Intelligent tire evaluation method and system

Abstract

The application relates to an intelligent tire evaluation method and system, which comprises the following steps: acquiring three-dimensional point cloud data and two-dimensional images of a tire side through a line structure light acquisition system, the line structure light acquisition system comprising a plurality of line structure light cameras, a roller fixing device and a motor-driven rotating platform, the rotating platform driving the tire to rotate at a preset rotating speed, so that full-surface non-dead-angle scanning of the tire side is realized; performing text detection on the two-dimensional images by using a PP-OCRv3 model, dividing the tire side surface into a character area and a non-character area based on the detection result, and acquiring position information of the character area; respectively performing defect detection on the character area and the non-character area, and extracting quantized defect features; inputting the defect features, including a character inclination angle, a character quantity, an average height, and a type and a bounding box area of a defect in the non-character area, into a preset evaluation model, and generating a tire evaluation report containing a safety level and defect information.

Description

Technical Field

[0001] This application relates to the field of tire evaluation technology, and in particular to an intelligent tire evaluation method and system. Background Technology

[0002] With the rapid development of the automotive industry, tires, as a key component affecting vehicle driving safety, have become increasingly important in terms of quality inspection. The tire sidewall surface not only contains character information for identification and traceability, but may also have manufacturing defects such as missing rubber, air bubbles, and damage. Traditional manual visual inspection methods are inefficient, labor-intensive, and the results are easily affected by subjective factors and fatigue, making it difficult to meet the stringent requirements of modern intelligent manufacturing for consistent quality and high efficiency.

[0003] Currently, a machine vision-based tire defect detection method has been proposed in existing technologies. This method typically uses an industrial camera to acquire two-dimensional images of the tire surface, and then employs a general object detection model based on deep learning to identify and locate defects in the images. For characters on the tire sidewall, optical character recognition technology is used for reading and verification to achieve preliminary detection of the character regions.

[0004] However, the above-mentioned technical methods have obvious limitations. The general target detection model is not adaptable enough to the complex and varied background textures and low-contrast defects on the tire sidewall, and the feature extraction is insufficient, which affects the detection accuracy.

[0005] Based on this, this application provides an intelligent tire evaluation method and system. Summary of the Invention

[0006] To address the shortcomings of general target detection models in adapting to complex and varied background textures and low-contrast defects on tire sides, as well as insufficient feature extraction that affects detection accuracy, this application provides an intelligent tire evaluation method and system.

[0007] Firstly, this application provides an intelligent tire evaluation method, which adopts the following technical solution: including: The tire sidewall's three-dimensional point cloud data and two-dimensional image are acquired through a line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, a roller fixing device, and a motor-driven rotating platform. The rotating platform drives the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots. The PP-OCRv3 model is used to perform text detection on the two-dimensional image. Based on the detection results, the tire side surface is divided into character regions and non-character regions, and the position information of the character regions is obtained. Defect detection is performed on the character region and the non-character region respectively, and quantized defect features are extracted; The defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, are input into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

[0008] Preferably, after acquiring the three-dimensional point cloud data and two-dimensional image of the tire sidewall through the line structured light acquisition system, the method further includes: The CLAHE algorithm is used to perform local adaptive histogram equalization on the character region in the two-dimensional image, and the character edge clarity is improved by combining a contrast enhancement algorithm with local mean filtering. The three-dimensional point cloud data is filled with missing data from laser scanning using bilinear interpolation, and the sidewall surface interference is eliminated using a curvature-guided adaptive height correction algorithm. We perform bounding box-level data augmentation on defective sample images, including random rotation, brightness adjustment, and noise addition, which significantly expands the dataset size to improve the model's generalization ability.

[0009] Preferably, the step of using the PP-OCRv3 model to perform text detection on the two-dimensional image, dividing the tire sidewall surface into character regions and non-character regions based on the detection results, and obtaining the position information of the character regions includes: The two-dimensional image is input into the text detection module of the PP-OCRv3 model to obtain one or more detection bounding boxes of text lines. Each detection bounding box is defined by its coordinates in the image coordinate system. The pixel regions covered by all the detected bounding boxes on the two-dimensional image are merged, and the merged connected region is defined as the character region; The coordinates of the bounding rectangle of the character region or the set of coordinates of all the detection bounding boxes contained therein are used as the position information of the character region; All regions in the two-dimensional image other than the character region are defined as the non-character region.

[0010] Preferably, the step of performing defect detection on the character region and the non-character region respectively, and extracting the quantized defect features, includes: Using the character recognition function of the PP-OCRv3 model, character recognition is performed on the character region, and the character tilt angle is calculated and the number of characters is counted based on the recognition results; Based on the position information of the character region, the corresponding depth information is extracted from the 3D point cloud data; the depth information is processed using a curvature-guided adaptive height correction algorithm to obtain corrected depth data; based on the corrected depth data, the average height of the character region is calculated. The non-character regions in the two-dimensional image are input into the improved YOLOv8 model to detect defects such as missing glue, bubbles, and breakage. The improved YOLOv8 model outputs the type, location, and bounding box area of ​​the defects in the non-character regions. The improved YOLOv8 model is trained by replacing the original C2f module with the C2f_SCConv module and introducing the FCA attention mechanism and MPDIoU loss function.

[0011] Preferably, the curvature-guided adaptive height correction algorithm is used to process the depth information to obtain corrected depth data, including: Calculate the local curvature of the height field constructed from the depth information, and obtain the curvature components in the X and Y directions respectively; Based on the curvature components in the X and Y directions, an adaptive weighting factor that is positively correlated with the curvature magnitude is dynamically generated through a predefined curvature threshold and slope coefficient. Based on the height value of a preset reference point in the height field, the height field is nonlinearly weighted by an adaptive weighting factor to eliminate the influence of global curvature and retain the depth features of local characters, thus obtaining the corrected depth data.

[0012] Preferably, the defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, are input into a preset evaluation model to generate a tire evaluation report containing safety level and defect information, including: For each of the aforementioned defect features, based on the type and value, a preset scoring rule library is queried to calculate the severity score of the character region defect and the severity score of the non-character region defect respectively. The severity score of the character region defect is calculated based on the character tilt angle, the number of missing characters, and the difference between the average height and the standard height. The severity score of the non-character region defect is calculated based on the defect type, bounding box area, and depth information. The severity scores of defects in the character area and the severity scores of defects in the non-character area are weighted and fused to obtain the overall defect score of the tire. The comprehensive defect score is compared with the preset safety level threshold range to determine and output the final safety level. Generate a structured tire assessment report, which includes at least the safety level, the type, location, quantitative value, and corresponding defect characteristics of all detected defects.

[0013] Preferably, before generating the structured tire evaluation report, the process further includes: The comprehensive defect score and the key influencing factors selected from the defect features are used together to construct the lifetime prediction input vector. The key influencing factors include at least: the total area of ​​glue-deficient defects in the non-character region, the bounding box area of ​​the largest breakage defect, and the difference between the average height of the character region and the standard height. The life prediction input vector is input into a pre-trained tire life prediction model, which outputs a quantified remaining life prediction value. The tire life prediction model is trained by a regression algorithm based on a dataset containing historical tire defect data and their corresponding actual service life. The remaining life prediction is mapped to a preset life level range to obtain an intuitive life level, which is used to characterize the tire's continued usability potential.

[0014] Secondly, this application discloses an intelligent tire evaluation device, which adopts the following technical solution, including: The data processing module is used to acquire three-dimensional point cloud data and two-dimensional images of the tire sidewall through a line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, roller fixing devices and motor-driven rotating platforms. The rotating platforms drive the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots. The text detection module is used to perform text detection on the two-dimensional image using the PP-OCRv3 model, divide the tire side surface into character regions and non-character regions based on the detection results, and obtain the position information of the character regions. The defect detection module is used to perform defect detection on the character region and the non-character region respectively, and extract the quantized defect features; The report generation module is used to input the defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

[0015] Thirdly, this application also provides a control device, the device comprising: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed, such as the intelligent tire evaluation method described above.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in an intelligent tire evaluation method.

[0017] In summary, this application acquires high-precision 3D point cloud data and 2D images of the tire sidewall using a line structured light acquisition system, providing a complete information foundation that integrates 3D morphology and 2D texture for subsequent analysis. Subsequently, the PP-OCRv3 model is used to intelligently analyze the 2D images, accurately locating and segmenting the character and non-character regions on the tire sidewall surface. To address the characteristics of different regions, a divide-and-conquer defect detection strategy is adopted: For character regions, not only is PP-OCR v3 used to identify character tilt and missing parts, but more importantly, 3D point cloud data is combined with a curvature-guided adaptive height correction algorithm to process depth information and accurately quantify character-related defects. For non-character regions, a specially improved YOLOv8 model is used for detection. This model significantly enhances the feature extraction capability and positioning accuracy of defects such as missing parts, bubbles, and breaks in complex backgrounds by introducing the C2f_SCConv module, FCA attention mechanism, and MPDIoU loss function. Finally, all extracted quantified defect features are input into a pre-set comprehensive evaluation model. This model calculates defect severity scores, weighted fusion to determine the safety level, and can further output the remaining life value based on a historical data-driven prediction model. Ultimately, a structured evaluation report integrating defect details, safety level, and life prediction is generated. This application, through multi-dimensional information fusion, precise regional detection, and intelligent evaluation decision-making, achieves a leap from traditional yes / no judgment to precise quantitative evaluation, comprehensively improving the automation, accuracy, and intelligence level of tire quality inspection. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an intelligent tire evaluation method.

[0019] Figure 2 This is a structural block diagram of an intelligent tire evaluation device. Detailed Implementation

[0020] The following combination Figures 1-2 This application will be described in further detail.

[0021] The core implementation process of this invention mainly includes four collaborative modules: a high-precision data acquisition and preprocessing module, a tire sidewall area intelligent segmentation module, a zoned defect detection and feature quantification module, and a comprehensive evaluation and report generation module. First, the system acquires first-hand three-dimensional and two-dimensional data of the tire sidewall using a carefully designed line structured light acquisition device. Then, advanced OCR technology is used to identify and segment character and non-character areas. Subsequently, based on the characteristics of different areas, deeply optimized dedicated algorithm models are invoked to accurately detect defects and extract quantified features. Finally, a comprehensive evaluation model integrates all information and outputs an authoritative report containing safety level and lifespan predictions.

[0022] Reference Figure 1 The embodiments of this application include at least steps S10 to S40.

[0023] S10 acquires three-dimensional point cloud data and two-dimensional images of the tire sidewall through a line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, roller fixing devices, and a motor-driven rotating platform. The rotating platform drives the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots.

[0024] S20. The PP-OCRv3 model is used to perform text detection on the two-dimensional image. Based on the detection results, the tire side surface is divided into character regions and non-character regions, and the position information of the character regions is obtained.

[0025] S30 performs defect detection on both character and non-character regions and extracts the quantized defect features.

[0026] S40 inputs defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

[0027] Specifically, the system integrates multiple line structured light cameras to simultaneously acquire 3D point cloud and 2D image data of the tire sidewall surface during tire rotation. Then, the PP-OCRv3 model is used to perform text detection and localization on the 2D images, precisely dividing the tire sidewall into two main regions: character and non-character areas. Subsequently, a partitioned detection strategy is implemented, performing character recognition and 3D depth analysis on the character areas to quantify tilt, missing, and glue-deficient defects. Simultaneously, an improved YOLOv8 model is used to detect and quantify glue-deficient, bubble, and breakage defects in the non-character areas. Finally, all extracted quantitative features are input into the evaluation model, and a comprehensive evaluation report containing a clear safety level and detailed defect information is generated through fusion calculation, thus achieving fully automated evaluation from multi-dimensional data acquisition to intelligent decision output.

[0028] In some embodiments, the data preprocessing steps are as follows: Local adaptive histogram equalization is performed on the character regions in the two-dimensional image using the CLAHE algorithm, and the character edge clarity is improved by combining a contrast enhancement algorithm with local mean filtering; Bilinear interpolation is used to fill in the missing data from the laser scan on the three-dimensional point cloud data, and curvature-guided adaptive height correction algorithm is used to eliminate interference from the tire sidewall surface; Boundary box-level data augmentation is performed on the defective sample images, including random rotation, brightness adjustment, and noise addition operations, significantly expanding the dataset size to improve the model's generalization ability.

[0029] Specifically, the CLAHE algorithm combined with local mean filtering is applied sequentially to the 2D image of the character region for contrast enhancement, effectively solving the problems of low contrast and blurred edges of the characters on the tire sidewall. For the 3D point cloud data, bilinear interpolation and curvature-guided height correction are performed sequentially to ensure the integrity and accuracy of the depth data and eliminate surface measurement errors. Simultaneously, bounding box-level data augmentation is applied to the defect samples, significantly expanding the scale and diversity of the training data by simulating morphological and lighting changes in real-world scenarios. These three preprocessing steps work synergistically, comprehensively improving the accuracy and robustness of subsequent defect detection and assessment from three dimensions: image quality improvement, 3D data optimization, and sample diversity enhancement.

[0030] In some embodiments, step S20 specifically includes the following steps: inputting a two-dimensional image into the text detection module of the PP-OCRv3 model to obtain detection bounding boxes of one or more text lines, each detection bounding box being defined by its coordinates in the image coordinate system; merging the pixel regions covered by all detection bounding boxes on the two-dimensional image, defining the merged connected region as a character region; using the coordinates of the bounding rectangle of the character region or the set of coordinates of all detection bounding boxes contained therein as the position information of the character region; defining all regions in the two-dimensional image other than the character region as non-character regions.

[0031] Specifically, the PP-OCRv3 model is used to intelligently segment the tire sidewall region: First, its text detection module accurately locates all text lines in the image and generates detection bounding boxes; then, the pixel areas covered by these bounding boxes are merged into a unified character region, ensuring complete coverage of the character region; next, the coordinates of the bounding rectangle of this character region or the original set of bounding boxes are extracted as precise location information; finally, the remaining part of the image is defined as a non-character region. This process achieves automatic recognition and precise positioning of character regions, establishing a reliable spatial benchmark for subsequent partition detection, while ensuring the completeness and omission-free detection of both character and non-character regions, thus facilitating differentiated defect detection.

[0032] In some embodiments, step S30 specifically includes the following steps: using the character recognition function of the PP-OCRv3 model to perform character recognition on the character region, calculating the character tilt angle and counting the number of characters based on the recognition result; extracting the corresponding depth information from the three-dimensional point cloud data based on the position information of the character region; processing the depth information using a curvature-guided adaptive height correction algorithm to obtain corrected depth data; calculating the average height of the character region based on the corrected depth data; inputting the non-character region in the two-dimensional image into the improved YOLOv8 model to detect defects such as missing glue, bubbles, and breakage, and the improved YOLOv8 model outputs the type, location, and bounding box area of ​​the defects in the non-character region; wherein, the improved YOLOv8 model is a model trained by replacing the original C2f module with the C2f_SCConv module and introducing the FCA attention mechanism and MPDIoU loss function.

[0033] Furthermore, a curvature-guided adaptive height correction algorithm is used to process the depth information to obtain corrected depth data. This includes: calculating the local curvature of the height field constructed from the depth information to obtain the curvature components in the X and Y directions; dynamically generating an adaptive weighting factor positively correlated with the curvature magnitude based on the curvature components in the X and Y directions through a predefined curvature threshold and slope coefficient; and performing nonlinear weighted calculation on the height field based on the height value of a preset reference point in the height field, combined with the adaptive weighting factor, to eliminate the influence of global curvature and retain the depth features of local characters, thereby obtaining corrected depth data.

[0034] Specifically, for character areas, the tilt angle and number of characters are calculated based on the recognition results of PP-OCRv3. Simultaneously, the average height is calculated based on 3D point cloud data after curvature correction, effectively detecting character shape and glue defects. For non-character areas, an improved YOLOv8 model integrating the C2f_SCConv module, FCA attention mechanism, and MPDIoU loss function is used to accurately locate glue defects, bubbles, and breakage defects in complex backgrounds and output quantitative data on type, location, and area. Thus, through a technical approach combining 2D recognition and 3D measurement, and complementing general detection with specialized optimization, multi-dimensional and accurate quantification of various defects on the tire sidewall surface is achieved, providing feature data support for subsequent comprehensive evaluation.

[0035] In some embodiments, step S40 specifically includes the following steps: For each defect feature, based on the type and value, query a preset scoring rule library to calculate the character area defect severity score and the non-character area defect severity score respectively; the character area defect severity score is calculated based on the character tilt angle, the number of missing characters, and the difference between the average height and the standard height, while the non-character area defect severity score is calculated based on the defect type, bounding box area, and depth information; the character area defect severity score and the non-character area defect severity score are weighted and fused to obtain the tire's comprehensive defect score; the comprehensive defect score is compared with a preset safety level threshold range to determine and output the final safety level; a structured tire evaluation report is generated, which includes at least the safety level, the type, location, quantitative value, and corresponding defect features of all detected defects.

[0036] Specifically, based on the defect type and numerical query scoring rule library, the severity scores for the character area and non-character area are calculated separately. Then, a weighted fusion is performed to obtain a comprehensive defect score, which is compared with a preset safety level threshold to determine the final safety level. Finally, a structured report containing the safety level, defect type, location, and quantified numerical values ​​is generated. This evaluation mechanism transforms multi-dimensional defect characteristics into a unified quality scoring system, achieving a leap from single defect detection to overall safety assessment. It provides an objective and accurate decision-making basis for tire quality judgment, significantly enhancing the practical value of the test results.

[0037] In some embodiments, considering the assessment of tire remaining life, the corresponding processing steps are as follows: A comprehensive defect score and key influencing factors selected from defect features are jointly constructed as a life prediction input vector. The key influencing factors include at least: the total area of ​​missing rubber defects in the non-character region, the bounding box area of ​​the largest breakage defect, and the difference between the average height and the standard height of the character region; the life prediction input vector is input into a pre-trained tire life prediction model, which outputs a quantified remaining life prediction value. The tire life prediction model is trained using a regression algorithm based on a dataset containing historical tire defect data and their corresponding actual service life; the remaining life prediction value is mapped to a preset life level range to obtain an intuitive life level, which characterizes the tire's continued usability potential.

[0038] Specifically, by constructing a life prediction input vector, integrating key influencing factors such as comprehensive defect scores, total area of ​​missing rubber, maximum damage area, and character height difference, this vector is input into a regression model trained based on historical data to obtain a quantified remaining life value, which is then mapped to an intuitive life level. This solution achieves a leap from static defect detection to dynamic life prediction, providing support for tire use and maintenance through correlation analysis of multi-dimensional defect data and historical failure patterns.

[0039] The implementation principle of the intelligent tire evaluation method in this application is as follows: High-precision three-dimensional point cloud data and two-dimensional images of the tire sidewall are acquired through a line structured light acquisition system, providing a complete information foundation integrating three-dimensional morphology and two-dimensional texture for subsequent analysis. Subsequently, the PP-OCR v3 model is used to intelligently analyze the two-dimensional image, accurately locating and segmenting the character and non-character regions on the tire sidewall surface. To address the characteristics of different regions, a divide-and-conquer defect detection strategy is adopted: For character regions, not only is PP-OCR v3 used to identify character tilt and missing parts, but more importantly, 3D point cloud data is combined with a curvature-guided adaptive height correction algorithm to process depth information and accurately quantify character-related defects. For non-character regions, a specially improved YOLOv8 model is used for detection. This model significantly enhances the feature extraction capability and positioning accuracy of defects such as missing parts, bubbles, and breaks in complex backgrounds by introducing the C2f_SCConv module, FCA attention mechanism, and MPDIoU loss function. Finally, all extracted quantified defect features are input into a pre-set comprehensive evaluation model. This model calculates defect severity scores, weighted fusion to determine the safety level, and can further output the remaining life value based on a historical data-driven prediction model. Ultimately, a structured evaluation report integrating defect details, safety level, and life prediction is generated. This application, through multi-dimensional information fusion, precise regional detection, and intelligent evaluation decision-making, achieves a leap from traditional yes / no judgment to precise quantitative evaluation, comprehensively improving the automation, accuracy, and intelligence level of tire quality inspection.

[0040] Figure 1 This is a flowchart illustrating an intelligent tire evaluation method in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0041] Based on the same technical concept, referring to Figure 2 This application also provides an intelligent tire evaluation device, which adopts the following technical solution: the device includes: The data processing module is used to process multi-source data. The data processing module is used to acquire three-dimensional point cloud data and two-dimensional images of the tire sidewall through the line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, roller fixing devices and motor-driven rotating platforms. The rotating platforms drive the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots. The text detection module is used to perform text detection on two-dimensional images using the PP-OCRv3 model. Based on the detection results, the tire side surface is divided into character regions and non-character regions, and the position information of the character regions is obtained. The defect detection module is used to perform defect detection on character regions and non-character regions respectively, and extract quantified defect features; The report generation module is used to input defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

[0042] In some embodiments, the data processing module is specifically used to perform local adaptive histogram equalization on the character regions in the two-dimensional image using the CLAHE algorithm, and to improve the clarity of character edges by combining a contrast enhancement algorithm with local mean filtering. Bilinear interpolation is used to fill in the missing data from laser scanning of the 3D point cloud data, and curvature-guided adaptive height correction algorithm is used to eliminate interference from the tire sidewall surface. We perform bounding box-level data augmentation on defective sample images, including random rotation, brightness adjustment, and noise addition, which significantly expands the dataset size to improve the model's generalization ability.

[0043] In some embodiments, the text detection module is specifically used to input a two-dimensional image into the text detection module of the PP-OCRv3 model to obtain detection bounding boxes of one or more lines of text, each detection bounding box being defined by its coordinates in the image coordinate system; Merge the pixel regions covered by all detected bounding boxes on the 2D image, and define the merged connected region as the character region; The coordinates of the bounding rectangle of the character region or the set of coordinates of all the detection bounding boxes contained therein are used as the position information of the character region; All regions in a two-dimensional image other than character regions are defined as non-character regions.

[0044] In some embodiments, the defect detection module is specifically used to utilize the character recognition function of the PP-OCRv3 model to perform character recognition on the character region, calculate the character tilt angle based on the recognition result, and count the number of characters. Based on the positional information of the character region, the corresponding depth information is extracted from the 3D point cloud data; a curvature-guided adaptive height correction algorithm is used to process the depth information to obtain corrected depth data; based on the corrected depth data, the average height of the character region is calculated. The non-character regions in a 2D image are input into an improved YOLOv8 model to detect defects such as missing glue, bubbles, and breakage. The improved YOLOv8 model outputs the type, location, and bounding box area of ​​the defects in the non-character regions. The improved YOLOv8 model is trained by replacing the original C2f module with the C2f_SCConv module and introducing the FCA attention mechanism and MPDIoU loss function.

[0045] In some embodiments, the defect detection module is specifically used to calculate the local curvature of the height field constructed by the depth information, and obtain the curvature components in the X and Y directions respectively. Based on the curvature components in the X and Y directions, an adaptive weighting factor that is positively correlated with the curvature magnitude is dynamically generated through a predefined curvature threshold and slope coefficient. Based on the height value of a preset reference point in the height field, the height field is nonlinearly weighted by an adaptive weighting factor to eliminate the influence of global curvature and retain the depth features of local characters, thus obtaining the corrected depth data.

[0046] In some embodiments, the report generation module is specifically used to query a preset scoring rule library for each defect feature, based on its type and value, and calculate the character region defect severity score and the non-character region defect severity score respectively; the character region defect severity score is calculated based on the character tilt angle, the number of missing characters, and the difference between the average height and the standard height, while the non-character region defect severity score is calculated based on the defect type, bounding box area, and depth information. The severity scores of defects in the character area and the severity scores of defects in the non-character area are weighted and fused to obtain the overall defect score of the tire. The comprehensive defect score is compared with the preset safety level threshold range to determine and output the final safety level. Generate a structured tire assessment report, which includes at least the safety level, the type, location, quantitative value, and corresponding defect characteristics of all detected defects.

[0047] In some embodiments, the report generation module is also used to construct a lifetime prediction input vector by combining the comprehensive defect score and key influencing factors selected from the defect features. The key influencing factors include at least: the total area of ​​glue-deficient defects in the non-character region, the bounding box area of ​​the largest breakage defect, and the difference between the average height of the character region and the standard height. The tire life prediction input vector is fed into a pre-trained tire life prediction model, which outputs a quantified remaining life prediction value. The tire life prediction model is trained by a regression algorithm based on a dataset containing historical tire defect data and their corresponding actual service life. The remaining life prediction is mapped to a preset life level range to obtain an intuitive life level, which is used to characterize the tire's continued usability potential.

[0048] This application also discloses a control device.

[0049] Specifically, the control device includes a memory and a processor, the memory storing a computer program that can be loaded and executed by the processor to perform the aforementioned intelligent tire evaluation method.

[0050] This application also discloses a computer-readable storage medium.

[0051] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed as described above in an intelligent tire evaluation method. The computer-readable storage medium includes, for example, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0052] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An intelligent tire evaluation method, characterized in that, include: The tire sidewall's three-dimensional point cloud data and two-dimensional image are acquired through a line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, a roller fixing device, and a motor-driven rotating platform. The rotating platform drives the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots. The PP-OCRv3 model is used to perform text detection on the two-dimensional image. Based on the detection results, the tire side surface is divided into character regions and non-character regions, and the position information of the character regions is obtained. Defect detection is performed on the character region and the non-character region respectively, and quantized defect features are extracted; The defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, are input into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

2. The intelligent tire evaluation method according to claim 1, characterized in that, After acquiring the three-dimensional point cloud data and two-dimensional image of the tire sidewall through the line structured light acquisition system, the method further includes: The CLAHE algorithm is used to perform local adaptive histogram equalization on the character region in the two-dimensional image, and the character edge clarity is improved by combining a contrast enhancement algorithm with local mean filtering. The three-dimensional point cloud data is filled with missing data from laser scanning using bilinear interpolation, and the sidewall surface interference is eliminated using a curvature-guided adaptive height correction algorithm. We perform bounding box-level data augmentation on defective sample images, including random rotation, brightness adjustment, and noise addition, which significantly expands the dataset size to improve the model's generalization ability.

3. The intelligent tire evaluation method according to claim 2, characterized in that, The step of using the PP-OCRv3 model to perform text detection on the two-dimensional image, dividing the tire sidewall surface into character regions and non-character regions based on the detection results, and obtaining the position information of the character regions includes: The two-dimensional image is input into the text detection module of the PP-OCRv3 model to obtain one or more detection bounding boxes of text lines. Each detection bounding box is defined by its coordinates in the image coordinate system. The pixel regions covered by all the detected bounding boxes on the two-dimensional image are merged, and the merged connected region is defined as the character region; The coordinates of the bounding rectangle of the character region or the set of coordinates of all the detection bounding boxes contained therein are used as the position information of the character region; All regions in the two-dimensional image other than the character region are defined as the non-character region.

4. The intelligent tire evaluation method according to claim 3, characterized in that, The step of performing defect detection on the character region and the non-character region respectively, and extracting the quantized defect features, includes: Using the character recognition function of the PP-OCRv3 model, character recognition is performed on the character region, and the character tilt angle is calculated and the number of characters is counted based on the recognition results; Based on the position information of the character region, the corresponding depth information is extracted from the 3D point cloud data; the depth information is processed using a curvature-guided adaptive height correction algorithm to obtain corrected depth data; based on the corrected depth data, the average height of the character region is calculated. The non-character regions in the two-dimensional image are input into the improved YOLOv8 model to detect defects such as missing glue, bubbles, and breakage. The improved YOLOv8 model outputs the type, location, and bounding box area of ​​the defects in the non-character regions. The improved YOLOv8 model is trained by replacing the original C2f module with the C2f_SCConv module and introducing the FCA attention mechanism and MPDIoU loss function.

5. The intelligent tire evaluation method according to claim 4, characterized in that, The curvature-guided adaptive height correction algorithm processes the depth information to obtain corrected depth data, including: Calculate the local curvature of the height field constructed from the depth information, and obtain the curvature components in the X and Y directions respectively; Based on the curvature components in the X and Y directions, an adaptive weighting factor that is positively correlated with the curvature magnitude is dynamically generated through a predefined curvature threshold and slope coefficient. Based on the height value of a preset reference point in the height field, the height field is nonlinearly weighted by an adaptive weighting factor to eliminate the influence of global curvature and retain the depth features of local characters, thus obtaining the corrected depth data.

6. The intelligent tire evaluation method according to claim 1, characterized in that, The defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, are input into a preset evaluation model to generate a tire evaluation report containing safety level and defect information, including: For each of the aforementioned defect features, based on the type and value, a preset scoring rule library is queried to calculate the severity score of the character region defect and the severity score of the non-character region defect respectively. The severity score of the character region defect is calculated based on the character tilt angle, the number of missing characters, and the difference between the average height and the standard height. The severity score of the non-character region defect is calculated based on the defect type, the bounding box area, and the depth information. The severity scores of defects in the character area and the severity scores of defects in the non-character area are weighted and fused to obtain the overall defect score of the tire. The comprehensive defect score is compared with the preset safety level threshold range to determine and output the final safety level. Generate a structured tire assessment report, which includes at least the safety level, the type, location, quantitative value, and corresponding defect characteristics of all detected defects.

7. The intelligent tire evaluation method according to claim 6, characterized in that, Prior to generating the structured tire evaluation report, the following is also included: The comprehensive defect score and the key influencing factors selected from the defect features are used together to construct the lifetime prediction input vector. The key influencing factors include at least: the total area of ​​glue-deficient defects in the non-character region, the bounding box area of ​​the largest breakage defect, and the difference between the average height and the standard height of the character region. The life prediction input vector is input into a pre-trained tire life prediction model, which outputs a quantified remaining life prediction value. The tire life prediction model is trained by a regression algorithm based on a dataset containing historical tire defect data and their corresponding actual service life. The remaining life prediction is mapped to a preset life level range to obtain an intuitive life level, which is used to characterize the tire's continued usability potential.

8. An intelligent tire evaluation device, characterized in that, The device includes: The data processing module is used to acquire three-dimensional point cloud data and two-dimensional images of the tire sidewall through a line structured light acquisition system. The line structured light acquisition system includes multiple line structured light cameras, roller fixing devices and motor-driven rotating platforms. The rotating platforms drive the tire to rotate at a preset speed to achieve full-surface scanning of the tire sidewall without blind spots. The text detection module is used to perform text detection on the two-dimensional image using the PP-OCRv3 model, divide the tire side surface into character regions and non-character regions based on the detection results, and obtain the position information of the character regions. The defect detection module is used to perform defect detection on the character region and the non-character region respectively, and extract the quantized defect features; The report generation module is used to input the defect features, including character tilt angle, number of characters, average height, and type and bounding box area of ​​non-character area defects, into a preset evaluation model to generate a tire evaluation report containing safety level and defect information.

9. A control device, characterized in that, The device includes: A memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images