Vehicle appearance quality damage detection method and device, electronic equipment and storage medium

By using a multi-layer network structure object detection model in vehicle appearance quality and loss detection, combining image acquisition components and light sources, obtaining vehicle appearance data from multiple angles, extracting and fusion feature information, the efficiency and accuracy in the prior art are solved The problem of low quality and loss detection of vehicle appearance can be achieved quickly and accurately.

CN119090838APending Publication Date: 2024-12-06FAW LOGISTICS CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411185961.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing vehicle appearance quality-destructive testing methods are inefficient and have low accuracy, which cannot meet the needs of large-scale production, and have shortcomings in data processing and analysis, resulting in frequent misjudgment and missed inspections.

Method used

The image acquisition components and light sources in the preset confined space are used to obtain the initial image data set of the target vehicle's appearance from multiple angles, and detect it through the target detection model. The multi-layer network structure is used to extract and fuse feature information to generate the vehicle's appearance quality. Destructive test results. The model includes the first network for feature extraction, the second network for feature fusion, and the third network for detection, combining the focus module, bottleneck module, upsampling module, feature fusion module and path aggregation structure to achieve rapid and accurate identification of vehicles Breakage of appearance.

Benefits of technology

It improves the accuracy and efficiency of vehicle appearance quality and loss testing, reduces the subjectivity and time cost of manual inspection, and can quickly and accurately identify whether there is damage in the vehicle appearance, improving the accuracy and efficiency of inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090838A_ABST
    Figure CN119090838A_ABST
Patent Text Reader

Abstract

The invention discloses a vehicle appearance quality damage detection method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring an initial image data set of the appearance of a target vehicle from a plurality of shooting angles by using an image acquisition assembly and a light source arranged in a preset closed space; the initial image data set is input to a target detection model for detection, a target detection result is obtained, the target detection model at least comprises a first network and a second network, the first network is used for extracting first feature information corresponding to the initial image data set, and the second network is used for fusing the first feature information; and generating a vehicle appearance quality damage detection result based on the target detection result. According to the invention, the technical problem of low efficiency and accuracy of a vehicle appearance quality damage detection method in the prior art is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a method, device, electronic equipment and storage medium for detecting vehicle appearance damage. Background Art

[0002] Vehicle exterior damage inspection is a crucial component of the automotive manufacturing and repair industries. With the rapid development of the automotive industry, the requirements for accuracy and efficiency in vehicle exterior damage inspection are becoming increasingly stringent. However, existing vehicle exterior damage inspection technologies have several shortcomings that limit their performance in practical applications.

[0003] First, traditional manual inspection methods suffer from inefficiency and low accuracy. Manual inspection relies on the experience and skills of the inspector, making it prone to missed detections and misjudgments. Furthermore, manual inspection is time-consuming and cannot meet the demands of large-scale production.

[0004] Secondly, existing vehicle exterior damage inspection technologies also have shortcomings in data processing and analysis. Many systems lack the ability to deeply mine and analyze inspection data. This often leads to misjudgments and missed detections when dealing with damage issues on complex surfaces and hidden areas, resulting in low vehicle exterior damage inspection accuracy.

[0005] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0006] Embodiments of the present invention provide a vehicle appearance damage detection method, device, electronic device, and storage medium to at least solve the technical problem of low efficiency and accuracy of vehicle appearance damage detection methods in related technologies.

[0007] According to one embodiment of the present invention, a method for detecting vehicle exterior damage is provided, comprising: acquiring an initial image dataset of a target vehicle exterior from multiple shooting angles using an image acquisition component and a light source provided in a preset enclosed space; inputting the initial image dataset into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network being used to extract first feature information corresponding to the initial image dataset, and the second network being used to fuse the first feature information; and generating a vehicle exterior damage detection result based on the target detection result.

[0008] Optionally, the target detection model further includes a third network, and the initial image data set is input into the target detection model for detection, and the target detection result is obtained, including: preprocessing the initial image data set to obtain a target image data set; using the first network to extract features of the target image data set to obtain first feature information corresponding to the initial image data set; using the second network to fuse the first feature information to obtain target feature information; and detecting the target feature information based on the third network to obtain a target detection result.

[0009] Optionally, the first network includes at least a focus module and a bottleneck module, and using the first network to extract features of the target image data set to obtain first feature information corresponding to the initial image data set includes: inputting the target image data set into the focus module; using the focus module to perform dimensionality reduction processing on the target image data set to obtain second feature information; using the bottleneck module to extract features from the second feature information to obtain first feature information.

[0010] Optionally, the bottleneck module includes at least a cross-stage partially connected layer and a spatial pyramid pooling layer, and using the bottleneck module to extract features from the second feature information to obtain the first feature information includes: using the first cross-stage partially connected layer to extract features from the second feature information to obtain first sub-feature information; using the second cross-stage partially connected layer to extract features from the first sub-feature information to obtain second sub-feature information; using the spatial pyramid pooling layer to convert the second word feature information into a preset length to obtain third sub-feature information; and using the third cross-stage partially connected layer to extract features from the third sub-feature information to obtain fourth sub-feature information.

[0011] Optionally, the second network includes at least an upsampling module, a feature fusion module and a path aggregation structure. The second network is used to fuse the feature information to obtain the target feature information, including: using the upsampling module to sample the fourth sub-feature information to obtain third feature information; using the feature fusion module to fuse the third feature information with the second sub-feature information to obtain fourth feature information; using the feature fusion module to fuse the fourth feature information with the first sub-feature information to obtain fifth feature information; and performing feature extraction on the fifth feature information based on the path aggregation structure to obtain target feature information.

[0012] Optionally, the third network includes a convolution module, and the target feature information is detected based on the third network to obtain the target detection result, including: inputting the target feature information into the convolution module for prediction to obtain candidate anchor box information; performing weighted matching on the candidate anchor box information to obtain a matching result; and filtering out the target detection result from the candidate anchor box information based on the matching result.

[0013] Optionally, preprocessing the initial image dataset to obtain the target image dataset includes: expanding the initial image dataset using a mosaic data enhancement method to obtain a first image dataset; and adaptively scaling samples in the first image dataset to obtain the target image dataset.

[0014] Optionally, generating a vehicle appearance quality damage inspection result based on the target detection result includes: analyzing the target detection result to obtain an analysis result; in response to the analysis result indicating that the target vehicle is damaged, the vehicle appearance quality damage inspection result indicates that the target vehicle has failed the vehicle appearance quality damage inspection.

[0015] Optionally, obtaining an initial image dataset of the target vehicle's appearance from multiple shooting angles includes: in response to the target vehicle entering a damage inspection area, obtaining unique identifier information of the target vehicle; comparing the unique identifier information with vehicle information in a preset vehicle information list to obtain a comparison result; in response to the comparison result indicating that the unique identifier information passes the comparison, obtaining an initial image dataset of the target vehicle's appearance from multiple shooting angles.

[0016] According to one embodiment of the present invention, a vehicle appearance quality damage detection device is also provided, including: an acquisition module, used to use an image acquisition component and a light source set in a preset confined space to obtain an initial image data set of the target vehicle appearance from multiple shooting angles; a detection module, inputting the initial image data set into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image data set, and the second network is used to fuse the first feature information; a generation module, generating a vehicle appearance quality damage detection result based on the target detection result.

[0017] Optionally, the detection module is also used to: preprocess the initial image data set to obtain a target image data set; use the first network to extract features of the target image data set to obtain first feature information corresponding to the initial image data set; use the second network to fuse the first feature information to obtain target feature information; detect the target feature information based on the third network to obtain a target detection result.

[0018] Optionally, the detection module is also used to: input the target image data set into the focus module; use the focus module to perform dimensionality reduction processing on the target image data set to obtain second feature information; use the bottleneck module to perform feature extraction on the second feature information to obtain first feature information.

[0019] Optionally, the detection module is further used to: use the first cross-stage partially connected layer to perform feature extraction on the second feature information to obtain first sub-feature information; use the second cross-stage partially connected layer to perform feature extraction on the first sub-feature information to obtain second sub-feature information; use the spatial pyramid pooling layer to convert the second word feature information into a preset length to obtain third sub-feature information; and use the third cross-stage partially connected layer to perform feature extraction on the third sub-feature information to obtain fourth sub-feature information.

[0020] Optionally, the detection module is also used to: use the upsampling module to sample and process the fourth sub-feature information to obtain third feature information; use the feature fusion module to fuse the third feature information with the second sub-feature information to obtain fourth feature information; use the feature fusion module to fuse the fourth feature information with the first sub-feature information to obtain fifth feature information; and perform feature extraction on the fifth feature information based on the path aggregation structure to obtain target feature information.

[0021] Optionally, the acquisition module is also used to input the target feature information into the convolution module for prediction to obtain candidate anchor frame information; the detection module is also used to perform weighted matching on the candidate anchor frame information to obtain a matching result; the vehicle appearance quality damage detection device also includes a processing module for filtering out target detection results from the candidate anchor frame information based on the matching result.

[0022] Optionally, the processing module is further configured to: expand the initial image dataset using a mosaic data enhancement method to obtain a first image dataset; and perform adaptive scaling processing on samples in the first image dataset to obtain a target image dataset.

[0023] Optionally, the generation module is also used to: analyze the target detection result to obtain an analysis result; in response to the analysis result indicating that the target vehicle is damaged, the vehicle appearance damage inspection result indicates that the target vehicle has failed the vehicle appearance damage inspection.

[0024] Optionally, the acquisition module is also used to obtain the unique identifier information of the target vehicle in response to the target vehicle entering the damage inspection area; the processing module is also used to compare the unique identifier information with the vehicle information in a preset vehicle information list to obtain a comparison result; the acquisition module is also used to obtain an initial image data set of the appearance of the target vehicle from multiple shooting angles in response to the comparison result indicating that the unique identifier information has passed the comparison.

[0025] According to one embodiment of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the above-mentioned vehicle appearance quality damage detection method is executed when the program is running.

[0026] According to one embodiment of the present invention, a computer-readable storage medium is further provided, which includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the above-mentioned vehicle appearance quality damage detection method.

[0027] According to one embodiment of the present invention, a computer program product is also provided, including a computer program, which implements the above-mentioned vehicle appearance quality damage detection method when executed by a processor.

[0028] In an embodiment of the present invention, an image acquisition component and a light source provided in a preset enclosed space are used to obtain an initial image data set of the target vehicle's appearance from multiple shooting angles, and the initial image data set is input into a target detection model for detection to obtain a target detection result. Finally, a vehicle appearance damage detection result is generated based on the target detection result. This achieves the purpose of quickly and accurately identifying whether the vehicle's appearance is damaged, thereby realizing the technical effect of improving the accuracy of the vehicle appearance damage detection method, and further solving the technical problem of low efficiency and accuracy of the vehicle appearance damage detection method in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0030] Figure 1 is a flow chart of a vehicle appearance damage detection method according to one embodiment of the present invention;

[0031] Figure 2 is a schematic diagram of a vehicle appearance damage detection method according to one embodiment of the present invention;

[0032] Figure 3 2 is a schematic diagram of another method for detecting vehicle appearance damage according to one embodiment of the present invention;

[0033] Figure 4 is a schematic diagram of another vehicle appearance damage detection method according to one embodiment of the present invention;

[0034] Figure 5 The figure is a structural block diagram of a vehicle appearance quality damage detection device according to one embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0037] According to an embodiment of the present invention, a method embodiment of a vehicle appearance quality damage detection method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] The method embodiment can be executed in an electronic device or similar computing device that includes a memory and a processor. Taking running on a vehicle terminal as an example, the vehicle terminal may include one or more processors (processors may include but are not limited to central processing units (CPUs), graphics processing units (GPUs), digital signal processing (DSP) chips, microcontroller units (MCUs), field programmable gate arrays (FPGAs), neural network processors (NPUs), tensor processing units (TPUs), artificial intelligence (AI) type processors, etc.) and a memory for storing data. Optionally, the vehicle terminal may also include transmission equipment, input and output equipment, and display equipment for communication functions. Those skilled in the art will understand that the above structural description is only illustrative and does not limit the structure of the vehicle terminal. For example, the vehicle terminal may also include more or fewer components than described above, or have a configuration different from that described above.

[0039] The memory can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the vehicle appearance quality damage detection method in the embodiment of the present invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, realizing the above-mentioned vehicle appearance quality damage detection method. The memory may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to the mobile terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, corporate intranet, local area network, mobile communication network, and combinations thereof.

[0040] The transmission device is used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by the mobile terminal's communications provider. In one embodiment, the transmission device includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In one embodiment, the transmission device may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0041] The display device can be, for example, a touch-screen liquid crystal display (LCD) and a touch display (also referred to as a "touch screen" or "touch display screen"). The liquid crystal display can enable the user to interact with the user interface of the mobile terminal. In some embodiments, the mobile terminal has a graphical user interface (GUI), and the user can interact with the GUI by finger contacts and / or gestures on the touch-sensitive surface. The human-computer interaction functions here optionally include the following interactions: creating web pages, drawing, word processing, making electronic documents, games, video conferencing, instant messaging, sending and receiving emails, call interfaces, playing digital videos, playing digital music and / or web browsing, etc. The executable instructions for performing the above-mentioned human-computer interaction functions are configured / stored in a computer program product or readable storage medium executable by one or more processors.

[0042] Figure 1 FIG. 1 is a flow chart of a method for detecting vehicle appearance damage according to one embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0043] Step S10, using an image acquisition component and a light source provided in a preset enclosed space, to acquire an initial image dataset of the exterior of the target vehicle from multiple shooting angles;

[0044] Specifically, a damage inspection room was set up. It had no windows and no natural light. Instead, artificial light sources and high-definition cameras were deployed inside the room. The cameras and light sources were positioned on the front, back, left, right, top, and bottom of the room. The exterior of the target vehicle was photographed from multiple angles to generate an initial image dataset.

[0045] Step S12: Inputting the initial image dataset into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image dataset, and the second network is used to fuse the first feature information;

[0046] In step S12, the first network is used to represent the backbone network of the target detection model.

[0047] The second network is used to represent the neck network (Neck) of the target detection model.

[0048] Specifically, the obtained initial image data set is input into the target detection model, the first network in the target detection model performs feature extraction on the initial image data set to obtain first feature information, the second network in the target detection model performs feature fusion on the first feature information, and uses the fused first feature information to perform detection to obtain a target detection result, wherein the target detection result is used to characterize whether the appearance of the target vehicle is damaged.

[0049] Step S14: generating a vehicle appearance quality damage detection result based on the target detection result.

[0050] Specifically, the vehicle appearance damage inspection result is generated based on whether the target vehicle appearance is damaged in the target detection result. If the target vehicle appearance is damaged in the target detection result, the vehicle appearance damage inspection result is that the target vehicle has failed the vehicle appearance damage inspection, and the damage type is recorded. The vehicle appearance damage inspection result is provided to the staff to improve the efficiency of subsequent maintenance; if the target vehicle appearance is not damaged in the target detection result, the vehicle appearance damage inspection result is that the target vehicle has passed the vehicle appearance damage inspection.

[0051] Specifically, Figure 2 FIG. 1 is a schematic diagram of a vehicle appearance quality damage detection method according to one embodiment of the present invention. Figure 2 As shown, the target vehicle first drives from the vehicle waiting area to the quality inspection room. The vehicle waiting area serves as a waiting area before the target vehicle officially enters the inspection process. This area is typically clearly marked with signs and instructions to guide vehicles to queue in an orderly manner. This area may also be staffed with managers to ensure that vehicles enter the inspection process in an orderly manner. On its way to the quality inspection room, the target vehicle passes through the induction gate sensing area. When the target vehicle approaches the induction gate, sensors on the ground or on the gate detect its presence and trigger the door control system. The induction gate opens quickly and smoothly, allowing the target vehicle to enter the quality inspection room. This process is typically automated, requiring no human intervention.

[0052] Next, the target vehicle enters the inspection room. This is the core area for exterior damage inspection of the target vehicle. The interior environment is carefully designed to ensure ample light and no shadows. It is equipped with a professional industrial air-conditioning system to maintain a constant temperature and humidity. This environment is conducive to industrial cameras capturing clear and accurate vehicle images. Clear indicator lights are located at the entrance and exit of the inspection room. A red light indicates no entry (or waiting), while a green light indicates entry. This helps control vehicle flow and avoid collisions or congestion. Furthermore, restricted areas are set within the inspection room to protect the safety of testing equipment and personnel.

[0053] Then, after the target vehicle enters the quality inspection room, the industrial camera will take all-round photos of the target vehicle to obtain an initial image dataset of the target vehicle's appearance. The initial image dataset will be transmitted to the server and detected using the target detection model. The target detection model will automatically identify subtle flaws, scratches, dents and other problems on the target vehicle's appearance and obtain the target detection results. This process is fast and efficient, which can improve detection accuracy and efficiency. During the photo-taking and analysis process, the target vehicle needs to be parked in the designated quality inspection area, where a grating is used to detect whether the target vehicle is parked correctly. After the target vehicle is parked in the designated quality inspection area, the reader will identify the radio frequency identification (RFID) tag on the target vehicle to obtain the vehicle identification number (VIN) of the target vehicle. During this process, the target vehicle should remain stationary so that the industrial camera can take clear photos.

[0054] Once the inspection is complete, the target detection results are promptly transmitted to the cloud server for storage. The cloud server boasts powerful data storage and computing capabilities, ensuring data security and accessibility. It also supports remote access and data analysis, providing strong support for subsequent vehicle management and maintenance. Furthermore, LED screens are installed inside the inspection room or at the exit to display the target vehicle's VIN, target detection results, and vehicle exterior damage inspection results in real time. By viewing the information on the LED screen, staff can quickly understand the target vehicle's exterior damage inspection status.

[0055] Finally, when the target vehicle completes the exterior damage inspection and is ready to leave, the sensor door opens quickly and smoothly, allowing the vehicle to exit the inspection room. Based on the exterior damage inspection results, qualified and damaged vehicles are directed to separate areas for further processing. Qualified vehicles are sent to the qualified vehicle warehouse for storage, while damaged vehicles are sent to the damaged vehicle warehouse for storage and, depending on the specific situation, repair or other treatment. This process ensures controllable and traceable quality of the target vehicles.

[0056] Based on the above steps S10 to S14, an image acquisition component and a light source set in a preset enclosed space are used to obtain an initial image data set of the target vehicle's appearance from multiple shooting angles, and the initial image data set is input into a target detection model for detection to obtain a target detection result. Finally, a vehicle appearance damage detection result is generated based on the target detection result. This achieves the purpose of quickly and accurately identifying whether the vehicle's appearance is damaged, thereby realizing the technical effect of improving the accuracy of the vehicle appearance damage detection method, and further solving the technical problem of low efficiency and accuracy of the vehicle appearance damage detection method in the related art.

[0057] Optionally, in step S12, the target detection model further includes a third network, and the initial image dataset is input into the target detection model for detection, and the target detection result obtained includes:

[0058] Step S121, preprocessing the initial image dataset to obtain a target image dataset;

[0059] Step S122, using the first network to perform feature extraction on the target image dataset to obtain first feature information corresponding to the initial image dataset;

[0060] Step S123, using the second network to perform fusion processing on the first feature information to obtain target feature information;

[0061] Step S124: Detect target feature information based on the third network to obtain a target detection result.

[0062] In step S124, the third network is used to represent the head network of the object detection model. The head network is located at the end of the object detection model and receives the feature map from the second network. It is responsible for predicting the category and bounding box of the object in the initial image dataset.

[0063] Specifically, the acquired initial image data of the target vehicle's appearance is preprocessed by expansion and adaptive scaling to enable the initial image data to more fully reflect the appearance quality damage of the target vehicle. A first network is used to extract features from the image to obtain first feature information. A second network is used to fuse the first feature information to obtain target feature information. Feature Pyramid Networks (FPN) and Path Aggregation Networks (PAN) are used to fuse feature maps of different scales, improving the target detection model's ability to detect targets of different sizes. A third network receives the target feature information from the second network and, based on the target feature information, predicts the category and bounding box of the target in the initial image dataset to obtain the target detection result.

[0064] Based on the above steps S121 to S124, the initial image data set is preprocessed to obtain the target image data set; the first network is used to extract features of the target image data set to obtain first feature information corresponding to the initial image data set; the second network is used to fuse the first feature information to obtain target feature information; the target feature information is detected based on the third network to obtain target detection results, thereby achieving the purpose of quickly and accurately identifying whether the vehicle appearance is damaged, thereby achieving the technical effect of improving the accuracy of the vehicle appearance damage detection method, and further solving the technical problem of low efficiency and accuracy of the vehicle appearance damage detection method in the related art.

[0065] Optionally, in step S122, the first network includes at least a focus module and a bottleneck module, and the first network is used to perform feature extraction on the target image dataset to obtain first feature information corresponding to the initial image dataset, including:

[0066] Step S1221, inputting the target image data set into the focus module;

[0067] In step S1221, the focus module is used to represent the Focus() layer. As the input layer, the Focus module directly reduces the spatial resolution of the target image dataset through dimensionality reduction, thereby reducing the computational effort of subsequent convolutional layers. This approach improves processing speed while preserving key information from the target image dataset.

[0068] Step S1222: using the focus module to perform dimensionality reduction processing on the target image dataset to obtain second feature information;

[0069] Step S1223: Using the bottleneck module to perform feature extraction on the second feature information to obtain the first feature information.

[0070] In step S1223, the bottleneck module is used to characterize the image composed of several cross-stage partial connection layers and spatial pyramid pooling layers (Spatial Pyramid Pooling, SSP). Inside the module, the number of channels is first reduced by convolution to form a bottleneck, and then feature extraction is performed through a series of convolution layers, and then the original number of channels is restored. The cross-stage partial connection layer and SSP() layer inside the bottleneck module can further extract the deep features of the image while maintaining the feature richness. In addition, the module also applies the cross-stage partial network (Cross Stage Partial Networks, CSP) strategy, which divides the input second feature information into two parts for processing and then merges them to obtain the first feature information, which can enhance the learning ability of the first network, because each part can learn different feature representations, and the merged feature information is richer and more comprehensive.

[0071] Based on the above steps S1221 to S1223, the target image dataset is input into the focus module; the focus module is used to perform dimensionality reduction processing on the target image dataset to obtain second feature information; the bottleneck module is used to perform feature extraction on the second feature information to obtain first feature information, which can improve the processing speed while retaining the key information of the target image dataset.

[0072] Optionally, in step S1223, the bottleneck module includes at least a cross-stage partial connection layer and a spatial pyramid pooling layer. The bottleneck module is used to perform feature extraction on the second feature information to obtain the first feature information including:

[0073] Step S12231, using the first cross-stage partial connection layer to perform feature extraction on the second feature information to obtain first sub-feature information;

[0074] In step S12331, the first cross-stage partially connected layer includes a bottleneck cross-stage partially connected layer (BottleneckCSP()) and a convolutional layer (Conv()).

[0075] Specifically, the target image dataset is input into the Focus() layer, the output of the Focus() layer is input into the Conv() layer, the output of the Conv() layer is input into the BottleneckCSP() layer, the output of the BottleneckCSP() layer is input into the Conv() layer in the first cross-stage partially connected layer, the output of this layer is input into the BottleneckCSP() layer in the first cross-stage partially connected layer, and the output of this layer is used as the first sub-feature information.

[0076] Step S12232: extract features from the first sub-feature information using the second cross-stage partial connection layer to obtain second sub-feature information;

[0077] In step S12332, the second cross-stage partial connection layer includes a BottleneckCSP() layer and a Conv() layer.

[0078] Specifically, the first sub-feature information is input to the Conv() layer in the second cross-stage partially connected layer, the output of this layer is input to the BottleneckCSP() layer in the second cross-stage partially connected layer, and the output of this layer is used as the second sub-feature information.

[0079] Step S12233, using a spatial pyramid pooling layer to convert the second character feature information into a preset length to obtain third sub-feature information;

[0080] Specifically, the second sub-feature information is input to the Conv() layer, whose output is input to the SPP() layer. The SPP() layer converts the second sub-feature information to a preset length, and the output of the SPP() layer serves as the third sub-feature information. The SPP() layer enables the first network to process input features of different scales and fuses multi-scale feature information by performing pooling operations at multiple scales. This multi-scale feature fusion helps the first network better cope with the large size variations in object detection tasks.

[0081] Step S12234: Use the third cross-stage partially connected layer to perform feature extraction on the third sub-feature information to obtain fourth sub-feature information.

[0082] In step S12334, the third cross-stage partial connection layer includes a BottleneckCSP() layer.

[0083] Specifically, the third sub-feature information is input to the BottleneckCSP() layer in the third cross-stage partial connection layer, and the output of the layer is used as the fourth sub-feature information.

[0084] Based on the above steps S12231 to S12234, the first cross-stage partially connected layer is used to perform feature extraction on the second feature information to obtain first sub-feature information; the second cross-stage partially connected layer is used to perform feature extraction on the first sub-feature information to obtain second sub-feature information; the spatial pyramid pooling layer is used to convert the second word feature information into a preset length to obtain third sub-feature information; the third cross-stage partially connected layer is used to perform feature extraction on the third sub-feature information to obtain fourth sub-feature information. The Conv() layer is used to perform basic convolution operations to extract local features, while the BottleneckCSP() layer further refines features through its special structure, reducing the amount of calculation while maintaining efficient feature extraction capabilities.

[0085] Optionally, in step S123, the second network includes at least an upsampling module, a feature fusion module, and a path aggregation structure. The second network is used to perform fusion processing on the feature information to obtain target feature information including:

[0086] Step S1231: using an upsampling module to sample and process the fourth sub-feature information to obtain third feature information;

[0087] In step S1231, the above-mentioned upsampling module is used to represent the Upsample() layer.

[0088] Step S1232: using a feature fusion module to fuse the third feature information with the second sub-feature information to obtain fourth feature information;

[0089] In step S1232, the feature fusion module is used to characterize the concatenation layer (Concat()).

[0090] Specifically, the fourth sub-feature information is input into the Conv() layer, the output of which is input into the Upsample() layer, and the output of which is used as the third feature information. The third feature information and the second sub-feature information are input into the Concat() layer, the output of which is input into the BottleneckCSP() layer, the output of which is input into the Conv() layer, and the output of which is used as the fourth feature information. The fourth sub-feature information is upsampled by the Upsample() layer so that it is consistent with the second sub-feature information in spatial resolution, thereby enabling feature fusion. This multi-scale feature fusion method helps the second network capture target information of different scales and improve the accuracy of target detection.

[0091] Step S1233: using a feature fusion module to fuse the fourth feature information with the first sub-feature information to obtain fifth feature information;

[0092] Specifically, the fourth feature information is input to the Upsample() layer, the output of the Upsample() layer and the first sub-feature information are input to the Concat() layer, and the output of the Concat() layer is used as the fifth feature information.

[0093] Step S1234: extract the fifth feature information based on the path aggregation structure to obtain target feature information.

[0094] In step S1234, the path aggregation structure is a Path Aggregation Network (PAN) structure. Through bottom-up feature transfer and top-down feature fusion, the expressive power of the feature graph is further enhanced. This structure enables the model to fully utilize feature information at different levels and effectively detect objects of different sizes. Furthermore, the PAN structure reduces information loss, improves the detection performance of the second network, and enhances computational efficiency.

[0095] Specifically, the fifth feature information is extracted using the PAN structure to obtain target feature information. Figure 3 Schematic diagram of another vehicle appearance quality damage detection method according to one embodiment of the present invention, as shown in FIG. Figure 3 As shown in FIG, as an optional embodiment, in the first network, the target image data set is input to the Focus (3, 64, 1, 1) layer, the output of the Focus (3, 64, 1, 1) layer is input to the Conv (64, 128, 3, 2) layer, the output of the Conv (64, 128, 3, 2) layer is input to the BottleneckCSP (128, 128) x3 layer, the output of the BottleneckCSP (128, 128) x3 layer is input to the Conv (128, 256, 3, 2) layer, the output of the Conv (128, 256, 3, 2) layer is input to the BottleneckCSP (256, 256) x9 layer, the output of the BottleneckCSP (256, 256) x9 layer is used as the first sub-feature information, and the first sub-feature information is used as the first sub-feature information. The feature information is input to the Conv(256,512,3,2) layer, the output of the Conv(256,512,3,2) layer is input to the BottleneckCSP(512,512)x9 layer, the output of the BottleneckCSP(512,512)x9 layer is used as the second sub-feature information, the second sub-feature information is input to the Conv(512,1024,3,2) layer, the output of the Conv(512,1024,3,2) layer is input to the SPP(1024) layer, the output of the SPP(1024) layer is used as the third sub-feature information, the third sub-feature information is input to the BottleneckCSP(1024,1024)x3 layer, and the output of the BottleneckCSP(1024,1024)x3 layer is used as the fourth sub-feature information.

[0096] In the second network, the fourth sub-feature information is input to the Conv(1024,512,1,1) layer, the output of the Conv(1024,512,1,1) layer is input to the Upsample(512) layer, and the output of the Upsample(512) layer is used as the third feature information. The third feature information and the second sub-feature information are input to the Concat(1024) layer, the output of the Concat(1024) layer is input to the BottleneckCSP(1024,512)x3 layer, the output of the BottleneckCSP(1024,512)x3 layer is input to the Conv(512,256,1,1) layer, and the output of the Conv(512,256,1,1) layer is used as the fourth feature information. The fourth feature information is input to the Upsample(256) layer. The output of the Upsample(256) layer and the first sub-feature information are input to the Concat(512) layer. The output of the Concat(512) layer is used as the fifth feature information. The fifth feature information is input to the BottleneckCSP(512,256)x3 layer. On the one hand, the output of the BottleneckCSP(512,256)x3 layer is input to the Conv2d(256,na*(nc+5),1,1) layer of the third network for prediction. On the other hand, the output of the BottleneckCSP(512,256)x3 layer is input to the Conv(256,256,3,2) layer. The output of the Conv(256,256,3,2) layer and the output of the Conv(512,256,1,1) layer are input to the Concat(512) layer, and the output of the Concat(512) layer is input to the BottleneckCSP(512,512)x3 layer. On the one hand, the output of the BottleneckCSP(512,512)x3 layer is input to the Conv2d(512,na*(nc+5),1,1) layer of the third network for prediction. On the other hand, the output of the BottleneckCSP(512,512)x3 layer is input to the Conv2d(512,na*(nc+5),1,1) layer of the third network for prediction. The output of the layer is input to the Conv(512,512,3,2) layer, the output of the Conv(512,512,3,2) layer and the output of the Conv(1024,512,1,1) layer are input to the Concat(1024) layer, the output of the Concat(1024) layer is input to the BottleneckCSP(1024,1024)x3 layer, and the output of the BottleneckCSP(1024,1024))x3 layer is input to the Conv2d(1024,na*(nc+5),1,1) layer of the third network for prediction.

[0097] In the third network, the Conv2d(256,na*(nc+5),1,1) layer, Conv2d(512,na*(nc+5),1,1) layer, and Conv2d(1024,na*(nc+5),1,1) layer are used to make predictions and obtain the target detection results.

[0098] It should be noted that the present invention does not limit the parameters of the Focus() layer, BottleneckCSP() layer, Conv() layer, Upsample() layer, Concat() layer, and Conv2d() layer.

[0099] Based on the above steps S1231 to S1234, the fourth sub-feature information is sampled and processed using an upsampling module to obtain third feature information; the third feature information is fused with the second sub-feature information using a feature fusion module to obtain fourth feature information; the fourth feature information is fused with the first sub-feature information using a feature fusion module to obtain fifth feature information; and feature extraction is performed on the fifth feature information based on the path aggregation structure to obtain target feature information. This can fuse feature maps of different scales and improve the target detection model's ability to detect targets of different sizes.

[0100] Optionally, in step S124, the third network includes a convolution module, and the target feature information is detected based on the third network, and the target detection result obtained includes:

[0101] Step S1241: Input the target feature information into the convolution module for prediction to obtain candidate anchor box information;

[0102] Step S1242: weighted matching is performed on the candidate anchor box information to obtain a matching result;

[0103] Step S1243: Filter out target detection results from the candidate anchor box information based on the matching results.

[0104] Specifically, in the third network, target feature information is input into the convolution module for prediction, generating multiple candidate anchor boxes. These candidate anchor boxes are preliminary predictions of areas in the initial image dataset that may contain targets. The non-maximum suppression (NMS) algorithm traverses all bounding boxes. For each candidate anchor box, the degree of overlap with other candidate anchor boxes is calculated, namely the generalized intersection over union (GIoU). After obtaining the GIoU, the candidate anchor boxes are weighted to further optimize the detection results. The weighted matching process involves multiple factors, such as the confidence of the anchor box, the GIoU value between the anchor box and the ground-truth box, and the size and shape of the anchor box. Through weighted matching, the importance of each candidate anchor box can be more accurately assessed, thereby providing more valuable information for subsequent object detection. After completing the weighted matching, the third network will filter the candidate anchor boxes to obtain the final object detection result based on the matching results. During this process, a threshold is set, and only anchor boxes with weighted matching results above the threshold are retained as detection results. The final output detection results will include the category and location of the vehicle's exterior damage.

[0105] Based on the above steps S1241 to S1243, the target feature information is input into the convolution module for prediction to obtain candidate anchor box information; the candidate anchor box information is weighted and matched to obtain the matching result; based on the matching result, the target detection result is filtered out from the candidate anchor box information. The weighted matching strategy takes into account multiple factors, so that the target detection model can work more stably when facing complex scenes, thereby improving the robustness of detection.

[0106] Optionally, in step S121, preprocessing the initial image dataset to obtain the target image dataset includes:

[0107] Step S121, expanding the initial image dataset using a mosaic data enhancement method to obtain a first image dataset;

[0108] In step S121, in the above-mentioned mosaic data enhancement method, the data of four pictures are calculated simultaneously, and the four pictures are flipped, scaled, and color gamut changed respectively. On this basis, they are arranged in four directions to obtain a first image data set, thereby enriching the detection background.

[0109] Step S1212: Perform adaptive scaling processing on the samples in the first image dataset to obtain a target image dataset.

[0110] Specifically, the minimum black edges are adaptively added to the samples in the first image data set. Therefore, since the black edges at both ends of the image height are reduced, the amount of calculation will be reduced during inference, and the speed of target detection will be improved.

[0111] Based on the above steps S1211 to S1212, the initial image dataset is expanded using a mosaic data enhancement method to obtain a first image dataset; the samples in the first image dataset are adaptively scaled to obtain a target image dataset, thereby enriching the diversity of the detection dataset and improving the target detection speed.

[0112] Optionally, in step S14, generating a vehicle appearance quality damage detection result based on the target detection result includes:

[0113] Step S141, analyzing the target detection result to obtain an analysis result;

[0114] Step S142 , in response to the analysis result indicating that the target vehicle is damaged, the vehicle appearance damage inspection result indicates that the target vehicle has failed the vehicle appearance damage inspection.

[0115] Specifically, the target detection results are analyzed based on whether the target vehicle is damaged and the type of damage. If the analysis results indicate damage to the target vehicle, the vehicle appearance damage inspection result indicates that the target vehicle has failed the vehicle appearance damage inspection, and repairs or other treatments are performed based on the target detection results. If the analysis results indicate no damage to the target vehicle, the vehicle appearance damage inspection result indicates that the target vehicle has passed the vehicle appearance damage inspection.

[0116] Based on the above steps S141 to S142, the target detection results are analyzed to obtain analysis results; in response to the analysis results indicating that the target vehicle is damaged, the vehicle appearance damage inspection result indicates that the target vehicle has failed the vehicle appearance damage inspection. Through the automated target detection algorithm, the damage to the vehicle appearance can be quickly and accurately identified, reducing the subjectivity and time cost of manual inspection, and can quickly obtain a conclusion on whether the vehicle has passed the appearance damage inspection, thereby improving the inspection efficiency.

[0117] Optionally, in step S10, acquiring an initial image dataset of the target vehicle appearance from multiple shooting angles includes:

[0118] Step S101, in response to a target vehicle entering a damage inspection area, obtaining unique identifier information of the target vehicle;

[0119] In step S101 , the unique identifier information of the target vehicle is used to represent the VIN code of the target vehicle.

[0120] Step S102, comparing the unique identifier information with the vehicle information in the preset vehicle information list to obtain a comparison result;

[0121] Step S103 : in response to the comparison result indicating that the unique identifier information passes the comparison, an initial image dataset of the appearance of the target vehicle is acquired from multiple shooting angles.

[0122] Specifically, when the target vehicle enters the quality damage inspection area, the reader will identify the RFID tag on the target vehicle and obtain the VIN code of the target vehicle. The obtained VIN code is compared with the vehicle information in the existing vehicle information list. If the VIN code matches the vehicle information in the existing vehicle information list, it means that the VIN code of the target vehicle has passed the comparison, and the shooting equipment can obtain the initial image data set of the appearance of the target vehicle from multiple shooting angles. If the reader does not identify the RFID tag on the target vehicle, or the VIN code of the target vehicle does not match the vehicle information in the existing vehicle information list, it means that the vehicle is an abnormal vehicle, and the information of the vehicle is stored in the vehicle information list and marked as an abnormal vehicle. At the same time, the information is displayed on the LED screen to remind the staff to take the next instruction.

[0123] Based on the above steps S101 to S103, in response to the target vehicle entering the quality damage inspection area, the unique identifier information of the target vehicle is obtained; the unique identifier information is compared with the vehicle information in the preset vehicle information list to obtain a comparison result; in response to the comparison result indicating that the unique identifier information passes the comparison, an initial image data set of the target vehicle's appearance is obtained from multiple shooting angles, and the read VIN code is automatically compared with the preset vehicle information list, which saves time for manual search and confirmation of vehicle information and improves inspection efficiency. Moreover, once the comparison is passed, the shooting equipment can immediately shoot the vehicle appearance from multiple angles without waiting for manual confirmation, thereby speeding up the entire inspection process.

[0124] Figure 4 FIG. 1 is a schematic diagram of another method for detecting vehicle appearance damage according to one embodiment of the present invention. Figure 4 As shown, the method includes the following steps: First, log in to the vehicle exterior damage inspection system and activate the equipment used in the damage inspection process. The target vehicle is controlled to drive from the vehicle waiting area to the inspection room. As the target vehicle enters the inspection room, it passes through the sensing area of ​​the induction door. When the target vehicle approaches the induction door, sensors on the ground or on the door detect the target vehicle's presence and trigger the door control system. The induction door opens quickly and smoothly, allowing the target vehicle to enter the inspection room. This process is typically automated and requires no human intervention.

[0125] Next, the target vehicle enters the inspection room and continues driving until it is parked in a designated inspection area. A light barrier is used to verify the vehicle's correct parking position. Once parked in the designated inspection area, a reader reads the information from the vehicle's RFID tag. If the reader detects the RFID tag, the vehicle is deemed normal. The vehicle's VIN number is then retrieved from the RFID tag and displayed on an LED screen within the inspection room. The retrieved VIN number is then compared with the vehicle information in the existing vehicle information list. If the VIN number matches the existing vehicle information list, the comparison is successful, and the camera captures an initial image dataset of the vehicle's exterior from multiple angles. If the reader fails to recognize the RFID tag, or if the retrieved VIN number does not match the existing vehicle information list, the vehicle is deemed abnormal. The vehicle's information is stored in the vehicle information list and marked as abnormal. This information is also displayed on the LED screen, prompting staff to proceed with further instructions. With the consent of the staff, the shooting equipment can obtain an initial image dataset of the target vehicle's appearance from multiple shooting angles.

[0126] The camera then takes 360° images of the target vehicle, obtaining an initial image dataset of its exterior. This dataset is then stored in the damage inspection system, where it is inspected using the target detection model within the damage inspection system. The inspection process is displayed in real time on an LED screen. The target detection model automatically identifies subtle defects, such as scratches and dents, on the target vehicle's exterior and generates target detection results. This process is fast and efficient, improving inspection accuracy and efficiency. During the image capture and analysis process, the target vehicle must remain stationary to allow the industrial camera to capture clear images.

[0127] Finally, a vehicle exterior damage inspection report is generated based on the target inspection results. When the target vehicle completes the exterior damage inspection and is ready to leave, an indicator light illuminates, indicating that the target vehicle is about to exit the scanning area. The target vehicle then exits. The sensor door opens quickly and smoothly, allowing the target vehicle to exit the inspection room. Once the target vehicle leaves the inspection room, the automatic door closes.

[0128] Based on the above embodiments, an image acquisition component and a light source set in a preset enclosed space are used to obtain an initial image data set of the target vehicle's appearance from multiple shooting angles, and the initial image data set is input into a target detection model for detection to obtain a target detection result. Finally, a vehicle appearance damage detection result is generated based on the target detection result. This achieves the purpose of quickly and accurately identifying whether the vehicle's appearance is damaged, thereby realizing the technical effect of improving the accuracy of the vehicle appearance damage detection method, and further solving the technical problem of low efficiency and accuracy of the vehicle appearance damage detection method in the related art.

[0129] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0130] The present invention also provides a vehicle exterior damage detection device for implementing the aforementioned embodiments and preferred implementations. Details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. While the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0131] Figure 5 FIG. 1 is a structural block diagram of a vehicle appearance quality damage detection device according to one embodiment of the present invention. Figure 5 As shown, the device includes:

[0132] An acquisition module 501 is configured to acquire an initial image dataset of the exterior of a target vehicle from multiple shooting angles using an image acquisition component and a light source provided in a preset enclosed space;

[0133] The detection module 502 inputs the initial image dataset into the target detection model for detection to obtain target detection results, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image dataset, and the second network is used to fuse the first feature information;

[0134] The generation module 503 generates a vehicle appearance quality damage detection result based on the target detection result.

[0135] Optionally, the detection module 502 is also used to: preprocess the initial image data set to obtain a target image data set; use the first network to extract features of the target image data set to obtain first feature information corresponding to the initial image data set; use the second network to fuse the first feature information to obtain target feature information; and detect the target feature information based on the third network to obtain a target detection result.

[0136] Optionally, the detection module 502 is further used to: input the target image data set into the focus module; use the focus module to perform dimensionality reduction processing on the target image data set to obtain second feature information; use the bottleneck module to perform feature extraction on the second feature information to obtain first feature information.

[0137] Optionally, the detection module 502 is further configured to: perform feature extraction on the second feature information using a first cross-stage partially connected layer to obtain first sub-feature information; perform feature extraction on the first sub-feature information using a second cross-stage partially connected layer to obtain second sub-feature information; convert the second word feature information into a preset length using a spatial pyramid pooling layer to obtain third sub-feature information; and perform feature extraction on the third sub-feature information using a third cross-stage partially connected layer to obtain fourth sub-feature information.

[0138] Optionally, the detection module 502 is also used to: use the upsampling module to sample and process the fourth sub-feature information to obtain third feature information; use the feature fusion module to fuse the third feature information with the second sub-feature information to obtain fourth feature information; use the feature fusion module to fuse the fourth feature information with the first sub-feature information to obtain fifth feature information; and perform feature extraction on the fifth feature information based on the path aggregation structure to obtain target feature information.

[0139] Optionally, the acquisition module 501 is also used to input the target feature information into the convolution module for prediction to obtain candidate anchor frame information; the detection module 502 is also used to perform weighted matching on the candidate anchor frame information to obtain a matching result; the vehicle appearance quality damage detection device also includes a processing module 504, which is used to filter out target detection results from the candidate anchor frame information based on the matching result.

[0140] Optionally, the processing module 504 is further configured to: expand the initial image dataset using a mosaic data enhancement method to obtain a first image dataset; and perform adaptive scaling processing on samples in the first image dataset to obtain a target image dataset.

[0141] Optionally, the generation module 503 is further used to: analyze the target detection result to obtain an analysis result; in response to the analysis result indicating that the target vehicle is damaged, the vehicle appearance damage inspection result indicates that the target vehicle has failed the vehicle appearance damage inspection.

[0142] Optionally, the acquisition module 501 is also used to obtain the unique identifier information of the target vehicle in response to the target vehicle entering the quality damage inspection area; the processing module 504 is also used to compare the unique identifier information with the vehicle information in the preset vehicle information list to obtain a comparison result; the acquisition module 501 is also used to obtain an initial image data set of the appearance of the target vehicle from multiple shooting angles in response to the comparison result indicating that the unique identifier information has passed the comparison.

[0143] It should be noted that the above modules can be implemented through software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0144] According to one embodiment of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the above-mentioned vehicle appearance quality damage detection method is executed when the program is running.

[0145] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0146] Step S1, using an image acquisition component and a light source set in a preset enclosed space to obtain an initial image dataset of the target vehicle's exterior from multiple shooting angles;

[0147] Step S2: inputting the initial image dataset into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image dataset, and the second network is used to fuse the first feature information;

[0148] Step S3: generating a vehicle appearance quality damage detection result based on the target detection result.

[0149] According to one embodiment of the present invention, a computer-readable storage medium is further provided, which includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the above-mentioned vehicle appearance quality damage detection method.

[0150] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0151] Step S1, using an image acquisition component and a light source set in a preset enclosed space to obtain an initial image dataset of the target vehicle's exterior from multiple shooting angles;

[0152] Step S2: inputting the initial image dataset into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image dataset, and the second network is used to fuse the first feature information;

[0153] Step S3: generating a vehicle appearance quality damage detection result based on the target detection result.

[0154] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0155] According to one embodiment of the present invention, a computer program product is also provided, including a computer program, which implements the above-mentioned vehicle appearance quality damage detection method when executed by a processor.

[0156] Optionally, in this embodiment, the computer program product may be configured as a computer program for executing the following steps:

[0157] Step S1, using an image acquisition component and a light source set in a preset enclosed space to obtain an initial image dataset of the target vehicle's exterior from multiple shooting angles;

[0158] Step S2: inputting the initial image dataset into a target detection model for detection to obtain a target detection result, wherein the target detection model includes at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image dataset, and the second network is used to fuse the first feature information;

[0159] Step S3: generating a vehicle appearance quality damage detection result based on the target detection result.

[0160] Optionally, specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0161] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0162] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0163] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0164] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0165] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0166] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0167] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A vehicle appearance quality damage detection method, characterized in that: include: Using the image acquisition components and light sources set in a preset confined space, an initial image data set of the target vehicle's appearance is acquired from multiple shooting angles; Inputting the initial image data set into a target detection model for detection to obtain a target detection result, wherein the target detection model comprises at least a first network and a second network, the first network is used to extract first feature information corresponding to the initial image data set, and the second network is used to fuse the first feature information; A vehicle appearance quality damage detection result is generated based on the target detection result.

2. The vehicle appearance quality damage detection method according to claim 1, characterized in that: The target detection model further includes a third network, and the initial image data set is input into the target detection model for detection, and the target detection result obtained includes: Preprocessing the initial image data set to obtain a target image data set; Using the first network to perform feature extraction on the target image data set to obtain first feature information corresponding to the initial image data set; Using the second network to perform fusion processing on the first feature information to obtain target feature information; The target feature information is detected based on the third network to obtain the target detection result.

3. The vehicle appearance quality damage detection method according to claim 1, characterized in that: The first network at least includes a focus module and a bottleneck module. The first network is used to extract features from the target image data set, and the first feature information corresponding to the initial image data set is obtained, including: inputting the target image data set into the focus module; Using the focus module to perform dimensionality reduction processing on the target image data set to obtain second feature information; The bottleneck module is used to perform feature extraction on the second feature information to obtain the first feature information.

4. The vehicle appearance quality damage detection method according to claim 3, characterized in that: The bottleneck module at least includes a cross-stage partial connection layer and a spatial pyramid pooling layer. The bottleneck module is used to extract features from the second feature information to obtain the first feature information, including: Using the first cross-stage partial connection layer to perform feature extraction on the second feature information to obtain first sub-feature information; Using a second cross-stage partial connection layer to perform feature extraction on the first sub-feature information to obtain second sub-feature information; The second character feature information is converted into a preset length by using the spatial pyramid pooling layer to obtain third sub-feature information; The third cross-stage partially connected layer is used to perform feature extraction on the third sub-feature information to obtain fourth sub-feature information.

5. The vehicle appearance quality damage detection method according to claim 4, characterized in that: The second network at least includes an upsampling module, a feature fusion module and a path aggregation structure. The feature information is fused using the second network to obtain target feature information including: Using the up-sampling module to sample and process the fourth sub-feature information to obtain third feature information; Using the feature fusion module to fuse the third feature information with the second sub-feature information to obtain fourth feature information; Using the feature fusion module to fuse the fourth feature information with the first sub-feature information to obtain fifth feature information; Feature extraction is performed on the fifth feature information based on the path aggregation structure to obtain the target feature information.

6. The vehicle appearance quality damage detection method according to claim 2, characterized in that: The third network includes a convolution module, and the target feature information is detected based on the third network to obtain the target detection result, including: Inputting the target feature information into the convolution module for prediction to obtain candidate anchor box information; Performing weighted matching on the candidate anchor box information to obtain a matching result; The target detection result is filtered out from the candidate anchor box information based on the matching result.

7. The vehicle appearance quality damage detection method according to claim 2, characterized in that: Preprocessing the initial image data set to obtain the target image data set includes: Expanding the initial image data set by using a mosaic data enhancement method to obtain a first image data set; Adaptively scale the samples in the first image data set to obtain the target image data set.

8. The vehicle appearance quality damage detection method according to claim 1, characterized in that: Generating a vehicle appearance quality damage detection result based on the target detection result includes: Analyzing the target detection result to obtain an analysis result; In response to the analysis result indicating that the target vehicle is damaged, the vehicle appearance quality damage inspection result indicates that the target vehicle has failed the vehicle appearance quality damage inspection.

9. The vehicle appearance quality damage detection method according to claim 1, characterized in that: The initial image dataset for obtaining the target vehicle's appearance from multiple shooting angles includes: In response to the target vehicle entering a quality damage inspection area, acquiring unique identifier information of the target vehicle; Comparing the unique identifier information with the vehicle information in a preset vehicle information list to obtain a comparison result; In response to the comparison result indicating that the unique identifier information passes the comparison, an initial image data set of the appearance of the target vehicle is acquired from multiple shooting angles.

10. A vehicle appearance quality damage detection device, characterized in that: include: An acquisition module is used to acquire an initial image data set of the appearance of the target vehicle from multiple shooting angles using an image acquisition component and a light source set in a preset confined space; A detection module, inputting the initial image data set into a target detection model for detection to obtain a target detection result, wherein the target detection model at least includes a first network and a second network, the first network is used to extract first feature information corresponding to the initial image data set, and the second network is used to fuse the first feature information; A generation module generates a vehicle appearance quality damage detection result based on the target detection result.

11. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the vehicle appearance quality damage detection method described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the vehicle appearance quality damage detection method according to any one of claims 1 to 9.

13. A computer program product, characterized in that It comprises a computer program, which, when executed by a processor, implements the vehicle appearance quality damage detection method according to any one of claims 1 to 9.

Citation Information

Patent Citations

  • Vehicle damage detection method, device and equipment and storage medium

    CN116452802A

  • Vehicle body paint surface defect rapid detection method, image processing equipment and readable medium

    CN116883313A

  • Auxiliary method for capturing damage assessment image of vehicle, device, and apparatus

    WO2020042800A1

  • Work vehicle detection and tracking method and system

    WO2023065395A1