model, and the acquisition equipment parameters can be adjusted based on this parameter and a more accurate vehicle appearance image can be re-collected to achieve accurate detection of the vehicle appearance.

[0110] Specifically, the vehicle appearance detection process in the embodiment of the present application can be divided into two parts, including a model training part and a model application part. In the model training part, the model obtains sample appearance images and sample environmental parameters from the training data set, and inputs them into the image-generated video sub-model to generate a vehicle appearance video. In the generated video, each frame of the reference appearance image and the corresponding reference environmental parameters will be used for comparative analysis. The parameter-fixed sub-model compares these generated video frames with the standard appearance image and calculates the image similarity. If the image similarity is lower than the preset threshold, the model will analyze the difference between the reference environmental parameters and the sample environmental parameters, and adjust the internal parameters of the model according to these differences. By continuously iterating this process, the model will gradually optimize its prediction ability and can better adapt to the image generation requirements under different environmental conditions. In the model application part, the trained parameter prediction model receives the appearance image and initial environmental parameters of the vehicle to be detected, generates the corresponding appearance video data using the image-generated video sub-model, and analyzes the generated appearance video data in combination with the parameter-fixed sub-model to obtain the desired target environmental parameters. Then, these target environmental parameters are used to adjust the various indicators of the acquisition equipment to collect vehicle appearance images that meet the desired conditions, thereby improving the accuracy and reliability of the detection.

[0111] In addition, it should be noted that the artificial neural network model in the embodiment of the present application can be trained online or offline, and can be adjusted again according to user feedback after the training is completed, which is not specifically limited here. In this article, offline training is used as an example to illustrate that through offline training, the model can better perform feature learning and parameter optimization when processing a large amount of historical data, ensuring a higher detection accuracy in the application.

[0112] Below, the model training part in the embodiment of the present application is first described:

[0113] See also Figure 2 As shown, it is an implementation flow chart of a model training method provided in an embodiment of the present application. Taking the server as the execution subject as an example, the specific implementation process of the method is as follows S21~S22:

[0114] S21: Obtain a training sample set, each training sample in the training sample set includes: a sample image and a sample environmental parameter corresponding to a sample vehicle; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: the environmental conditions when the sample appearance image is collected.

[0115] Among them, sample images refer to image data used for training, standard appearance images refer to vehicle images taken under standardized conditions and used as benchmark images; sample appearance images refer to vehicle appearance images used in the model training process. These images are used together with sample environmental parameters to train parameter prediction models, thereby helping the model learn and generate high-quality prediction results.

[0116] Environmental parameters refer to the environmental conditions that describe the vehicle appearance images when they are collected. Sample environmental parameters refer to the environmental conditions that are used along with sample appearance images during model training to help the model learn how to generate or optimize appearance images in different environments.

[0117] A standard appearance image refers to a vehicle image taken under standardized conditions and used as a reference image. Vehicles with the same appearance configuration (regardless of whether they are the same model) will share one or a group of standard appearance images, also called standard appearance drawings or standard photos.

[0118] In the embodiment of the present application, the standard appearance image may be a standard photo during the vehicle inspection process, including but not limited to the overall appearance photo of the vehicle, standard photos of each part of the vehicle body, standard photos from a specific perspective, and standard photos of iconic parts. The overall appearance photo of the vehicle shows the overall appearance of the vehicle from a standard perspective, usually covering panoramic images of the front, side, and rear of the vehicle to provide a benchmark for the overall appearance of the vehicle; standard photos of each part of the vehicle body include standard images of the front, rear, doors, wheels, windows and other parts of the vehicle to ensure detailed inspection and comparison of the appearance of each part; standard photos from a specific perspective are vehicle images taken at specific standard angles, such as a 45-degree front angle, a 45-degree side angle, etc., to capture the appearance characteristics of the vehicle from different perspectives; standard photos of iconic parts include detailed images of key parts such as vehicle logos, lights, license plates, etc., which are used to compare the status and details of the iconic parts of the vehicle; and so on.

[0119] The standard appearance images listed above are just simple examples, and this article does not make any specific restrictions on them. It is sufficient to ensure that the standard appearance images, the appearance images to be detected, and the sample appearance images of the model training part and the model application part are unified.

[0120] For example, in the model training part of this article, the standard appearance image of the input model can be a standard photo group consisting of four images: a 45-degree photo of the left front of the vehicle, a 45-degree photo of the right rear of the vehicle, a photo directly above the front of the vehicle, and a photo directly above the rear of the vehicle. Correspondingly, the sample appearance image can be a sample appearance photo group consisting of a 45-degree photo of the left front of the real vehicle and a 45-degree photo of the right rear. Correspondingly, in the subsequent model application part, the appearance image to be tested of the input model can also be a group of appearance photos to be tested consisting of a 45-degree photo of the left front of the real vehicle and a 45-degree photo of the right rear.

[0121] In an optional embodiment, the 45-degree photo of the left front of the vehicle and the 45-degree photo of the right rear of the vehicle in the standard photo group should meet the following requirements: First, the photo is taken at a slight downward "side tilt" at 45 degrees from the left front of the vehicle and 45 degrees from the right rear of the vehicle, and the side tilt angle is between 15-30 degrees. Second, the height of the camera from the ground is between 1.5-2.5 meters. Third, the vehicle screen content occupies more than 2 / 3 of the entire image screen content, and the distance between the body contour (including accessories) and the upper and lower edges of the picture is greater than 150 pixels, and the distance from the left and right edges of the picture is greater than 100 pixels. Fourth, the horizontal pixel of the photo is not less than 1280, and the vertical pixel is not less than 960. Fifth, the photo is a three-channel red, green and blue (RGB) color picture stored in JPG format, and the size does not exceed 300KB. Sixth, the photo background should be simple, without text, icons or other logos with advertising nature, and the background color and the body color should be clearly distinguishable. Seventh, the photos should clearly identify the basic features such as the vehicle logo, grille, headlights, wheels, exhaust pipe outlet (if any), and no protective film should be present. Eighth, when the external light conditions cannot meet the requirements, soft auxiliary lighting equipment should be used according to the on-site environmental conditions, and the vehicle body should have uniform brightness, no obvious reflections, and a natural and real color tone.

[0122] In an optional embodiment, the photos taken from the front and rear of the vehicle in the standard photo group should meet the following requirements: First, the photos are taken from the front and rear of the vehicle, with a pitch angle between 30-45 degrees and a horizontal rotation angle of less than 10 degrees. Second, the horizontal distance monitored by the front and rear cameras is not less than the installation height. times. Third, the vehicle content occupies more than 2 / 3 of the entire image content. There can only be one vehicle in a single picture, and the photo should include the complete front or rear view of the vehicle. Fourth, the horizontal pixel and vertical pixel ratio of the photo should be 4:3 and the number of photo pixels should not be less than 2 million. Fifth, the photo is a three-channel RGB color picture stored in JPG format, with a size not exceeding 300KB. Sixth, the road background should be a single light color. Seventh, the vehicle brand logo, body color, model, and headlights should be identifiable, and the grille should be clearly distinguishable in the front photo. Eighth, the light source is located directly above and should not cause obvious reflections or other interference to the body.

[0123] In an optional embodiment, the 45-degree left front photo and the 45-degree right rear photo of the vehicle in the sample appearance photo group should meet the following requirements: First, the photo is taken at a slight downward "sideways" angle of 45 degrees from the left front of the vehicle and 45 degrees from the right rear of the vehicle. Second, the 45-degree left front photo can clearly show the front appearance of the vehicle and the condition of the tires, and the 45-degree right rear photo can clearly show the rear appearance and the condition of the tires. Third, the photo is a three-channel RGB color picture stored in JPG format. Fourth, the background of the photo should be simple and should not contain text, icons or other logos with advertising nature. Fifth, the photo should be able to clearly identify basic features such as the car logo, grille, headlights, wheels, body color, model, exhaust pipe port (if any), etc. Sixth, when the external light conditions cannot meet the requirements, soft auxiliary lighting equipment should be used according to the on-site environmental conditions, and the body brightness should be uniform, without obvious reflections, and in a natural and real color tone.

[0124] Correspondingly, the appearance photo group to be inspected is the same as the current appearance photo group, and will not be repeated here.

[0125] In an embodiment of the present application, environmental parameters may be various conditions when capturing images of the vehicle's exterior, including but not limited to: shooting time, ambient lighting direction, ambient lighting intensity, marking parameters of surrounding objects, color of surrounding objects, curtain color, etc.

[0126] For example, the sample environmental parameters in this application and the initial environmental parameters of the model application part can be collected by the camera algorithm service in combination with the camera hardware device. These environmental parameters record the actual environmental conditions during the shooting process to ensure that the captured appearance images can accurately reflect the actual appearance of the vehicle and can serve as an important reference in subsequent processing and analysis.

[0127] After obtaining the training samples, the model can be trained based on the training samples. The following will describe in detail:

[0128] S22: For each training sample, perform the following operations S221~S222 respectively:

[0129] S221: Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle;

[0130] Among them, the input of the image-generated video sub-model is the sample appearance image and sample environmental parameters, and the output is the vehicle appearance video corresponding to the corresponding sample vehicle.

[0131] In the embodiment of the present application, the image-to-video sub-model can be an image-to-video generator XL (I2VGen-XL), which is based on deep learning image generation technology and generates simulated vehicle appearance videos by combining sample appearance images and sample environmental parameters. The I2VGen-XL model can efficiently process input data and generate high-quality video output, helping to achieve accurate simulation and prediction of vehicle appearance.

[0132] In addition, the image-generated video sub-model can also be other generative models, such as the Generative Adversarial Network (GAN), which uses adversarial training to generate high-quality video sequences, or the Variational Autoencoder (VAE), which generates videos with similar appearance to the input samples by learning the latent space of sample data.

[0133] The above is just a simple example of the specific structure of the image-generated video sub-model, and this article does not make any specific limitations.

[0134] After introducing the image-generated video sub-model, the fixed-parameter sub-model will be introduced below.

[0135] S222: Input the vehicle appearance video and the corresponding standard appearance image into the fixed-parameter sub-model in the parameter prediction model to be trained, compare each frame of the reference appearance image in the vehicle appearance video with the corresponding standard appearance image based on the fixed-parameter sub-model, and adjust the parameters of the image-generated video sub-model according to the comparison results.

[0136] Among them, the trained parameter prediction model is used to predict the target environmental parameters for the vehicle to be detected (also the present vehicle). The target environmental parameters (i.e., the first target environmental parameters mentioned above) represent the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the target environmental parameters and perform vehicle appearance detection.

[0137] In the parameter prediction model processing of the above process, two sub-models are mainly involved: the image-generated video sub-model and the fixed parameter sub-model.

[0138] like Figure 3 As shown, it is a processing logic diagram of a parameter prediction model provided by an embodiment of the present application. Among them, the input of the parameter determination sub-model is each frame of the reference appearance image and the corresponding standard appearance image in the vehicle appearance video, and the output is the second target environment parameter. The vehicle appearance video is generated by the image generation video sub-model based on the input sample appearance image and sample environment parameters.

[0139] Specifically, the parameter-determining sub-model compares each frame of the reference appearance image with the corresponding standard appearance image to calculate the image similarity, and adjusts the parameters of the image-generated video sub-model based on the comparison results. Ultimately, through these adjustments, the parameter-determining sub-model can help determine the target environment parameters suitable for the sample vehicle, that is, the second target environment parameters.

[0140] In the above process, the parameter prediction model in the embodiment of the present application achieves efficient optimization of vehicle appearance detection through the cooperation of the image-generated video sub-model and the fixed parameter sub-model. The image-generated video sub-model can combine the sample appearance image and environmental parameters to generate high-quality vehicle appearance videos, thereby improving the accuracy of appearance simulation. The fixed parameter sub-model compares the video frame with the standard appearance image, calculates the image similarity, and extracts accurate target environmental parameters. In this way, the vehicle appearance detection method in the embodiment of the present application can more accurately predict the vehicle appearance and adapt to different environmental conditions, thereby achieving more efficient and accurate vehicle appearance detection.

[0141] After introducing the parameter prediction model, the processing of the two sub-models of the parameter prediction model will be specifically explained below.

[0142] In the embodiment of the present application, an optional implementation of S222 is as follows: Figure 4 As shown, it is a flow chart of the implementation of the parameter adjustment process of a parameter prediction model provided in an embodiment of the present application, including S40~S45:

[0143] S40: Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle.

[0144] S41: Input the vehicle appearance video and the corresponding standard appearance image into the fixed parameter sub-model in the parameter prediction model to be trained, and split and decode each frame of the vehicle appearance video and the corresponding reference environment parameters based on the fixed parameter sub-model.

[0145] Specifically, in the model training part, the vehicle appearance video output by the image-generated video sub-model refers to the high-quality appearance image video of the sample vehicle in the corresponding training sample. Similarly, in the model application part, the vehicle appearance video output by the image-generated video sub-model refers to the high-quality appearance image video of the corresponding vehicle to be detected. Specifically, the vehicle appearance video in this application may include the appearance image of the vehicle from one or more perspectives.

[0146] In an embodiment of the present application, after receiving the vehicle appearance video, the fixed parameter sub-model first divides the vehicle appearance video into frames, analyzes each video frame (that is, the reference appearance image in this article) frame by frame, and extracts the image content of each frame. Then, through the decoding module in the fixed parameter sub-model, these image contents are decoded, and the environmental parameters of each frame of the reference appearance image are extracted and recorded as reference environmental parameters. In addition, the decoding module can associate the video frame information with its corresponding environmental parameters, thereby outputting each frame of the reference appearance image and its corresponding reference environmental parameters.

[0147] S42: Compare each frame of the reference appearance image with the corresponding standard appearance image to obtain the image similarity corresponding to each frame of the reference appearance image.

[0148] Among them, image similarity refers to the degree of match between the reference appearance image and the standard appearance image in terms of visual features, which is used to evaluate the similarity between the two. Specifically, image similarity is measured by comparing the feature information (such as color, shape, texture, etc.) of the two images. Its value is usually expressed as a percentage or a specific scoring mechanism, indicating the degree of similarity between the reference appearance image and the standard appearance image.

[0149] When the standard appearance image is a standard photo group including multiple standard photos at different viewing angles, then in S42, each frame of the reference appearance image (also referred to as the reference appearance photo) needs to be compared with the standard photo at the corresponding viewing angle. At this time, the vehicle appearance video generated by the image-generated video sub-model can be understood as a reference appearance photo group.

[0150] In an embodiment of the present application, image similarity may be determined based on structural similarity index (SSIM), for example, by comprehensively evaluating features such as brightness, contrast, and structure of the image to determine the similarity between the reference appearance image and the standard appearance image.

[0151] In another embodiment of the present application, the image similarity may also be determined based on the mean square error (MSE), for example, by calculating the average square difference between the image pixel values ​​to quantify the degree of difference between the reference appearance image and the standard appearance image.

[0152] It should be noted that the two methods of calculating image similarity listed above are just simple examples, and this application does not make any specific limitations on the specific image similarity algorithm.

[0153] In the embodiment of the present application, the degree of image similarity directly affects the adjustment and optimization process of the model. When the image similarity is higher than the preset threshold, it means that the reference appearance image and the standard appearance image have a high degree of match, and it can be determined that the reference appearance image is consistent with the standard appearance image. At this time, it can be considered that the prediction of the model is relatively accurate. On the contrary, if the image similarity is not higher than the threshold, it is necessary to further adjust the parameters of the model to improve the accuracy of the prediction.

[0154] The following will introduce the situations where the image similarity is not higher than the preset threshold and the image similarity is higher than the threshold respectively:

[0155] S43: Determine whether the obtained similarities of each image are not higher than a preset threshold, if so, execute S44, otherwise execute S45.

[0156] like Figure 5 As shown, it is a schematic diagram of image similarity comparison between a reference appearance photo group and a standard photo group provided in an embodiment of the present application.

[0157] Among them, each reference appearance photo in the reference appearance photo group is compared with the standard photo of the corresponding perspective in the standard photo group.

[0158] For example, if the reference appearance photo group contains 20 reference appearance photos, then the first reference appearance photo in the reference appearance photo group can be compared with the corresponding standard photo in the standard photo group to obtain image similarity S1. Next, the second reference appearance photo in the reference appearance photo group is compared with the corresponding standard photo in the standard photo group to obtain image similarity S2, and so on, until all reference appearance photos are compared, and image similarities are obtained respectively: S1, S2, S3, S4, S5, S6, ..., S20.

[0159] In an optional embodiment, the preset threshold is 95%. Assuming S1=92%, S2=90%, S3=93%, S4=91%, S5=94%, S6=89%, ..., S20=93%, the similarity of each image is not higher than the preset threshold, so step S44 is executed.

[0160] In an optional embodiment, the preset threshold is 95%. Assume that S1=92%, S2=97%, S3=93%, S4=98%, S5=94%, S6=96%, ..., S20=92%. Because there are multiple similarities (S2, S4, S6) higher than 95, there is an image similarity higher than the preset threshold among the image similarities, so step S45 is executed.

[0161] Next, steps S44 and S45 will be described in detail.

[0162] S44: The reference appearance image with the highest image similarity and the corresponding reference environment parameters are used as new sample appearance images and sample environment parameters corresponding to the sample vehicle, and the process returns to S40 until the image similarity corresponding to each frame of the reference appearance image in the latest vehicle appearance video has an image similarity higher than a preset threshold.

[0163] Specifically, if the similarities of each image obtained are not higher than the preset threshold, it means that the vehicle appearance video generated by the image generation sub-model fails to meet the expected similarity requirements under the current conditions. In this case, the image generation video sub-model needs to be further trained.

[0164] Specifically, when the similarities of each image obtained are no higher than a preset threshold, the model uses the reference appearance image and reference environmental parameters with the highest image similarity as new sample appearance images and sample environmental parameters, and inputs them again into the image-generated video sub-model in the parameter prediction model to be trained to generate a new vehicle appearance video.

[0165] Then, based on the differences between the reference environment parameters and the corresponding sample environment parameters, the model will make a series of parameter adjustments to the image-generated video sub-model.

[0166] Specifically, the model analyzes the differences between the reference environmental parameters and the sample environmental parameters, and based on these differences, readjusts the internal parameters of the image-generated video sub-model to better adapt to the current environmental conditions and ensure that the generated vehicle appearance video can more realistically reflect the actual situation. This adjustment process is iterative, and after each adjustment, the model regenerates the vehicle appearance video and calculates the image similarity. If the image similarity of all frames in the generated video still does not reach the preset threshold, the model will continue to adjust the parameters. This process will continue until the image similarity reaches or exceeds the preset threshold, or the number of adjustments reaches the preset upper limit (such as 10 times). Through this repeated adjustment, the performance of the image-generated video sub-model will be gradually optimized to ensure its stability and accuracy under different environmental conditions.

[0167] Through this repeated adjustment and iterative optimization, the image-generated video sub-model can gradually improve its ability to generate high-similarity videos under different environmental conditions. This method effectively improves the adaptability of the model in complex or changing environments, ensuring that the generated videos more realistically reflect the actual situation. Ultimately, the system can not only more accurately generate standard vehicle appearance videos, but also significantly enhance the ability to resist interference from changes in external conditions, thereby improving detection accuracy and overall results. This continuous optimization process ensures the stability and accuracy of the model, making detection results more reliable in various environments.

[0168] In the embodiment of the present application, only the parameters of the image-generated video sub-model need to be adjusted. Specifically, when adjusting, all the parameters of the sub-model can be adjusted. Of course, in order to improve the model training efficiency, only some parameters of the sub-model can be adjusted. An optional adjustment method is as follows:

[0169] The parameters of the low-rank matrix corresponding to the image-generated video sub-model are adjusted, and the low-rank matrix is ​​obtained by performing low-rank decomposition on the weight matrix to be adjusted in the image-generated video sub-model.

[0170] The above method means adjusting the parameters of the model through the low-rank adaptive method (LoRA).

[0171] Taking the image-generated video sub-model as I2VGen-XL as an example, the process of adjusting the parameters of the model through the low-rank adaptive method (LoRA) is as follows:

[0172] (1) Weight decomposition: First, decompose the weight matrix of the I2VGen-XL model, find the weight part for parameter adjustment, and perform low-rank decomposition on it, usually by decomposing it into the product of two low-rank matrices.

[0173] Among them, the weights that need to be adjusted are those weight matrices that are closely related to environmental parameters. These weight matrices directly affect how the model handles environmental factors such as light intensity, light direction, and background color when generating videos. By performing low-rank decomposition and fine-tuning training on these weight matrices, the model can better adapt to different environmental conditions, thereby generating high-quality videos that meet expectations and improving image similarity.

[0174] (2) Fine-tuning training: Next, most of the weights of the graph-generated video sub-model I2VGen-XL are frozen, and only the decomposed low-rank matrix is ​​trained.

[0175] During training, the sample appearance image and sample environmental parameters are input into the image-generated video sub-model I2VGen-XL in the parameter prediction model to be trained, the vehicle appearance video output by I2VGen-XL is obtained, and the reference environmental parameters are determined in the above manner. Then, the parameters of the low-rank matrix can be adjusted according to the difference between the sample environmental parameters and the corresponding reference environmental parameters to complete the fine-tuning training.

[0176] In an optional embodiment, the fine-tuning process mainly relies on hyperparameter optimization tools, such as Bayesian optimization and genetic algorithms, to quickly find the best parameter combination in a larger parameter space. Adaptive optimizers are also used to dynamically adjust the learning rate during training so that the model can better adapt to changing gradients. In addition, gradient clipping is used to control the range of gradients during backpropagation to further optimize the stability and accuracy of the model.

[0177] (3) Model reconstruction: After fine-tuning, the decomposed low-rank matrix is ​​reassembled into a weight matrix and integrated back into the I2VGen-XL model, resulting in a new I2VGen-XL model that can generate high-quality image videos containing multiple sets of environmental parameter combinations. These environmental parameter combinations can correspond to the environmental conditions of each frame of the video.

[0178] In the above process, the image-generated video model I2VGen-XL is fine-tuned and trained through LoRA. After fine-tuning and training, the I2VGen-XL model has the ability to autonomously learn based on sample appearance images and sample environmental parameters, and can output high-quality image videos containing multiple sets of environmental parameter combinations. This process improves the accuracy and environmental adaptability of the model-generated image videos, thereby improving the stability and detection accuracy of the entire system under different environmental conditions.

[0179] The above describes the case where the image similarity is lower than the preset threshold, the model adjusts the parameters and retrains. The following describes the case where the image similarity is higher than the preset threshold:

[0180] S45: Using the reference environment parameters corresponding to the reference appearance image with the highest image similarity as the second target environment parameters corresponding to the sample vehicle, the second target environment parameters representing the desired environment conditions when the appearance image of the sample vehicle is currently collected.

[0181] Specifically, if the similarities of the images obtained are higher than the preset threshold, it means that the vehicle appearance generated by the image generation sub-model has reached the expected similarity requirement under the current conditions. In this case, new sample vehicle images and sample environmental parameters can be selected to continue model training to ensure that the model can still maintain high accuracy and stability when processing different samples.

[0182] Specifically, we will continue to select new training samples and repeat Figure 4 The training is performed according to the process shown in the figure. This process will continue until enough training samples are selected, or all samples meet the preset image similarity requirements, or the model converges, etc., and the model training is completed.

[0183] In an embodiment of the present application, the training sample set may include training samples corresponding to different sample vehicles. In this way, the model can gradually optimize its adaptability to different vehicle samples, ensuring that high-quality vehicle appearance images and videos can still be generated under diverse environmental conditions.

[0184] In the above process, by continuously adjusting the sample appearance images and sample environmental parameters and generating new vehicle appearance videos, the image similarity can be gradually optimized to ensure that the similarity between the generated vehicle appearance videos and the standard appearance images continues to increase. This helps to achieve more accurate appearance simulation and improve the prediction accuracy of the model. The model after parameter adjustment has the ability to autonomously learn based on appearance photos and environmental parameters, and can output high-quality image videos containing multiple sets of environmental parameter combinations.

[0185] In addition, during the training process, the validation set can be used to evaluate the performance of the model.

[0186] For example, in practical applications, a training sample set and a verification set can be formed by collecting environmental parameters and photo data of historical project standard photos and exterior photos. The photo data includes multiple sets of four standard photos (front left, rear right, front straight, rear straight) and two exterior photos (front left, rear right).

[0187] In the embodiment of the present application, the validation set is used to track the performance of the model during training to ensure that the model can maintain good predictive ability on unseen data. The calculation logic of the model during verification is the same as the above model training part, which will not be repeated here. Through the feedback of the validation set, the model parameters can be further fine-tuned to make the model more robust. After the training is completed, the model verified by the validation set will be more reliable and can be effectively applied to the actual vehicle appearance detection task.

[0188] After the above parameter prediction model is trained, the trained model can be used to detect the actual vehicle appearance, which is the specific model application part.

[0189] The following is a detailed description of the vehicle appearance detection method in the model application part:

[0190] See also Figure 6 As shown, it is an implementation flow chart of a vehicle appearance detection method provided in an embodiment of the present application. Taking the server as the execution subject as an example, the specific implementation process of the method is as follows S61~S64:

[0191] S61: Acquire the image to be detected and initial environment parameters corresponding to the vehicle to be detected.

[0192] The images to be detected include: a standard appearance image of the vehicle to be detected and an appearance image to be detected, and the initial environmental parameters represent: the environmental conditions when the appearance image to be detected is collected.

[0193] Specifically, the standard appearance image is the same as the standard appearance image in the above-mentioned model training part, the appearance image to be detected is the same as the sample appearance image in the above-mentioned model training part, and the initial environment parameters are the same as the sample environment parameters in the above-mentioned model training part. The different names are to distinguish between training and application. The actual meanings are the same, but the scenes are different. The specific standard appearance images, appearance images to be detected, and the initial environment parameters are not repeated here.

[0194] S62: Input the image to be detected and the initial environmental parameters into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model.

[0195] The target environment parameters represent the expected environment conditions when the appearance image of the vehicle to be detected is currently collected, and the parameter prediction model is trained based on the sample images and sample environment parameters corresponding to the sample vehicles.

[0196] Specifically, the target environment parameters are the same as the first target environment parameters mentioned above. The different names are to distinguish between training and application. The actual meanings are the same, but the scenarios are different. The specific target environment parameters and parameter prediction models will not be repeated here.

[0197] Optionally, the parameter prediction model includes two parts: an image-generated video sub-model and a fixed parameter sub-model. Accordingly, an optional implementation of S62 is as follows, including S621~S623 ( Figure 6 not shown):

[0198] S621: Input the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and obtain the vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model.

[0199] In the embodiment of the present application, the relevant model of the image-generated video sub-model can be found in the model training part, which will not be repeated here.

[0200] Optionally, S621 may be implemented according to the following process, including S6211 to S6213 ( Figure 6 not shown):

[0201] S6211: Input the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and extract the feature vectors corresponding to the appearance image to be detected and the initial environmental parameters respectively through the image-generated video sub-model.

[0202] In this process, the image-generated video sub-model processes the appearance image to be detected and the initial environmental parameters respectively, and extracts their corresponding feature vectors. The feature vector is a digital expression of the image and environmental parameters, which can reflect the key features of the image and the specific conditions of the environment.

[0203] S6212: Fuse the extracted feature vectors to obtain a fused vector.

[0204] The fusion process is to integrate the feature information of the image and environmental parameters to generate a fusion vector. The fusion vector is a comprehensive expression of the input image and environmental conditions, which can better reflect the correlation between the two.

[0205] In an embodiment of the present application, the extracted feature vectors can be fused by vector summation, such as extracting a 128-dimensional image feature vector and a 128-dimensional environmental parameter feature vector respectively, adding the elements at corresponding positions in the two vectors to obtain a fused vector.

[0206] Of course, in addition to how to sum vectors, the fusion process can also be achieved through other methods, such as vector splicing, weighted averaging, etc., which will not be elaborated in this article.

[0207] S6213: Generate a vehicle appearance video corresponding to the vehicle to be detected based on the fusion.

[0208] The vehicle appearance video is generated based on the information in the fusion vector, which can more realistically reflect the appearance characteristics of the vehicle to be detected under specific environmental conditions. Through the generated vehicle appearance video, the vehicle appearance can be further detected and analyzed to ensure that it meets the expected standards.

[0209] In addition, the length of the vehicle appearance video is not specifically limited in this article.

[0210] Based on the above implementation, the image-generated video sub-model can generate a high-quality simulated vehicle appearance video, which can truly present the appearance characteristics of the vehicle to be tested in some specific environments and accurately reflect the impact of environmental conditions on the vehicle appearance. This simulated video helps to conduct a comprehensive inspection and analysis of the vehicle appearance, thereby ensuring that the vehicle appearance meets the standard requirements.

[0211] After introducing how to obtain the vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model, the following will introduce the parameter-determining sub-model:

[0212] S622: Input the vehicle appearance video and the standard appearance image into the fixed parameter sub-model, and segment and decode the vehicle appearance video through the fixed parameter sub-model to obtain each frame of the vehicle appearance video and the corresponding reference environment parameters.

[0213] In an embodiment of the present application, segmenting and decoding the vehicle appearance video refers to decomposing the video into reference appearance images frame by frame, and decoding the reference environment parameters corresponding to the frame reference appearance images. For the specific implementation method, please refer to the relevant description of the model training part, which will not be repeated here.

[0214] S623: Compare each frame of the reference appearance image with the standard appearance image respectively, and determine the target environment parameters from the obtained reference environment parameters according to the comparison result.

[0215] In the embodiment of the present application, when the standard appearance image can be a standard photo group including multiple standard photos at different viewing angles, then in S623, each frame of the reference appearance image (also referred to as the reference appearance photo) needs to be compared with the standard photo at the corresponding viewing angle. At this time, the vehicle appearance video generated by the image-generated video sub-model can be understood as a reference appearance photo group. The specific comparison method can be referred to the relevant description of the model training part, which will not be repeated here.

[0216] Considering the above comparison method, there may be multiple image similarity comparison results. Therefore, S623 can be implemented according to the following process, including S6231~S6232 ( Figure 6 not shown):

[0217] S6231: Compare each frame of the reference appearance image with the standard appearance image to obtain an image similarity corresponding to each frame of the reference appearance image;

[0218] S6232: Using the reference environment parameters corresponding to the reference appearance image with the highest image similarity as the target environment parameters.

[0219] By performing image similarity comparison according to the above process and selecting the reference environment parameters corresponding to the reference appearance image with the highest image similarity, it can be ensured that the determination of the target environment parameters is closest to the actual situation. This method can effectively improve the accuracy and reliability of the target environment parameters because the best matching reference image is selected, thereby reducing the potential errors caused by other images above the threshold. This method can not only optimize the results of vehicle appearance detection, but also improve the overall detection performance and robustness of the system.

[0220] In addition, if there are multiple reference appearance images with the highest image similarity, one reference appearance image can be selected from them and its corresponding reference environment parameters can be used as target environment parameters. In practical applications, there are many ways to select a reference appearance image from these multiple reference appearance images with the highest similarity:

[0221] In an optional embodiment, a reference appearance image may be randomly selected from the multiple reference appearance images with the highest similarity, and the corresponding reference environment parameters may be used as the target environment parameters.

[0222] In another optional embodiment, the selection may also be made based on the matching degree of the viewing angle. For example, the reference appearance image closest to the standard viewing angle may be selected from the multiple reference appearance images with the highest similarity, and the corresponding reference environment parameters may be used as the target environment parameters, which helps to ensure the detection accuracy at a specific viewing angle.

[0223] The several selection methods listed above are just simple examples. In addition, other methods are also applicable to the embodiments of the present application and will not be described in detail here.

[0224] S63: adjusting the initial acquisition device parameters according to the target environment parameters to obtain target acquisition device parameters for the vehicle to be detected.

[0225] The initial acquisition device parameters include: various index parameters in the acquisition device when acquiring the appearance image to be detected.

[0226] The acquisition device refers to the device used to acquire the vehicle appearance image, such as a camera, a mobile phone with a camera, a video camera, etc. The acquisition device parameters refer to the technical indicators that need to be set and adjusted during the acquisition process, including but not limited to some or all of the following:

[0227] Resolution: Resolution refers to the number of pixels that can be displayed in an image, usually expressed as the number of pixels in width and height (for example, 1920x1080). The higher the resolution, the more detailed the image will be, capturing more details of the vehicle's exterior. High resolution is important for detecting subtle defects or features on the vehicle's exterior, but it also increases the size of the image file.

[0228] Exposure time: Exposure time refers to the length of time the camera shutter is open, usually expressed in seconds or minutes and seconds. Longer exposure times allow more light to enter the camera, which is suitable for use in low-light environments; while shorter exposure times can reduce image blur and are suitable for capturing moving objects. The choice of exposure time directly affects the brightness and clarity of the image.

[0229] Aperture size: The aperture is the part of the camera lens that controls the amount of light that enters, and is usually expressed in f-numbers (such as f / 2.8, f / 5.6). A larger aperture (i.e., a smaller f-number) allows more light to enter the lens, making it suitable for use in low-light conditions, while also allowing for a shallower depth of field effect, highlighting certain features of the vehicle; a smaller aperture can increase the depth of field, making the entire exterior of the vehicle clearly visible.

[0230] Focal length: Focal length refers to the distance from the center of the lens to the imaging plane, usually expressed in millimeters (mm). The length of the focal length affects the shooting angle and magnification. A short focal length (wide-angle lens) is suitable for shooting a wide range of scenes and can fully capture the appearance of the vehicle; a long focal length (telephoto lens) is suitable for shooting specific parts or details of the vehicle and magnifying distant objects.

[0231] White balance setting: White balance is used to adjust the color balance of an image so that the image presents natural colors under different light sources. Common white balance settings include daylight, shadow, fluorescent light, incandescent light, etc. In vehicle appearance inspection, the correct white balance setting can ensure that the color of the image is restored realistically without causing color difference problems due to different light sources.

[0232] Sensitivity (ISO): Sensitivity refers to the sensitivity of the camera's photosensitive element to light, usually expressed as an ISO value. A higher ISO value can produce brighter images in low-light conditions, but will increase image noise; a lower ISO value can produce clearer, less noisy images, but requires better lighting conditions. Appropriate ISO settings can achieve the best image quality under different lighting conditions.

[0233] In an embodiment of the present application, the initial acquisition device parameters are adjusted according to the target environment parameters. Taking the acquisition device as a camera as an example, the adjustment process is to input the target environment parameters into the camera algorithm service. The camera algorithm service will fine-tune the camera hardware and software parameters through the target environment parameters, that is, the target camera parameters (that is, the target acquisition device parameters). The target camera parameters (that is, the camera hardware and software parameters) mainly include but are not limited to adjusting the exposure parameters, exposure range, balance parameters, focus parameters, noise parameters, and image enhancement parameters of the acquisition device to adapt to the lighting, humidity and temperature conditions of the target environment, optimize the resolution and focus settings of the device, and ensure the clarity of the image and the ability to capture details.

[0234] In addition, in an optional embodiment, the white balance setting of the acquisition device can be adjusted according to the impact of the environment on color performance to ensure the color accuracy of the image. At the same time, considering the impact of background noise, the noise suppression function of the acquisition device can be adjusted to improve image quality.

[0235] In the above process, the target environment parameters are obtained through the image-generated video sub-model and the fixed parameter sub-model of the parameter prediction model, and the target camera parameters are obtained through the camera algorithm service, which realizes the automatic update of the camera's software and hardware parameters, thereby reducing the impact of environmental factors on the appearance photos collected by the camera.

[0236] After adjusting the initial acquisition device parameters, the target acquisition device parameters that meet the desired environmental parameters can be obtained, thereby executing the following process:

[0237] S64: collecting a target appearance image of the vehicle to be detected according to the target acquisition device parameters, and performing vehicle appearance detection based on the target appearance image to obtain a corresponding detection result.

[0238] Among them, performing vehicle appearance detection based on a target appearance image refers to comparing the target appearance image with a standard appearance image to obtain a detection result.

[0239] Specifically, if the detection result indicates that the target appearance image is inconsistent with the standard appearance image, the current target appearance image and target environment parameters can be used as new appearance images to be detected and initial environment parameters, and re-input into the parameter prediction model. An optional implementation method is:

[0240] The current target appearance image and target environment parameters are used as new appearance image to be detected and initial environment parameters corresponding to the vehicle to be detected, and the process returns to S62 until the latest detection result indicates that the latest target appearance image is consistent with the standard appearance image.

[0241] In the embodiment of the present application, first, according to the adjusted target acquisition device parameters, the target appearance image of the vehicle to be detected is acquired using the acquisition device. These parameters ensure the quality and accuracy of the image under actual environmental conditions. The acquired target appearance image is then compared with the standard appearance image. For specific comparison methods, see the relevant description in the above model training section. The purpose of this step is to identify whether the appearance of the vehicle to be detected is consistent with the appearance of the standard vehicle. The detection result may be that the target appearance image is consistent with the standard appearance image or inconsistent.

[0242] Specifically, the standard appearance image may be a standard photo group containing multiple standard photos from different perspectives, and correspondingly, the appearance image to be detected may be a group of appearance photos to be detected. When comparing the photo groups, the target appearance image of each perspective needs to be compared with the standard appearance image of the corresponding perspective. If in the comparison of a certain perspective, the similarity between the target appearance image and the standard appearance image is lower than a preset threshold, it indicates that the image of that perspective is inconsistent. For the entire photo group, if the similarity between the target appearance image and the standard appearance image of any perspective is lower than a preset threshold, the entire photo group is judged to be inconsistent. This means that although other perspectives may meet the standards, as long as one perspective does not meet the standards, the appearance inspection results of the entire vehicle will be judged as unqualified.

[0243] On the contrary, if the similarity between the target appearance image of each perspective and the standard appearance image of the corresponding perspective is higher than the preset threshold, that is, only when the similarity between the target appearance image of each perspective and the corresponding standard appearance image reaches or exceeds the set threshold, the consistency of the entire photo group can be determined.

[0244] If the test result indicates that the target appearance image is inconsistent with the standard appearance image, it means that the acquisition result of the vehicle appearance in the current environment does not meet expectations. There may be improper settings of environmental parameters or acquisition equipment parameters. Therefore, the current target appearance image and target environmental parameters are used as new images to be tested and initial environmental parameters. These new parameters will be input into the trained parameter prediction model, and the model will predict new target environmental parameters. Then, the acquisition equipment is adjusted using the predicted target environmental parameters, and a new target appearance image is acquired again. Repeat the comparison test until the test result shows that the latest target appearance image is consistent with the standard appearance image.

[0245] In the above process, the automatic update of the acquisition equipment parameters is realized, which reduces the influence of environmental factors on the acquisition of the appearance image to be detected by the acquisition equipment, thereby improving the overall efficiency and reliability of vehicle appearance detection.

[0246] Take the acquisition device as a camera as an example, Figure 7 As shown, it is an implementation diagram of a vehicle appearance detection method provided in an embodiment of the present application.

[0247] In the application scenario of the embodiment of the present application, different times of the day are used as examples to illustrate how to cope with changes in environmental conditions during vehicle appearance inspection. Assume that there are multiple vehicles to be inspected that need to undergo vehicle appearance inspection in one day, and the appearance image of each vehicle needs to be compared with the corresponding standard appearance image.

[0248] Specifically, in practical applications, there are data on appearance images to be detected and standard appearance images and initial environmental parameters corresponding to the appearance images to be detected that need to be used for vehicle appearance detection, which are specifically manifested as a small amount of production data in different project implementation scenarios. In the application, a group of appearance images to be detected and the corresponding standard appearance images and their corresponding initial environmental parameters are input into the parameter prediction model each time to obtain the target environmental parameters. Then, the target environmental parameters can be passed to the camera algorithm service, and the camera algorithm service will add the target environmental parameters to the parameter update record table. After that, the camera algorithm service will load the latest environmental parameters and adjust various index parameters in the camera, such as exposure parameters, exposure range, balance parameters, focus parameters, noise parameters, image enhancement parameters, etc., to obtain the target camera parameters for the group of vehicles to be detected after adjusting the initial camera parameters, and then collect the target appearance images of the vehicles to be detected according to the target camera parameters, perform vehicle appearance detection based on the target appearance images, and compare the target appearance images with the standard appearance images.

[0249] If the comparison is consistent, it means that the appearance image of the vehicle to be tested matches the standard appearance image, and the pre-inspection test process is successfully completed. At this time, the camera algorithm service can continue to collect the appearance image of the next vehicle to be tested according to the current target camera parameters, and repeat the above test process until the appearance test results of all vehicles are consistent, and the pre-inspection test of all vehicles to be tested is completed.

[0250] However, due to different environmental conditions throughout the day, such as soft light in the morning, strong light at noon, and light changes in the afternoon, the image taken at the previous time may not be applicable at this time. If the target appearance image of a vehicle is found to be inconsistent with the standard appearance image during the inspection process, this may be because the current environmental conditions are not suitable for shooting at this time. At this time, the current target appearance image and target environmental parameters will be used as new images to be detected and initial environmental parameters, and will be input into the parameter prediction model again to generate new target environmental parameters. Subsequently, the image will be taken again according to these updated parameters, and the vehicle appearance inspection will continue. This process will be repeated until the inspection results of all vehicles are consistent.

[0251] In a possible embodiment of the present application, the environmental parameters of the standard appearance image used can also be recorded, referred to as standard environmental parameters. If it is detected that the standard environmental parameters at a certain time on a certain day in the future are similar to the standard environmental parameters at a certain time on a certain day in the past, then the target environmental parameters in the parameter update record table at the corresponding time on the previous day can be directly used to perform vehicle appearance detection. This means that when the environmental conditions do not change much, the previous target environmental parameters can be directly reused, reducing the time for readjustment and improving detection efficiency.

[0252] In the above process, dynamic adjustment and repeated optimization are used to ensure that the appearance inspection results of each vehicle to be inspected under different environmental conditions can meet the standard requirements. At the same time, under similar environmental conditions, existing data can be used for optimization to further improve the flexibility and practicality of vehicle appearance inspection.

[0253] The embodiment of the present application ensures the quality and accuracy of the image under actual environmental conditions by adjusting the parameters of the target acquisition device. Subsequently, the target appearance image is compared with the standard appearance image to accurately determine whether the appearance of the vehicle to be detected is consistent with the standard. If the test result shows inconsistency, the current target appearance image and target environment parameters are used as new images to be detected and initial environment parameters, and are input into the trained parameter prediction model to predict new target environment parameters. In this way, the system can dynamically adjust the parameters of the acquisition device and repeat the test until the target appearance image is consistent with the standard appearance image, thereby improving the qualified rate of the acquisition of the appearance image to be detected during the pre-inspection of the vehicle, and improving the qualified rate of the comparison between the appearance image to be detected and the standard appearance image.

[0254] This process uses the parameter prediction model to automatically update the parameters of the acquisition equipment, reducing the impact of environmental factors on the appearance images to be detected by the acquisition equipment, thereby improving the overall efficiency and reliability of vehicle appearance detection. At the same time, it is applicable to different on-site environments and can automatically optimize the parameters of the acquisition equipment. The use of automated appearance image capture reduces project implementation costs and manual problem handling, greatly optimizing costs.

[0255] Based on the same inventive concept, the present application also provides a vehicle appearance detection device. Figure 8 As shown, it is a schematic diagram of the structure of a vehicle appearance detection device 800, which may include:

[0256] The data acquisition unit 801 is used to acquire the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected; wherein the image to be detected includes: the standard appearance image of the vehicle to be detected and the appearance image to be detected, and the initial environmental parameters represent: the environmental conditions when the appearance image to be detected is collected;

[0257] The environmental parameter prediction unit 802 is used to input the image to be detected and the initial environmental parameters into the trained parameter prediction model to obtain the target environmental parameters predicted by the parameter prediction model; wherein the target environmental parameters represent: the environmental conditions expected when the appearance image of the vehicle to be detected is currently collected; the parameter prediction model is trained based on the sample image and sample environmental parameters corresponding to the sample vehicle;

[0258] The device parameter adjustment unit 803 is used to adjust the initial acquisition device parameters according to the target environment parameters to obtain the target acquisition device parameters for the vehicle to be detected; wherein the initial acquisition device parameters include: various index parameters in the acquisition device when acquiring the appearance image to be detected;

[0259] The appearance detection unit 804 is used to collect the target appearance image of the vehicle to be detected according to the target acquisition device parameters, and perform vehicle appearance detection based on the target appearance image to obtain corresponding detection results.

[0260] Optionally, the appearance detection unit 804 is specifically used for:

[0261] Comparing the target appearance image with the standard appearance image to obtain the detection result;

[0262] If the detection result indicates that the target appearance image is inconsistent with the standard appearance image, the appearance detection unit 804 is further configured to:

[0263] The current target appearance image and target environment parameters are used as the new appearance image to be detected and initial environment parameters corresponding to the vehicle to be detected, and the step of inputting the image to be detected and the initial environment parameters into the trained parameter prediction model to obtain the target environment parameters predicted by the parameter prediction model is returned, until the latest detection result indicates that the latest target appearance image is consistent with the standard appearance image.

[0264] Optionally, the parameter prediction model includes an image-generated video sub-model and a fixed parameter sub-model; the environmental parameter prediction unit 802 is specifically used for:

[0265] Inputting the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and obtaining a vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model;

[0266] Inputting the vehicle appearance video and the standard appearance image into the fixed parameter sub-model, segmenting and decoding the vehicle appearance video through the fixed parameter sub-model, and obtaining each frame of the vehicle appearance video reference appearance image and corresponding reference environment parameters;

[0267] Each frame of the reference appearance image is compared with the standard appearance image, and the target environment parameters are determined from the obtained reference environment parameters according to the comparison result.

[0268] Optionally, the environmental parameter prediction unit 802 is specifically used for:

[0269] Inputting the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and extracting feature vectors corresponding to the appearance image to be detected and the initial environmental parameters respectively through the image-generated video sub-model;

[0270] The extracted feature vectors are fused to obtain a fused vector;

[0271] Generating a vehicle appearance video corresponding to the vehicle to be detected based on the fusion;

[0272] and, comparing each frame of the reference appearance image with the standard appearance image to obtain an image similarity corresponding to each frame of the reference appearance image;

[0273] The reference environment parameters corresponding to the reference appearance image with the highest image similarity are used as the target environment parameters.

[0274] Based on the same inventive concept, the present application also provides a training device for a parameter prediction model. Fig. 9 As shown, it is a schematic diagram of the structure of a training device 900 for a parameter prediction model, which may include:

[0275] The sample acquisition unit 901 is used to acquire a training sample set, each training sample in the training sample set includes: a sample image corresponding to a sample vehicle and a sample environmental parameter; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: an environmental condition when the sample appearance image is collected;

[0276] The training unit 902 is used to perform the following operations for each training sample:

[0277] Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle;

[0278] The vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained; based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video is compared with the corresponding standard appearance image, and the parameters of the image-generated video sub-model are adjusted according to the comparison result; wherein the trained parameter prediction model is used to predict the first target environmental parameters for the sample vehicle, the first target environmental parameters representing the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the first target environmental parameters and perform vehicle appearance detection.

[0279] Optionally, the training unit 902 is specifically used for:

[0280] Based on the fixed parameter sub-model, the vehicle appearance video is segmented and decoded to obtain a reference appearance image of each frame in the vehicle appearance video and corresponding reference environment parameters;

[0281] Comparing each frame of the reference appearance image with the corresponding standard appearance image to obtain the image similarity corresponding to each frame of the reference appearance image;

[0282] If the obtained similarities of each image are not higher than a preset threshold, the parameters of the image-generated video sub-model are adjusted according to the difference between the reference environment parameters and the corresponding sample environment parameters.

[0283] Optionally, the training unit 902 is specifically used for:

[0284] Parameters of a low-rank matrix corresponding to the image-generated video sub-model are adjusted, wherein the low-rank matrix is ​​obtained by performing low-rank decomposition on a weight matrix to be adjusted in the image-generated video sub-model.

[0285] Optionally, the parameter prediction model training device 900 is further used for:

[0286] If the similarities of the obtained images are not higher than the preset threshold, the reference appearance image and the corresponding reference environment parameters with the highest image similarity are used as the new sample appearance image and sample environment parameters corresponding to the sample vehicle, and the step of inputting the corresponding sample image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained to obtain the vehicle appearance video corresponding to the corresponding sample vehicle is returned, until the image similarities corresponding to each frame of the reference appearance image in the latest vehicle appearance video have an image similarity higher than the preset threshold;

[0287] If there is an image similarity higher than the preset threshold among the obtained image similarities, the reference environmental parameters corresponding to the reference appearance image with the highest image similarity are used as the second target environmental parameters corresponding to the sample vehicle. The second target environmental parameters represent the expected environmental conditions when the appearance image of the sample vehicle is currently being collected.

[0288] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0289] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0290] After introducing the vehicle appearance detection method and device according to an exemplary embodiment of the present application, next, an electronic device according to another exemplary embodiment of the present application is introduced.

[0291] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0292] Based on the same inventive concept as the above method embodiment, an electronic device is also provided in the embodiment of the present application. In one embodiment, the electronic device may be a server, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Fig.10 As shown, it includes a memory 1001 , a communication module 1003 and one or more processors 1002 .

[0293] The memory 1001 is used to store computer programs executed by the processor 1002. The memory 1001 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0294] The memory 1001 may be a volatile memory, such as a random-access memory (RAM); the memory 1001 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 1001 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1001 may be a combination of the above memories.

[0295] The processor 1002 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 1002 is used to implement the above-mentioned vehicle appearance detection method when calling the computer program stored in the memory 1001 .

[0296] The communication module 1003 is used to communicate with terminal devices and other servers.

[0297] The specific connection medium between the memory 1001, the communication module 1003 and the processor 1002 is not limited in the embodiment of the present application. Fig.10 In the embodiment, the memory 1001 and the processor 1002 are connected via a bus 1004. The bus 1004 is connected to the processor 1002 via a bus 1004. Fig.10 The connections between the other components are only for illustration and are not intended to be limiting. The bus 1004 can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Fig.10 The diagram shows that only one thick line is used, but this does not mean that there is only one bus or only one type of bus.

[0298] The memory 1001 stores a computer storage medium, and the computer storage medium stores computer executable instructions, and the computer executable instructions are used to implement the vehicle appearance detection method of the embodiment of the present application. The processor 1002 is used to execute the above-mentioned vehicle appearance detection method, such as Figure 6 shown.

[0299] In another embodiment, the electronic device may also be other electronic devices, such as Figure 1 The terminal device 110 shown in FIG. 1 is a terminal device 110 shown in FIG. 1 . In this embodiment, the structure of the electronic device can be as follows: Fig.11 As shown, it includes: a communication component 1110, a memory 1120, a display unit 1130, a camera 1140, a sensor 1150, an audio circuit 1160, a Bluetooth module 1170, a processor 1180 and other components.

[0300] The communication component 1110 is used to communicate with the server. In some embodiments, a wireless fidelity (WiFi) module may be included. The WiFi module belongs to a short-range wireless transmission technology. The electronic device can help the user to send and receive information through the WiFi module.

[0301] The memory 1120 can be used to store software programs and data. The processor 1180 executes various functions and data processing of the terminal device 110 by running the software programs or data stored in the memory 1120. The memory 1120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. The memory 1120 stores an operating system that enables the terminal device 110 to run. In the present application, the memory 1120 can store an operating system and various application programs, and can also store a computer program for executing the vehicle appearance detection method or the parameter prediction model training method of the embodiment of the present application.

[0302] The display unit 1130 can also be used to display information input by the user or information provided to the user and a graphical user interface (GUI) of various menus of the terminal device 110. Specifically, the display unit 1130 may include a display screen 1132 disposed on the front of the terminal device 110. The display screen 1132 may be configured in the form of a liquid crystal display, a light emitting diode, etc. The display unit 1130 may be used to display the vehicle appearance detection user interface in the embodiment of the present application, etc.

[0303] The display unit 1130 can also be used to receive input digital or character information, and generate signal input related to user settings and function control of the terminal device 110. Specifically, the display unit 1130 may include a touch screen 1131 arranged on the front of the terminal device 110, which can collect user touch operations on or near it, such as clicking a button, dragging a scroll box, etc.

[0304] The touch screen 1131 can be covered on the display screen 1132, or the touch screen 1131 and the display screen 1132 can be integrated to realize the input and output functions of the terminal device 110, and the integrated display screen can be referred to as a touch display screen. In this application, the display unit 1130 can display the application and the corresponding operation steps.

[0305] The camera 1140 can be used to capture still images, and the user can publish the images taken by the camera 1140 through the application. The camera 1140 can be one or more. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the optical signal into an electrical signal, and then transmits the electrical signal to the processor 1180 to convert it into a digital image signal.

[0306] The terminal device may further include at least one sensor 1150, such as an acceleration sensor 1151, a distance sensor 1152, a fingerprint sensor 1153, and a temperature sensor 1154. The terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, a light sensor, and a motion sensor.

[0307] The audio circuit 1160, the speaker 1161, and the microphone 1162 can provide an audio interface between the user and the terminal device 110. The audio circuit 1160 can transmit the electrical signal converted from the received audio data to the speaker 1161, which is converted into a sound signal for output. The terminal device 110 can also be equipped with a volume button for adjusting the volume of the sound signal. On the other hand, the microphone 1162 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1160 and converted into audio data, and then the audio data is output to the communication component 1110 to be sent to, for example, another terminal device 110, or the audio data is output to the memory 1120 for further processing.

[0308] The Bluetooth module 1170 is used to exchange information with other Bluetooth devices having Bluetooth modules through the Bluetooth protocol. For example, the terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smart watch) that also has a Bluetooth module through the Bluetooth module 1170 to exchange data.

[0309] The processor 1180 is the control center of the terminal device. It uses various interfaces and lines to connect various parts of the entire terminal. It executes various functions of the terminal device and processes data by running or executing software programs stored in the memory 1120 and calling data stored in the memory 1120. In some embodiments, the processor 1180 may include one or more processing units; the processor 1180 may also integrate an application processor and a baseband processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the baseband processor mainly processes wireless communications. It is understandable that the above-mentioned baseband processor may not be integrated into the processor 1180. In the present application, the processor 1180 can run the operating system, application programs, user interface display and touch response, as well as the vehicle appearance detection method or parameter prediction model training method of the embodiment of the present application. In addition, the processor 1180 is coupled to the display unit 1130.

[0310] In some optional embodiments, various aspects of the vehicle appearance detection method provided by the present application may also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the vehicle appearance detection method or the parameter prediction model training method according to various exemplary embodiments of the present application described above in this specification. For example, the electronic device may execute the following steps: Figure 2 or Figure 6 Follow the steps shown in .

[0311] The program product may adopt any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0312] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can be run on an electronic device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which can be used by or in combination with a command execution system, apparatus, or device.

[0313] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0314] The computer program embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0315] The computer program for performing the operation of the present application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and also conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer program can be executed entirely on the user electronic device, partially on the user electronic device, as an independent software package, partially on the user electronic device and partially on the remote electronic device, or completely on the remote electronic device or server. In the case of involving a remote electronic device, the remote electronic device can be connected to the user electronic device through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external electronic device (e.g., using an Internet service provider to connect through the Internet).

[0316] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0317] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0318] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain a computer-usable computer program.

[0319] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program commands. These computer program commands can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0320] These computer program commands may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the commands stored in the computer readable memory produce an article of manufacture comprising a command device, the command device implementing the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0321] These computer program commands can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the commands executed on the computer or other programmable device provide the instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0322] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0323] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A vehicle appearance detection method, characterized in that: The method comprises: Acquire the image to be detected and the initial environmental parameters corresponding to the vehicle to be detected; wherein the image to be detected includes: the standard appearance image of the vehicle to be detected and the appearance image to be detected, and the initial environmental parameters represent: the environmental conditions when the appearance image to be detected is collected; Input the image to be detected and the initial environmental parameters into the image-generated video sub-model in the trained parameter prediction model, and obtain the vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model; input the vehicle appearance video and the standard appearance image into the fixed parameter sub-model in the parameter prediction model, and segment and decode the vehicle appearance video through the fixed parameter sub-model to obtain each frame of the reference appearance image and the corresponding reference environmental parameters in the vehicle appearance video; compare each frame of the reference appearance image with the standard appearance image, and determine the target environmental parameters from the obtained reference environmental parameters according to the comparison results; wherein the target environmental parameters represent: the expected environmental conditions when the appearance image of the vehicle to be detected is currently collected; the parameter prediction model is trained based on the sample images and sample environmental parameters corresponding to the sample vehicles; The initial acquisition equipment parameters are adjusted according to the target environment parameters to obtain the target acquisition equipment parameters for the vehicle to be detected; wherein the initial acquisition equipment parameters include: various index parameters in the acquisition equipment when acquiring the appearance image to be detected; The target appearance image of the vehicle to be detected is acquired according to the target acquisition device parameters, and the vehicle appearance detection is performed based on the target appearance image to obtain a corresponding detection result.

2. The method according to claim 1, characterized in that The performing vehicle appearance detection based on the target appearance image to obtain corresponding detection results includes: Comparing the target appearance image with the standard appearance image to obtain the detection result; If the detection result indicates that the target appearance image is inconsistent with the standard appearance image, the method further includes: The current target appearance image and target environment parameters are used as the new appearance image to be detected and initial environment parameters corresponding to the vehicle to be detected, and the step of inputting the image to be detected and the initial environment parameters into the trained parameter prediction model to obtain the target environment parameters predicted by the parameter prediction model is returned, until the latest detection result indicates that the latest target appearance image is consistent with the standard appearance image.

3. The method according to claim 1, characterized in that The step of inputting the image to be detected and the initial environmental parameters into a picture-generated video sub-model in a trained parameter prediction model, and obtaining a vehicle appearance video corresponding to the vehicle to be detected through the picture-generated video sub-model, comprises: Inputting the appearance image to be detected and the initial environmental parameters into the image-generated video sub-model, and extracting feature vectors corresponding to the appearance image to be detected and the initial environmental parameters respectively through the image-generated video sub-model; The extracted feature vectors are fused to obtain a fused vector; Generate a vehicle appearance video corresponding to the vehicle to be detected based on the fusion vector; And, comparing each frame of the reference appearance image with the standard appearance image, and determining the target environment parameter from each reference environment parameter obtained according to the comparison result, comprises: Comparing each frame of the reference appearance image with the standard appearance image to obtain an image similarity corresponding to each frame of the reference appearance image; The reference environment parameters corresponding to the reference appearance image with the highest image similarity are used as the target environment parameters.

4. A training method for a parameter prediction model, characterized in that: The method comprises: Acquire a training sample set, wherein each training sample in the training sample set includes: a sample image corresponding to a sample vehicle and a sample environmental parameter; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: an environmental condition when the sample appearance image is collected; For each training sample, perform the following operations: Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle; The vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained; based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video corresponding to the sample vehicle is compared with the corresponding standard appearance image, and the parameters of the image-generated video sub-model are adjusted according to the comparison result; the trained parameter prediction model is used to obtain the vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model, and the vehicle appearance video corresponding to the vehicle to be detected is segmented and decoded through the fixed-parameter sub-model to obtain each frame of the reference appearance image and the corresponding reference environmental parameters in the vehicle appearance video corresponding to the vehicle to be detected; each frame of the reference appearance image in the vehicle appearance video corresponding to the vehicle to be detected is compared with the corresponding standard appearance image, and the first target environmental parameter is determined from the obtained reference environmental parameters according to the comparison result; the first target environmental parameter represents the expected environmental condition when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the first target environmental parameter, and perform vehicle appearance detection.

5. The method according to claim 4, characterized in that The method of comparing each frame of the reference appearance image in the vehicle appearance video corresponding to the sample vehicle with the corresponding standard appearance image based on the fixed parameter sub-model, and adjusting the parameters of the image-generated video sub-model according to the comparison result, includes: Based on the fixed-parameter sub-model, the vehicle appearance video corresponding to the sample vehicle is segmented and decoded to obtain a reference appearance image of each frame in the vehicle appearance video corresponding to the sample vehicle and corresponding reference environment parameters; Respectively comparing each frame of the reference appearance image in the vehicle appearance video corresponding to the sample vehicle with the corresponding standard appearance image to obtain the image similarity corresponding to each frame of the reference appearance image; If the similarities of each image obtained are not higher than a preset threshold, the parameters of the image-generated video sub-model are adjusted according to the difference between the reference environment parameters of each frame of the reference appearance image in the vehicle appearance video corresponding to the sample vehicle and the corresponding sample environment parameters.

6. The method according to claim 5, characterized in that The method further comprises: If the similarities of each image obtained are not higher than the preset threshold, the reference appearance image and the corresponding reference environment parameters with the highest image similarity in the vehicle appearance video corresponding to the sample vehicle are used as new sample appearance images and sample environment parameters corresponding to the sample vehicle, and the step of inputting the corresponding sample image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained to obtain the vehicle appearance video corresponding to the corresponding sample vehicle is returned, until the image similarities corresponding to each frame of the reference appearance image in the latest vehicle appearance video have an image similarity higher than the preset threshold; If there is an image similarity higher than the preset threshold among the obtained image similarities, the reference environmental parameters corresponding to the reference appearance image with the highest image similarity in the vehicle appearance video corresponding to the sample vehicle are used as the second target environmental parameters corresponding to the sample vehicle. The second target environmental parameters represent the expected environmental conditions when the appearance image of the sample vehicle is currently being collected.

7. A vehicle appearance detection device, characterized in that: The device comprises: A data acquisition unit, used to acquire an image to be detected and initial environmental parameters corresponding to the vehicle to be detected; wherein the image to be detected includes: a standard appearance image of the vehicle to be detected and an appearance image to be detected, and the initial environmental parameters represent: environmental conditions when the appearance image to be detected is collected; An environmental parameter prediction unit is used to input the image to be detected and the initial environmental parameters into a trained image-generated video sub-model in a parameter prediction model, and obtain a vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model; input the vehicle appearance video and the standard appearance image into a fixed parameter sub-model in the parameter prediction model, and segment and decode the vehicle appearance video through the fixed parameter sub-model to obtain a reference appearance image of each frame in the vehicle appearance video and corresponding reference environmental parameters; compare each frame of the reference appearance image with the standard appearance image, and determine target environmental parameters from the obtained reference environmental parameters according to the comparison results; wherein the target environmental parameters represent: the environmental conditions expected when the appearance image of the vehicle to be detected is currently collected; the parameter prediction model is trained based on sample images and sample environmental parameters corresponding to sample vehicles; The device parameter adjustment unit is used to adjust the initial acquisition device parameters according to the target environment parameters to obtain the target acquisition device parameters for the vehicle to be detected; wherein the initial acquisition device parameters include: various index parameters in the acquisition device when acquiring the appearance image to be detected; The appearance detection unit is used to collect the target appearance image of the vehicle to be detected according to the target acquisition device parameters, and perform vehicle appearance detection based on the target appearance image to obtain corresponding detection results.

8. A training device for a parameter prediction model, characterized in that: The device comprises: A sample acquisition unit is used to acquire a training sample set, each training sample in the training sample set includes: a sample image corresponding to a sample vehicle and a sample environmental parameter; wherein the sample image includes: a standard appearance image and a sample appearance image of the sample vehicle, and the sample environmental parameter represents: an environmental condition when the sample appearance image is collected; The training unit is used to perform the following operations on each training sample: Inputting the corresponding sample appearance image and sample environment parameters into the image-generated video sub-model in the parameter prediction model to be trained, and obtaining the vehicle appearance video corresponding to the corresponding sample vehicle; The vehicle appearance video and the corresponding standard appearance image are input into the fixed-parameter sub-model in the parameter prediction model to be trained; based on the fixed-parameter sub-model, each frame of the reference appearance image in the vehicle appearance video corresponding to the sample vehicle is compared with the corresponding standard appearance image, and the parameters of the image-generated video sub-model are adjusted according to the comparison result; the trained parameter prediction model is used to obtain the vehicle appearance video corresponding to the vehicle to be detected through the image-generated video sub-model, and the vehicle appearance video corresponding to the vehicle to be detected is segmented and decoded through the fixed-parameter sub-model to obtain each frame of the reference appearance image and the corresponding reference environmental parameters in the vehicle appearance video corresponding to the vehicle to be detected; each frame of the reference appearance image in the vehicle appearance video corresponding to the vehicle to be detected is compared with the corresponding standard appearance image, and the first target environmental parameter is determined from the obtained reference environmental parameters according to the comparison result; the first target environmental parameter represents the expected environmental condition when the appearance image of the vehicle to be detected is currently collected, so as to collect the target appearance image of the vehicle to be detected based on the first target environmental parameter, and perform vehicle appearance detection.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 1 to 6.

Images