Crispy cone package detection system and data processing method applied by same

Abstract

The invention discloses a crispy cone package detection system and a data processing method applied by the same, and relates to the technical field of package detection, the system comprises a conveyor belt, a front sensor, a material stop lever, a camera and a detection processing unit; the conveying belt is used for continuously conveying crackling cone materials, the front sensor is arranged at the inlet end of the conveying belt and used for triggering the conveying belt to suspend when sensing that the crackling cone materials arrive, and the material stop lever is arranged at the suspend position of the conveying belt; the camera is arranged above the material stop lever so as to obtain a detection image of the crispy cone paper sleeve when the material stop lever triggers the conveying belt to suspend, and the lens axis of the camera is perpendicular to the axis of the crispy cone paper sleeve; and the detection processing unit analyzes the number of the paper sleeves of the crispy cone based on the detection image and outputs a paper sleeve number judgment result. According to the invention, the detection efficiency and accuracy of the package detection of the crispy cones are improved.

Description

Technical Field

[0001] This application relates to the field of packaging inspection technology, and in particular to a data processing method for a crispy tube packaging inspection system and its application. Background Technology

[0002] After the crispy cones are made, they usually need to be wrapped in an outer layer of packaging paper before being sold. Because the packaging paper is relatively thin, it often overlaps, so each packaged crispy cone needs to be re-inspected.

[0003] Currently, the main method of re-inspection is manual visual inspection, which involves observing whether two paper sleeves are used to package the same crispy tube at the same time. This method requires a high level of visual acuity and attention, and is prone to false detection. Therefore, it has the shortcomings of poor detection efficiency and accuracy in crispy tube packaging inspection, and urgently needs to be improved. Summary of the Invention

[0004] To improve the efficiency and accuracy of crispy cone packaging inspection, this application provides a crispy cone packaging inspection system and its application data processing method.

[0005] Firstly, the objective of this invention is achieved through the following technical solution: A crispy tube packaging inspection system includes a conveyor belt, a front sensor, a material stop bar, a camera, and an inspection and processing unit. The conveyor belt continuously transports crispy tube materials. The front sensor is located at the inlet end of the conveyor belt and triggers the conveyor belt to stop when crispy tube materials are detected. The material stop bar is located at the stop position of the conveyor belt. The camera is located above the material stop bar to acquire an inspection image of the crispy tube paper sleeve when the material stop bar triggers the conveyor belt to stop. The lens axis of the camera is perpendicular to the axis of the crispy tube paper sleeve. The detection and processing unit analyzes the number of paper sleeves on the crispy tube based on the detected image and outputs the result of the paper sleeve quantity judgment.

[0006] By adopting the above technical solution, the problem of easy misdetection due to manual visual inspection is solved, significantly improving the detection efficiency and accuracy of crispy tube packaging inspection. The material stop bar is used to prevent the crispy tube material from inertial displacement, which helps to ensure the fixed position of the camera and avoid misjudgment caused by image blurring. This application achieves high detection efficiency through automated conveyor belts and front-end sensors. The operating conditions of the crispy tube packaging inspection system include: when the conveyor belt transports the crispy tube material to the sensing position of the front-end sensor, it triggers the conveyor belt to stop, the camera takes a picture to detect whether the paper sleeves overlap, and the detection processing unit analyzes the detection image to determine whether the crispy tube paper sleeves overlap. That is, the detection result is based on the image analysis of the number of crispy tube paper sleeves. If there are two or more paper sleeves, it is determined that there are paper sleeve overlaps and the product packaging is unqualified; otherwise, it is determined that the product is qualified. The crispy tube packaging inspection system of this application can realize continuous crispy tube material processing, reduce manual intervention time, and accurately identify paper sleeve overlap by vertically shooting the paper sleeve axis with a camera and combining image analysis. The detection accuracy is high, and at the same time, the automated detection replaces manual visual inspection, solving the problem of high false detection rate.

[0007] In a preferred embodiment of this application, the detection processing unit further includes: Based on machine vision calibration methods, the intrinsic and extrinsic parameter matrices of the camera are determined, and the geometric projection relationship of the crispy paper sleeve is constructed. Based on the geometric projection relationship, the paper sleeve images of products of different sizes are normalized to a standard coordinate system, and a benchmark detection model is constructed. The original image of the pattern on the surface of the paper sleeve is collected and preprocessed. The texture feature vector of the preprocessed image is extracted and combined with the product pattern type to construct a pattern feature parameter set corresponding to a unique product identifier. The proportion of bright pixels in the image grayscale histogram is calculated to determine the reflectivity level of the paper cover material for different product types. Based on the reflectivity level, the segmentation compensation parameters of the corresponding paper cover area and background area are determined to obtain the reflectivity compensation dataset. In the benchmark detection model, based on the product identification, the pattern feature parameter set and the reflectivity compensation dataset are combined to perform multi-source parameter fusion optimization to determine the detection standard parameters and error allowable range for different product types. The detection processing unit performs the detection operation based on the detection standard parameters and the allowable error range.

[0008] By adopting the above technical solutions, the benchmark detection model normalizes paper sleeve images of different sizes to a standard coordinate system, solving the scale adaptation problem for multi-specification product detection. Further, through pattern feature differentiation analysis of the pattern feature parameter set and high reflectivity detection scenario analysis of the grayscale histogram, the system extracts the surface pattern texture feature vector and constructs a unique pattern feature parameter set corresponding to the product identifier, enabling the system to adaptively distinguish between different product models. Combining the proportion of bright pixels in the grayscale histogram to calculate reflectivity levels, segmentation compensation parameters for different materials are generated, effectively overcoming the problem of region segmentation failure caused by uneven gloss on the paper sleeve surface. This significantly improves the robustness and generalization ability of paper sleeve quantity recognition under complex working conditions, achieving intelligent and compatible detection of packaging across categories and multiple materials.

[0009] Secondly, the objective of this invention is achieved through the following technical solution: A method for detecting crispy cone packaging, applied to a crispy cone packaging detection system as described above, the method comprising: The brittle tube material is transported to the inspection station by a conveyor belt. The arrival of the brittle tube material is detected by a front-mounted sensor, which triggers the conveyor belt to stop. When the conveyor belt is paused, a detection image of the brittle paper sleeve is acquired by a camera positioned above the material stop bar, wherein the lens axis of the camera is perpendicular to the axis of the brittle paper sleeve. The detected image is transmitted to the detection processing unit, which analyzes and identifies the number of brittle paper sleeves in the image based on the detected image. The system outputs information indicating the number of paper sleeves based on the recognition results.

[0010] By adopting the above technical solution, a front-mounted sensor detects the arrival of the crispy tube material in real time and triggers the conveyor belt to stop, ensuring that the object under test remains spatially stationary at the moment of imaging. This fundamentally avoids the image ghosting or defocusing problems caused by material movement on high-speed production lines. When the conveyor belt stops, the camera located directly above the material stop bar immediately acquires an image of the paper sleeve end face. Its lens axis is designed to be strictly perpendicular to the paper sleeve axis, ensuring that the imaging plane is parallel to the target plane, which helps improve the detection accuracy of crispy tube packaging.

[0011] In a preferred embodiment, this application further includes: Acquire the original image data of the camera in the target crispy tube detection scene, and generate a number of image detection units in the detection scene corresponding to the pause position of the conveyor belt based on the original image data; Using each image detection unit as a basic unit, image quality analysis is performed according to a preset first image quality evaluation factor to obtain the corresponding basic image quality score, and the corresponding image quality stability data is calculated based on the basic image quality score. Using each image detection unit as a basic unit, image quality adaptability data is calculated based on a preset second image quality evaluation factor. Based on the image quality adaptability data and image quality stability data corresponding to each image detection unit, a comprehensive quality score result for the crispy tube detection image is obtained.

[0012] To further optimize image detection results and evaluate the applicability of the detected images to packaging inspection tasks, this application introduces an image quality monitoring and evaluation analysis method. Through unitized detection and multi-dimensional quality scoring, dynamic quality control of the crispy packaging tube detection process is achieved. Specifically, the entire detection image is divided into several image detection units representing sub-regions, and quality is evaluated from two dimensions: stability and adaptability. A basic image quality score is calculated using a first image quality evaluation factor, and image quality stability data is derived, which can be used to evaluate the consistency level of repeated imaging at the same detection point at different times or locations. A second image quality evaluation factor measures the image information fidelity. The combination of these two methods avoids misleading identification results from low-quality areas. This application achieves quantitative and hierarchical management of image usability.

[0013] In a preferred embodiment of this application: the acquisition of raw image data from a camera of the target crispy tube detection scene, and the generation of several image detection units in the detection scene corresponding to the conveyor belt pause position based on the raw image data, specifically includes: The raw image data captured by the camera is preprocessed to obtain preprocessed target image data; Based on the preprocessed target image data, the main body area of ​​the crispy paper sleeve and the surrounding packaging background area are identified by the semantic segmentation algorithm to generate the initial image detection unit; Based on the requirements of the crispy tube detection task, the initial image detection unit is subjected to boundary correction and size standardization to obtain the final image detection unit.

[0014] By adopting the above technical solution, the generation of the initial image detection unit relies on semantic understanding rather than simple grid segmentation, which is more in line with the needs of actual detection tasks. For example, the clarity of the paper sleeve edge area is monitored, while the evaluation intensity of the blank background area is reduced. Furthermore, the initial unit is subjected to boundary correction and size standardization processing according to the specific detection task requirements, such as paper sleeve quantity statistics and printing defect detection. This ensures that the detection units in images from different batches and locations have consistent spatial scale and semantic correspondence.

[0015] In a preferred embodiment of this application: the step of using each image detection unit as a basic unit, performing image quality analysis according to a preset first image quality evaluation factor, and obtaining a corresponding basic image quality score includes: The first image quality evaluation factor includes sharpness, contrast, noise level, and illumination uniformity; The image detection unit is divided into several sub-pixel blocks, and each sub-pixel block is assigned a value according to the first image quality evaluation factor to obtain the sub-pixel block quality score. The weighted average of the quality scores of all sub-pixel blocks in a single image detection unit is extracted to calculate the comprehensive score of the first evaluation factor of the single image detection unit, which is the basic score of image quality.

[0016] By employing the above technical solution, values ​​are assigned based on the first image quality evaluation factor to accurately locate blurred, overexposed, or low-contrast areas in the image. The comprehensive score of the first evaluation factor reflects the basic performance capability of the imaging system in local areas. Especially in scenes with uneven lighting or severe lens edge distortion, it can effectively identify weak imaging areas to monitor the sensitivity and response capability of the entire fragile lens detection system to fluctuations in the imaging environment.

[0017] In a preferred embodiment of this application, the step of calculating the corresponding image quality stability data based on the image quality baseline score includes: Acquire multiple frames of detection images of the same crispy tube sample under different imaging devices or different conveyor belt pause positions, and extract the basic image quality score of the corresponding image detection unit for each frame image; Calculate the standard deviation and coefficient of variation of the baseline quality scores of multiple frames of images, and generate image quality stability data based on the standard deviation and coefficient of variation.

[0018] By adopting the above technical solution, multiple sets of detection images of the same brittle tube sample under different imaging devices or multiple frame pause positions are obtained. The basic image quality score of the corresponding image detection unit of each frame is extracted, and then its standard deviation and coefficient of variation are calculated to simulate and analyze the imaging fluctuations caused by mechanical positioning errors, light source fluctuations or small camera displacements in the real production environment. This application comprehensively describes the imaging process of the camera in the long-term production process through image quality assessment to evaluate the reliability of image detection.

[0019] In a preferred embodiment of this application, the step of calculating image quality adaptation data based on a preset second image quality evaluation factor, using each image detection unit as a basic unit, includes: The second image quality evaluation factor includes texture preservation, color fidelity, and defect identifiability; For each image detection unit, the image quality adaptation data is obtained by weighting and summing according to the preset weight of the second image quality evaluation factor.

[0020] By employing the above technical solution, the second image quality evaluation factor is used to assess the information integrity and diagnostic value of image content. Unlike the basic imaging quality, the second image quality evaluation factor focuses more on whether the image is "useful." For example, even if the image is generally clear, if the pattern colors are distorted due to white balance imbalance, the product type may be misidentified; if texture details are lost, it will be difficult to distinguish the boundaries of adjacent paper sleeves. High adaptability means the image is sufficient to support accurate recognition, while low adaptability suggests the need to adjust the light source and retake the image.

[0021] In a preferred embodiment, this application states that obtaining a comprehensive quality score for the crispy tube detection image based on the image quality adaptability data and image quality stability data corresponding to each image detection unit specifically includes: The image quality adaptability data and image quality stability data are integrated using logical fusion rules to generate a comprehensive quality score. The image detection units are graded according to the comprehensive quality score threshold, and the overall quality level of the crispy tube detection image and the local quality score of each image detection unit are output.

[0022] By adopting the above technical solution, a comprehensive and multi-level evaluation of the detected images is achieved. Adaptability reflects whether the image content meets the recognition requirements, while stability reflects whether the imaging process is reliable.

[0023] In a preferred embodiment of this application, the method further includes: A high-precision checkerboard calibration plate is fixedly installed above the conveyor belt. The calibration plate contains multiple black and white checkerboard squares, and the calibration plate remains relatively stationary with respect to the brittle paper sleeve detection area. The camera captures images of the calibration board at a fixed frame rate, detects the grid feet of the chessboard and extracts the corner coordinates, and matches the corner coordinates with preset corner world coordinates to obtain the actual corner coordinates. Based on a pre-built camera pinhole projection model, calculate the theoretical projection coordinates of the camera pinhole projection; Calculate the Euclidean distance between the actual corner coordinates and the theoretical projected coordinates to obtain the single-point reprojection error, and calculate the overall reprojection error based on the single-point reprojection error. When the overall reprojection error of 5 consecutive frames exceeds the preset pose determination threshold, the camera is determined to have a significant pose shift, a camera pose prompt message is generated, and a dynamic parameter recalibration process is triggered.

[0024] By adopting the above technical solution, the pose of the camera is monitored in real time. By acquiring images of the calibration board with the camera at a fixed frame rate, online monitoring of the camera pose status is realized. Specifically, when the error of 5 consecutive frames exceeds the preset threshold, it is determined that the camera has a significant pose shift (such as changes in the installation angle caused by vibration, impact or thermal expansion and contraction). In actual production, once a shift is detected, a prompt message is generated and the parameter dynamic recalibration process is initiated, automatically updating the intrinsic or extrinsic parameter matrix of the benchmark detection model to ensure that the visual model is always aligned with the physical space.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. The linkage design between the conveyor belt and the front sensor enables automatic start and stop control of materials, ensuring the accuracy of the detection sequence. The operating conditions of the crispy tube packaging detection system include: when the conveyor belt transports crispy tube materials to the sensing position of the front sensor, it triggers the conveyor belt to stop, the camera takes a picture to detect whether the paper sleeves overlap, and the detection processing unit analyzes the detection image to determine whether the crispy tube paper sleeves overlap. That is, the detection result is based on the image analysis of the number of crispy tube paper sleeves. If there are two or more paper sleeves, it is determined that there are paper sleeve overlaps and the product packaging is unqualified; otherwise, the product is determined to be qualified.

[0026] 2. By dividing the entire detection image into several detection sub-regions, image detection units are used, and quality is evaluated from two dimensions: stability and adaptability. The basic image quality score is calculated using the first image quality evaluation factor, and image quality stability data is derived, which can be used to evaluate the consistency level of repeated imaging of the same detection point at different times or locations. The second image quality evaluation factor can measure the image information fidelity. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall structure of a crispy tube packaging detection system according to one embodiment of this application; Figure 2 This is a display interface diagram of the quantity judgment results of two types of paper sleeves in a crispy tube packaging detection system according to an embodiment of this application; Figure 3 This is a flowchart of a crispy tube packaging testing method according to one embodiment of this application; Figure 4 This is another flowchart of a crispy tube packaging detection method in one embodiment of this application. Detailed Implementation

[0028] The present application will be further described in detail below with reference to the accompanying drawings.

[0029] In one embodiment, such as Figure 1As shown, this application discloses a crispy tube packaging inspection system. The system includes a conveyor belt, a front sensor, a material stop bar, a camera, and a detection processing unit. The conveyor belt continuously transports the crispy tube material. The front sensor is located at the inlet end of the conveyor belt and triggers a pause when crispy tube material is detected. The material stop bar is located at the pause position of the conveyor belt. The camera, a CCD camera, is positioned above the material stop bar and is mounted 300mm directly above the stop bar. It acquires an image of the crispy tube paper sleeve when the material stop bar triggers the conveyor belt to pause. The camera's lens axis is perpendicular to the axis of the crispy tube paper sleeve. The detection processing unit analyzes the number of crispy tube paper sleeves based on the detection image and outputs a result indicating the number of paper sleeves.

[0030] In this embodiment, the front sensor can be an infrared photoelectric sensor, installed 50mm above the conveyor belt inlet, with a detection distance adjustable from 10-100mm. When the arrival of the brittle drum material is detected, a trigger signal is sent to the PLC controller of the conveyor belt to stop it. The material stop is a cylinder-driven aluminum alloy baffle that extends when the conveyor belt stops, blocking the material movement by ≤2mm to eliminate inertial deviation. Figure 1 The image shows the working state of the material stop bar. The cylinder drives the material stop bar to extend or retract back and forth on the conveyor belt to complete the inspection of multiple batches of brittle tube materials.

[0031] In this embodiment, the detection and processing unit is an embedded industrial control computer. The paper sleeve region segmentation of the detection and processing unit is based on the HSV color space threshold segmentation of the paper sleeve region. The number of paper sleeves is identified by eliminating noise through morphological opening operations and counting the number of paper sleeve contours through connected component analysis, wherein the contour area threshold is 300-500px. 2 If the number of paper sleeves detected is greater than or equal to 2, an NG signal is output and an audible and visual alarm is triggered; if the number is 1, an OK signal is output. Figure 2 As shown, the human-machine interface of the embedded industrial computer displays two different test result interfaces.

[0032] Furthermore, the detection and processing unit also includes determining the intrinsic and extrinsic parameter matrices of the camera based on machine vision calibration methods, and constructing the geometric projection relationship of the crispy paper sleeve; based on the geometric projection relationship, normalizing the paper sleeve images of products of different sizes to a standard coordinate system, and constructing a benchmark detection model; specifically, using the Zhang Zhengyou calibration method, a 10×10 checkerboard calibration board (grid spacing 20mm) is used to calibrate the intrinsic parameter matrix. A camera pinhole projection model of the camera is constructed based on the geometric projection relationship. Where (X) w Y w Z w(u, v) represents the coordinates of a point on the paper sleeve surface in the world coordinate system, and (u, v) represents the pixel coordinates of the image. Images of paper sleeves of different sizes are mapped to the 1920×1080 standard coordinate system through projection transformation for normalization.

[0033] The original image of the paper sleeve surface pattern is acquired and preprocessed. Texture feature vectors are extracted from the preprocessed image. A unique pattern feature parameter set corresponding to the product's pattern type is constructed. Specifically, the original image undergoes Gaussian filtering to remove noise, and then contrast is enhanced using the CLAHE algorithm before texture feature extraction. The paper sleeve's ROI region is divided into 8×8 sub-blocks, and the LBP (Local Binary Pattern) feature of each sub-block is calculated to obtain the LBP feature vector, which serves as the texture feature. Based on different texture features, the corresponding crispy tube product ID is associated and bound to construct a feature vector library, thereby obtaining the pattern feature parameter set.

[0034] The proportion of bright pixels in the image's grayscale histogram is calculated to determine the reflectivity level of the paper sleeve material for different product types. Based on the reflectivity level, the corresponding segmentation compensation parameters for the paper sleeve area and background area are determined, resulting in a reflectivity compensation dataset. Specifically, bright pixels in the image's grayscale histogram are pixels with a brightness > 220. The proportion of bright pixels is the ratio of the number of bright pixels to the total number of pixels. The reflectivity level is divided into three levels: low reflectivity, medium reflectivity, and high reflectivity, based on the proportion of bright pixels. For example, low reflectivity is defined as a bright pixel proportion ≤ 10%, medium reflectivity as a bright pixel proportion > 10% and ≤ 30%, and high reflectivity as a bright pixel proportion > 30%. For example, the segmentation thresholds for the segmentation compensation parameters are: a segmentation threshold of +5 for low reflectivity, +10 for medium reflectivity, and +15 for high reflectivity.

[0035] In the benchmark detection model, multi-source parameter fusion optimization is performed based on product identification, combined with pattern feature parameter sets and reflectivity compensation datasets, to determine the detection standard parameters and allowable error ranges for different product types. The detection processing unit performs detection operations based on the detection standard parameters and allowable error ranges. Specifically, the multi-source parameter fusion optimization adopts a multi-factor weighted fusion optimization method, where the weight factors include pattern feature parameters, reflectivity compensation parameters, and size parameters, with associated weight coefficients of 0.6, 0.3, and 0.1, respectively. The allowable error range includes the number of paper sleeves and the position of the paper sleeves. The number of paper sleeves requires absolute accuracy, with an allowable error range of ±0. The allowable error range for the position of the paper sleeves is ±2 pixels.

[0036] In another embodiment, such as Figure 3As shown, this application also discloses a method for detecting crispy cone packaging, which is applied to a crispy cone packaging detection system. The crispy cone packaging detection method specifically includes the following steps: S1: The brittle tube material is conveyed to the inspection station via a conveyor belt. The arrival of the brittle tube material is detected by a front-mounted sensor, which triggers the conveyor belt to stop.

[0037] S2: When the conveyor belt stops, a detection image of the brittle paper sleeve is obtained by a camera set above the material stop bar, wherein the lens axis of the camera is perpendicular to the axis of the brittle paper sleeve.

[0038] S3: The detection image is transmitted to the detection processing unit, which analyzes and identifies the number of brittle paper sleeves in the image based on the detection image.

[0039] S4: Output the judgment information of the number of paper sleeves based on the recognition result.

[0040] In this embodiment, when the number of crispy paper sleeves in the image is 1, OK is displayed on the HMI interface; when the number is greater than or equal to 2, NG is output and an audible and visual alarm is triggered.

[0041] In one embodiment, such as Figure 4 As shown, a method for detecting crispy tube packaging also includes: S10: Acquire the original image data of the camera in the target brittle tube detection scene, and generate several image detection units in the detection scene corresponding to the pause position of the conveyor belt based on the original image data.

[0042] In this embodiment, the original image is divided into several independent analysis regions to obtain image detection units. The smallest unit size of the image detection unit is 192×108 pixels. Each image detection unit corresponds to a local feature region of a crispy paper sleeve. Adaptive division is performed based on the main structural features of the crispy paper sleeve, ensuring coverage of key detection regions. Image detection units are generated using a semantic segmentation network based on U-Net++.

[0043] Specifically, step S10 includes: S101: Perform data preprocessing on the raw image data captured by the camera to obtain preprocessed target image data.

[0044] In this embodiment, the raw image data is an RGB image captured by the camera when the conveyor belt is paused. Data preprocessing includes noise filtering, grayscale normalization, and color enhancement.

[0045] S102: Based on the preprocessed target image data, the main body area of ​​the crispy paper sleeve and the surrounding packaging background area are identified by the semantic segmentation algorithm to generate the initial image detection unit.

[0046] In this embodiment, the initial image detection unit is the smallest bounding rectangle region containing the complete paper sleeve body. A lightweight segmentation network based on U-Net++ is used, based on 5000 (or more) images... Figure 2 OK or NG labeled image samples are used for model parameter tuning and optimization training to obtain a paper cover region probability map that can perform network inference and threshold segmentation through preprocessed images. Then, a segmentation network model with a binary mask is generated by using a probability threshold (e.g., a probability threshold of 0.85). Paper cover regions with an area greater than or equal to 500 pixels are selected as initial image detection units.

[0047] S103: Based on the requirements of the brittle tube detection task, the initial image detection unit is subjected to boundary correction and size standardization to obtain the final image detection unit.

[0048] In this embodiment, boundary correction refers to eliminating jagged edges or holes in the segmentation edges, filling holes using morphological closing operations, and applying Gaussian filtering (σ = 1.5) to soften the edges and reduce jagged effects. Size standardization refers to unifying the size of the detection units, for example, scaling them to 256×256 pixels while maintaining the aspect ratio, and filling blank areas with black.

[0049] Specifically, the requirements for brittle tube inspection include counting the number of paper sleeves, detecting crease defects, and detecting printing misalignment. The counting objective for paper sleeves is to count the number of paper sleeve outlines; boundary correction prioritizes outline integrity; and the size standardization strategy involves global scaling to maintain proportionality. The crease defect detection objective is to identify creases on the paper sleeve surface; boundary correction focuses on local texture enhancement; and the size standardization strategy involves cropping and enlarging local areas, such as cropping the crease area to 200×200 pixels, and locating the crease position based on texture analysis. The printing misalignment detection objective is to check for pattern positional deviations; boundary correction focuses on edge sharpening, performed using Canny edge detection and Hough line correction; and the size standardization strategy involves proportional scaling of key areas.

[0050] S20: Using each image detection unit as a basic unit, perform image quality analysis based on the preset first image quality evaluation factor to obtain the corresponding basic image quality score, and calculate the corresponding image quality stability data based on the basic image quality score.

[0051] In this embodiment, a four-factor approach is used to cover the core dimensions of image quality, addressing the problem that traditional single indicators cannot comprehensively evaluate industrial inspection scenarios. The first image quality evaluation factor includes sharpness, contrast, noise level, and illumination uniformity. The Sobel operator gradient magnitude mean is used to calculate image sharpness; histogram entropy is used to calculate image contrast; and signal-to-noise ratio (SNR) = 20log(μ / σ) is used to calculate image noise level, where μ is the sub-block pixel mean, reflecting signal strength, and σ is the pixel standard deviation, quantifying noise fluctuation. The SNR score is S. noise SNR values ​​are taken, with values ​​greater than 35dB considered high SNR and values ​​less than 20dB considered low SNR. The illumination uniformity of the image is calculated using the regional brightness variance calculation method to analyze the stability of the light source.

[0052] Specifically, step S20 includes: S201: Divide a single image detection unit into several sub-pixel blocks, assign a value to each sub-pixel block according to the first image quality evaluation factor, and obtain the sub-pixel block quality score.

[0053] In this embodiment, a sub-pixel block refers to dividing a detection unit (e.g., 256×256 pixels) into a 16×16 grid (e.g., a single block size of 16×16 pixels). The sub-pixel block quality score is based on a 0-100 scale and is a four-factor quantitative score for each sub-block. The central area of ​​the grid (8×8 sub-blocks) is the main area of ​​the paper cover, and the edge areas are the background transition areas.

[0054] For example, the gradient magnitude G is calculated using the Sobel operator: Where I represents the input image, referring to the original image data to be processed, and G... x and G y G represents the gradient components of an image in the horizontal and vertical directions. x convolution kernel Used to detect vertical edges. For G y Convolutional kernels are used to detect horizontal edges; the sharpness is divided into... Where k1 = 0.5 is the normalization coefficient.

[0055] Contrast ratio is calculated by calculating the entropy value of the grayscale histogram: ε = 10 -7 Where, p i Let ε be the probability of gray level i appearing; ε = 10 -7 S is a smoothing term used to prevent log(0) from being returned; contrast The larger the entropy value, the more dispersed the response distribution; S contrast Values ​​greater than 6 are considered high entropy values.

[0056] The formula for calculating illumination uniformity is: S illum =100-σ lum ×k2,k2=2.5, where σ lum σ is the standard deviation of brightness in a 4×4 micro-area; k2 is the variance amplification factor. lum ∈[0, 40] mapped to S illum ∈[0, 100]. σ lum A variance less than 10 indicates low variance and uniform illumination. σ lum A value greater than 20 indicates high variance, suggesting the presence of localized shadows or overexposure.

[0057] Then, the formula for calculating the sub-pixel block quality score is: S202: Extract the weighted average of the quality scores of all sub-pixel blocks in a single image detection unit, and calculate the comprehensive score of the first evaluation factor of the single image detection unit, which is the basic score of image quality.

[0058] Specifically, a Gaussian weight matrix W is constructed for spatial weight allocation. 16x16 : The center has the largest weight, such as W(8,8)=0.78; the edge has the smallest weight, W(1,1)=0.02.

[0059] The formula for calculating the weighted average is:

[0060] Furthermore, the crispy tube detection system also incorporates a dynamic compensation mechanism for its detection parameters. For example, if the average score of the central 8×8 area is less than 40, the position of the trigger lever is finely adjusted, such as by a displacement of ±1mm, and / or the brightness of the ring light source of the camera is increased. If the illumination uniformity score is less than 50, multi-frame fusion shooting is initiated, and three consecutive frames are captured to obtain the median brightness value.

[0061] S203: Obtain multiple frames of detection images of the same brittle tube sample under different imaging devices or different conveyor belt pause positions, and extract the basic image quality score of the corresponding image detection unit for each frame image.

[0062] In this embodiment, different imaging devices, such as 5-megapixel or 2-megapixel industrial cameras, and different lenses, such as 25mm / 35mm, are used; different conveyor belt pause positions refer to offset imaging within the range of ±2mm of the stop bar positioning error.

[0063] Specifically, three cameras of different models were used to fix the light source and environment, such as the camera's ring LED 1000 lux and color temperature 6000K, to simulate position offset, control the lever displacement ±2mm, and the conveyor belt pause position deviation ±1mm. Each device acquired 5 frames of images at each position, for a total of 15 frames.

[0064] S204: Calculate the standard deviation and coefficient of variation of the baseline quality scores of multiple frames of images, and generate image quality stability data based on the standard deviation and coefficient of variation.

[0065] In this embodiment, the image quality stability data is an evaluation report generated from the standard deviation and coefficient of variation; the standard deviation σ1 is used to quantify the degree of score fluctuation across multiple frames of the same sample. N1 is the number of image frames, μ1 is the mean standard deviation, and i1 is the image frame variable index; a standard deviation σ1 greater than 5 indicates system instability requiring calibration. Coefficient of variation. A stability rating of CV ≤ 10% is excellent, requiring no intervention. A CV between 10% and 20% is good, requiring monitoring of the detection system. A CV > 20% is poor, requiring immediate calibration of the camera or material stop. A CV > 20% automatically triggers the calibration process. Equipment B's score drops by 12% when there is a +2mm offset; in this case, it is recommended to check the air valve of the material stop's drive motor.

[0066] S30: Using each image detection unit as a basic unit, calculate the image quality adaptability data based on the preset second image quality evaluation factor.

[0067] In this embodiment, the second image quality evaluation factor includes texture retention, color fidelity, and defect identifiability. Texture retention quantifies the integrity of the pattern on the paper sleeve surface, color fidelity is used to evaluate the consistency of the printed colors of the paper sleeve with those of the standard sample, and defect identifiability measures the visual salience of defects such as scratches and wrinkles.

[0068] For example, texture preservation is calculated by extracting high-frequency texture components using wavelet transform, employing the db4 wavelet basis, and using a 3-level decomposition approach to calculate the structural similarity index (SSIM): Where, μ x and μ y Let σ be the pixel mean of image patch x and y. xy Let σ be the covariance of image patch x and y. x and σ y Let x and y be the standard deviations of the image patch; C1 = (0.01L) 2 C1 is the luminance stability constant; C2 = (0.03L) 2C2 is the contrast stability constant; L = 255 is the dynamic range. Texture retention = SSIM × 100, 0-100 scale. SSIM values ​​need to be normalized. SSIM is always in the range of [-1, 1], and the closer the value is to 1, the higher the similarity.

[0069] Color fidelity is calculated by converting the image to the CIELAB color space and then calculating the average color difference ΔE: The standard color chart data is L = 54.2, a = 35.8, and b = 42.1. Each image detection unit is divided into 8×8 sub-blocks, and ΔE is calculated for each sub-block. Outliers with ΔE > 10 are removed. k' is the scaling factor, with a value of 1.5.

[0070] The defect identifiability is calculated by outputting defect confidence scores from a pre-trained Faster R-CNN model, and then calculating the defect identifiability based on these confidence scores. Specifically, the mean confidence score is calculated based on the defect confidence score, and then mapped to a score between 0 and 100. The Faster R-CNN model identifies four types of defects—scratches, wrinkles, overlaps, and stains—based on 256×256 normalized units. A defect is considered a true defect when Conf > 0.7.

[0071] In actual computation, candidate boxes are generated through inference using the Faster R-CNN model, and the average confidence score (Conf) of all candidate boxes is calculated. mean Defect identifiability = Conf mean ×100.

[0072] Specifically, step S30 includes: S301: For each image detection unit, perform weighted summation according to the preset weight of the second image quality evaluation factor to obtain image quality adaptability data.

[0073] Specifically, the image quality fit data (0-100 points) uses a weighted summation formula, with weight coefficients of 0.5, 0.3, and 0.2 for texture preservation, color fidelity, and defect identifiability, respectively. Based on the image quality fit data, if the score is greater than or equal to 85, the sample is directly deemed qualified, and an "OK" message is displayed. If the score is greater than 70 but less than 84, manual review is required. If the score is less than 70, a retake is triggered, and the camera re-acquires the image.

[0074] S40: Based on the image quality adaptability data and image quality stability data corresponding to each image detection unit, obtain the comprehensive quality score result of the crispy tube detection image.

[0075] Specifically, step S40 includes: S401: Integrate image quality adaptability data and image quality stability data through logical fusion rules to generate a comprehensive quality score.

[0076] In this embodiment, the logical fusion rule is a non-linear fusion function. The overall quality score is a global quality index ranging from 0 to 100, reflecting the overall usability of the detected image.

[0077] Specifically, the image quality adaptability data and image quality stability data are first normalized: the image quality adaptability data (S... adapt Mapping to [0, 1]: Image quality stability data (S) stable ) converted to S' stable =e -0.05×CV . The nonlinear fusion function is: Where α is the adaptability weight, representing the intensity task matching degree, with a value of 0.7; β is the stability weight, used to suppress fluctuations in the brittle barrel detection system; and γ is the gain coefficient, with a value of 5. The nonlinear fusion function is defined in S'... adapt When the value is greater than 0.7, the output saturates rapidly; when the CV is greater than 20%, S' stable If the score is less than 0.37, the rating will be forcibly lowered.

[0078] Furthermore, if CV is greater than 15%, material stop lever position calibration is triggered; if S adapt If the value is less than 70, increase the camera's light source brightness: L new =L old ×(1+0.01(70-S adapt )).

[0079] S402: Classify the image detection units according to the comprehensive quality score threshold, and output the overall quality level of the crispy tube detection image and the local quality score of each image detection unit.

[0080] Specifically, based on industrial quality inspection standards, three quality levels are defined, with thresholds of 85 and 70. A score above 85 indicates excellent quality, a score below 85 but above 70 indicates good quality, and a score below 70 indicates poor quality. A tiered response mechanism is established based on the different quality level classifications. For excellent quality, an OK label is directly generated, and the test result is uploaded to a control terminal such as a MES (Manufacturing Execution System). For good quality, the corresponding image detection unit with a low score is highlighted on the HMI (Hybrid Management Interface), and this function outputs an audible and visual prompt for re-inspection. For poor quality, the response is to automatically pause the production line for the brittle drum and send a calibration command to the corresponding connected PLC control terminal.

[0081] In one embodiment, a method for detecting crispy tube packaging further includes: S100: A high-precision checkerboard calibration plate is fixedly installed above the conveyor belt. The calibration plate contains multiple black and white checkerboard patterns, and the calibration plate remains relatively stationary with respect to the brittle paper sleeve detection area. In this embodiment, the chessboard consists of 10×10 alternating black and white squares, each square measuring 20mm×20mm. Each square contains a 1mm diameter circular positioning mark.

[0082] Specifically, the calibration plate is fixed directly above the testing station by a magnetic base, and the relative stillness between it and the brittle paper sleeve testing area means that the positional deviation is ≤0.1mm.

[0083] S200: The camera captures images of the calibration board at a fixed frame rate, detects the grid feet of the chessboard and extracts the corner coordinates, and matches the corner coordinates with the preset corner world coordinates to obtain the actual corner coordinates.

[0084] Specifically, the camera captures images of the calibration board at a fixed frame rate (10 frames per second), and the corner points of the chessboard are detected using OpenCV's findChessboardCorners function to extract the X coordinates of the corner points. img As the actual corner coordinates, and compared with the preset corner world coordinates X world Perform the matching. Adjust the exposure time to 5ms.

[0085] S300: Calculates the theoretical projection coordinates of the camera pinhole projection based on a pre-built camera pinhole projection model.

[0086] In this embodiment, the camera pinhole projection model is a camera pinhole projection model obtained by optimizing and training model parameters based on the principle of camera pinhole projection. The theoretical projection coordinates u are calculated using the camera pinhole projection model. proj =K[R|T]X c Where K is the intrinsic parameter matrix of the camera, [R|T] is the extrinsic parameter matrix of the camera, and X c These are the three-dimensional coordinates of the corner point.

[0087] Specifically, Among them, f x f y c is the focal length in pixels. x c y These are the coordinates of the stationary point. Where R is a 3×3 rotation matrix and T is a 3×1 translation vector. With the center of the calibration plate as the origin, the Z-axis is perpendicular to the plane of the calibration plate.

[0088] S400: Calculate the Euclidean distance between the actual corner coordinates and the theoretical projected coordinates to obtain the single-point reprojection error, and calculate the overall reprojection error based on the single-point reprojection error.

[0089] In this embodiment, the actual corner coordinates X are calculated. img With theoretical projected coordinates u proj The Euclidean distance between them yields the single-point reprojection error e. i =‖X img -u proj ‖2, and calculate the overall reprojection error. Where N is the total number of corner points detected.

[0090] S500: When the overall reprojection error of 5 consecutive frames exceeds the preset pose determination threshold, it determines that the camera has a significant pose shift, generates camera pose prompt information, and triggers the parameter dynamic recalibration process.

[0091] In this embodiment, the preset pose determination threshold is set to When 5 consecutive frames of images e r Greater than When a significant pose shift is detected by the camera, a dynamic parameter recalibration process is triggered. The detection and processing unit records the time and direction of the shift, generates a pose shift alarm log, and sends it to the user terminal.

[0092] The dynamic recalibration process is deeply integrated with the camera pinhole projection model and feature compensation module, specifically including: S5001: Upon receiving the camera pose prompt, the detection processing unit pauses the current detection task and initiates the parameter recalibration sub-process: the control conveyor belt pauses transport, and the robotic arm moves the calibration plate to the center of the detection station; the camera re-acquires a high-resolution image of the calibration plate, and the corner positioning accuracy is improved to 0.1px through a sub-pixel corner detection algorithm. Abnormal corners are removed using the RANSAC algorithm, the camera extrinsic parameter matrix [R′∣T′] is re-estimated, and new camera pinhole projection model parameters are calculated.

[0093] Specifically, the location coordinates of the inspection station are known within the inspection processing unit. The high-resolution calibration plate image is 4096×3072 pixels.

[0094] S5002: Based on the new camera pinhole projection model, the geometric projection relationship of the fragile paper sleeve is recalculated, and the normalized standard coordinate system transformation matrix M is updated. norm =K -1 [R′∣T′].

[0095] Specifically, K -1 The inverse of the intrinsic parameter matrix. S5003: The feature compensation module re-extracts the paper sleeve surface texture feature vector T based on the updated projection model. s And associate it with the product pattern type, and correct the pattern feature parameter set P. tex .

[0096] Specifically, if the pose offset is mainly a rotational component, i.e., the change in R′ > 0.01 rad, then adjust P. tex The texture orientation distribution parameter Δθ in the defect detection CNN network is used to recalculate the texture orientation weight coefficients of the texture branch. If the pose offset is mainly the pose offset translation component Δd, i.e., the change in T′ is greater than 5px, then P is updated. tex The texture scale parameter is adapted to the changes in texture density caused by changes in the image field of view.

[0097] For example, the updated texture direction weight coefficient = original texture direction weight coefficient × (1 + 0.1Δθ); the texture scale parameter is adjusted to the texture scale scaling factor, wherein the adjusted texture scale scaling factor = original texture scale scaling factor × (1 + 0.05Δd).

[0098] S5004: The feature compensation module recalculates the image grayscale histogram, updates the reflectivity level determination based on the new normalized image, and corrects the reflectivity compensation parameter set P. ref .

[0099] Specifically, if the pose shift causes a change in the brightness variance of the image grayscale histogram of >10%, it is determined that it leads to an increase in illumination non-uniformity. In this case, the Otsu segmentation threshold offset ΔT is increased, with an adjustment range of +5 to +20, to improve the accuracy of reflective area segmentation.

[0100] Specifically, if the pose shift causes the appearance of local highlight areas, which are defined as an increase in the proportion of image pixels with a brightness greater than 240 pixels in the image grayscale histogram, then the grayscale stretching coefficient of the highlight areas needs to be adjusted, with an adjustment range of 1.1 to 1.5, in order to suppress highlight interference.

[0101] For example, the segmentation threshold offset ΔT = original segmentation threshold offset + 5 × (adjusted luminance variance / original luminance variance - 1).

[0102] S5005: The detection and processing unit re-executes multi-source parameter fusion optimization based on the corrected pattern feature parameter set and reflection compensation parameter set.

[0103] In this embodiment, the multi-source parameter fusion optimization includes: updating the defect detection confidence threshold and size measurement parameters in the benchmark detection model, calculating the performance difference before and after the parameter update, and if the detection accuracy is improved by ≥2% or the false detection rate is reduced by ≥1%, then the new parameter set is solidified.

[0104] Specifically, the defect detection confidence threshold Th defect The dynamic update formula is: Among them, Th defect_old The previous defect detection confidence threshold; Acc new For the updated detection accuracy; Acc old The accuracy rate of the detection before the update.

[0105] Furthermore, after the parameter recalibration is completed, the detection processing unit generates a calibration report, which includes the pose offset, parameter adjustment details and expected results, and sends it to the user terminal, after which the normal detection process is resumed.

[0106] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0107] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0108] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A crispy tube packaging inspection system, characterized in that, Includes conveyor belt, front sensor, material stop bar, camera and detection and processing unit; The conveyor belt is used to continuously transport the crispy tube material. The front sensor is set at the inlet end of the conveyor belt and is used to trigger the conveyor belt to stop when the crispy tube material is detected. The material stop bar is set at the stop position of the conveyor belt. The camera is set above the material stop bar to acquire a detection image of the crispy tube paper sleeve when the material stop bar triggers the conveyor belt to stop. The lens axis of the camera is perpendicular to the axis of the crispy tube paper sleeve. The detection and processing unit analyzes the number of paper sleeves on the crispy tube based on the detected image and outputs the result of the paper sleeve quantity judgment.

2. The crispy tube packaging inspection system according to claim 1, characterized in that, The detection and processing unit further includes: Based on machine vision calibration methods, the intrinsic and extrinsic parameter matrices of the camera are determined, and the geometric projection relationship of the crispy paper sleeve is constructed. Based on the geometric projection relationship, the paper sleeve images of products of different sizes are normalized to a standard coordinate system, and a benchmark detection model is constructed. The original image of the pattern on the surface of the paper sleeve is collected and preprocessed. The texture feature vector of the preprocessed image is extracted and combined with the product pattern type to construct a pattern feature parameter set corresponding to a unique product identifier. Calculate the proportion of bright pixels in the image grayscale histogram, determine the reflectivity level of paper sleeve material for different product types, and determine the segmentation compensation parameters of the corresponding paper sleeve area and background area based on the reflectivity level to obtain the reflectivity compensation dataset. In the benchmark detection model, based on the product identification, multi-source parameter fusion optimization is performed by combining the pattern feature parameter set and the reflectivity compensation dataset to determine the detection standard parameters and error allowable range for different product types. The detection processing unit performs the detection operation based on the detection standard parameters and the allowable error range.

3. A method for detecting crispy tube packaging, characterized in that, The method, applied to the crispy tube packaging inspection system as described in any one of claims 1 to 2, comprises: The brittle tube material is transported to the inspection station by a conveyor belt. The arrival of the brittle tube material is detected by a front-mounted sensor, which triggers the conveyor belt to stop. When the conveyor belt is paused, a detection image of the brittle paper sleeve is acquired by a camera positioned above the material stop bar, wherein the lens axis of the camera is perpendicular to the axis of the brittle paper sleeve. The detected image is transmitted to the detection processing unit, which analyzes and identifies the number of brittle paper sleeves in the image based on the detected image. The system outputs information indicating the number of paper sleeves based on the recognition results.

4. The method for detecting crispy tube packaging according to claim 3, characterized in that, The method also includes: Acquire the original image data of the camera in the target crispy tube detection scene, and generate a number of image detection units in the detection scene corresponding to the pause position of the conveyor belt based on the original image data; Using each image detection unit as a basic unit, image quality analysis is performed according to a preset first image quality evaluation factor to obtain the corresponding basic image quality score, and the corresponding image quality stability data is calculated based on the basic image quality score. Using each image detection unit as a basic unit, image quality adaptability data is calculated based on a preset second image quality evaluation factor. Based on the image quality adaptability data and image quality stability data corresponding to each image detection unit, a comprehensive quality score result for the crispy tube detection image is obtained.

5. The method for detecting crispy tube packaging according to claim 4, characterized in that, The process involves acquiring raw image data from a camera in the target crispy tube detection scenario, and generating several image detection units based on this raw image data in the detection scenario corresponding to the conveyor belt pause position. Specifically, this includes: The raw image data captured by the camera is preprocessed to obtain preprocessed target image data; Based on the preprocessed target image data, the main body area of ​​the crispy paper sleeve and the surrounding packaging background area are identified by the semantic segmentation algorithm to generate the initial image detection unit; Based on the requirements of the crispy tube detection task, the initial image detection unit is subjected to boundary correction and size standardization to obtain the final image detection unit.

6. The method for detecting crispy tube packaging according to claim 4, characterized in that, The step of taking each image detection unit as a basic unit and performing image quality analysis according to a preset first image quality evaluation factor to obtain a corresponding basic image quality score includes: The first image quality evaluation factor includes sharpness, contrast, noise level, and illumination uniformity; The image detection unit is divided into several sub-pixel blocks, and each sub-pixel block is assigned a value according to the first image quality evaluation factor to obtain the sub-pixel block quality score. The weighted average of the quality scores of all sub-pixel blocks in a single image detection unit is extracted to calculate the comprehensive score of the first evaluation factor of the single image detection unit, which is the basic score of image quality.

7. The method for detecting crispy tube packaging according to claim 6, characterized in that, The step of calculating the corresponding image quality stability data based on the image quality baseline score includes: Acquire multiple frames of detection images of the same crispy tube sample under different imaging devices or different conveyor belt pause positions, and extract the basic image quality score of the corresponding image detection unit for each frame image; Calculate the standard deviation and coefficient of variation of the baseline quality scores of multiple frames of images, and generate image quality stability data based on the standard deviation and coefficient of variation.

8. The method for detecting crispy tube packaging according to claim 4, characterized in that, The step of calculating image quality adaptability data based on a preset second image quality evaluation factor, using each image detection unit as a basic unit, includes: The second image quality evaluation factor includes texture preservation, color fidelity, and defect identifiability; For each image detection unit, the image quality adaptation data is obtained by weighting and summing according to the preset weight of the second image quality evaluation factor.

9. The method for detecting crispy tube packaging according to claim 4, characterized in that, The step of obtaining a comprehensive quality score for the crispy tube detection image based on the image quality adaptability data and image quality stability data corresponding to each image detection unit specifically includes: The image quality adaptability data and image quality stability data are integrated using logical fusion rules to generate a comprehensive quality score. The image detection units are graded according to the comprehensive quality score threshold, and the overall quality level of the crispy tube detection image and the local quality score of each image detection unit are output.

10. The method for detecting crispy tube packaging according to claim 4, characterized in that, The method further includes: A high-precision checkerboard calibration plate is fixedly installed above the conveyor belt. The calibration plate contains multiple black and white checkerboard squares, and the calibration plate remains relatively stationary with respect to the brittle paper sleeve detection area. The camera captures images of the calibration board at a fixed frame rate, detects the grid feet of the chessboard and extracts the corner coordinates, and matches the corner coordinates with preset corner world coordinates to obtain the actual corner coordinates. Based on a pre-built camera pinhole projection model, calculate the theoretical projection coordinates of the camera pinhole projection; Calculate the Euclidean distance between the actual corner coordinates and the theoretical projected coordinates to obtain the single-point reprojection error, and calculate the overall reprojection error based on the single-point reprojection error. When the overall reprojection error of 5 consecutive frames exceeds the preset pose determination threshold, the camera is determined to have a significant pose shift, a camera pose prompt message is generated, and a dynamic parameter recalibration process is triggered.

Images