A method and system for intelligent detection of sports net morphology anomalies based on surveillance images

Abstract

This invention relates to an intelligent detection method and system for abnormal morphology of sports nets based on monitoring images, belonging to the field of sports net production inspection technology. The method includes: acquiring an image to be inspected; classifying it using a calibration model to construct a normal image library; and extracting calibration parameters from normal images in the library; acquiring an image to be identified; comparing its calibration parameters with those of the normal image library to preliminarily determine whether it is a normal or abnormal image; when it is an abnormal image, dividing the abnormal image into regions to obtain offset regions; calculating the relative offset of each offset region to accumulate and generate a correction degree; and comparing the correction degree with a preset correction threshold to determine whether morphological abnormalities exist and outputting control information. By constructing a dynamic normal image library to encompass process fluctuations and employing a matching-then-comparison strategy to identify subtle defects, the method not only improves the accuracy and sensitivity of detection but also enables quantitative analysis of abnormal morphologies, providing data support for the automated and precise control of the production process.

Description

Technical Field

[0001] This invention belongs to the field of sports net production and inspection technology, specifically relating to an intelligent detection method and system for sports net morphological anomalies based on monitoring images. Background Technology

[0002] As a key component of sports facilities and safety systems, the quality and structural integrity of sports nets directly affect the fairness of sports and the safety of users' lives and property. In the industrial production process of sports nets, automated mechanical weaving is usually employed. Therefore, efficient and accurate quality inspection of the shape and structure of sports nets during the production process to eliminate defective products with structural abnormalities is a core link in ensuring the safety and reliability of the final product and is of great significance for improving product quality and enterprise competitiveness.

[0003] However, traditional inspection methods largely rely on manual visual inspection, which is not only labor-intensive and inefficient, but also susceptible to the influence of subjective factors and fatigue on the inspection personnel. This leads to inconsistent inspection standards, high false alarm and missed detection rates, and makes it difficult to meet the needs of large-scale, high-standard production. Furthermore, some automated inspection solutions using traditional machine vision employ relatively rudimentary techniques, typically based on simple image binarization, edge detection, or template matching algorithms. These algorithms are sensitive to changes in lighting, background interference, and subtle differences in the color and material of the mesh itself, exhibiting poor robustness. More importantly, they struggle to effectively identify complex defects with inconspicuous features, such as slight thread deformation, loose nodes, or uneven local tension, resulting in severely insufficient accuracy in identifying complex defects. In addition, on high-speed production lines, continuously acquired high-definition images generate massive amounts of data. Traditional centralized processing architectures face enormous data transmission bandwidth pressure and computational load, making it difficult to achieve real-time response and analysis of anomalies. This processing delay often prevents production equipment from adjusting in a timely manner, potentially causing continuous batches of defective products, leading to serious material waste and economic losses.

[0004] Based on the above-mentioned technical problems, this application proposes an intelligent detection method and system for abnormal sports net morphology based on surveillance images. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent detection method and system for abnormal sports net morphology based on monitoring images, so as to solve the problems of misjudgment or omission caused by manual labeling of detection in the prior art, and the production schedule delay caused by the lack of a rapid response control mechanism.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for intelligent detection of sports net morphology anomalies based on surveillance images includes:

[0008] Acquire the image to be recognized;

[0009] In response to acquiring the image to be identified, the correction degree of the image to be identified is determined based on a comparison with a pre-built normal image library, wherein the normal image library contains normal images and corresponding calibration parameters;

[0010] The correction degree is compared with a preset correction threshold, and when the correction degree is lower than the preset correction threshold, control information indicating the presence of morphological abnormalities is output.

[0011] Preferably, the step of determining the correction degree of the image to be identified based on comparison with a pre-built normal image library includes:

[0012] The calibration parameters of the image to be identified are compared with the calibration parameters in the normal image library to preliminarily determine whether the image to be identified is a normal image or an abnormal image.

[0013] When the image to be identified is initially determined to be an abnormal image, the abnormal image is divided into regions to obtain at least one offset region;

[0014] Calculate the relative offset of each of the aforementioned offset regions;

[0015] The relative offset of the at least one offset region is calculated cumulatively to generate the correction degree.

[0016] Preferably, the step of pre-building a normal image library includes:

[0017] Acquire surveillance images as images to be inspected;

[0018] The image to be inspected is processed by a calibration model to classify it as a normal image or an abnormal image.

[0019] The images to be examined that are classified as normal images are stored in the normal image library as normal images, and the calibration parameters of each normal image are extracted, including the calibration angle.

[0020] Preferably, the step of comparing the calibration parameters of the image to be identified with the calibration parameters in the normal image library to initially determine whether the image to be identified is a normal image or an abnormal image includes:

[0021] An image matching function is used to match the image to be identified with each normal image in the normal image library to determine the optimal matching image;

[0022] The calibration parameters of the image to be identified are compared with the calibration parameters of the best matching image to generate a difference.

[0023] When the difference is not within the preset normal offset range, the image to be identified is initially determined to be an abnormal image.

[0024] Preferably, the image matching function includes a color matching function and an edge alignment function;

[0025] The matching result obtained by the image matching function is a weighted sum of the color matching result and the edge alignment result;

[0026] The color matching result is calculated based on pixel color difference, and the edge alignment result is calculated based on edge contour position deviation.

[0027] Preferably, when the image to be identified is initially determined to be an abnormal image, the step of dividing the abnormal image into regions includes:

[0028] Obtain the total offset angle of the abnormal image and determine the reference angle from the normal image library;

[0029] Calculate the difference between the total offset angle and the reference angle;

[0030] The difference is divided by a preset unit angle to determine the number of offset regions and to divide the abnormal image into regions.

[0031] Preferably, the step of cumulatively calculating the relative offset of the at least one offset region includes:

[0032] The relative offset within each offset region is fitted to a first-order function template to generate an offset function;

[0033] And the offset function is integrated within the domain of each offset region to obtain a cumulative result as the correction degree.

[0034] This invention also discloses an intelligent detection system for abnormal sports net morphology based on surveillance images, comprising:

[0035] The image acquisition module is used to acquire the image to be inspected for building a baseline and the image to be identified for real-time detection.

[0036] The benchmark calibration module is used to classify the images to be inspected acquired by the image acquisition module in order to construct and update a normal image library containing normal images and their corresponding calibration parameters.

[0037] The morphological diagnosis module is configured to, in response to acquiring the image to be identified, determine the correction degree of the image to be identified by comparing the image to be identified with the normal image library;

[0038] The control decision module is used to generate and output control information when a morphological abnormality is determined, based on the comparison result between the correction degree determined by the morphological diagnosis module and the preset correction threshold.

[0039] Preferably, comparing the image to be identified with the normal image library to determine the correction degree of the image to be identified includes:

[0040] The calibration parameters of the image to be identified are compared with the calibration parameters in the normal image library to preliminarily determine whether the image to be identified is a normal image or an abnormal image.

[0041] When the image to be identified is initially determined to be an abnormal image, the abnormal image is divided into regions to obtain at least one offset region, and the relative offset of each offset region is calculated.

[0042] The relative offset of the at least one offset region is calculated cumulatively to generate the correction degree.

[0043] Preferably, the system further includes:

[0044] Image acquisition devices constitute the image acquisition module;

[0045] An image processing platform, wherein the reference calibration module, the morphological diagnosis module, and the control decision module are implemented by computer-executable instructions running on the image processing platform;

[0046] The result output terminal is used to output the control information.

[0047] Beneficial effects

[0048] This invention acquires images to be inspected, classifies them using a calibration model to construct a normal image library, and extracts calibration parameters from the normal images in the library, thereby establishing a dynamic standard for normal morphology. By constructing and utilizing this normal image library, this invention can form a reference based on multiple normal images, rather than relying on a single fixed ideal template. This effectively accommodates process fluctuations during production, avoids misjudgments caused by such fluctuations, and improves the accuracy and environmental adaptability of sports net morphology anomaly detection.

[0049] When determining the image to be identified, this invention uses an image matching function to determine the optimal matching image in a normal image library. Then, it compares the calibration parameters of the image to be identified with the calibration parameters of the optimal matching image to generate a difference for determination. By adopting a matching-then-comparison strategy, it ensures that the image to be identified is evaluated under the most comparable benchmark. Compared with techniques that directly compare with a single template or average standard, this invention can accurately identify deviations caused by morphological anomalies, thereby effectively improving the sensitivity and accuracy of detection and enabling the identification of more subtle morphological anomalies.

[0050] After determining that the image to be identified is an abnormal image, this invention further divides the abnormal image into regions to obtain offset regions, calculates the relative offset of each offset region, and performs cumulative calculations to generate a correction degree. Finally, the correction degree is compared with a preset correction threshold to output control information. Through this quantitative analysis process, this invention can not only determine whether there is a morphological abnormality, but also accurately assess the severity of the morphological abnormality through the correction degree. Thus, it can provide a correction degree as quantitative data for closed-loop control, overcoming the limitation of traditional detection methods that can only output qualitative alarm signals. This provides a direct basis for realizing the automation and precise control of the production process, thereby improving product yield and production efficiency. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0052] Figure 2 This is a schematic diagram of the processing flow of abnormal images according to the present invention;

[0053] Figure 3 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0055] Example 1

[0056] See Figures 1-2 This embodiment provides an intelligent detection method for abnormal morphology of sports nets based on monitoring images, applied to the production quality control of sports nets. It utilizes monitoring images collected during mass production to identify and judge the structural morphology of the sports nets, thereby discovering potential morphological anomalies during production. Based on correction degree analysis, corresponding production process adjustments are made. The specific method includes the following steps:

[0057] S1, Production Process Data Acquisition

[0058] Monitoring cameras are deployed during the mass production of sports nets to capture images of the sports net shapes at each stage of the production line in real time. To ensure the consistency and accuracy of subsequent processing, the monitoring images are subjected to format normalization, sharpness enhancement and noise reduction in sequence. Background difference processing based on reference background images or color segmentation processing based on the known color range of sports net materials are used to remove irrelevant backgrounds or impurities from the images, thereby accurately preserving the main area that can reflect the shape of the sports net.

[0059] To reduce data redundancy and improve subsequent processing efficiency, the process of summarizing the monitoring images also includes performing image deduplication on the monitoring images. Image deduplication is achieved by generating a compact digital fingerprint, i.e., a perceptual hash value, for each image, or by calculating a pixel-level difference matrix between the image and adjacent image frames.

[0060] When the fingerprint difference or the sum of pixel differences is below a preset similarity threshold, the current image is determined to be a duplicate image generated due to a brief pause in the production line or an excessively high camera frame rate and is discarded. The filtered non-duplicate images are collected to generate images to be examined, which are then used for subsequent feature analysis and image library construction.

[0061] S2, Gallery Construction

[0062] The images to be inspected are processed one by one. Image segmentation processes such as threshold segmentation or edge detection are used to separate the mesh structure. Then, feature extraction is performed to identify and quantify the key structural features in each image.

[0063] The key structural features refer to a set of core parameters extracted from the sports net image, i.e. the image to be inspected, which can quantify its shape and quality. Specifically, they include the structural nodes of the sports net, i.e. the intersections of the net, the net texture, i.e. the density or direction of the net, and the corner intersections. All extracted feature parameter values ​​are submitted to a multi-standard evaluation process.

[0064] The multi-criteria evaluation process is a comprehensive image classification mechanism that applies a set of preset classification rules to evaluate images. Each rule corresponds to a specific quality indicator. Specifically:

[0065] The structural integrity rule is used to check whether the number of detected structural nodes is within a standard range. It assesses the integrity of the sports net structure by checking whether the number of detected structural nodes in the image to be inspected is within a preset standard range.

[0066] The node sharpness rule is used to evaluate the average gradient value of the area around each node. It quantifies the sharpness of the node by calculating the average gradient value of the local area around each structural node. The higher the gradient value, the sharper the node.

[0067] Texture consistency rules are used to calculate the variance of the mesh texture across the entire image. By calculating the variance of the mesh texture features across the entire image, they assess the uniformity and consistency of the sports net texture, where the mesh texture features are the mesh spacing or orientation.

[0068] Based on whether the various parameters of the image meet the combination conditions of the classification rules, a classification result is generated, classifying the image to be inspected as a normal image or an abnormal image. According to the classification result, the image to be inspected that is classified as a normal image is stored in the normal image library, and the image to be inspected that is classified as an abnormal image is stored in the abnormal image library. This process is compatible with batch processing and can efficiently build a normal image library for subsequent comparison.

[0069] S3, Calibration Parameter Extraction

[0070] Each normal image in the normal image library is processed to extract its calibration parameters, thereby establishing a parameter index for fast retrieval and comparison;

[0071] The calibration parameters include:

[0072] The calibration angle is used to characterize the overall orientation of the sports net, and it represents the angle formed by the main structural line of the sports net and the reference direction of the production trajectory in each normal image;

[0073] Offset count, which represents the number of structural lines or structural nodes counted within a predefined specific area in a standard morphological sports net image, is used as a benchmark measure of structural density.

[0074] The method for obtaining the calibration angle is as follows: edge detection processing is applied to outline the contour within the sports net image, and then a line detection calculation process is used to identify the dominant line in the image. This dominant line is defined as the main structural line of the sports net, and the angle between it and the preset production trajectory reference direction is calculated. This angle is the calibration angle of the image. The line detection calculation process is preferably based on the Hough transform calculation. The preset production trajectory reference direction is preferably the movement direction of the production line conveyor belt or the horizontal or vertical direction in the image coordinate system.

[0075] The offset amount is determined by counting feature points or detecting lines within a predefined area. All extracted calibration parameters are associated with and stored with the corresponding normal image to form a parameter index table for subsequent image recognition. The predefined area is preferably a specific quadrant of the image or a strip-shaped area along the edge of the production line.

[0076] S4. Image Processing to be Recognized

[0077] The image to be identified is obtained, which is the sports network image on the current production line captured in real time by the image acquisition device. Before the shape determination, the image to be identified is also processed to extract its corresponding calibration parameters. This processing process applies the same edge detection processing, line detection calculation process and feature counting processing sequence as in step S3, so as to obtain the calibration angle and offset of the current image to be identified. The calibration parameters are synchronously organized and prepared for subsequent comparison.

[0078] S5. Determination of Morphological Differences

[0079] The image to be identified is used as input. In order to achieve accurate comparison, the best matching image needs to be found in the normal image library. This process executes a quantitative similarity calculation program, which calculates the image to be identified with each normal image in the normal image library one by one to obtain a set of matching differences.

[0080] The procedure for calculating the quantified similarity includes:

[0081] A color similarity calculation step generates a color matching result based on pixel color difference by accumulating the color difference values ​​of corresponding pixels in the RGB or HSV color space of two images. Specifically, the color matching result is generated by calculating and accumulating the color difference values ​​between the image to be identified and the normal image at corresponding pixel positions. The color matching result represents the degree of difference between the two images in overall color, and is a value generated by the color similarity calculation step.

[0082] An edge alignment calculation step first extracts the edge contour maps of two images using an edge contour extraction process based on the Canny or Sobel operator. Then, an edge alignment result is generated by calculating the positional deviation of the edge contour maps. The edge alignment result represents the degree of alignment and matching of the two images on the structural contour, and is a value generated by the edge alignment calculation step.

[0083] The specific steps are as follows:

[0084] The matching difference is a weighted sum of the color matching result and the edge alignment result. The system sorts the set of matching differences to determine the normal image with the highest matching degree, that is, the normal image with the smallest matching difference, and defines the normal image as the optimal matching image. Specifically, the optimal matching image is regarded as the reference sample that is closest to the shape of the image to be identified.

[0085] Based on the optimal matching image, a comparison region is determined in the image to be identified. The division of the comparison region is based on the key structural nodes of the sports network, and the analysis range is delineated around them by equidistant radial lines or concentric circles.

[0086] Then, the calibration parameters of the image to be identified are retrieved from the result of step S4, and the calibration parameters of the best matching image are retrieved from the parameter index table. The two sets of calibration parameters are compared to generate a difference.

[0087] Based on the difference, a morphological difference determination is made. The difference is compared with a preset normal offset range determined based on the parameter fluctuation statistics of historical normal samples. If the difference is within the preset normal offset range, the image to be identified is determined to be a normal image. If the difference is not within the preset normal offset range, the image to be identified is determined to be an abnormal image, and the subsequent abnormal assessment process is triggered.

[0088] The preset normal offset range refers to a predefined numerical interval, which is determined based on the statistical fluctuations of calibration parameters from a large number of historical normal samples. This interval is used to determine whether the new parameter difference is within an acceptable range of normal variation. The calibration parameters for historical normal samples are the calibration angle and the number of offsets, and the statistical fluctuations are the mean plus or minus three standard deviations.

[0089] S6. Anomaly Quantitative Assessment

[0090] Anomaly quantification assessment is used to accurately quantify the degree of morphological anomaly in an image by dividing the offset region, establishing the offset function, and finally calculating the correction degree. It is an analysis process that is initiated after an image is determined to be abnormal.

[0091] If the image to be identified is determined to be an anomalous image, the system initiates anomaly quantification evaluation, dividing the offset distribution of the anomalous image into regions to obtain at least one offset region. The offset region refers to the sector division of the anomalous image for quantifying local deformation. Each region represents a specific angular range, and its number is determined by the difference between the total offset angle and the reference angle. This division process aims to classify deformations from an angular dimension, where the offset distribution describes the specific manifestation of deformation at different positions and directions, specifically referring to the distribution of the geometric deviation of the anomalous image relative to its normal shape throughout the entire image space.

[0092] The specific steps are as follows:

[0093] By performing a line detection calculation on the abnormal image to determine the dominant direction of its structural lines, the total offset angle of the abnormal image is obtained. The total offset angle represents the degree of rotational deviation of the image as a whole, specifically referring to the current direction angle of the main structural line of the sports net determined by the line detection algorithm in the abnormal image;

[0094] Obtain a reference angle determined from the normal image library. This reference angle can be the average of the calibration angles of all normal images, specifically an angle value representing an ideal or standard direction.

[0095] Calculate the difference between the total offset angle and the reference angle;

[0096] Divide the difference by a preset unit angle to determine the number of offset regions, wherein the preset unit angle is used as a unit of measurement for dividing the offset regions, preferably 15 degrees.

[0097] Based on the number of offset regions, the abnormal image is divided into sectors with the image center as the origin. For each offset region, a mathematical expression describing its local deformation is generated. Specifically, within each offset region, the relative offset path of the sports net structure is extracted by comparing the positions of corresponding structural points in the abnormal image and the optimal matching image. This relative offset path refers to the set of displacement vectors of multiple structural points relative to their corresponding positions in the optimal matching image within a certain offset region, used to describe the local deformation trajectory of that region. These displacement vectors, i.e., their magnitudes, are the relative offsets, representing the distance a structural point has shifted from its standard position. Using these displacement vectors and their spatial positions as input, a linear regression calculation is performed. This calculation fits the data to a preset linear relationship, i.e., a first-order function template, which refers to a linear mathematical model in the form of... In this scheme, the relationship between the relative offset and its spatial location within a specific region is fitted; thereby generating a mathematical expression that describes the displacement trend within that region. This expression is defined as the offset function for that region, and the input to the offset function is the offset region. The location of each structural point inside. and its relative offset The output is a relative offset describing the region. How to define a linear function that varies with spatial location The offset function is:

[0098]

[0099] In the formula: The offset function for region k represents the function that, in the first region k, is used to express the offset function for the region k. Within each offset region, the magnitude of the relative offset varies with its radial distance from the region's center point. A changing linear relationship;

[0100] Represents radial distance, which means the distance of a structure point relative to the center point of its offset region in the polar coordinate system of the image;

[0101] The offset growth rate is represented by a coefficient calculated through linear regression, indicating the value at the 1st... Within each offset region, the relative offset varies with radial distance. The rate of change;

[0102] This represents the base offset, which is the intercept calculated through linear regression, indicating the position of the first element. The center point of each offset region ( The theoretical relative offset at () position;

[0103] Subscript Indicates the relationship with the first Parameters related to each offset region;

[0104] Subscript Used for indexing individual structure points within the offset region.

[0105] The step of accumulating the relative offsets of the at least one offset region to generate a correction degree specifically involves: performing numerical integration on each offset function within its corresponding offset region to obtain the cumulative result for that region; and summing the cumulative results of all offset regions to generate the correction degree.

[0106] Specifically, the correction degree refers to the overall quantitative evaluation index of abnormal morphology in the entire sports network image. It is obtained by integrating and summing the offset functions of all offset regions. The larger the value, the more serious the morphological abnormality, and the greater the production adjustment required.

[0107] S7, Control Information Output

[0108] The output correction degree is dynamically monitored and compared with a preset correction threshold determined based on historical data or engineering experience. This threshold represents the critical point at which production intervention is required. Based on historical production data or engineering experience, if the correction degree exceeds the preset correction threshold, it is determined that there is a morphological abnormality in the current production process, and control information is output. This control information is an adjustment signal or alarm signal sent to the downstream production execution system or maintenance personnel to trigger actions such as parameter fine-tuning, equipment pause, and position reset of the production line, thereby ensuring the structural quality of the subsequent production network. If the correction degree does not exceed the threshold, the current offset is considered to be within an acceptable range, and no control is triggered.

[0109] Example 2

[0110] See Figure 3 This embodiment provides an intelligent detection system for abnormal sports net morphology based on monitoring images. It can accurately identify the presence of abnormal morphology by automatically analyzing sports net images and output control information in a timely manner.

[0111] In its specific implementation, this system may include image acquisition equipment as the hardware foundation, an image processing platform, and a result output terminal. The image acquisition equipment, such as a fixedly installed surveillance camera, is responsible for capturing images of the sports network. The image processing platform, such as a server, personal computer, or embedded system, is the core of the system, and the computer-executable instructions running on it constitute the various logical functional modules of the system. The result output terminal, such as a display screen, alarm, or mobile device application, is used to present the detection results or alarms. Logically, the system can be divided into the following collaborative modules:

[0112] Image acquisition module

[0113] This module can be physically implemented by the image acquisition device, which is configured to perform two image acquisition tasks:

[0114] Firstly, during the system initialization or baseline update phase, monitoring images used to build a normal image library are acquired; these images are referred to as images to be inspected.

[0115] Secondly, during the real-time detection phase of normal system operation, images that need to be morphologically diagnosed are acquired. These images are referred to as images to be identified. This module ensures a stable source of raw image data required for all subsequent analysis and comparison.

[0116] Reference calibration module

[0117] The core of this module is to build and dynamically maintain a normal image library as a comparison benchmark. It receives the images to be inspected provided by the image acquisition module. For each image to be inspected, the module calls a pre-trained calibration model to process it. The calibration model can analyze various features of the image and classify it as a normal image or an abnormal image. The various features of the image include the integrity of the grid, the straightness of the edges, and the overall tension.

[0118] When an image to be inspected is classified as a normal image, the module stores it in the normal image library. At the same time, the module extracts and stores a set of corresponding calibration parameters for each stored normal image. The calibration parameters include at least a calibration angle that can reflect the image shooting angle or the rotation angle of the sports net itself. Through this process, the system establishes a normal image library containing sports net images under various normal conditions and their precise parameters.

[0119] Morphological Diagnosis Module

[0120] This module is the core of morphological anomaly diagnosis. Once an image to be identified is acquired, this module is activated and determines the correction level of the image through a series of rigorous steps. Its specific workflow is as follows:

[0121] A preliminary determination is made by comparing the calibration parameters of the image to be identified with the calibration parameters stored in the normal image library. In one specific embodiment, the module may use an image matching function to perform matching calculations between the image to be identified and each normal image in the normal image library to determine the optimal matching image that is most similar to the image to be identified.

[0122] The image matching function is a comprehensive evaluation function, the result of which is a weighted sum of color matching results and edge alignment results;

[0123] The color matching result is calculated based on pixel color difference, while the edge alignment result is calculated based on edge contour position deviation. After finding the optimal matching image, the module extracts its calibration parameters and compares them with the calibration parameters of the image to be identified, generating a difference value. If the difference value exceeds a preset normal offset range, the module initially determines the image to be identified as an abnormal image and initiates subsequent refined analysis; otherwise, it is determined to be normal, and the diagnostic process ends.

[0124] When the image to be identified is determined to be an abnormal image, the module divides it into regions. In order to achieve the division, the module obtains the total offset angle of the abnormal image and determines a reference angle from the normal image library, where the normal image library is the best matching image determined in the previous step. By calculating the difference between the total offset angle and the reference angle and dividing it by a preset unit angle, the module can determine the number of offset regions into which the abnormal image needs to be divided, and complete the region division of the image accordingly.

[0125] Calculate the relative offset for each of the offset regions, which represents the degree of deviation of the local region from the normal state;

[0126] The relative offsets of all the offset regions are cumulatively calculated to generate a final correction degree. In a preferred embodiment, in order to make the correction degree more accurately reflect the overall offset trend, the relative offsets in each of the offset regions are fitted with a first-order function template to generate an offset function for each region. Then, the correction degree is obtained by integrating each offset function within the domain of its corresponding offset region and summing all the integration results. This correction degree can comprehensively reflect the degree of overall morphological anomaly.

[0127] Regulation and decision-making module

[0128] This module is the system's final decision-making and output unit. It receives the correction degree calculated by the morphology diagnosis module. This module has a preset correction threshold, which represents the upper limit of morphological deviation that the system can tolerate. The control decision module compares the correction degree with the preset correction threshold. When the correction degree exceeds the preset correction threshold, the system determines that there is a morphological abnormality in the current sports net. At this time, the module generates and outputs the control information. This control information can be an alarm signal indicating the abnormality, sent to the result output terminal for display or sounding; it can also be a diagnostic report containing detailed information such as the location and degree of the abnormality for maintenance personnel to refer to; or it can be a control command that triggers automatic or semi-automatic adjustment devices.

[0129] Through the collaborative work of the image acquisition module, benchmark calibration module, morphology diagnosis module, and control decision module, this embodiment can achieve automated, intelligent, and high-precision detection of sports net morphology. By establishing an accurate normal benchmark and combining it with a coarse-to-fine diagnostic logic, it can not only determine whether an abnormality exists, but also quantify the severity of the abnormality.

Claims

1. A method for intelligent detection of abnormal sports net morphology based on surveillance images, characterized in that, include: Acquire the image to be recognized; In response to acquiring the image to be identified, the correction degree of the image to be identified is determined based on a comparison with a pre-built normal image library, wherein the normal image library contains normal images and corresponding calibration parameters; The correction degree is compared with a preset correction threshold, and when the correction degree is lower than the preset correction threshold, control information indicating the presence of morphological abnormalities is output. The step of determining the correction degree of the image to be identified based on comparison with a pre-built normal image library includes: The calibration parameters of the image to be identified are compared with the calibration parameters in the normal image library to preliminarily determine whether the image to be identified is a normal image or an abnormal image. When the image to be identified is initially determined to be an abnormal image, the abnormal image is divided into regions to obtain at least one offset region; Calculate the relative offset of each of the aforementioned offset regions; The relative offset of the at least one offset region is calculated cumulatively to generate the correction degree.

2. The intelligent detection method for sports net morphology anomalies based on surveillance images according to claim 1, characterized in that, The steps for pre-building a normal image library include: Acquire surveillance images as images to be inspected; The image to be inspected is processed by a calibration model to classify it as a normal image or an abnormal image. The images to be examined that are classified as normal images are stored in the normal image library as normal images, and the calibration parameters of each normal image are extracted, including the calibration angle.

3. The intelligent detection method for sports net morphology anomalies based on surveillance images according to claim 2, characterized in that, The step of comparing the calibration parameters of the image to be identified with the calibration parameters in the normal image library to initially determine whether the image to be identified is a normal image or an abnormal image includes: An image matching function is used to match the image to be identified with each normal image in the normal image library to determine the optimal matching image; The calibration parameters of the image to be identified are compared with the calibration parameters of the best matching image to generate a difference. When the difference is not within the preset normal offset range, the image to be identified is initially determined to be an abnormal image.

4. The intelligent detection method for sports net morphology anomalies based on surveillance images according to claim 3, characterized in that, The image matching function includes a color matching function and an edge alignment function; The matching result obtained by the image matching function is a weighted sum of the color matching result and the edge alignment result; The color matching result is calculated based on pixel color difference, and the edge alignment result is calculated based on edge contour position deviation.

5. The intelligent detection method for sports net morphology anomalies based on surveillance images according to claim 2, characterized in that, When the image to be identified is initially determined to be an abnormal image, the steps for dividing the abnormal image into regions include: Obtain the total offset angle of the abnormal image and determine the reference angle from the normal image library; Calculate the difference between the total offset angle and the reference angle; The difference is divided by a preset unit angle to determine the number of offset regions and to divide the abnormal image into regions.

6. The intelligent detection method for abnormal sports net morphology based on surveillance images according to claim 2, characterized in that, The step of cumulatively calculating the relative offset of the at least one offset region includes: The relative offset within each offset region is fitted to a first-order function template to generate an offset function; And the offset function is integrated within the domain of each offset region to obtain a cumulative result as the correction degree.

7. An intelligent detection system for abnormal sports net morphology based on surveillance images, characterized in that, include: The image acquisition module is used to acquire the image to be inspected for building a baseline and the image to be identified for real-time detection. The benchmark calibration module is used to classify the images to be inspected acquired by the image acquisition module in order to construct and update a normal image library containing normal images and their corresponding calibration parameters. The morphological diagnosis module is configured to, in response to acquiring the image to be identified, determine the correction degree of the image to be identified by comparing the image to be identified with the normal image library; The control decision module is used to generate and output control information when a morphological abnormality is determined, based on the comparison result between the correction degree determined by the morphological diagnosis module and the preset correction threshold.

8. The intelligent detection system for abnormal sports net morphology based on monitoring images according to claim 7, characterized in that, The process of comparing the image to be identified with the normal image library to determine the correction degree of the image to be identified includes: The calibration parameters of the image to be identified are compared with the calibration parameters in the normal image library to preliminarily determine whether the image to be identified is a normal image or an abnormal image.

9. The intelligent detection system for abnormal sports net morphology based on monitoring images according to claim 7, characterized in that, The process of comparing the image to be identified with the normal image library to determine the correction degree of the image to be identified further includes: When the image to be identified is initially determined to be an abnormal image, the abnormal image is divided into regions to obtain at least one offset region, and the relative offset of each offset region is calculated. The relative offset of the at least one offset region is calculated cumulatively to generate the correction degree.

10. The intelligent detection system for abnormal sports net morphology based on monitoring images according to claim 7, characterized in that, The system also includes: Image acquisition devices constitute the image acquisition module; An image processing platform, wherein the reference calibration module, the morphological diagnosis module, and the control decision module are implemented by computer-executable instructions running on the image processing platform; The result output terminal is used to output the control information.

Images