Automobile part gluing glue path visual detection method and system

Abstract

The application belongs to the technical field of visual detection, and discloses a kind of automobile parts glue road visual detection method and system, by obtaining the overview image containing glue road and carrying out preliminary analysis, dynamically determine the applicable image acquisition parameter set and applicable detection rule for different regions according to the analysis result, and carry out local special camera image acquisition and detection based on the determined parameter set, so as to solve the detection area inaccuracy caused by part pose deviation under high-speed production rhythm and the differentiated demand of different regions or defect types on imaging and detection parameters, and achieve the effect of stable, accurate and comprehensive detection of glue road.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visual detection, in particular to a visual detection method and system for glue path of automobile parts. BACKGROUND

[0002] In modern automobile manufacturing high-speed production lines, the quality detection of key automobile parts after the glue application operation is of great importance. After the glue application, these parts will quickly enter the automated visual detection station. This station is usually equipped with high-speed industrial cameras, programmable light source systems, and high-performance industrial computers, which are used to quickly capture images of the glue application area and evaluate the macro-geometric characteristics (such as width, position, continuity, shape) and micro-defects (such as bubbles, shrinkage, impurities, scratches) of the glue path through complex image analysis algorithms. In order to ensure the stability and accuracy of the detection, the parameters of the camera, light source, algorithm, etc. of the visual system need to be accurately set.

[0003] However, in the actual high-speed production environment, the extremely short production cycle poses a severe challenge to the efficiency of the visual detection system. At the same time, due to mechanical transmission, part tolerances or slight relative displacement with the fixture, the actual position and posture of the part to be detected within the camera's field of view often deviates slightly and randomly from the pre-set standard pose. This pose deviation may cause the fixed image capture region (ROI) set based on the standard pose to fail to accurately cover the actual glue application area on the current part, or cause the pixel coordinates of the glue path in the image to deviate from the image processing template or reference contour established based on the standard pose, seriously affecting the effectiveness and robustness of the image processing algorithm which relies on accurate region definition and template matching.

[0004] In addition, with the increasing quality requirements of automobile products, the detection of micro-defects with small sizes is increasingly demanded. These micro-defects are extremely sensitive to imaging conditions, and different types of defects exhibit significant differences under different combinations of lighting and camera parameters. For example, a set of parameters optimized for the overall width detection of the glue path may not effectively highlight the slight shrinkage of the glue path edge or the internal micro-bubbles. Different regions of the glue path (such as straight segments, corners, starting points, and ending points) may also be prone to different types of defects, requiring differentiated imaging conditions. In the prior art, the visual system usually uses a fixed set of global parameters for detection, which is difficult to dynamically adjust the detection region according to the actual pose of the part, and cannot flexibly select the optimal local imaging parameters and detection rules according to the characteristics of different regions of the glue path or the potential defect types, resulting in difficulty in ensuring comprehensive and accurate identification of all potential defects under high-speed cycles, affecting the robustness and reliability of the detection.

[0005] In view of the above problems, the prior art needs to be improved. SUMMARY

[0006] The purpose of the present application is to provide a vehicle part glue path visual detection method and system, which can adapt to part pose deviation, dynamically select optimal imaging parameters and detection rules according to local area characteristics, and improve the accuracy and robustness of detection.

[0007] In a first aspect, the present application provides a vehicle part glue path visual detection method, which is used for visual detection of vehicle part glue path based on a visual sensor. The steps of the method include:

[0008] A1. An overview image of a vehicle part to be detected containing a glue path is obtained;

[0009] A2. According to the overview image, the actual pixel range of a plurality of ROI regions set in advance based on a standard pose in the overview image is determined; each ROI region contains a part of the glue path;

[0010] A3. According to the determined actual pixel range, the region image corresponding to each ROI region in the overview image is preliminarily analyzed to obtain image feature data of each ROI region;

[0011] A4. For each ROI region, according to the corresponding image feature data, a suitable image acquisition parameter set and a suitable detection rule are determined from a plurality of pre-set image acquisition parameter sets and a plurality of pre-set detection rules corresponding to the ROI region;

[0012] A5. The suitable image acquisition parameter set of each ROI region is sequentially determined according to the actual pixel range, and the corresponding special camera image is obtained by image acquisition of the vehicle part to be detected; the relative pose of the visual sensor and the vehicle part to be detected is the same when the special camera image and the overview image are acquired;

[0013] A6. For each ROI region, according to the determined actual pixel range, the region image corresponding to the ROI region in the corresponding special camera image is detected for glue path defects by using the corresponding suitable detection rule;

[0014] A7. The glue path defect detection results of each ROI region are summarized to obtain the overall detection result of the vehicle part to be detected.

[0015] In a second aspect, the present application provides a vehicle part glue path visual detection system, which is used for visual detection of vehicle part glue path based on a visual sensor. The system includes an image acquisition device, a positioning table and an industrial computer;

[0016] The positioning table is used for positioning the vehicle part to be detected after gluing;

[0017] The image acquisition device is provided with a visual sensor and an illuminating light source, and is used to acquire images of the automobile part to be detected and send the images to the industrial computer;

[0018] The industrial computer is installed with a visual detection program, which is configured with:

[0019] An overview image acquisition module is used to acquire an overview image of the automobile part to be detected containing a rubber coating path by using the image acquisition device;

[0020] A region mapping module is used to determine the actual pixel range of a plurality of ROI regions set in advance based on a standard pose in the overview image according to the overview image; each ROI region contains a part of the rubber coating path;

[0021] A region feature analysis module is used to preliminarily analyze the region image corresponding to each ROI region in the overview image according to the determined actual pixel range, and acquire image feature data of each ROI region;

[0022] A parameter rule screening module is used to determine, for each ROI region, an applicable image acquisition parameter set and an applicable detection rule from a plurality of preset image acquisition parameter sets and a plurality of preset detection rules corresponding to the ROI region according to the corresponding image feature data;

[0023] A special image acquisition module is used to acquire images of the automobile part to be detected according to the determined applicable image acquisition parameter set for each ROI region in sequence by using the image acquisition device, so as to obtain special images corresponding to each ROI region; the relative pose of the visual sensor is the same as that of the automobile part to be detected when the special images are acquired and the overview image is acquired;

[0024] A detection module is used to perform rubber coating path defect detection on the region image corresponding to each ROI region in the corresponding special image according to the determined actual pixel range and the corresponding applicable detection rule for each ROI region;

[0025] A result summarizing module is used to summarize the rubber coating path defect detection results of each ROI region, so as to obtain the overall detection result of the automobile part to be detected.

[0026] Advantages: The automobile part rubber coating path visual detection method and system provided by the application can determine the actual ROI region by acquiring the overview image, dynamically determine the local optimal image acquisition parameter and detection rule according to the preliminary analysis result of the ROI region, and then perform defect detection by using the special image, which has the advantages of being able to adapt to the part pose deviation, dynamically selecting the optimal imaging parameter and detection rule according to the local region characteristics, and improving the accuracy and robustness of detection. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The flow chart of the automobile part glue coating visual detection method provided by the embodiment of the application is shown.

[0028] Figure 2 The structure schematic diagram of the automobile part glue coating visual detection system provided by the embodiment of the application is shown.

[0029] Figure 3 The function module configuration diagram of the visual detection program is shown.

[0030] Label explanation: 1, image acquisition device; 101, visual sensor; 102, illumination light source; 103, mechanical arm; 104, cloud platform mechanism; 2, positioning table; 3, industrial computer; 301, overview image acquisition module; 302, region mapping module; 303, region feature analysis module; 304, parameter rule screening module; 305, special camera image acquisition module; 306, detection module; 307, result summary module. DETAILED DESCRIPTION

[0031] The technical solutions in the application will be described in detail below with reference to the drawings in the application. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments. The components of the application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the application.

[0032] It should be noted that: similar labels and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the application, the terms "first", "second" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0033] REFERENCE Figure 1 The application provides an automobile part glue coating visual detection method, which is used for visual detection of automobile part glue coating based on a visual sensor. The steps of the method include:

[0034] A1. obtaining an overview image of a to-be-detected automobile part containing glue coating;

[0035] A2. Determine actual pixel ranges of a plurality of ROI regions pre-set based on a standard pose in the overview image according to the overview image; each of the ROI regions contains a part of the glue path;

[0036] A3. Perform preliminary analysis on region images corresponding to each of the ROI regions in the overview image according to the determined actual pixel ranges, to obtain image feature data of each of the ROI regions;

[0037] A4. For each of the ROI regions, determine an applicable image acquisition parameter set and an applicable detection rule from a plurality of pre-set image acquisition parameter sets and a plurality of pre-set detection rules corresponding to the ROI region according to the corresponding image feature data;

[0038] A5. Perform image acquisition on the automobile part to be detected according to each of the determined applicable image acquisition parameter sets in sequence, to obtain special camera images corresponding to each of the ROI regions; the relative pose of the visual sensor to the automobile part to be detected is the same when the special camera images are acquired and when the overview image is acquired;

[0039] A6. For each of the ROI regions, perform glue path defect detection on region images corresponding to the ROI region in the corresponding special camera image according to the determined actual pixel range and the corresponding applicable detection rule;

[0040] A7. Aggregate glue path defect detection results of each of the ROI regions to obtain an overall detection result of the automobile part to be detected.

[0041] Wherein, determining the actual pixel range of the ROI region refers to adjusting the pre-set standard ROI region to accurate pixel coordinates and size according to the actual position and attitude of the part in the overview image, which can be realized by recognition and positioning technology based on part key structural features, pose estimation technology based on template matching or target detection technology based on deep learning, and the main purpose is to compensate for the influence of part pose deviation on detection region positioning, to ensure that subsequent operations for the ROI region are performed at accurate positions.

[0042] Wherein, performing preliminary analysis on the region image of the ROI region refers to quickly processing and feature extraction on the image content of each ROI region on the overview image, to obtain data reflecting the characteristics or macro features of the region image, which can be realized by calculating the average brightness, contrast, gray histogram, texture features or performing simple edge detection, and the main purpose is to obtain preliminary information reflecting the imaging condition or potential problems of the current region, to provide a basis for subsequent targeted selection of image acquisition parameters and detection rules.

[0043] Wherein, determining the applicable image acquisition parameter set and the applicable detection rule means that according to the image feature data obtained through the preliminary analysis, from the multiple image acquisition parameter combinations and multiple detection algorithms and parameters preset for the ROI region, the combination most suitable for high-precision image acquisition and defect detection of the current region is selected, which can be realized by using a decision tree based on a preset rule, a mapping relationship based on a lookup table or classification and selection based on a machine learning model, and the main purpose is to enable subsequent high-precision image acquisition and defect detection to be self-adaptive and optimized according to the regional characteristics and potential defect types, thereby improving the detection effect.

[0044] Wherein, sequentially acquiring images of the part according to the determined applicable image acquisition parameter set to obtain the special images means that for each ROI region, the optimal image acquisition parameter combination selected for it is used to acquire high-resolution or special imaging condition images of the region by using a visual sensor, which can be realized by adjusting the color, angle, brightness of the light source, adjusting the exposure time, gain, focal length of the camera and / or using different illumination modes, and the main purpose is to highlight the glue path features or potential defects of the ROI region in the best imaging conditions to provide high-quality images for high-precision detection.

[0045] Wherein, using the corresponding applicable detection rule to detect the region image of the corresponding ROI region in the special image for the glue path defect means that the optimal detection algorithm and parameters selected for the ROI region are used to analyze the image content of the corresponding region in the acquired special image, and to judge whether there is a glue path defect, which can be realized by using edge detection-based glue path width measurement, template matching-based glue path position deviation detection, gray or texture analysis-based bubble or impurity detection, or deep learning-based defect classification and positioning, and the main purpose is to use the most effective and robust method for high-precision defect detection according to the characteristics and potential defect types of the region.

[0046] The core innovation of the present application is that by acquiring an overview image containing the glue path and performing preliminary analysis, the applicable image acquisition parameter set and the applicable detection rule for different regions are dynamically determined according to the analysis result, and local special image acquisition and detection are performed based on the determined parameter set, thereby solving the problems of inaccurate detection region caused by part pose deviation under high-speed production rhythm and the differentiated needs of different regions or defect types for imaging and detection parameters, and achieving the effect of stable, accurate and comprehensive detection of the glue path.

[0047] In particular, the method first obtains a global view of the detection area by acquiring an overview image of the automobile part to be detected containing the rubber coating path, laying a foundation for subsequent accurate positioning and region division. Then, using the global information provided by the overview image, the key features on the part are identified through image processing techniques, the pose deviation of the part relative to the standard position is calculated, and the pre-set standard ROI region is accurately mapped to the actual pixel range in the overview image according to the deviation. This ensures that all subsequent operations on the ROI region are based on the actual position of the part, effectively solving the problem of inaccurate detection area caused by pose deviation. Subsequently, a quick preliminary analysis is performed on each ROI region in the overview image that has a determined actual pixel range, and data reflecting the image characteristics or macro features of the region are extracted. These data reflect the current imaging conditions and some macro features of the region, providing a preliminary basis for selecting the optimal image acquisition parameters and detection rules for the region subsequently. Based on the image feature data obtained through preliminary analysis, the system intelligently selects the parameter set and rules most suitable for high-precision image acquisition and defect detection for each ROI region from the pre-set multiple image acquisition parameter sets and detection rules. The selection based on image feature data enables the detection process to adaptively cope with the characteristics of different regions and the types of defects that may occur, solving the problem that traditional methods using fixed parameters cannot meet all detection needs. Then, according to the applicable image acquisition parameter set determined for each ROI region, the part is sequentially imaged to obtain special images for final detection. When acquiring special images, the relative pose between the vision sensor and the part remains consistent with that when acquiring the overview image, simplifying system design and control and avoiding additional complexity or errors introduced by changes in pose, while ensuring that the spatial correspondence between the special images and the overview image is easy to establish. By using different parameter sets, the features of the rubber coating path or potential defects can be highlighted in the best imaging conditions for different regions. After obtaining the special images, for each ROI region, the most suitable detection rules selected for it are used to perform high-precision defect detection on the image content of the corresponding region in the special image. Using the corresponding applicable detection rules for detection ensures that the most effective and robust detection algorithms and parameters are used for the characteristics and potential defect types of the region, improving the accuracy and reliability of defect detection. Finally, the local detection results of all ROI regions are integrated to form the final judgment and report on the overall rubber coating quality of the automobile part to be detected.

[0048] Through the above-described solution, this application effectively addresses the positional deviations of automotive parts on high-speed production lines, ensuring the accuracy of the inspection area. Simultaneously, by conducting preliminary analysis of different areas and adaptively selecting the optimal image acquisition parameters and inspection rules, it can specifically highlight the adhesive path characteristics or potential defects in different areas, improving the detection capability for various defects, especially microscopic defects. Overall, this application achieves stable, accurate, and comprehensive visual inspection of adhesive paths on automotive parts under high-speed production cycles.

[0049] In some implementations, step A1 includes:

[0050] A101. Obtain imaging characteristic information of preset key structural features on the automotive part to be detected; the imaging characteristic information includes imaging performance data of the preset key structural features under different combinations of image acquisition parameters, and the imaging performance data includes contrast, shadow features, specular emission parameters and / or surface texture features;

[0051] A102. Based on the imaging characteristic information, determine the combination of image acquisition parameters that is conducive to the clear presentation of the preset key structural features in the image, and use it as the target image acquisition parameter set;

[0052] A103. Based on the target image acquisition parameter set, acquire an overview image of the automotive part to be detected, including the adhesive coating path;

[0053] A104. The overview image is preprocessed to enhance the contrast and / or edge information of the preset key structural features to obtain the final overview image.

[0054] Among them, the preset key structural features refer to specific geometric shapes, marks or areas on automotive parts used for positioning or identification, which can be achieved by the edges of the parts, holes, bolts, or pre-painted positioning points.

[0055] Imaging characteristic information refers to a set of data describing the visual performance of key structural features under different imaging conditions (i.e., a set of imaging performance data), which can be stored and managed using a database or lookup table. Imaging performance data refers to specific data types, such as contrast, shadow features, specular emission parameters, and surface texture features. This data can be extracted and quantified from acquired images using image processing algorithms. Image acquisition parameter combinations refer to a set of camera and light source parameters that affect image quality. These can include specific configurations of camera exposure time, gain, white balance, and light source parameters such as brightness, color, and illumination angle. The target image acquisition parameter set refers to the optimal parameter combination selected based on the imaging characteristic information; it can be a configuration set containing specific values ​​for camera and light source parameters.

[0056] The preprocessing refers to an enhancement operation on the image, which can be implemented by using image processing techniques such as grayscale, filtering, histogram processing, and / or edge enhancement algorithm.

[0057] The present scheme aims to solve the problem that simple acquisition of the overview image may lead to unclear key structural features, thereby affecting the accuracy of subsequent positioning and preliminary analysis. Specifically, first, the imaging characteristic information of the preset key structural features under different image acquisition parameter combinations is obtained. This provides a basis for the performance data of the key features under various imaging conditions. Then, according to the imaging characteristic information, the image acquisition parameter combination that can make the preset key structural features clearly present in the image is analyzed and determined, and it is set as the target image acquisition parameter set. This process is based on data analysis, avoiding empirical or fixed parameter selection. Then, based on the determined target image acquisition parameter set, the overview image of the automobile part to be detected is acquired. By using the parameters optimized for the key features for acquisition, the clarity of the key structural features in the obtained original overview image is improved. Finally, the acquired overview image is preprocessed to further enhance the contrast and / or edge information of the preset key structural features. The preprocessing step finely adjusts the image, making the key features more prominent at the pixel level. The entire process ensures that the acquired overview image can clearly and reliably present the key structural features for subsequent positioning and analysis through data-driven parameter optimization and image enhancement. This provides a high-quality input image for subsequent determination of the actual pixel range of the ROI region and preliminary analysis of the ROI region, thereby improving the accuracy and robustness of these subsequent steps.

[0058] In one embodiment, acquiring imaging characteristic information can be performed in a system debugging phase. A standard part is placed in a detection position, and at least one of camera exposure, gain, light source brightness, angle, and the like is adjusted through an automated program loop to collect a large number of images. For each image, an image analysis algorithm is run to calculate at least one of local contrast, shadow intensity, highlight area size, surface texture definition, and the like of a key structural feature (for example, a part edge or a positioning hole), and these indicators are recorded in a database in combination with corresponding parameters. Determining a target parameter set can be performed before actual detection or at system initialization. The parameter combination associated with a specific key structural feature that can maximize its contrast or edge definition can be found from the database or calculated through a model. For example, for a certain part, it is found that the edge of its positioning hole is sharpest when the exposure time is 10 ms, the gain is 10 dB, the light source brightness is 80%, the light source color is white, and the illumination angle is 30 degrees, and then this set of parameters is set as the target image acquisition parameter set. When acquiring an overview image based on the target parameter set, at the detection station, after the part is in place, the camera and the light source are configured according to the set target image acquisition parameter set, and then a frame of overview image is triggered to be collected. When preprocessing the overview image, the collected overview image can first be subjected to grayscale processing (if a color image is collected), and then an adaptive histogram equalization algorithm is applied to enhance local contrast, or a non-sharpening mask filter is applied to enhance edge details, to obtain the final overview image for subsequent analysis.

[0059] By acquiring and utilizing the imaging characteristic information of the preset key structural feature to determine the optimized image acquisition parameters and pre-process the collected image, the scheme can ensure that the key structural feature used for positioning and analysis is clearly presented in the overview image. This overcomes the problem that fixed parameters may cause the key feature to be blurred when collecting images, and improves the accuracy and robustness of pose calculation, ROI accurate mapping, and regional preliminary analysis based on the overview image.

[0060] In some embodiments, step A2 comprises:

[0061] A201. In the overview image, a preset key structural feature on the automobile part to be detected is identified;

[0062] A202. According to the identified preset key structural feature, a pose deviation of the automobile part to be detected relative to a preset standard pose is calculated;

[0063] A203. According to the pose deviation, the preset standard coordinate information of a plurality of ROI regions is converted into actual pixel ranges in the overview image.

[0064] Among them, identifying the preset key structural features on the automobile part to be detected refers to locating and determining the preset key structural features on the automobile part to be detected in the obtained overview image through image processing algorithm. The method of identifying these features can adopt a template matching-based technology to determine the position by searching for a similar area to the preset key structural feature template in the overview image; can also adopt a feature point detection and matching-based method, such as SIFT, SURF or ORB algorithm, to extract and match the feature points in the overview image and the standard reference image; or adopt an edge detection and shape matching-based method to identify and match the contour or shape of the key structure. The purpose of identification is to provide a reliable reference benchmark for subsequent calculation of the actual pose of the part.

[0065] Among them, calculating the pose deviation of the automobile part to be detected relative to the preset standard pose according to the identified preset key structural features refers to comparing the actual position and attitude information of the preset key structural features in the overview image with the preset position and attitude information of these key structural features when the part is in the preset standard pose after the actual position and attitude information of the preset key structural features in the overview image are identified. Through this comparison, the translation and rotation deviation of the current automobile part to be detected relative to the ideal standard pose can be quantified. The method of calculating the pose deviation can be selected according to the type and number of the identified key structural features. For example, if a group of feature points are identified, the pose deviation can be calculated by solving the transformation relationship between the point sets, and the commonly used methods include affine transformation or rigid body transformation calculation based on least squares method. If geometric shapes are identified, the pose deviation can be determined by comparing the position and direction difference between the actual shape and the standard shape. The calculated pose deviation usually includes the translation along the image coordinate axis and the rotation angle around an axis, which accurately describes the difference between the actual space position and direction of the part on the current detection station and the standard set position and direction.

[0066] In the method, the preset standard coordinate information of the plurality of ROI regions is converted into actual pixel ranges in the overview image according to the pose deviation, which means that the standard coordinate information of the plurality of ROI regions preset based on the part in the standard pose is corrected by using the calculated pose deviation information of the part to be detected. The preset standard coordinate information is relative to the reference coordinate system when the part is in the standard pose. By applying the calculated displacement and rotation deviation to the standard coordinates, the ROI regions in the standard coordinate system can be accurately mapped to the pixel coordinate system of the overview image with the current pose deviation. This coordinate conversion is usually realized by two-dimensional geometric transformation, for example, by applying an affine transformation matrix or a rigid body transformation matrix containing translation and rotation components to convert the standard coordinate points (for example, the corner point coordinates defining the ROI boundary) of each ROI region to the corresponding pixel coordinates in the overview image. The actual pixel range obtained after conversion can accurately frame the actual glue path region on the part to be detected in the current overview image, even if the part has a certain translation and rotation relative to the standard pose.

[0067] The present application solves the problem of accurately determining the actual pixel range of the preset ROI region in the overview image when the part to be detected has a pose deviation by introducing a pose recognition and correction method based on key structural features. Specifically, first, the preset key structural features on the part to be detected are recognized in the overview image, which serve as a reliable reference for calculating the actual pose of the part. Then, according to the reference information of the recognized key structural features and the standard pose, the pose deviation of the part to be detected relative to the preset standard pose is accurately calculated, quantifying the difference between the actual position and direction of the part and the ideal position and direction. Subsequently, the standard coordinate information of the plurality of ROI regions preset in advance is converted by using the calculated pose deviation, accurately mapping the ROI regions in the standard pose to the current overview image to obtain the actual pixel range that can accurately cover the actual glue path region. This process ensures that the subsequent image analysis and detection can accurately focus on the actual target region. After obtaining the overview image of the part to be detected containing the glue path, the present application can accurately determine the actual position of the ROI region required for subsequent analysis and detection, providing an accurate positioning basis for subsequent preliminary analysis of the image of these regions, obtaining image feature data, and determining the applicable image acquisition parameter set and detection rules based on these data. Accurate ROI positioning enables the preliminary analysis to effectively extract features related to the glue path, and also ensures that the subsequent special image acquisition for a specific ROI region can focus on the correct target region, thereby enabling the final glue path defect detection to be accurately performed in the corresponding special image region. This accurate region positioning capability can be achieved even when the part has a pose deviation, significantly improving the robustness and reliability of the entire glue path visual detection method.

[0068] In some embodiments, step A3 comprises:

[0069] A301. Obtain preset region characteristic information of each ROI region; the preset region characteristic information includes information of glue path geometry shape within the region, information of prone defect type, and information of region detection requirement;

[0070] A302. Determine the image feature type to be extracted for preliminary analysis of each ROI region according to the preset region characteristic information;

[0071] A303. Select an analysis method for preliminary analysis of the region image of each ROI region according to the determined image feature type;

[0072] A304. For each ROI region, based on the selected analysis method, perform preliminary analysis on the region image corresponding to the ROI region in the overview image to obtain image feature data of the ROI region corresponding to the determined image feature type.

[0073] Wherein, the preset region characteristic information refers to specific attribute data about each ROI region pre-configured and stored according to the analysis of process characteristics, potential defect patterns and quality control standards of different ROI regions before the actual detection starts. These information can be stored in configuration files, databases or lookup tables.

[0074] Wherein, the image feature type refers to a quantitative indicator used to describe the visual characteristics of the local or overall image, such as at least one of edge feature, texture feature, color feature, brightness distribution feature, shape feature, etc. Determining these types can be based on preset rules, expert knowledge base or machine learning model.

[0075] Wherein, the analysis method refers to an image processing algorithm or technique used to extract specific image features from image data, such as edge detection algorithm, texture analysis algorithm, color space conversion and analysis, histogram analysis, etc. The selection method can be based on the preset mapping relationship or algorithm library.

[0076] Through the synergistic effect of the above steps, the scheme realizes a targeted regional image preliminary analysis process. First, the pre-set regional characteristic information is obtained, which is the basis for distinguishing the characteristics of different regions. Then, based on these characteristic information, it is determined for each region which type of image feature needs to be extracted, which makes the preliminary analysis have a clear purpose and only focuses on the features related to the potential problems or detection focus of the region. Then, according to the determined feature type, the image analysis algorithm most suitable for extracting these features is selected to ensure the efficiency and accuracy of feature extraction. Finally, the selected analysis method is applied to the image data of the corresponding region, so as to obtain image feature data that can better reflect the actual situation and potential problems of the region.

[0077] This way of differential analysis according to regional characteristics overcomes the limitations of uniform analysis method, and can provide more valuable and targeted information support for subsequent intelligent selection of the most suitable image acquisition parameters and detection rules for each region. In this way, the scheme can more effectively cope with the detection challenges of different regions and improve the adaptability and accuracy of the detection effect of the entire detection method.

[0078] In order to more specifically illustrate the implementation mode of the scheme, an example is provided below. Assuming that there are two ROI regions on the part to be detected, ROI-A is a curved corner region prone to bubbles, and ROI-B is a straight line region focusing on the width of the glue path. Before preliminary analysis, the system obtains the pre-set regional characteristic information. For ROI-A, the pre-set regional characteristic information includes: the geometric shape is "curved corner", the easy-to-occur defect type is "bubble, shrinkage", and the detection requirement is "high-precision edge detection". For ROI-B, the pre-set regional characteristic information includes: the geometric shape is "straight line", the easy-to-occur defect type is "glue break, width abnormality", and the detection requirement is "fast width measurement". According to this information, the system determines that the image feature types that need to be extracted for ROI-A include edge features, texture features and local contrast features, while the feature types that need to be extracted for ROI-B include edge features and connectivity features. Subsequently, the system selects the analysis method according to the determined feature type. For ROI-A, Canny edge detection algorithm, LBP texture analysis algorithm and local contrast calculation algorithm can be selected. For ROI-B, Sobel edge detection algorithm and connected component analysis algorithm can be selected. Finally, the system applies the selected algorithms to the regional images of ROI-A and ROI-B in the overview image respectively to obtain their corresponding image feature data, such as the edge density, texture descriptor and contrast average of ROI-A, and the edge point set and the number of connected regions of ROI-B.

[0079] Preferably, step A302 can include:

[0080] For each of the ROI regions, determine the candidate image feature types respectively related to the glue path geometry information, the defect type information, and the region detection requirement information in the region by table lookup;

[0081] For each of the ROI regions, aggregate the determined candidate image feature types and eliminate the repeated candidate image feature types to obtain a candidate image feature type set;

[0082] For each of the ROI regions, according to the preset priority of each image feature type, the preset analysis efficiency value, and the preset preliminary analysis time limit requirement, screen the candidate image feature types in the candidate image feature type set to obtain the image feature types required for preliminary analysis.

[0083] Wherein, the table lookup refers to a data query method, which quickly retrieves and obtains the corresponding value in a preset data structure by inputting information, which can be realized by using hash table or B-tree index structure. The glue path geometry-related image feature type query table, the defect type-related image feature type query table, and the region detection requirement-related image feature type query table can be pre-set, the glue path geometry-related image feature type query table is queried according to the glue path geometry information in the region, the defect type-related image feature type query table is queried according to the defect type information, and the region detection requirement-related image feature type query table is queried according to the region detection requirement information to obtain the related candidate image feature types. Each query table can be formulated according to experiments or expert experience.

[0084] Wherein, the candidate image feature type refers to a set of related image feature categories initially obtained according to a specific input, such as at least one of gray feature, texture feature, shape feature, color feature, and gradient feature. The candidate image feature type set refers to a set of all candidate image feature types determined from different sources for a specific region, wherein the repeated items are eliminated, which can be a list or a set data structure.

[0085] Wherein, the preset priority refers to the weight or ordering rule pre-set for different image feature types when performing feature screening, which can be a numerical value or a rule. The preset priority is a global parameter, and the preset priority of the same image feature type is the same for different ROI regions. The preset priority can be determined by analyzing the influence of selecting each image feature type for preliminary analysis on the final product quality (quality index related to glue coating) based on historical production data, or set according to expert experience.

[0086] Wherein, the preset analysis efficiency value represents the speed of extracting and preliminarily analyzing each image feature type, which can be represented by the inverse of the required processing time. The preset analysis efficiency value can be determined by data statistics.

[0087] The preset preliminary analysis time limit requirement refers to a time constraint set for the entire preliminary analysis process, which can be a time threshold.

[0088] In the screening of the candidate image feature types in the candidate image feature type set according to the preset priority of each image feature type, the preset analysis efficiency value, and the preset preliminary analysis time limit requirement, the candidate image feature types in the candidate image feature type set can be sorted in descending order according to the preset priority, and then each candidate image feature type is traversed in order according to the sorting order. The required processing time of the candidate image feature type currently traversed is determined according to the preset analysis efficiency value of the candidate image feature type. If the required processing time of the candidate image feature type currently traversed is added to the total required processing time of the image feature types selected for preliminary analysis (i.e., the sum of the required processing times of the selected image feature types), the total required processing time does not exceed the preset preliminary analysis time limit requirement, the candidate image feature type currently traversed is added to the image feature types selected for preliminary analysis.

[0089] The scheme provides a specific method for determining the image feature types required for preliminary analysis for each ROI region, aiming to address the problem of how to balance efficiency and accuracy when determining image feature types. The method first determines the candidate image feature types related to the characteristics of each ROI region according to the preset regional characteristic information of the ROI region by table lookup. Through table lookup, the relationship between the known regional characteristic information and the related image feature types can be quickly established, providing a basis for subsequent screening. Then, all candidate image feature types determined from different regional characteristic information for the same ROI region are summarized and duplicates are removed to form a complete candidate image feature type set, ensuring that all relevant characteristic information is considered and omission is avoided. Finally, the candidate set is further screened based on the preset priority of each image feature type, the preset analysis efficiency value, and the preset preliminary analysis time limit requirement. By screening the candidate set based on a comprehensive evaluation of these factors, the image feature types used for preliminary analysis are finally obtained. This screening mechanism based on priority, efficiency, and time limit makes the selection process more optimized, enabling efficient analysis while ensuring the acquisition of effective information. As a specific implementation of the preliminary analysis step in the overall detection method, this scheme selects feature types associated with regional characteristics and meeting efficiency requirements, enabling efficient and accurate preliminary analysis, and providing effective data for subsequent parameter rule selection, so that the overall detection method can better adapt to part pose deviation and local detection demand differences, and cope with detection challenges under high speed.

[0090] According to the technical solution, the image feature type required for preliminary analysis of different ROI regions can be efficiently and accurately selected according to the characteristics of the ROI regions. The correlation between the region characteristics and the features is quickly established through the table lookup, the completeness of the candidate set is ensured by eliminating the repeated items, and the selected feature type can reflect the region characteristics, support the selection of subsequent parameter rules, and meet the efficiency requirement under the high-speed rhythm through the comprehensive screening based on the priority, efficiency and time limit. This improves the efficiency and accuracy of the preliminary analysis, provides a basis for the subsequent detection process, and thus solves the problem of how to efficiently and accurately select the image feature type in the visual detection of the glue path of the automobile parts on the high-speed production line to balance the analysis efficiency and the accuracy of the subsequent parameter rule selection.

[0091] In some embodiments, step A4 comprises:

[0092] A401. According to the image feature data of the ROI region, the attention feature type of the ROI region and the corresponding attention degree are identified;

[0093] A402. For each identified attention feature type, relevant candidate image acquisition parameter sets and candidate detection rules are screened from the multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region;

[0094] A403. According to the preset priority of the identified attention feature type and the corresponding attention degree, the candidate image acquisition parameter sets and the candidate detection rules are further screened to determine the applicable image acquisition parameter sets and the applicable detection rules.

[0095] The image feature data refers to the quantitative information obtained by preliminary analysis of the region image of the ROI region, such as brightness distribution, contrast, texture feature, edge information, color information, geometric feature parameter, etc., which can be obtained by histogram analysis, gray level co-occurrence matrix calculation, edge detection algorithm, statistical quantity calculation, etc.

[0096] The attention feature type refers to the category identified according to the image feature data analysis, which indicates that the ROI region may have specific visual characteristics or potential problems, such as “brightness anomaly”, “insufficient contrast”, “texture blur”, “edge missing”, “suspected bubble” or “glue path narrowing”, etc., which can be identified by rule-based expert system, machine learning classifier or pattern matching algorithm, etc.

[0097] The attention degree refers to a quantitative evaluation of the possibility, severity or significance of the identified attention feature type existing in the current ROI region, which can be represented in the form of a probability value, a confidence score, a severity level index or a deviation amount from the normal condition, and is determined by calculating the matching degree of the image feature data and the preset threshold or model output.

[0098] The preset image acquisition parameter set refers to a set of parameter configurations defined in advance for guiding the image acquisition of the visual sensor, for example, a combination of parameters including light source type, color, angle, brightness, camera exposure time, gain, focal length and / or field of view range, which can be stored in a database or configuration file.

[0099] The preset detection rule refers to a series of logic or algorithm configurations defined in advance for guiding the analysis and judgment of the special image by the defect detection algorithm, for example, including a specific image processing step sequence, a feature extraction method, a defect judgment standard, a threshold setting or a classification model, which can be stored in the form of a script, an algorithm flowchart or a model file.

[0100] The candidate image acquisition parameter set refers to a parameter set selected from a plurality of preset image acquisition parameter sets and related to a specific attention feature type, which is considered to be able to effectively highlight or facilitate the detection of the attention feature type. The candidate detection rule refers to a detection rule selected from a plurality of preset detection rules and related to a specific attention feature type, which is designed for detecting or analyzing the attention feature type. The relevance or degree of relevance between each feature type and each preset image acquisition parameter set and preset detection rule can be calibrated in advance to form a relevance query table. In actual work, the relevant preset image acquisition parameter set and preset detection rule with a relevance or degree of relevance exceeding a preset relevance threshold can be obtained in the relevance query table according to the identified attention feature type, as the relevant candidate image acquisition parameter set and candidate detection rule.

[0101] The preset priority refers to the importance level or weight preset for different attention feature types for decision-making or trade-off when there are multiple attention feature types, which can be an integer value, a floating point number or an enumeration type, stored in a configuration table. The preset priority of each attention feature type can be a global parameter, i.e., the same attention feature type is the same in all ROI regions, or the preset priority of each attention feature type can be set separately for each ROI region.

[0102] Here, the applicable image acquisition parameter set is a set of parameter configurations finally determined for acquiring the special camera image of the current ROI region after further screening. The applicable detection rule is one or more detection rules finally determined for detecting the special camera image of the current ROI region after further screening.

[0103] The method identifies specific types of features of interest and their degrees in the region by further interpreting the image feature data obtained from preliminary analysis, thereby converting the general feature data into information with clear business implications. Based on these clear points of interest, the system can selectively screen candidate solutions related to these points of interest from a large set of preset parameters and rule libraries, greatly reducing the selection range. Further, by considering the preset importance of different feature types of interest and their actual significant degrees in the current region, the candidate solutions are weighed and optimized to finally determine the image acquisition parameter set and detection rule that best fit the actual situation of the current region. This process enables the subsequent special camera image acquisition to better highlight the key features or potential defects in the region, and the defect detection can apply the most suitable algorithm and standard, thereby significantly improving the accuracy and reliability of the detection. Compared with simple mapping or selection based on image feature data, this dynamic and intelligent selection mechanism based on the type, degree and priority of points of interest makes the entire detection process more adaptive and robust, effectively coping with complex and variable actual detection scenarios, especially when multiple potential defects or regional characteristics need to be considered, and making better decisions.

[0104] By identifying the feature types of interest and corresponding degrees of interest of the region according to the image feature data of the ROI region, and selectively screening and determining the image acquisition parameter set and detection rule based on this information, the method can make the subsequent special camera image acquisition more effectively highlight the key visual information or potential defect features in the region, while enabling the defect detection to apply the most suitable algorithm and standard for the characteristics of the current region. This solves the problem of being unable to selectively choose the optimal parameters and rules based on general image feature data, improving the effectiveness of subsequent special camera image acquisition and the accuracy of defect detection.

[0105] Preferably, step A401 can include:

[0106] obtaining a preset feature type association rule associated with the ROI region; the preset feature type association rule defines the correspondence between the image feature data and the preset feature type of interest and the triggering condition;

[0107] According to the image feature data, the triggering condition of any preset feature type of interest is evaluated according to the preset feature type association rule, and the degree of satisfaction is determined.

[0108] According to the evaluation result, all preset attention feature types meeting the trigger condition are identified, the attention feature types of the ROI region are obtained, and the attention degree of each identified attention feature type is determined according to the degree of meeting the condition.

[0109] wherein the preset feature type association rule refers to a set of logic rules pre-established for mapping image feature data to a specific attention feature type and its significance level. These rules can be defined in various forms, such as a set of threshold-based conditional statements, a decision tree model, a lookup table, or a more complex machine learning model. The purpose of introducing these rules is to provide a structured and configurable way to interpret the image feature data obtained from preliminary analysis, so as to accurately identify the potential problem type and its severity. The trigger condition refers to a specific condition or combination of conditions defined in the preset feature type association rule that the image feature data needs to meet to determine the existence of a certain preset attention feature type in the region. The trigger condition can be a single image feature value reaching a certain range, or a complex relationship between multiple image feature values. The degree of meeting the condition refers to the matching degree or intensity of the image feature data meeting the trigger condition of the preset attention feature type. This degree can be binary (satisfying or not satisfying), or a quantitative value, such as a confidence score, a distance metric, or a normalized matching index. The degree of meeting the condition is determined to quantify the significance of the potential problem, providing a basis for determining the attention degree. The preset attention feature type refers to a pre-defined category of potential problems or region characteristics that need special attention from the system. These types can correspond to specific defect types (such as bubbles, glue breakage, glue overflow), specific region states (such as strong reflection, abnormal texture), or situations requiring special handling.

[0110] The method provided in the present application provides a systematic mechanism to identify the attention feature types and the attention degrees according to the image feature data obtained through preliminary analysis by introducing preset feature type association rules. First, preset feature type association rules associated with the current ROI region are obtained. These rules are pre-configured for different region characteristics, ensuring the pertinence of the identification process. These rules clearly define which combinations or states of image feature data correspond to which preset attention feature types, and what degree of image feature data meets the triggering condition. Then, the image feature data obtained through preliminary analysis is input into these rules for evaluation. According to the preset feature type association rules, the system will determine whether the current image feature data meets the triggering condition of any preset attention feature type. This evaluation process not only determines whether the condition is met, but also quantifies the degree to which the condition is met. For example, if the rule defines that a certain texture feature intensity exceeding a certain threshold indicates the presence of a certain defect, the evaluation process will calculate the texture feature intensity of the current region and determine whether it exceeds the threshold, as well as the degree to which it exceeds the threshold. Finally, according to the evaluation results, the system identifies all preset attention feature types that meet the triggering condition and takes these types as the attention feature types of the current ROI region. At the same time, according to the quantified degree of condition satisfaction, a corresponding attention degree is determined for each identified attention feature type. For example, the higher the degree of condition satisfaction, the higher the attention degree. In this way, the system can flexibly and quantitatively identify multiple potential problems and their significance in the current region based on image feature data.

[0111] This identification method based on configurable rules can handle more complex feature combinations and triggering logic than simple threshold judgments, thus more accurately and comprehensively reflecting the actual state of the ROI region, providing more refined input information for subsequent targeted selection of optimal image acquisition parameters and detection rules. Combined with the preliminary analysis of the overview image and the extraction of image feature data of the ROI region in the previous step, this method can fully utilize these preliminary information, avoiding complex and time-consuming defect detection in the preliminary analysis stage, and focusing on identifying regions and problem types that require special attention, thereby improving the efficiency of the overall detection process while ensuring detection effectiveness.

[0112] Preferably, step A403 can include:

[0113] B1. According to the preset priority of the identified attention feature types and the corresponding attention degrees, calculate the comprehensive evaluation value of each attention feature type, and sort the identified attention feature types in descending order according to the comprehensive evaluation value;

[0114] B2. Initialize the applicable image acquisition parameter set collection and the applicable detection rule set.

[0115] B3. sequentially traversing the sorted attention feature types, evaluating the compatibility between the candidate image acquisition parameter set and the candidate detection rule corresponding to the currently traversed attention feature type and the currently determined applicable image acquisition parameter set collection and the applicable detection rule set;

[0116] B4. according to the compatibility evaluation result and the comprehensive evaluation value of the currently traversed attention feature type, selecting one or more schemes from the corresponding candidate image acquisition parameter set and the corresponding candidate detection rule to join the applicable image acquisition parameter set collection and the applicable detection rule set;

[0117] B5. according to the preset detection rhythm constraint, performing final screening and optimization on the applicable image acquisition parameter set collection and the applicable detection rule set to obtain the final applicable image acquisition parameter set and the final applicable detection rule.

[0118] Wherein, the comprehensive evaluation value refers to a numerical value for quantifying the importance and significance of each attention feature type, which can be calculated by weighting and summing or multiplying the preset priority and attention degree.

[0119] Wherein, the compatibility refers to the ability of different image acquisition parameter sets or detection rules to interfere or conflict with each other when applied simultaneously, which can be evaluated by a preset compatibility matrix, rule conflict list or by simulation execution.

[0120] Wherein, the detection rhythm constraint refers to the maximum time allowed to complete the image acquisition and detection task of a ROI region, which can be represented by a fixed time threshold or a time window dynamically calculated according to the production line speed.

[0121] Wherein, the applicable image acquisition parameter set collection refers to the combination of image acquisition parameter sets accumulated in the iterative selection process and considered to be able to effectively deal with the processed attention feature types, which can be stored in a list or set data structure. The applicable detection rule set refers to the combination of detection rules accumulated in the iterative selection process and considered to be able to effectively deal with the processed attention feature types, which can be stored in a list or set data structure.

[0122] The present scheme provides a systematic method for determining the final applicable image acquisition parameter set and detection rules for each ROI region, taking into account multiple attention features, candidate scheme compatibility, and detection beat constraints. Specifically, first, by calculating the comprehensive evaluation value of each attention feature type and sorting, it ensures that in the subsequent scheme selection process, more important or more significant features are processed first, providing an optimized processing order for iterative selection. Then, initialize the applicable scheme set to prepare for the gradual construction of the final scheme. When iterating through the sorted attention feature types, for the candidate scheme of the current feature, evaluate its compatibility with the currently accumulated applicable scheme set, which can identify potential parameter conflicts or rule execution conflicts, avoiding selecting mutually interfering or simultaneously executable scheme combinations. Then, according to the compatibility evaluation result and the importance of the current feature, select the appropriate scheme from the candidate scheme to join the applicable set, gradually accumulating compatible schemes that can effectively solve different attention feature problems. Finally, considering the strict detection beat constraints of high-speed production lines, the accumulated scheme set is finally screened and optimized, which can include merging similar parameter settings, optimizing image acquisition order, simplifying non-critical detection rules, etc., to ensure that the finally determined scheme set can efficiently complete the detection task within the allowed time.

[0123] Through this iterative, compatibility evaluation and global optimization process, the present scheme can comprehensively determine one or a group of optimized, detection effect and efficiency taking into account applicable image acquisition parameter set and detection rules from the candidate schemes selected for a single feature. This method, combined with the pre-step of determining ROI region from overview image, obtaining regional image feature data, and identifying attention feature type and degree, etc., makes the entire detection process able to dynamically adjust image acquisition and detection strategies according to the actual state and local features of the part, thereby improving the robustness and efficiency of detection.

[0124] For example, in one specific implementation, step B1 can be implemented as follows: for each identified feature type of interest, look up a pre-defined feature priority table to obtain its priority value, and also obtain the degree of interest from the image feature data analysis. The comprehensive evaluation value can be calculated as the product of the priority and the degree of interest. Then sort all identified feature types of interest in descending order according to the calculated comprehensive evaluation values. Step B2 can simply create two empty lists, one for storing the applicable image acquisition parameter sets and the other for storing the applicable detection rules. Step B3 can be implemented as follows: for the currently traversed feature type of interest, obtain its corresponding list of candidate image acquisition parameter sets and list of candidate detection rules. For each candidate parameter set, check if it has a conflict with any existing parameter set in the current set of applicable image acquisition parameter sets (e.g., one requires red ring light and the other requires blue coaxial light). For each candidate detection rule, check if it has an execution conflict with any existing rule in the current set of applicable detection rules (e.g., two rules require processing the same image region but rely on different preprocessing steps and cannot be parallelized). Determine the compatibility according to the conflict check results. Step B4 can be implemented as follows: if a candidate parameter set and corresponding candidate detection rule combination is compatible with the current applicable set and the combination can effectively deal with the current feature type of interest (e.g., the combination has a detection rate for the feature higher than a threshold in offline testing), then add it to the applicable set. The solution with good compatibility and effectiveness can be selected according to the comprehensive evaluation values. Step B5 can be implemented as follows: after all feature types of interest are processed, check if the total execution time of all parameter sets and rules in the applicable set exceeds a pre-defined detection beat constraint. If it does, adjust according to a pre-defined optimization strategy, e.g., remove rules for low-priority features, merge similar parameter settings, or select a parameter set and rule combination that can handle multiple features but has the shortest detection time, until the beat requirement is met. The resulting set is the final applicable image acquisition parameter set and applicable detection rules.

[0125] Preferably, step B3 can include:

[0126] B301. For each candidate image acquisition parameter set corresponding to the currently traversed feature type of interest, evaluate the parameter item conflict degree between it and each applicable image acquisition parameter set in the set of applicable image acquisition parameter sets;

[0127] B302. For each candidate detection rule corresponding to the currently traversed feature type of interest, evaluate the rule execution conflict degree between it and each applicable detection rule in the set of applicable detection rules;

[0128] B303. According to the parameter item conflict degree and the rule execution conflict degree, determine the compatibility evaluation result between the candidate image acquisition parameter set and the candidate detection rule corresponding to the concerned feature type currently traversed and the currently determined applicable image acquisition parameter set collection and the applicable detection rule set.

[0129] Wherein, the parameter item conflict degree refers to the degree of measuring the mutual contradiction, mutual interference or the inability to satisfy simultaneously between two or more image acquisition parameter sets in the setting value or setting range of the parameter items (such as exposure time, gain, light source brightness, light source color, light source angle, etc.) contained in each other. It can be realized by using a quantitative scoring mechanism, a rule-based conflict judgment logic or a query based on a preset conflict matrix.

[0130] Wherein, the rule execution conflict degree refers to the degree of measuring the mutual contradiction, mutual interference or the inability to simultaneously effectively execute between two or more detection rules in the execution order, dependency relationship, requirement for image preprocessing (such as one rule needs sharpening and another needs smoothing), occupation of computing resources or interpretation of detection results. It can be realized by using rule dependency graph-based analysis, a query based on a preset conflict list or a conflict detection based on simulated execution.

[0131] Wherein, the compatibility evaluation result refers to the determination conclusion of whether the current candidate image acquisition parameter set and the candidate detection rule can work together, whether there is a significant conflict or potential risk according to the evaluation of the parameter item conflict degree and the rule execution conflict degree. It can be realized by using a comprehensive score based on weighted summation, a classification judgment based on a decision tree or a comprehensive reasoning based on fuzzy logic.

[0132] The scheme provides a meticulous and quantitative compatibility evaluation method through the above steps. For each feature type of interest traversed and each candidate image acquisition parameter set and each candidate detection rule corresponding to the feature type, the scheme first evaluates the parameter item conflict degree between the candidate image acquisition parameter set and each member of the current determined set of applicable image acquisition parameter sets. This ensures that the new image acquisition scheme does not produce irreconcilable contradictions with the determined scheme in hardware settings or imaging conditions. At the same time, the rule execution conflict degree between the candidate detection rule and each member of the current determined set of applicable detection rules is evaluated. This ensures that the new detection rule does not conflict with the determined rule in software logic or processing flow. Finally, the overall compatibility evaluation result of the current candidate scheme and the set of determined schemes is determined by comprehensively considering the conflict degree of the two aspects. This evaluation result is then used in step B4 to guide the selection of the scheme. Through this meticulous and quantitative compatibility evaluation, the scheme can avoid selecting candidate schemes that are effective for the current feature of interest but conflict with other determined schemes, thereby ensuring that the finally selected set of applicable image acquisition parameter sets and set of applicable detection rules is a coordinated and consistent whole that can work together to effectively support the detection task for all feature types of interest. This meticulous compatibility evaluation combined with the sorting selection process based on the priority and attention degree of the feature of interest enables the system to prioritize important feature detection requirements while considering the compatibility and coordination of the overall scheme, thereby maximizing the coverage and accurate detection of multiple types of rubberized road features and defects within a limited detection time.

[0133] In step A5, if there are applicable image acquisition parameter sets that require adjustment of the focal length of the vision sensor and applicable image acquisition parameter sets that do not require adjustment of the focal length of the vision sensor in the determined set of applicable image acquisition parameter sets, the feature images corresponding to the applicable image acquisition parameter sets that do not require adjustment of the focal length of the vision sensor are collected first, and then the remaining feature images are collected. For the feature images corresponding to the applicable image acquisition parameter sets that do not require adjustment of the focal length of the vision sensor, since the relative pose of the vision sensor and the automobile part to be detected and the focal length of the vision sensor are the same when collecting the feature images and collecting the overview image, the actual pixel range of each ROI region in the overview image and the feature image is exactly the same, the range of the ROI region in the feature image can be determined directly according to the actual pixel range determined in the overview image, and the processing efficiency is improved.

[0134] Reference Figure 2 and Figure 3 The present application provides a visual detection system for rubberized road of automobile parts, which is used for visual detection of rubberized road of automobile parts based on a vision sensor, and the system comprises an image acquisition device 1, a positioning table 2 and an industrial computer 3.

[0135] The positioning table 2 is used for positioning the placed glued automobile parts to be detected;

[0136] The image acquisition device 1 is provided with a visual sensor 101 and an illumination light source 102, and is used for acquiring images of the automobile parts to be detected and sending to the industrial computer 3;

[0137] The industrial computer 3 is installed with a visual detection program, which is configured with:

[0138] An overview image acquisition module 301 is configured to acquire an overview image of the automobile parts to be detected containing a glue path by using the image acquisition device 1 (for details, refer to the step A1 in the foregoing description);

[0139] A region mapping module 302 is configured to determine the actual pixel range of a plurality of ROI regions in the overview image according to the overview image, wherein the plurality of ROI regions are set based on a standard pose; each of the ROI regions contains a part of the glue path (for details, refer to the step A2 in the foregoing description);

[0140] A region feature analysis module 303 is configured to preliminarily analyze the region image corresponding to each of the ROI regions in the overview image according to the determined actual pixel range, and acquire image feature data of each of the ROI regions (for details, refer to the step A3 in the foregoing description);

[0141] A parameter rule screening module 304 is configured to determine, for each of the ROI regions, a suitable image acquisition parameter set and a suitable detection rule from a plurality of preset image acquisition parameter sets and a plurality of preset detection rules corresponding to the ROI region according to the corresponding image feature data (for details, refer to the step A4 in the foregoing description);

[0142] A special image acquisition module 305 is configured to acquire special images corresponding to each of the ROI regions by sequentially acquiring images of the automobile parts to be detected according to the determined suitable image acquisition parameter set of each of the ROI regions by using the image acquisition device 1; the relative pose of the visual sensor 101 and the automobile parts to be detected is the same when acquiring the special images and acquiring the overview image (for details, refer to the step A5 in the foregoing description);

[0143] A detection module 306 is configured to perform glue path defect detection on the region image corresponding to each of the ROI regions in the corresponding special image according to the determined actual pixel range and the corresponding suitable detection rule (for details, refer to the step A6 in the foregoing description);

[0144] The result summary module 307 is used to summarize the detection results of adhesive path defects in each ROI area to obtain the overall detection results of the automotive parts to be inspected (for details, refer to step A7 above).

[0145] The image acquisition device 1 may further include a robotic arm 103, and a vision sensor 101 and an illumination source 102 may be disposed at the end of the robotic arm 103 (e.g., Figure 2 As shown, the position and orientation of the vision sensor 101 and the illumination source 102 can be adjusted by the robotic arm 103 to acquire images from a suitable position. Preferably, a gimbal mechanism 104 is also provided at the end of the robotic arm 103, and the illumination source 102 is mounted on the gimbal mechanism 104 so that the angle of the illumination source 102 can be adjusted independently. When the applicable image acquisition parameter set includes the illumination angle of the illumination source 102, it can be ensured that the illumination angle of the illumination source 102 can still be accurately adjusted to the illumination angle required by the applicable image acquisition parameter set under different poses of the robotic arm 103.

[0146] Preferably, the illumination source 102 is an array-type light source, which can independently adjust the working state, brightness, and color of each light point in the array according to actual needs. Furthermore, the array-type light source is concentrically arranged with the vision sensor 101, and a clearance hole is provided in the center of the array-type light source to avoid obstructing the vision sensor 101, so as to obtain coaxial illumination light coaxial with the vision sensor 101 according to actual needs.

[0147] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A visual inspection method for adhesive application paths on automotive parts, used for visual inspection of adhesive application paths on automotive parts based on a visual sensor, characterized in that, The steps of this method include: A1. Obtain an overview image of the automotive part to be inspected, including the adhesive application path; A2. Based on the overview image, determine the actual pixel range of multiple ROI regions in the overview image that are pre-set based on a standard pose; each ROI region contains a portion of the adhesive application path; A3. Based on the determined actual pixel range, perform preliminary analysis on the region images corresponding to each ROI region in the overview image to obtain image feature data of each ROI region; A4. For each ROI region, based on the corresponding image feature data, determine the applicable image acquisition parameter set and applicable detection rule from multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region; A5. Sequentially acquire images of the vehicle part to be detected according to the determined sets of applicable image acquisition parameters to obtain special images corresponding to each ROI region; when acquiring the special images and the overview images, the visual sensor and the vehicle part to be detected have the same relative pose; A6. For each ROI region, based on the determined actual pixel range, the corresponding applicable detection rules are used to perform glue path defect detection on the region image of the corresponding ROI region in the special effects image; A7. Summarize the detection results of adhesive path defects in each ROI area to obtain the overall detection results of the automotive parts to be inspected; Step A3 includes: A301. Obtain preset region characteristic information for each of the ROI regions; the preset region characteristic information includes the geometric shape information of the adhesive path within the region, information on common defect types, and information on region detection requirements; A302. Based on the preset region characteristic information, determine the type of image features to be extracted for preliminary analysis of each ROI region; A303. Based on the determined image feature type, select an analysis method for preliminary analysis of the region image of each ROI region; A304. For each ROI region, based on the selected analysis method, perform preliminary analysis on the region image corresponding to the ROI region in the overview image to obtain image feature data of the ROI region corresponding to the determined image feature type; Step A4 includes: A401. Based on the image feature data of the ROI region, identify the type of attention feature of the ROI region and the corresponding degree of attention; A402. For each type of feature of interest identified, relevant candidate image acquisition parameter sets and candidate detection rules are selected from multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region; A403. Based on the preset priority of the identified interest feature type and the corresponding interest level, further filter the selected candidate image acquisition parameter set and candidate detection rules to determine the applicable image acquisition parameter set and the applicable detection rules.

2. The method for visual inspection of adhesive application paths on automotive parts according to claim 1, characterized in that, Step A1 includes: A101. Obtain imaging characteristic information of preset key structural features on the automotive part to be detected; the imaging characteristic information includes imaging performance data of the preset key structural features under different combinations of image acquisition parameters, and the imaging performance data includes contrast, shadow features, specular emission parameters and / or surface texture features; A102. Based on the imaging characteristic information, determine the combination of image acquisition parameters that is conducive to the clear presentation of the preset key structural features in the image, and use it as the target image acquisition parameter set; A103. Based on the target image acquisition parameter set, acquire an overview image of the automotive part to be detected, including the adhesive coating path; A104. The overview image is preprocessed to enhance the contrast and / or edge information of the preset key structural features to obtain the final overview image.

3. The method for visual inspection of adhesive application paths on automotive parts according to claim 1, characterized in that, Step A2 includes: A201. In the overview image, identify the preset key structural features on the automotive part to be detected; A202. Based on the identified preset key structural features, calculate the pose deviation of the automotive part to be detected relative to the preset standard pose; A203. Based on the pose deviation, convert the preset standard coordinate information of multiple ROI regions into the actual pixel range in the overview image.

4. The method for visual inspection of adhesive application paths on automotive parts according to claim 1, characterized in that, Step A302 includes: For each ROI region, candidate image feature types related to the geometric shape information of the adhesive path within the region, the information on the types of common defects, and the information on the region detection requirements are determined by looking up a table. For each ROI region, the obtained candidate image feature types are summarized and duplicate candidate image feature types are removed to obtain a set of candidate image feature types; For each ROI region, the candidate image feature types in the candidate image feature type set are filtered according to the preset priority, preset analysis efficiency value and preset preliminary analysis time limit of each image feature type to obtain the image feature types to be extracted for preliminary analysis.

5. The method for visual inspection of adhesive application paths on automotive parts according to claim 1, characterized in that, Step A401 includes: Obtain a preset feature type association rule associated with the ROI region; the preset feature type association rule defines the correspondence between the image feature data and the preset interest feature type and the triggering conditions; Based on the image feature data and according to the preset feature type association rules, the ROI region is evaluated to determine whether it meets the triggering conditions of any preset interest feature type, and the degree to which the conditions are met is determined. Based on the evaluation results, all preset attention feature types that meet the triggering conditions are identified, the attention feature types of the ROI region are obtained, and the attention level of each identified attention feature type is determined according to the degree to which the conditions are met.

6. The method for visual inspection of adhesive application paths on automotive parts according to claim 1, characterized in that, Step A403 includes: B1. Based on the preset priority of the identified interest feature type and the corresponding interest level, calculate the comprehensive evaluation value of each interest feature type, and sort the identified interest feature types in descending order according to the comprehensive evaluation value; B2. Initialize the set of applicable image acquisition parameters and the set of applicable detection rules; B3. Iterate through the sorted feature types in sequence, and evaluate the compatibility between the candidate image acquisition parameter set and the candidate detection rule corresponding to the currently traversed feature type and the currently determined set of applicable image acquisition parameter set and set of applicable detection rule; B4. Based on the compatibility evaluation results and the comprehensive evaluation value of the currently traversed feature type of interest, select one or more schemes from the corresponding candidate image acquisition parameter set and the corresponding candidate detection rule, and add them to the applicable image acquisition parameter set and the applicable detection rule set; B5. Based on the preset detection cycle constraints, the set of applicable image acquisition parameters and the set of applicable detection rules are finally filtered and optimized to obtain the final set of applicable image acquisition parameters and the final set of applicable detection rules.

7. A visual inspection method for adhesive application paths on automotive parts according to claim 6, characterized in that, Step B3 includes: B301. For each candidate image acquisition parameter set corresponding to the currently traversed feature type of interest, evaluate the degree of parameter conflict between it and each applicable image acquisition parameter set in the applicable image acquisition parameter set set; B302. For each candidate detection rule corresponding to the currently traversed feature type, evaluate the degree of rule execution conflict between it and each applicable detection rule in the set of applicable detection rules; B303. Based on the degree of conflict of the parameter items and the degree of conflict of the rule execution, determine the compatibility evaluation result between the candidate image acquisition parameter set and the candidate detection rule corresponding to the currently traversed feature type of interest and the currently determined set of applicable image acquisition parameter sets and the set of applicable detection rules.

8. A visual inspection system for adhesive application paths on automotive parts, used for visual inspection of adhesive application paths on automotive parts based on a visual sensor, characterized in that, The system includes an image acquisition device, a positioning platform, and an industrial control computer; The positioning platform is used to position and place the automotive parts to be inspected after the adhesive has been applied. The image acquisition device is equipped with a vision sensor and a lighting source. The image acquisition device is used to acquire images of the automotive parts to be inspected and send them to the industrial control computer. The industrial control computer is equipped with a vision inspection program, which is configured with: The overview image acquisition module is used to acquire an overview image of the automotive part to be inspected, including the adhesive coating path, using the image acquisition device. The region mapping module is used to determine the actual pixel range of multiple ROI regions in the overview image based on the overview image, which are pre-set based on a standard pose; each ROI region contains a portion of the adhesive application path; The region feature analysis module is used to perform preliminary analysis on the region images corresponding to each of the ROI regions in the overview image based on the determined actual pixel range, and to obtain the image feature data of each of the ROI regions; The parameter rule filtering module is used to determine the applicable image acquisition parameter set and applicable detection rule for each ROI region based on the corresponding image feature data from multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region. The special effects image acquisition module is used to acquire images of the automobile parts to be detected by the image acquisition device in sequence according to the determined sets of applicable image acquisition parameters, so as to obtain special effects images corresponding to each of the ROI regions; When acquiring the special effects image and the overview image, the visual sensor and the car part to be detected are in the same relative pose. The detection module is used to perform glue path defect detection on the corresponding area image of the special effect image in the ROI region according to the determined actual pixel range and the corresponding applicable detection rules for each ROI region. The results summary module is used to summarize the detection results of adhesive path defects in each ROI region to obtain the overall detection results of the automotive parts to be inspected. When the region feature analysis module performs preliminary analysis on the region images corresponding to each ROI region in the overview image based on the determined actual pixel range, and obtains the image feature data of each ROI region, it executes the following: A301. Obtain preset region characteristic information for each of the ROI regions; the preset region characteristic information includes the geometric shape information of the adhesive path within the region, information on common defect types, and information on region detection requirements; A302. Based on the preset region characteristic information, determine the type of image features to be extracted for preliminary analysis of each ROI region; A303. Based on the determined image feature type, select an analysis method for preliminary analysis of the region image of each ROI region; A304. For each ROI region, based on the selected analysis method, perform preliminary analysis on the region image corresponding to the ROI region in the overview image to obtain image feature data of the ROI region corresponding to the determined image feature type; When the parameter rule filtering module determines the applicable image acquisition parameter set and applicable detection rule for each ROI region based on the corresponding image feature data from multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region, it performs the following: A401. Based on the image feature data of the ROI region, identify the type of attention feature of the ROI region and the corresponding degree of attention; A402. For each type of feature of interest identified, relevant candidate image acquisition parameter sets and candidate detection rules are selected from multiple preset image acquisition parameter sets and multiple preset detection rules corresponding to the ROI region; A403. Based on the preset priority of the identified interest feature type and the corresponding interest level, further filter the selected candidate image acquisition parameter set and candidate detection rules to determine the applicable image acquisition parameter set and the applicable detection rules.

Images