Automobile detection system and detection method based on machine vision

Abstract

The invention relates to the technical field of machine vision, in particular to an automobile detection system and method based on machine vision, and the system comprises an image processing module, an automobile body feature extraction module, a defect positioning module, a defect analysis module, a decision support module and a detection evaluation module. According to the method, the U-Net model is combined with the conditional random field, various vehicle body defects such as scratches and recesses can be finely recognized and accurately positioned, the recognition capability of different defect types is enhanced by extracting complex geometric and textural features from the image, the defect classification is more accurate, and the accuracy of vehicle body defect classification is improved. The defect boundary is continuously optimized, the detection parameters are automatically adjusted, the reliability of decision support is improved, the repair process is more targeted and efficient, human errors are reduced through automatic detection and parameter adjustment, the operation cost is reduced, and potential maintenance cost and risks caused by defect omission are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and in particular to a vehicle detection system and detection method based on machine vision. Background Art

[0002] Machine vision technology is a technology that automatically detects and analyzes the environment through cameras and computer vision systems. It integrates knowledge from fields such as digital image processing, mechanical engineering, optics, and computer science to capture images through visual sensors (such as cameras) and then use software algorithms to process and analyze the images. Machine vision has a wide range of applications, including industrial and non-industrial fields such as quality control, production automation, and robot navigation. It can provide fast and accurate visual inspections to support decision-making and operations.

[0003] Among them, the automobile inspection system of machine vision is mainly used to detect defects in automobile bodies and surfaces. It uses high-resolution cameras to capture images of the automobile surface, and then uses algorithms to analyze the images to identify defects such as scratches, dents, rust or coating problems. It is widely used in automobile production lines and maintenance industries to ensure that the appearance quality of the automobile meets the standards. Through the automated inspection process, it can not only improve the inspection efficiency, but also reduce the cost and time loss caused by omissions in manual inspections.

[0004] Existing technologies for automotive defect detection primarily rely on basic image capture and primary analysis methods. Limited by image quality and analysis algorithm limitations, traditional technologies, without the use of deep learning technology, struggle to identify tiny defects against complex backgrounds, such as minor scratches or initial rust. This leads to errors in detection results and increases the difficulty and cost of subsequent repairs. Furthermore, traditional technologies lack flexibility when dealing with diverse vehicle models and are unable to adapt to the specific needs of different models, impacting detection efficiency and accuracy. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a vehicle detection system and detection method based on machine vision.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: A vehicle detection system based on machine vision includes: The image processing module collects vehicle body image data based on the camera equipment, performs distortion correction on the image, and processes the image data using histogram equalization to adjust the contrast and brightness of the image to obtain a processed image set; The body feature extraction module extracts the geometric shape and texture features of the body parts based on the processed image set using the U-Net model, and calculates the morphological and texture feature vectors of the body parts to obtain a component feature vector set; The defect location module uses a conditional random field to perform multi-scale analysis based on the component feature vector set to detect scratches and dents on the vehicle body surface, and performs edge detection and morphological analysis to determine the defect location and generate defect location information. The defect analysis module continuously optimizes the defect boundary based on the defect location information, classifies the detected defects, determines the type of each defect, and generates a defect type identification record; The decision support module analyzes the processing priority of the difference type defects based on the defect type identification record, sorts the detected defect priorities, and generates a repair priority list; The detection evaluation module evaluates the detection results according to the repair priority list, analyzes the detection accuracy of multiple types of defects, and automatically adjusts the detection parameters to optimize the detection process, generating a detection effect confirmation result.

[0007] As a further solution of the present invention, the processed image set includes a brightness balance image and a color adjustment image, the component feature vector set includes a geometric feature vector, a texture feature vector, and a component boundary vector, the defect location information includes an edge positioning index, a surface roughness parameter, and a defect depth index, the defect type identification record includes a defect shape classification, a defect size classification, and a defect color difference, the repair priority list includes emergency repair projects, medium priority repair projects, and low priority repair projects, and the detection effect confirmation result includes an error rate evaluation result, a detection consistency analysis result, and an adjustment effect evaluation result.

[0008] As a further solution of the present invention, the image processing module includes an image acquisition submodule, an image correction submodule, and an image enhancement submodule; The image acquisition submodule collects vehicle body image data based on the camera device. By automatically triggering the camera device at regular intervals, it collects continuous vehicle body image data streams to obtain the original image sequence. The image correction submodule corrects the image frame by frame based on the original image sequence by adjusting the image acquisition angle and light illumination conditions to obtain a corrected image sequence; The image enhancement submodule applies histogram equalization based on the corrected image sequence, adjusts the image contrast and brightness frame by frame, refines the dynamic range of each frame of image, and generates a processed image set.

[0009] As a further solution of the present invention, the vehicle body feature extraction module includes a feature modeling submodule, a feature calculation submodule, and a feature vector generation submodule; The feature modeling submodule locates the boundaries of the vehicle body parts based on the processed image set, highlights the geometric shape and texture details of the parts through edge enhancement, identifies and records the texture features of the parts, and obtains preliminary feature data; The feature calculation submodule evaluates the spatial relationship and structural layout of the component based on the preliminary feature data and the component morphology, numerically represents the component features, and extracts key feature vectors to obtain feature vector data; The feature vector generation submodule summarizes the feature vectors of the components based on the feature vector data, unifies the formats and standardizes the feature data, and generates a component feature vector set.

[0010] As a further solution of the present invention, the defect location module includes a multi-scale analysis submodule, a defect identification submodule, and a position determination submodule; The multi-scale analysis submodule uses a conditional random field based on the component feature vector set to process image data of different scales, analyze image details at each level, identify potential defect areas, and generate multi-scale defect analysis information; The defect recognition submodule refines the image edges based on the multi-scale defect analysis information, mines small scratches or depressions, and analyzes the defect morphology through morphology. It combines edge and morphological features to identify the defect type and obtain preliminary defect recognition results. The position determination submodule locates the coordinates of the identified defects based on the preliminary defect identification results, records the geometric positions of the defects, and generates defect position information.

[0011] As a further solution of the present invention, based on the multi-scale defect analysis information, image edge refinement and mining are performed on small scratches or depressions according to the formula: ;

[0012] Calculate defect detection values ​​after edge enhancement , where Representative The gradient value of each pixel, Representative The kernel enhancement factor corresponding to each pixel is is the total number of pixels involved in the calculation.

[0013] As a further solution of the present invention, the defect analysis module includes a boundary optimization submodule, a defect classification submodule, and a type determination submodule; The boundary optimization submodule adjusts the defect edge based on the defect position information, refines the boundary contrast and adjusts the light, optimizes the visual details of the defect, and obtains a refined defect boundary; The defect classification submodule compares and classifies the morphological features of the difference defects based on the refined defect boundaries, and refines the defects according to their size, shape, and depth to obtain defect morphological classification; The type determination submodule determines the defect category based on the defect morphology classification by evaluating the defect morphology and impact degree, records the classification attributes of each defect, and generates a defect type identification record.

[0014] As a further solution of the present invention, the decision support module includes a priority analysis submodule, a sorting submodule, and a list generation submodule; The priority analysis submodule quantitatively evaluates the severity and impact of defects based on the defect type identification records, assesses the emergency handling requirements of each defect, adjusts the weight factors to define the handling priority, and generates priority quantitative analysis results; The sorting submodule prioritizes the defects according to the urgency of processing and the difficulty of repair based on the priority quantitative analysis results, and updates the sorting order in real time to obtain a priority sorting list; The list generation submodule formulates a repair plan based on the priority list, including repair steps, required materials and estimated time, to obtain a repair priority list.

[0015] As a further solution of the present invention, the detection and evaluation module includes an evaluation and analysis submodule, a parameter adjustment submodule, and a result confirmation submodule; The evaluation and analysis submodule evaluates the detection accuracy and error rate of different defect types based on the repair priority list, analyzes the stability of the detection process through real-time data monitoring and historical data comparison, and generates a detection stability analysis record; The parameter adjustment submodule adjusts the detection parameters based on the detection stability analysis record, and adjusts the image analysis threshold and defect detection process according to the error feedback information to obtain the adjusted detection parameters; The result confirmation submodule executes the defect detection process again based on the adjusted detection parameters, evaluates the adjustment effect through the verification test set, and generates a detection effect confirmation result.

[0016] A vehicle detection method based on machine vision is implemented based on the above-mentioned vehicle detection system based on machine vision, and includes the following steps: S1: Collect vehicle body image data through a camera device, perform distortion correction on the image, and perform histogram equalization on the image data to adjust the image contrast and brightness to obtain a processed image set; S2: Based on the processed image set, using the U-Net model to extract the geometric shape and texture features of the vehicle body, calculate the morphological and texture feature vectors of the vehicle body parts, and generate a component feature vector set; S3: Based on the component feature vector set, a conditional random field is used to perform multi-scale analysis to detect scratches and dents on the vehicle body surface, including edge detection and morphological analysis, to determine the defect location and generate defect location information; S4: Based on the defect location information, continuously optimize the defect boundary, classify the detected defects, determine the type of each defect, and generate a defect type identification record; S5: Based on the defect type identification record, analyze the processing priority of different type defects, sort the defect priorities, evaluate the detection results according to the repair priority, analyze the detection accuracy of multiple types of defects, automatically adjust the detection parameters to optimize the detection process, and obtain the detection effect confirmation result.

[0017] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by combining the U-Net model with conditional random fields, it is possible to finely identify and accurately locate various body defects, such as scratches and dents. By extracting complex geometric and texture features from images, the ability to recognize different defect types is enhanced, making defect classification more accurate. Continuous optimization of defect boundaries and automatic adjustment of detection parameters improve the reliability of decision support, making the repair process more targeted and efficient. Automated detection and parameter adjustment reduce human errors, lower operating costs, and reduce potential repair costs and risks caused by missed defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a system flow chart of the present invention; Figure 2 Schematic diagram of the system framework of the present invention; Figure 3 This is a flow chart of the image processing module of the present invention; Figure 4 This is a flow chart of the vehicle body feature extraction module of the present invention; Figure 5 This is a flow chart of the defect location module of the present invention; Figure 6 This is a flow chart of the defect analysis module of the present invention; Figure 7 This is a flow chart of the decision support module of the present invention; Figure 8 This is a flow chart of the detection and evaluation module of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0021] Example 1 See also Figure 1 The present invention provides a technical solution: a vehicle detection system based on machine vision includes: The image processing module collects vehicle body image data based on the camera equipment, performs distortion correction on the image, and processes the image data using histogram equalization to adjust the contrast and brightness of the image to obtain a processed image set; The body feature extraction module uses the U-Net model to extract the geometric shape and texture features of the body parts based on the processed image set, and calculates the morphological and texture feature vectors of the body parts to obtain the component feature vector set. The body parts include doors, roof, front and rear bumpers, side skirts, and hood. The defect localization module uses conditional random fields to perform multi-scale analysis based on component feature vector sets to detect scratches and dents on the vehicle body surface. It also performs edge detection and morphological analysis to determine the defect location and generate defect location information. The defect analysis module continuously optimizes defect boundaries based on defect location information, classifies detected defects, determines the type of each defect, and generates defect type identification records; The decision support module identifies and records the defect types, analyzes the priority of handling the different defect types, ranks the detected defects, and generates a repair priority list; The detection evaluation module evaluates the detection results based on the repair priority list, analyzes the detection accuracy of multiple types of defects, and automatically adjusts the detection parameters to optimize the detection process and generate detection effect confirmation results.

[0022] The processed image set includes brightness balanced images and color adjusted images. The component feature vector set includes geometric feature vectors, texture feature vectors, and component boundary vectors. The defect location information includes edge positioning indicators, surface roughness parameters, and defect depth indicators. The defect type identification record includes defect shape classification, defect size classification, and defect color difference. The repair priority list includes emergency repair projects, medium priority repair projects, and low priority repair projects. The detection effect confirmation results include error rate evaluation results, detection consistency analysis results, and adjustment effect evaluation results.

[0023] See also Figure 2 and Figure 3 ,The image processing module includes an image acquisition submodule, an image correction submodule, and an image enhancement submodule; The image acquisition submodule collects vehicle body image data based on the camera device. By automatically triggering the camera device at regular intervals, it collects continuous vehicle body image data streams to obtain the original image sequence. Vehicle body image data is collected based on the camera equipment, and an automatic trigger mechanism is set up to automatically start the camera equipment when the vehicle enters the monitoring range. The time interval is set to determine the frequency of image collection, for example, the camera equipment is triggered every 5 seconds to ensure that the vehicle's operating status can be continuously recorded. By adjusting the camera's resolution and frame rate, the size and quality of the data are controlled to ensure that the effective information of the image data is maximized within the limited storage space. The relative position and angle of the camera and the vehicle must be accurately measured and debugged to avoid blind spots and ensure that the image covers all parts of the vehicle. The camera's aperture, exposure time and sensitivity also need to be adjusted according to the external light conditions to obtain clear images and continuously record a continuous stream of vehicle body image data.

[0024] The image correction submodule corrects the image frame by frame based on the original image sequence by adjusting the image acquisition angle and light illumination conditions to obtain a corrected image sequence; It starts by correcting each image frame by frame. By automatically analyzing the distortion in the original image, such as perspective deformation and optical distortion, it calculates the correction parameters. Then it adjusts the image acquisition angle and lighting conditions, applies the correction parameters to each image frame, and performs distortion correction, color correction and contrast adjustment to ensure the visual consistency and accuracy of each image frame. Through technical means, it gradually optimizes the quality of the entire image sequence to make it more suitable for subsequent processing requirements, and obtains a series of corrected image sequences. The correction process completely improves various image defects in the original image.

[0025] The image enhancement submodule applies histogram equalization based on the corrected image sequence, adjusts the image contrast and brightness frame by frame, refines the dynamic range of each frame, and generates a processed image set; Based on the corrected image sequence, the images are further enhanced and their contrast and brightness are adjusted using histogram equalization technology. Each frame undergoes a detailed dynamic range analysis to calculate the optimal adjustment parameters, such as gain and offset. These parameters are then applied to optimize the image frame by frame, refining the dynamic range of each frame. During this process, the histogram of each frame is recalculated and adjusted to achieve the best visual effect. The resulting processed image set has a significant visual improvement in clarity and contrast, allowing each frame to better display details.

[0026] See also Figure 2 and Figure 4 ,The vehicle body feature extraction module includes a feature modeling submodule, a feature ,computation submodule, and a feature vector generation submodule; The feature modeling submodule locates the boundaries of the body parts based on the processed image set, highlights the geometric shape and texture details of the parts through edge enhancement, identifies and records the texture features of the parts, and obtains preliminary feature data; In the process of locating the boundaries of body parts in the processed image set, an edge detection algorithm is used to enhance the edge information of the image, and the geometric shape and texture details of the parts are emphasized through filters. Different filter settings are applied to each part to adapt to its texture and shape characteristics, such as using Sobel or Canny algorithms to improve the accuracy of edge detection. The contours of body parts are extracted through technology, and the surface features of the parts are further detected using texture analysis technology, such as using gray-level co-occurrence matrix or wavelet transform method to identify and record the unique texture features of different parts. Combined with the shape and texture data, preliminary feature data is obtained, which includes key information such as the size, shape, and surface texture features of the parts.

[0027] The feature calculation submodule evaluates the spatial relationship and structural layout of the component based on the preliminary feature data and the component morphology, numerically represents the component features, and extracts key feature vectors to obtain feature vector data; First, the morphology of the components and their spatial relationships are analyzed. The relative positions of the components are evaluated through image analysis techniques such as distance measurement and angle calculation. Using this information, the numerical representation of the component features is calculated, including geometric properties such as area, perimeter, and geometric center. Next, key feature vectors are extracted through principal component analysis (PCA) or other statistical methods. The feature vectors cover the main morphological features and structural layout of the components, ensuring the comprehensiveness and representativeness of the features. Through a series of calculations, the feature vector data of the components is obtained.

[0028] The feature vector generation submodule aggregates the feature vectors of the components based on the feature vector data, unifies the format and standardizes the feature data, and generates a component feature vector set; The feature vectors of all components are first formatted in a unified manner to ensure consistency in the data structure, including adjusting the length and scale of all feature vectors to the same standard. The data is then normalized, such as by applying Z-score normalization or minimum-maximum normalization methods, to eliminate scale differences between features of different components, ensuring the comparability and applicability of the feature data. Through these processing steps, a unified and standardized set of component feature vectors is generated. The set contains a detailed feature description of each component of the vehicle body, providing accurate input data for subsequent machine learning and pattern recognition tasks.

[0029] See also Figure 2 and Figure 5 ,The defect location module includes a multi-scale analysis submodule, a defect identification submodule, and a location determination submodule; The multi-scale analysis submodule uses conditional random fields based on the component feature vector set to process image data of different scales, analyze image details at each level, identify potential defect areas, and generate multi-scale defect analysis information; When processing the component feature vector set, conditional random fields are used to analyze image data of different scales. First, the image data is classified according to scale levels, such as micro, meso, and macro scales. Different model parameter settings are used for images at each scale level. The parameters are optimized according to the requirements and characteristics of the image details to more effectively process and analyze the detailed information at each level. The conditional random field model captures the contextual relationship in the image through a probabilistic graphical model, thereby more accurately locating and analyzing potential defect areas in the image, such as cracks or scratches. Using this method, the image details of each scale are analyzed and extracted layer by layer, thereby obtaining detailed multi-scale defect analysis information, including the scale and location of each defect and its importance assessment relative to the entire component.

[0030] The defect recognition submodule refines image edges based on multi-scale defect analysis information, discovers small scratches or depressions, and analyzes the defect morphology through morphology. Combining edge and morphological features, it identifies the defect type and obtains preliminary defect recognition results. Based on the multi-scale defect analysis information, image edge refinement and mining of small scratches or dents are performed according to the formula: ;

[0031] Calculate defect detection values ​​after edge enhancement , where Representative The gradient value of each pixel, represents the nuclear enhancement factor corresponding to this point, is the total number of pixels involved in the calculation.

[0032] First, for each pixel, calculate its gradient To quantify the edge strength of the image at that point, the gradient is obtained by calculating the grayscale difference between the pixel and its surrounding pixels. For example, for the pixel and the pixel to its right The grayscale values ​​are and ,gradient Then, in order to more effectively identify defect features, a nuclear enhancement factor , adjust the gradient value, and this factor is set according to the relative position of the pixel point and its importance in the defect structure. For example, near the edge of the defect, Greater than in areas away from the defect.

[0033] If there is a 5x5 pixel area, where the grayscale values ​​of the central pixel and its immediate neighbors are , and the nuclear enhancement factors are set as , then the gradient They are ,calculate as follows: ; ;

[0034] The results show that the edge-enhanced image shows a higher defect detection response value in the analysis area, indicating that the edge refinement technology effectively improves the detection capability of small defects such as scratches or dents.

[0035] The position determination submodule locates the coordinates of the identified defects based on the preliminary defect identification results, records the geometric position of the defects, and generates defect location information; The coordinates of each defect in the preliminary defect identification results are located. First, the pixel position of the identified defect in the image is converted into actual geometric coordinates. The process involves coordinate mapping and scale conversion to ensure that the location information of each defect corresponds to the size and shape of the actual component. Then, the specific location of each defect is recorded, such as the distance relative to the boundary of the component and the important functional area. The position data is formatted and saved to form structured defect location information.

[0036] See also Figure 2 and Figure 6 ,The defect analysis module includes a boundary optimization submodule, a defect classification submodule, and a type determination submodule; The boundary optimization submodule adjusts the defect edge based on the defect location information, refines the boundary contrast, adjusts the light, optimizes the visual details of the defect, and obtains a refined defect boundary; Based on the known defect location information, each defect edge is analyzed in detail to identify the irregularity and ambiguity of the boundary. Edge detection algorithms such as Sobel or Laplacian algorithms are used to refine the defect boundary. By enhancing the pixel contrast around the defect, the defect and non-defect areas are clearly distinguished. Then, the light intensity and direction in the image are adjusted, and local lighting adjustment technology is used to improve the visual effect of the defect area, making the defect easier to observe and analyze. Through the processing steps, the optimized defect boundary is visually clearer, thereby providing more accurate basic data for subsequent defect classification and analysis.

[0037] The defect classification submodule compares and classifies the morphological features of different defects based on the refined defect boundaries, and refines them according to the defect size, shape and depth to obtain the defect morphological classification; The morphological characteristics of the difference defects were compared and classified according to the formula: ;

[0038] Calculate defect morphology classification scores , where Representative The size of the defect, Representative The depth of the defect, and are the scoring coefficients for defect size and depth, is the total number of defects.

[0039] Quantify the characteristics of each defect and convert it into a numerical value that can be used for further processing. The total defect classification score is calculated using a weighted average method. The size of each defect and depth Obtained through image analysis technology, involving edge detection and area measurement algorithms, scoring coefficient and Adjustments are made based on the type of defect and its expected impact. For example, for defect types that have a greater impact on vehicle safety, the corresponding coefficient will be set higher.

[0040] If there are three defects, large and small 15, 20 and 25 respectively, depth The rating coefficients of size are 3, 5 and 2 respectively. The depth rating coefficients are 0.6, 0.7 and 0.65 respectively. are 0.4, 0.3 and 0.35 respectively, calculate as follows: ; ; ; ;

[0041] The results show that by comprehensively considering the size and depth of the defects, the average morphological classification score of each defect is about 14.22, which can be used to further guide the priority and method of defect treatment, as well as for subsequent maintenance and quality control decisions.

[0042] The type determination submodule classifies defects based on their morphology. It determines the defect category by evaluating the defect's morphology and impact, and records the classification attributes of each defect to generate a defect type identification record. After the defect morphology classification is completed, the morphological characteristics and potential impact of each defect are evaluated, and a decision tree or neural network model is used to determine the defect category. The model is trained based on the classified morphological characteristics, and the severity and impact range of the defects are analyzed through algorithms. Each defect category is recorded and classified according to the degree of its potential impact on vehicle performance. The determined defect types include but are not limited to cracks, dents, and scratches. Through analysis and evaluation, a defect type identification record is generated.

[0043] See also Figure 2 and Figure 7 ,The decision support module includes the priority analysis submodule, the sorting submodule, and the list generation submodule; The priority analysis submodule quantifies the severity and impact of defects based on defect type identification records, assesses the emergency handling requirements of each defect, adjusts weight factors to define the handling priority, and generates priority quantitative analysis results; Based on the defect type identification records, a quantitative assessment is conducted on the severity and impact of various defects. First, quantitative criteria are defined, such as the size, location, and potential safety risks of the defect, and an initial score is assigned to each defect. Then, the weighting factors are adjusted based on the urgency of the defect handling requirements. For example, defects with a greater impact on driving safety are given a higher weight. The weighting factors are combined through calculation to form the handling priority of each defect.

[0044] The sorting submodule prioritizes defects according to urgency and difficulty of repair based on the priority quantitative analysis results, and updates the sorting order in real time to obtain a priority sorting list; Each defect is sorted according to its urgency and difficulty of repair. A dynamic sorting algorithm is used to update the sorting order based on real-time data. For example, when new defect information is input or the status of existing defects changes, the priority of the defects is automatically re-evaluated and updated to ensure that the repair work can be carried out on the issues that need the most attention in the current situation. Through this real-time update mechanism, a priority sorting list is generated.

[0045] The list generation submodule formulates a repair plan based on the priority list, including repair steps, required materials and estimated time, to obtain a repair priority list; Input a prioritized list and develop a specific repair plan, which includes detailed repair steps, a list of required materials, and an estimated completion time. First, determine the order of repair based on the prioritized list. Then, analyze the specific repair methods and materials required for each defect. By connecting to the material supply database, automatically calculate the cost and supply cycle of the required materials. Finally, integrate all information to generate a comprehensive repair priority list.

[0046] See also Figure 2 and Figure 8 ,The detection and evaluation module includes an evaluation and analysis submodule, a ,parameter adjustment submodule, and a result confirmation submodule; The evaluation and analysis submodule evaluates the detection accuracy and error rate of different defect types based on the repair priority list. Through real-time data monitoring and historical data comparison, it analyzes the stability of the detection process and generates detection stability analysis records. Based on the repair priority list, we perform detection accuracy and error ratio assessments for different defect types, integrate real-time data monitoring and historical data comparison to analyze the stability and consistency of the detection process. We calculate the accuracy and deviation of each defect detection through an algorithm, and evaluate the fluctuation of the detection results through statistical methods such as standard deviation and coefficient of variation, ensuring the objectivity and reliability of the evaluation results. Through the evaluation, we generate a detection stability analysis record, which lists the accuracy indicators and improvement suggestions for each defect detection.

[0047] The parameter adjustment submodule adjusts the detection parameters based on the detection stability analysis record, and adjusts the image analysis threshold and defect detection process according to the error feedback information to obtain the adjusted detection parameters; The detection parameters are adjusted based on the error feedback information in the records. First, the key parameters that need to be adjusted are determined, such as the image analysis threshold and the sensitivity setting of the defect recognition algorithm. Through simulation tests, the differences in detection results before and after each parameter adjustment are evaluated to find the optimal parameter configuration. The adjustment process includes but is not limited to modifying the filter settings and threshold parameters in the image processing algorithm to reduce false positive and false negative results. Through adjustment, a set of adjusted detection parameters are obtained. The parameters are applied to the actual defect detection process to achieve higher detection accuracy.

[0048] The result confirmation submodule re-executes the defect detection process based on the adjusted detection parameters, evaluates the adjustment effect through the verification test set, and generates the detection effect confirmation result; Re-execute the defect detection process to verify the effect of the adjustment, run the defect detection algorithm with the new parameter configuration, and introduce a verification test set. The test set contains sample data with known results to ensure the accuracy and consistency of the evaluation results. Compare the test results before and after the adjustment, analyze the specific impact of the parameter adjustment on the detection accuracy, and generate a confirmation result of the detection effect.

[0049] A vehicle detection method based on machine vision comprises the following steps: S1: Collect vehicle body image data through a camera device, perform distortion correction on the image, and perform histogram equalization on the image data to adjust the image contrast and brightness to obtain a processed image set; S2: Based on the processed image set, the U-Net model is used to extract the geometric shape and texture features of the vehicle body, calculate the morphological and texture feature vectors of the vehicle body parts, and generate a component feature vector set; S3: Based on the component feature vector set, conditional random fields are used to perform multi-scale analysis to detect scratches and dents on the vehicle body surface, including edge detection and morphological analysis, to determine the defect location and generate defect location information; S4: Based on the defect location information, continuously optimize the defect boundary, classify the detected defects, determine the type of each defect, and generate defect type identification records; S5: Based on the defect type identification records, analyze the processing priority of different type defects, sort the defects, evaluate the detection results according to the repair priority, analyze the detection accuracy of multiple types of defects, automatically adjust the detection parameters to optimize the detection process, and obtain the detection effect confirmation results.

[0050] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A car inspection system based on machine vision, characterized in that: The system comprises: The image processing module collects vehicle body image data based on the camera equipment, performs distortion correction on the image, and processes the image data using histogram equalization to adjust the contrast and brightness of the image to obtain a processed image set; The body feature extraction module extracts the geometric shape and texture features of the body parts based on the processed image set using the U-Net model, and calculates the morphological and texture feature vectors of the body parts to obtain a component feature vector set; The defect location module uses a conditional random field to perform multi-scale analysis based on the component feature vector set to detect scratches and dents on the vehicle body surface, and performs edge detection and morphological analysis to determine the defect location and generate defect location information. The defect analysis module continuously optimizes the defect boundary based on the defect location information, classifies the detected defects, determines the type of each defect, and generates a defect type identification record; The decision support module analyzes the processing priority of the difference type defects based on the defect type identification record, sorts the detected defect priorities, and generates a repair priority list; The detection evaluation module evaluates the detection results according to the repair priority list, analyzes the detection accuracy of multiple types of defects, and automatically adjusts the detection parameters to optimize the detection process, generating a detection effect confirmation result.

2. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The processed image set includes a brightness balance image and a color adjustment image; the component feature vector set includes a geometric feature vector, a texture feature vector, and a component boundary vector; the defect location information includes an edge positioning index, a surface roughness parameter, and a defect depth index; the defect type identification record includes a defect shape classification, a defect size classification, and a defect color difference; the repair priority list includes emergency repair projects, medium priority repair projects, and low priority repair projects; and the detection effect confirmation result includes an error rate evaluation result, a detection consistency analysis result, and an adjustment effect evaluation result.

3. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The image processing module includes an image acquisition submodule, an image correction submodule, and an image enhancement submodule; The image acquisition submodule collects vehicle body image data based on the camera device. By automatically triggering the camera device at regular intervals, it collects continuous vehicle body image data streams to obtain the original image sequence. The image correction submodule corrects the image frame by frame based on the original image sequence by adjusting the image acquisition angle and light illumination conditions to obtain a corrected image sequence; The image enhancement submodule applies histogram equalization based on the corrected image sequence, adjusts the image contrast and brightness frame by frame, refines the dynamic range of each frame of image, and generates a processed image set.

4. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The vehicle body feature extraction module includes a feature modeling submodule, a feature calculation submodule, and a feature vector generation submodule; The feature modeling submodule locates the boundaries of the vehicle body parts based on the processed image set, highlights the geometric shape and texture details of the parts through edge enhancement, identifies and records the texture features of the parts, and obtains preliminary feature data; The feature calculation submodule evaluates the spatial relationship and structural layout of the component based on the preliminary feature data and the component morphology, numerically represents the component features, and extracts key feature vectors to obtain feature vector data; The feature vector generation submodule summarizes the feature vectors of the components based on the feature vector data, unifies the formats and standardizes the feature data, and generates a component feature vector set.

5. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The defect location module includes a multi-scale analysis submodule, a defect identification submodule, and a position determination submodule; The multi-scale analysis submodule uses a conditional random field based on the component feature vector set to process image data of different scales, analyze image details at each level, identify potential defect areas, and generate multi-scale defect analysis information; The defect recognition submodule refines the image edges based on the multi-scale defect analysis information, mines small scratches or depressions, and analyzes the defect morphology through morphology. It combines edge and morphological features to identify the defect type and obtain preliminary defect recognition results. The position determination submodule locates the coordinates of the identified defects based on the preliminary defect identification results, records the geometric positions of the defects, and generates defect position information.

6. The machine vision-based automobile inspection system according to claim 5, characterized in that: Based on the multi-scale defect analysis information, image edge refinement and mining are performed on small scratches or depressions according to the formula: ； Calculate defect detection values ​​after edge enhancement , where Representative The gradient value of each pixel, Representative The kernel enhancement factor corresponding to each pixel is is the total number of pixels involved in the calculation.

7. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The defect analysis module includes a boundary optimization submodule, a defect classification submodule, and a type determination submodule; The boundary optimization submodule adjusts the defect edge based on the defect position information, refines the boundary contrast and adjusts the light, optimizes the visual details of the defect, and obtains a refined defect boundary; The defect classification submodule compares and classifies the morphological features of the difference defects based on the refined defect boundaries, and refines the defects according to their size, shape, and depth to obtain defect morphological classification; The type determination submodule determines the defect category based on the defect morphology classification by evaluating the defect morphology and impact degree, records the classification attributes of each defect, and generates a defect type identification record.

8. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The decision support module includes a priority analysis submodule, a sorting submodule, and a list generation submodule; The priority analysis submodule quantitatively evaluates the severity and impact of defects based on the defect type identification records, assesses the emergency handling requirements of each defect, adjusts the weight factors to define the handling priority, and generates priority quantitative analysis results; The sorting submodule prioritizes the defects according to the urgency of processing and the difficulty of repair based on the priority quantitative analysis results, and updates the sorting order in real time to obtain a priority sorting list; The list generation submodule formulates a repair plan based on the priority list, including repair steps, required materials and estimated time, to obtain a repair priority list.

9. The vehicle inspection system based on machine vision according to claim 1, characterized in that: The detection and evaluation module includes an evaluation and analysis submodule, a parameter adjustment submodule, and a result confirmation submodule; The evaluation and analysis submodule evaluates the detection accuracy and error rate of different defect types based on the repair priority list, analyzes the stability of the detection process through real-time data monitoring and historical data comparison, and generates a detection stability analysis record; The parameter adjustment submodule adjusts the detection parameters based on the detection stability analysis record, and adjusts the image analysis threshold and defect detection process according to the error feedback information to obtain the adjusted detection parameters; The result confirmation submodule executes the defect detection process again based on the adjusted detection parameters, evaluates the adjustment effect through the verification test set, and generates a detection effect confirmation result.

10. A vehicle detection method based on machine vision, characterized in that: The vehicle inspection system based on machine vision according to any one of claims 1 to 9 comprises the following steps: The vehicle body image data is collected by a camera device, distortion correction is performed on the image, and histogram equalization is performed on the image data to adjust the image contrast and brightness to obtain a processed image set; Based on the processed image set, using a U-Net model to extract geometric shape and texture features of the vehicle body, calculate the morphological and texture feature vectors of the vehicle body parts, and generate a component feature vector set; Based on the component feature vector set, conditional random fields are used to perform multi-scale analysis to detect scratches and dents on the vehicle body surface, including edge detection and morphological analysis, to determine the defect location and generate defect location information; Based on the defect location information, continuously optimize the defect boundary, classify the detected defects, determine the type of each defect, and generate a defect type identification record; Based on the defect type identification records, the processing priority of different type defects is analyzed, the defect priority is sorted, and the detection results are evaluated according to the repair priority. The detection accuracy of multiple types of defects is analyzed, the detection parameters are automatically adjusted to optimize the detection process, and the detection effect confirmation results are obtained.

Images