Monitoring Method and System for Poker Card Production Line

Abstract

The present invention provides a monitoring method and system for a playing card production line. The monitoring method for the playing card production line includes: receiving the original video of the playing card production line collected by a monitoring device; splitting the original video into multiple original video frames, and screening out at least two consecutive original video frames as target images based on an image feature matching algorithm, and respectively adjusting the size of each target image to be consistent with the size of a preset standard image to obtain multiple standard target images; splicing the multiple standard target images into a standard target image sequence according to the sequence of the time nodes corresponding to each standard target image; inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, and when the production status is abnormal, outputting an alarm message, thereby improving the monitoring efficiency, timeliness and accuracy of the playing card production line.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring of a playing card production line, and particularly relates to a monitoring method and system for a playing card production line. Background Art

[0002] A playing card production line is a highly automated production process, which involves multiple steps, from the preparation of raw materials to the packaging and quality inspection of the final product, and requires multi-faceted monitoring to ensure the normal operation of the production process and its safety during production.

[0003] Traditional monitoring methods for playing card production lines mainly rely on manual inspections and off-line detections. For example, repairs are only carried out when production equipment malfunctions, making it difficult to issue effective early warnings, and there are problems such as untimely, inaccurate, and inefficient monitoring.

[0004] In the technical solution with the application number CN201310597085.1, although it irradiates playing cards by generating light under set conditions, takes images of the playing cards and identifies them, and matches the target features extracted from the playing card images with the template features stored in the database to obtain the recognition result to achieve intelligent detection of playing cards. However, this technical solution only detects the quality of playing cards and does not monitor the entire production status of the playing card production line, resulting in poor monitoring effects.

[0005] In the technical solution with the application number CN 201880068678.X, it prints the suit and rank on one side of the paper stock for playing cards and the back pattern on the other side, prints different paper page IDs for each one or more paper stocks for playing cards onto the paper stock for playing cards, forms one or more decks of playing cards from the individual playing cards cut from the paper stock for playing cards using a cutting machine, shuffles one or more decks of playing cards using a shuffling machine and forms a group of shuffled playing cards, assigns a different shuffled playing card ID for each group as an ID code to the shuffled playing cards, and establishes an association between the paper page ID constituting the shuffled playing cards and the shuffled playing card ID in the database. Therefore, this technical solution only realizes the manufacturing of playing cards and does not realize the monitoring during the production process of playing cards. Summary of the Invention

[0006] The present invention provides a monitoring method and system for a playing card production line to improve the monitoring efficiency, timeliness, and accuracy of the playing card production line.

[0007] To solve the above problems, the present invention adopts the following technical solutions:

[0008] The present invention provides a monitoring method for a playing card production line, including:

[0009] Receive the original video of the poker production line collected by the monitoring device;

[0010] Segment the original video into multiple original video frames, and screen out at least two consecutive original video frames as target images based on an image feature matching algorithm, where the target images include at least one original video frame that matches the preset image features;

[0011] Adjust the size of each of the target images to be consistent with the size of the preset standard image to obtain multiple standard target images;

[0012] Stitch multiple standard target images in the order of the time nodes corresponding to each standard target image to generate a standard target image sequence;

[0013] Input the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line;

[0014] When it is detected that the production status of the poker production line is abnormal, output an alarm message.

[0015] Preferably, inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line includes:

[0016] Using a pre-trained YOLO model, evenly divide each standard target image in the standard target image sequence into multiple image blocks;

[0017] For each standard target image, respectively predict the bounding box and probability value of the corresponding each image block, where the bounding box contains the center point coordinates and size information of the image block, and the probability value is the probability that the target to be monitored falls into the bounding box, and the targets to be monitored include poker cards and production equipment on the poker production line;

[0018] After removing overlapping bounding boxes using the non-maximum suppression algorithm, remove the remaining bounding boxes with probability values lower than the preset probability value according to the threshold filtering method to obtain the target bounding box of each standard target image;

[0019] Analyze the targets to be monitored corresponding to the target bounding boxes of each standard target image respectively to obtain the production status of the poker production line.

[0020] Preferably, predicting the bounding box of the corresponding each image block respectively includes:

[0021] According to the edge detection algorithm, identify all edge segments of each image block corresponding to each standard target image;

[0022] Starting from a preset starting point, traverse each edge segment of each image block in a counterclockwise direction and assign a chain code value to each edge segment;

[0023] In a counterclockwise direction, calculate the difference between the chain code values of every two adjacent edge segments in sequence to obtain all the chain code differences corresponding to each image block;

[0024] Calculate the mean square error of all the chain code differences corresponding to each image block to obtain the edge smoothness of each image block, and the edge smoothness is used to characterize the smoothness of the edge formed by all the edge segments of the corresponding image block;

[0025] When there are abnormal image blocks with edge smoothness less than the preset edge smoothness, perform edge smoothing processing on the abnormal image blocks according to the edge smoothing algorithm until the edge smoothness of the abnormal image blocks is greater than or equal to the preset edge smoothness;

[0026] When the edge smoothness of each image block is greater than or equal to the preset edge smoothness, connect all the edge segments of each image block respectively and use them as the corresponding bounding boxes.

[0027] Preferably, when it is detected that there is an abnormality in the production status of the playing card production line, an alarm message is output, including:

[0028] When it is detected that there is an abnormality in the production status of the playing card production line, abnormal information and an alarm message are generated, and the abnormal information includes the position and type of the abnormal points on the playing card production line;

[0029] Generate an adjustment strategy for the playing card production line based on the abnormal information;

[0030] Compress and output the adjustment strategy and the alarm message.

[0031] Further, after inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, it further includes:

[0032] When it is detected that there is no abnormality in the production status of the playing card production line, receive the playing card simulation image obtained by the monitoring device taking pictures of the playing cards produced by the playing card production line;

[0033] Perform analog-to-digital conversion processing on the playing card simulation image to generate a playing card digital image;

[0034] Identify the playing card digital image and extract the image features of the playing card digital image;

[0035] Match the image features of the playing card digital image with the template image features pre-stored in the database to obtain the matching result of the playing card;

[0036] Classify and package the playing cards according to the matching result.

[0037] Further, after inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, it further includes:

[0038] When it is monitored that the production status of the playing card production line is normal, receive the playing card production line image captured by the monitoring device for the playing card production line;

[0039] Identify the pose features of the staff in the playing card production line image;

[0040] Compare the pose features of the staff with the abnormal pose features pre-stored in the database to obtain a comparison result;

[0041] Judge whether there is an abnormality in the pose of the staff according to the comparison result;

[0042] When it is determined that there is an abnormality in the pose of the staff, send a prompt message of abnormal pose to the staff.

[0043] Preferably, identifying the pose features of the staff in the playing card production line image includes:

[0044] Input the playing card production line image into a pre-constructed convolutional neural network model to obtain a target playing card production line image. The convolutional neural network model includes a plurality of residual blocks, and each residual block is used to mark at least one local feature of the playing card production line image. Between the residual blocks, the output result of the previous residual block is input to the input of the next adjacent residual block through a skip connection;

[0045] Extract target local features from all the marked local features of the target playing card production line image to obtain the pose features of the staff;

[0046] Comparing the pose features of the staff with the abnormal pose features pre-stored in the database to obtain a comparison result includes:

[0047] Align the format and dimension of the pose features of the staff with the abnormal pose features pre-stored in the database;

[0048] According to the Pr-VIPE algorithm, map the pose key points of the pose features and the pose key points of the abnormal pose features to the same embedding space with view invariance;

[0049] In the embedding space, calculate the Euclidean distance between the pose feature and the abnormal pose feature;

[0050] Compare the Euclidean distance with a preset threshold to generate a comparison result between the pose feature of the staff and the abnormal pose feature.

[0051] Further, when it is monitored that there is an abnormality in the production status of the poker production line, after outputting an alarm message, it further includes:

[0052] Receive the operation data of each production device in the poker production line collected by the acquisition device;

[0053] When it is determined that the target production device is abnormal according to the operation data, send a control instruction to the control device corresponding to the target production device, so that the control device corrects the operation parameters of the target production device according to the control instruction.

[0054] Further, before inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line, it further includes:

[0055] Obtain multiple video samples of the poker production line, where the video samples include normal video samples and abnormal video samples;

[0056] Segment each of the video samples respectively, and based on the image feature matching algorithm, screen out multiple training images from each of the video samples respectively;

[0057] For each of the video samples, adjust the size of each training image to be consistent with the size of a preset standard image to obtain multiple standard training images for each of the video samples;

[0058] For each of the video samples, splice the multiple standard training images in the order of the time nodes corresponding to each standard training image to generate a standard training image sequence for each of the video samples;

[0059] Use the YOLO algorithm to train the standard training image sequence of each video sample, and when the training result meets the requirements, obtain the YOLO model.

[0060] The present invention also provides a monitoring system for a poker production line, including:

[0061] A receiving module, configured to receive the original video of the poker production line collected by the monitoring device;

[0062] A segmentation module, configured to segment the original video into multiple original video frames, and screen out at least two consecutive original video frames as target images based on an image feature matching algorithm, where the target images include at least one original video frame that matches a preset image feature;

[0063] An adjustment module, configured to adjust the size of each of the target images to be consistent with the size of a preset standard image, to obtain multiple standard target images;

[0064] A splicing module, configured to splice the multiple standard target images in the order of the time nodes corresponding to each of the standard target images, to generate a standard target image sequence;

[0065] An analysis module, configured to input the standard target image sequence into a pre-trained YOLO model for analysis, to obtain the production status of the poker card production line;

[0066] An output module, configured to output an alarm message when it is detected that the production status of the poker card production line is abnormal.

[0067] Compared with the prior art, the technical solution of the present invention has at least the following advantages:

[0068] The monitoring method and system for a poker card production line provided by the present invention reduce the workload of manual monitoring through automated original video frame segmentation and target image screening, and also reduce the amount of image processing in subsequent computer analysis, improving the monitoring efficiency; by using an image feature matching algorithm, frames that match a preset image feature can be more accurately screened out from consecutive original video frames, improving the accuracy of target detection; by adjusting the size of the target images to be consistent with the preset standard image, it helps to unify the data format and facilitates the processing and analysis of the YOLO model; by splicing multiple standard target images into a standard target image sequence in the order of the time nodes and inputting them into a pre-trained YOLO model for analysis, the YOLO model can comprehensively consider the previous and subsequent standard target images during the image analysis process to improve the accuracy of production status analysis, and output an alarm message when an abnormality is detected, improving the real-time performance of monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flowchart of an embodiment of the monitoring method for a poker card production line of the present invention;

[0070] Figure 2 It is a flowchart of another embodiment of the monitoring method for a poker card production line of the present invention;

[0071] Figure 3 It is a flowchart of another embodiment of the monitoring method for a poker card production line of the present invention;

[0072] Figure 4 It is a flowchart of another embodiment of the monitoring method for the playing card production line of the present invention;

[0073] Figure 5 It is a structural block diagram of an embodiment of the monitoring system for the playing card production line of the present invention. Detailed implementation manners

[0074] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0075] In some processes described in the specification and claims of the present invention and the above-mentioned drawings, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as S11, S12, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0076] Those of ordinary skill in the art can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used here may include wireless connection or wireless coupling. The phrase "and / or" used here includes all or any unit and all combinations of one or more associated listed items.

[0077] Those of ordinary skill in the art can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used here have the same meaning as the general understanding of those of ordinary skill in the technical field to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.

[0078] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0079] Please refer to Figure 1 , the present invention provides a monitoring method for a poker production line, which may include the following steps:

[0080] S11. Receive the original video of the poker production line collected by the monitoring device;

[0081] S12. Divide the original video into multiple original video frames, and screen out at least two consecutive original video frames as target images based on an image feature matching algorithm, where the target images include at least one original video frame that matches the preset image features;

[0082] S13. Adjust the size of each of the target images to be consistent with the size of a preset standard image to obtain multiple standard target images;

[0083] S14. Stitch the multiple standard target images in the order of the time nodes corresponding to each of the standard target images to generate a standard target image sequence;

[0084] S15. Input the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line;

[0085] S16. When it is detected that the production status of the poker production line is abnormal, output an alarm message.

[0086] First, obtain video data from a monitoring camera installed on the poker production line to get the original video. This original video is used for subsequent analysis to monitor the operation status of the production line. A video usually consists of consecutive frames. In this embodiment, the original video can be decomposed into individual image frames to obtain multiple original video frames for individual analysis of each frame.

[0087] Then, according to the image feature matching algorithm in this embodiment, such as SIFT (Scale-Invariant Feature Transform), through feature matching, the original video frames containing the ones that match the preset features are screened out from the segmented multiple original video frames as the target images. The preset features can be specific production line signs, signs of abnormal events, etc.

[0088] Among them, SIFT is a feature extraction and description algorithm for detecting and describing local features in images. The features are invariant to image scaling, rotation, and brightness changes, and also have a certain degree of stability to perspective changes and noise.

[0089] Specifically, this embodiment can simulate the changes of the original video frames at different scales by constructing a Difference of Gaussian (DoG) pyramid, and then use the Hessian matrix to detect the extreme points, that is, the feature points of the original video frames at each scale. Then, a feature descriptor is generated for each feature point. The feature descriptor is a vector containing the gradient direction and magnitude information of the pixels in the neighborhood of the key point. By calculating the similarity between the feature descriptor of the original video frame and the corresponding vector of the preset image feature, the original video frames with a similarity greater than the preset similarity are used as the target images.

[0090] In addition, to ensure that the images input to the YOLO model have a consistent size, it is necessary to adjust the size of the screened target images, which helps the model to more accurately identify and analyze the image content. Then, the resized standard target images are stitched together in the order in which they appear in the original video to form a series of consecutive image sequences, which helps the YOLO model to capture the dynamic changes on the production line.

[0091] Finally, the pre-trained YOLO model can be used to analyze the stitched image sequences to identify and locate the objects and events in the images, so as to evaluate the operating conditions of the poker production line, such as each link of the production process, equipment status, product quality, production efficiency, etc. If the YOLO model identifies any abnormal situations, such as equipment jitter, operation errors, or other unexpected events, the system will automatically trigger an alarm to notify the relevant personnel to intervene, so that the production line manager can quickly respond to safety violation events and take measures to prevent potential safety accidents.

[0092] Among them, YOLO (You Only Look Once) is an object detection algorithm used to transform the object detection task into a regression problem and simultaneously perform object localization and classification through a single neural network to achieve real-time and efficient object detection, such as detecting the production status of the poker production line.

[0093] The monitoring method for the poker card production line provided by the present invention reduces the workload of manual monitoring through automated original video frame segmentation and target image screening. At the same time, it also reduces the amount of image processing during subsequent computer analysis, improving the monitoring efficiency. By using the image feature matching algorithm, frames that match the preset image features can be more accurately screened out from continuous original video frames, improving the accuracy of target detection. By adjusting the size of the target image to be consistent with the preset standard image, it helps to unify the data format and facilitates the processing and analysis of the YOLO model. By splicing multiple standard target images into a standard target image sequence in the order of time nodes and inputting it into a pre-trained YOLO model for analysis, the YOLO model can comprehensively consider the previous and subsequent standard target images during the image analysis process to improve the analysis accuracy of the production status and output an alarm message when an anomaly is detected, improving the real-time performance of the monitoring.

[0094] Preferably, please refer to Figure 2 , inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker card production line may specifically include:

[0095] S151. Using the pre-trained YOLO model, evenly divide each standard target image in the standard target image sequence into multiple image blocks;

[0096] S152. For each standard target image, respectively predict the bounding box and probability value of each corresponding image block. The bounding box includes the center point coordinates and size information of the image block, and the probability value is the probability that the target to be monitored falls into the bounding box. The target to be monitored includes poker cards and production equipment on the poker card production line;

[0097] S153. After removing overlapping bounding boxes using the non-maximum suppression algorithm, remove the remaining bounding boxes with probability values lower than the preset probability value according to the threshold filtering method to obtain the target bounding box of each standard target image;

[0098] S154. Analyze the target to be monitored corresponding to the target bounding box of each standard target image respectively to obtain the production status of the poker card production line.

[0099] In this embodiment, in order to improve the detection accuracy and efficiency, each standard target image in the standard target image sequence is divided into multiple small blocks to obtain multiple image blocks. The size and dimensions of all image blocks are the same, so that each image block can be individually detected for targets, which helps to identify targets in different regions of the image.

[0100] For each image patch, the YOLO model predicts a bounding box that defines the center point coordinates and size information of the image patch. At the same time, the YOLO model also assigns a probability value to each bounding box, indicating the probability that the target to be monitored exists within the bounding box.

[0101] This non-maximum suppression algorithm is used to remove overlapping bounding boxes and only retain the bounding box that is most likely to contain the target to be monitored. The threshold filtering method is to remove those bounding boxes with probability values lower than a preset probability value because the possibility of the bounding box containing the target to be monitored is relatively low, so as to ensure the accuracy of the final result.

[0102] Then, the filtered target bounding boxes are analyzed to determine the positions and states of the targets to be monitored (playing cards and production equipment) in each standard target image, so as to evaluate the operating conditions of the production line.

[0103] In this embodiment, by dividing and separately detecting image patches, the accuracy of target detection is improved; at the same time, by using the fast and accurate detection speed of the YOLO model, real-time monitoring of the playing card production line is realized; through the non-maximum suppression algorithm and the threshold filtering method, the possibility of false detection is reduced, and at the same time, the number of bounding boxes to be analyzed is reduced, further improving the processing efficiency of the YOLO model. In addition, this technical solution can also adapt to images of different sizes and shapes, improving the flexibility of the system.

[0104] For example, assume that on a playing card production line, it is necessary to monitor the quality of playing cards and the operating status of the machine. First, obtain the video stream from the monitoring camera, divide it into frames, and then adjust it to the standard size input by the YOLO model. Each frame image is divided into several small patches. For example, if the image resolution is high, it can be divided into 16 small patches. Then use the YOLO model to detect each small patch, predict the bounding boxes and probability values in each small patch, use the non-maximum suppression algorithm to remove overlapping bounding boxes, and then remove the remaining bounding boxes with probability values lower than the preset probability value according to the threshold filtering method, and analyze the remaining bounding boxes to determine the positions and states of the playing cards and the machine. For example, if an abnormality (such as abnormal jitter) is detected in a certain part of the machine, an alarm is triggered to immediately notify the operator to intervene, so as to effectively monitor the operating conditions of the playing card production line, timely discover and handle potential problems, and improve production efficiency and product quality.

[0105] In one embodiment, please refer to Figure 3 , predicting the bounding boxes of each corresponding image patch respectively may specifically include:

[0106] S1521. According to the edge detection algorithm, identify all edge segments of each image patch corresponding to each standard target image;

[0107] S1522. Starting from a preset starting point, traverse each edge segment of each image block in the counterclockwise direction, and assign a chain code value to each edge segment;

[0108] S1523. In the counterclockwise direction, calculate the difference between the chain code values of every two adjacent edge segments in sequence to obtain all the chain code differences corresponding to each image block;

[0109] S1524. Calculate the mean square error of all the chain code differences corresponding to each image block to obtain the edge smoothness of each image block, and the edge smoothness is used to characterize the smoothness of the edge formed by all the edge segments of the corresponding image block;

[0110] S1525. When there are abnormal image blocks with edge smoothness less than the preset edge smoothness, perform edge smoothing processing on the abnormal image blocks according to the edge smoothing algorithm until the edge smoothness of the abnormal image blocks is greater than or equal to the preset edge smoothness;

[0111] S1526. When the edge smoothness of each image block is greater than or equal to the preset edge smoothness, connect all the edge segments of each image block respectively as the corresponding bounding box.

[0112] In this embodiment, the system can use an edge detection algorithm (such as the Canny edge detection method) to process each image block corresponding to each standard target image and identify all the edge segments in the image block. This edge detection algorithm can identify the regions with drastic brightness changes in the image block, and this region usually corresponds to the contour or boundary of the object.

[0113] Then, determine a starting point (usually the first edge point identified by the edge detection algorithm), and then traverse each edge segment in each image block in the counterclockwise direction to ensure that the traversal order of the edges is consistent, which is convenient for subsequent processing. The chain code is a method of representing the edge direction with a digital sequence. For each edge segment, according to its direction relative to the previous edge segment, assign a chain code value. This process involves quantifying the edge direction into a limited number of discrete values, such as 0 - 7 or 0 - 15. In order to capture the local changes of the edge, calculate the chain code difference between adjacent edge segments in the counterclockwise direction. This chain code difference reflects the turning and curvature changes of the edge in the local area.

[0114] The mean square error is a statistic that measures the degree of dispersion of the chain code difference distribution. In this embodiment, the mean square error of all corresponding chain code differences in each image block can be calculated and used as an index to measure the edge smoothness. At the same time, a preset edge smoothness threshold is set. If the edge smoothness of a certain image block is lower than this threshold, it indicates that there may be noise or anomalies at the edge of the image block, and edge smoothing processing is required. At this time, edge smoothing algorithms (such as Gaussian filtering, median filtering, etc.) can be used to process the image blocks with edge smoothness lower than the threshold until their edge smoothness reaches or exceeds the preset threshold.

[0115] After the edge smoothness of all image blocks reaches the preset threshold, all edge segments of each image block are connected respectively to form the bounding box of the image block. This bounding box can be used for subsequent object detection and analysis.

[0116] In this embodiment, through the calculation of chain code differences and edge smoothness, the edges in the image can be identified and processed more accurately. At the same time, edge smoothing processing helps to reduce the influence of noise on the edge detection results and improve the robustness of object detection. In addition, accurate edge detection and smoothing processing can also improve the accuracy of subsequent object detection, especially in a complex background such as a poker production line.

[0117] In one embodiment, please refer to Figure 4 , when it is monitored that there is an abnormality in the production status of the poker production line, an alarm message is output, which may specifically include:

[0118] S161. When it is monitored that there is an abnormality in the production status of the poker production line, generate abnormal information and an alarm message, where the abnormal information includes the position and type of the abnormal point on the poker production line;

[0119] S162. Generate an adjustment strategy for the poker production line based on the abnormal information;

[0120] S163. Compress and output the adjustment strategy and the alarm message.

[0121] This embodiment can use a pre-trained YOLO model to analyze a standard target image sequence to identify abnormal situations on the poker production line, such as equipment failures, operation errors, or other unexpected events. Once an abnormality is detected, the system will record the position (such as a specific device or a specific area on the production line) and type (such as excessive jitter amplitude, abnormal speed, etc.) of the abnormal point.

[0122] This exception information is a detailed description of the abnormal situation on the production line, including the specific location and type of the abnormal point. This exception information is crucial for subsequent fault diagnosis and repair. Based on the exception information, the system will generate corresponding adjustment strategies. For example, if it is detected that the vibration amplitude of a certain machine is too large, the adjustment strategy may include reducing the machine running speed.

[0123] In addition, in order to reduce the data transmission volume and improve the response speed, the adjustment strategies and alarm information are compressed and then output to the control center or directly sent to the devices of relevant personnel.

[0124] The system of this embodiment can quickly identify the abnormal point and generate alarm information, which helps to respond to problems on the production line in a timely manner. At the same time, the exception information provides the specific location and type of the abnormal point, which helps to quickly locate the source of the problem. In addition, the adjustment strategies automatically generated based on the exception information can reduce human intervention and improve the automation level of the production line. By compressing the adjustment strategies and alarm information, the burden of data transmission is also reduced, and the response speed and efficiency of the system are improved.

[0125] For example, assume that in a playing card production line, an important detection step is to ensure the printing quality of each playing card. At this time, the YOLO model monitors the production line and identifies an abnormal situation where the printing of a playing card is blurred. The system records the abnormal location (such as a specific area of the playing card) and type (printing blur). According to the exception information, the system determines that it may be due to improper ink volume setting or insufficient printing pressure of the printing machine, and then generates adjustment strategies, such as adjusting the ink volume or increasing the printing pressure. The system compresses the adjustment strategy (adjusting the ink volume) and the alarm information (printing blur alarm), and then sends them to the production line control center. After receiving the compressed information, the control center decompresses and implements the adjustment strategy to adjust the settings of the printing machine. After the adjustment, the system continues to monitor the production line to ensure that the printing quality returns to normal, and records the adjustment effect to effectively monitor and manage the operation status of the playing card production line, promptly discover and handle abnormal situations, reduce production interruptions and product quality problems, and improve production efficiency and product quality.

[0126] In another embodiment, after inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, it may further include:

[0127] When it is monitored that there is no abnormality in the production status of the playing card production line, receive the playing card simulation image obtained by the monitoring device taking pictures of the playing cards produced by the playing card production line;

[0128] Perform analog-to-digital conversion processing on the playing card simulation image to generate a playing card digital image;

[0129] Identify the playing card digital image and extract the image features of the playing card digital image;

[0130] Match the image features of the playing card digital image with the template image features pre-stored in the database to obtain the matching result of the playing card;

[0131] Classify and package the playing cards according to the matching result.

[0132] In this embodiment, when the production line is running normally and no abnormalities are detected, the monitoring device (such as a high-definition camera) will take pictures of the playing cards on the production line to obtain the images of the playing cards. The image is in the form of an analog signal and needs further processing.

[0133] Analog-to-digital conversion (ADC) is the process of converting an analog image into a digital image. Through this process, the analog image is converted into a digital signal that can be processed and analyzed by a computer system. Image processing and computer vision techniques (such as edge detection, color recognition, texture analysis, etc.) can be used to extract image features from the digital image. The image features may include the suit, number, pattern, etc. of the playing card.

[0134] Then, use the template matching algorithm or the feature point matching algorithm to compare the extracted image features with the standard template image features pre-stored in the database. According to the matching result, the system can determine the exact identity of each playing card (such as Ace of Hearts, King of Spades, etc.) and guide the automated packaging line to sort the playing cards into the correct packaging boxes.

[0135] This embodiment can reduce the need for manual operations through automated image recognition and classification packaging, improve production efficiency, reduce human errors, and improve the accuracy of packaging. Fast image processing and matching algorithms can process a large number of playing cards in real time, improving the throughput of the production line. In addition, through precise image recognition, it can be ensured that each playing card is correctly classified, improving product quality.

[0136] For example, assume that during normal operation of a playing card production line, it is necessary to automatically classify and package the produced playing cards. First, a camera on the production line takes pictures of the playing cards being packaged to obtain images of the playing cards, and converts the analog images taken by the camera into digital images through analog-to-digital conversion for computer processing. Then, image processing software is used to identify the digital images of the playing cards, extract key features such as suits and numbers, and match the extracted features with the template features in the database to determine the specific type of each playing card. According to the matching results, an automated robotic arm sorts the playing cards into corresponding packaging boxes. For example, the Ace of Hearts is placed in the packaging box for the Ace of Hearts. Throughout the process, the system also monitors the quality of the playing cards, such as printing quality and whether the cutting is neat, to ensure that the final products meet the quality standards, thereby achieving automated quality control and classification packaging, improving production efficiency and product quality, and reducing labor costs and errors.

[0137] In one embodiment, after inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, it may further include:

[0138] When it is monitored that the production status of the playing card production line is normal, receive the playing card production line image captured by the monitoring device for the playing card production line;

[0139] Identify the posture features of the staff in the playing card production line image;

[0140] Compare the posture features of the staff with the abnormal posture features pre-stored in the database to obtain a comparison result;

[0141] Judge whether the posture of the staff is abnormal according to the comparison result;

[0142] When it is determined that the posture of the staff is abnormal, send a prompt message of abnormal posture to the staff.

[0143] When the production line is operating normally, a monitoring device (such as a camera) continuously captures real-time images of the production line to obtain playing card production line images. These images may include various aspects such as staff, machine operations, and product flow.

[0144] Then, computer vision technology, especially pose estimation algorithms (such as OpenPose, AlphaPose, etc.), is used to extract the human key points of the staff from the images of the poker production line, including the position information of parts such as the head, hands, and feet, so as to identify the pose characteristics of the staff. At the same time, the extracted pose characteristics are matched and compared with the predefined abnormal pose characteristics in the database to obtain the comparison result. The abnormal pose characteristics can be obtained based on safety specifications, operation standards, or historical data analysis. The comparison result includes the matching rate between the pose characteristics and each abnormal pose characteristic.

[0145] Based on the comparison result, the system can determine whether the pose of the staff meets the preset safety and operation standards. If the pose of the staff matches the abnormal pose characteristics, the system will determine it as abnormal. Once the system determines that there is an abnormality in the pose of the staff, various methods (such as display prompts, sound alarms, vibration devices, etc.) can be used to send a prompt message of abnormal pose to the staff, prompting them to immediately correct the improper pose to avoid possible safety accidents or operation errors.

[0146] This embodiment can monitor and analyze the poses of the staff in real time, timely discover and correct unsafe behaviors, reduce safety accidents in the workplace, ensure that the staff comply with the operating procedures, improve production efficiency and product quality. At the same time, through real-time feedback, it enhances the awareness of employees about safety and operation specifications, reduces incorrect operations caused by improper poses, and reduces the defective product rate.

[0147] For example, assume that on a poker production line, the staff needs to operate the machine in a specific posture to ensure safety and efficiency. First, the camera on the production line captures the operation images of the staff in real time. The pose estimation algorithm is used to extract the key point information of the staff from the images, identify their pose characteristics, and compare the extracted pose characteristics with the normal and abnormal pose characteristics pre-stored in the database. The system analyzes the comparison result and finds that the staff is bending over too much, which matches the abnormal pose characteristics in the database. The system immediately sends a prompt message of abnormal pose to the staff through the alarm system in the workshop. After receiving the prompt, the staff immediately adjusts the pose, so as to monitor the operation pose of the staff in real time, timely discover and correct the abnormal pose, and thus improve the safety and efficiency of the production line.

[0148] Preferably, identifying the pose characteristics of the staff in the images of the poker production line may specifically include:

[0149] Input the playing card production line image into a pre - constructed convolutional neural network model to obtain a target playing card production line image. The convolutional neural network model contains multiple residual blocks, and each residual block is used to mark at least one local feature of the playing card production line image. Between the residual blocks, the output result of the previous residual block is input to the input of the next adjacent residual block through a skip connection;

[0150] Extract the target local feature from all the marked local features of the target playing card production line image to obtain the posture feature of the staff.

[0151] In this embodiment, a pre - trained convolutional neural network (CNN) model can be used to process the playing card production line image obtained from the monitoring device. The CNN model is particularly suitable for image recognition tasks because it can automatically learn the features of the image.

[0152] A residual block is a structure in a CNN that solves the vanishing gradient problem in deep networks by introducing a skip connection. Each residual block contains multiple convolutional layers for extracting local features of the image.

[0153] In a CNN, each residual block is responsible for capturing specific local features in the playing card production line image, such as edges, textures, or shapes, etc. These local features are crucial for understanding the image content.

[0154] Among them, the skip connection allows the network to directly pass the output of one residual block to the next adjacent residual block to retain more information and helps the propagation of gradients in the network, making the training of deep networks more effective.

[0155] After all the local features of the target playing card production line image are marked, extract the specific local features related to the task from all the marked local features to obtain the target local feature. In this embodiment, the target local feature is the posture feature of the staff, and this feature is crucial for monitoring the behavior of personnel on the production line.

[0156] Finally, the system will identify and extract the posture feature of the staff, and this posture feature can be used to analyze whether the behavior of the staff conforms to safety specifications or operation standards.

[0157] This embodiment can enable the deep network to learn complex features more effectively through the combination of residual blocks and skip connections. At the same time, skip connections help solve the problem of gradient vanishing, making the training of the deep network easier. In addition, by learning local features, the model can better understand and generalize new scenarios. The quickly extracted pose features also enable the system to monitor the behavior of the staff in real time and provide instant feedback when anomalies are detected subsequently.

[0158] Preferably, comparing the pose features of the staff with the abnormal pose features pre-stored in the database to obtain a comparison result may specifically include:

[0159] Align the format and dimension of the pose features of the staff with the abnormal pose features pre-stored in the database;

[0160] According to the Pr-VIPE algorithm, map the pose key points of the pose features and the pose key points of the abnormal pose features into the same embedding space with view invariance;

[0161] In the embedding space, calculate the Euclidean distance between the pose features and the abnormal pose features;

[0162] Compare the Euclidean distance with a preset threshold to generate the comparison result between the pose features of the staff and the abnormal pose features.

[0163] In this embodiment, since the pose features extracted from the monitoring device and the abnormal pose features stored in the database may have different data formats and dimensions, they need to be converted to the same format and dimension for comparison. For example, convert them to the same format and dimension by means of operations such as normalization, scaling, or adjusting the data structure.

[0164] The Pr-VIPE (Probabilistic Viewpoint Invariant Human Pose Embedding) algorithm is a method for mapping the pose key points of pose features to a low-dimensional space with view invariance, which means that no matter from which view, the representations of the same human pose in the embedding space are similar.

[0165] In the embedding space, the system can measure the similarity between two points by calculating the Euclidean distance between them. The Euclidean distance is the actual distance between two points in a multi-dimensional space, and the calculation formula is the square root of the sum of the squares of the coordinate differences of the two points.

[0166] In addition, by comparing the calculated Euclidean distance with a preset threshold, it is possible to determine whether the posture of the staff member is similar to the abnormal posture characteristics. If the Euclidean distance is less than the preset threshold, it is considered that the posture characteristics match the abnormal posture characteristics; if the Euclidean distance is greater than the preset threshold, it is considered that the comparison result is a mismatch.

[0167] In this embodiment, by using the Pr-VIPE algorithm, the accuracy of posture recognition can be improved. Especially under different perspectives, the view invariance enables the system to stably recognize abnormal postures under different observation conditions. At the same time, the fast posture comparison can provide real-time feedback to the staff and correct unsafe postures in a timely manner. In addition, by timely detecting and correcting abnormal postures, workplace safety accidents can be reduced.

[0168] In one embodiment, when it is detected that there is an abnormality in the production status of the poker card production line, after outputting an alarm message, it may further include:

[0169] Receiving the operation data of each production device in the poker card production line collected by the acquisition device;

[0170] When it is determined according to the operation data that there is an abnormality in the target production device, a control instruction is sent to the control device corresponding to the target production device, so that the control device corrects the operation parameters of the target production device according to the control instruction.

[0171] In this embodiment, sensors and a data acquisition system (such as PLC, SCADA system, etc.) can be used to collect the real-time operation data of each production device on the poker card production line. The production devices may include printing presses, die-cutting machines, slitting and slicing machines, packaging lines, embossing machines, etc. The operation data may include key parameters such as temperature, speed, pressure, current, and voltage.

[0172] Among them, the printing press includes a four-color printing press for printing poker card patterns and texts; a two-color offset printing press for double-sided printing of poker cards. The die-cutting machine is used to die-cut the printed large sheets into individual poker cards. The slitting and slicing machine is used to slit and slice the die-cut poker cards. The packaging line is used for automatic packaging of poker cards. The embossing machine may include a full-automatic embossing machine for embossing processing of poker cards, realizing synchronous automatic detection of product quality and embossing.

[0173] Then, by performing real-time analysis on the collected operation data, it can be monitored whether the production device is operating as expected. If a certain parameter exceeds the preset normal range, or there is an abnormal correlation between multiple parameters, the system can determine that there is an abnormality in the target production device, and the target production device is the production device with abnormal operation data.

[0174] Once an anomaly is detected, the system automatically sends control instructions to the control system of the target production equipment (such as PLC or industrial PC). The control instructions are designed to adjust the operating parameters of the target production equipment to restore the normal operating state of the equipment.

[0175] After receiving the control instructions, the control equipment adjusts the operating parameters of the target production equipment, such as reducing the temperature, adjusting the speed, reallocating the load, etc., which helps the production equipment to resume normal operation and avoid production interruptions or product quality problems.

[0176] This embodiment can reduce the need for manual intervention through automated anomaly detection and response, improving the operating efficiency of the production line; at the same time, quickly identifying and responding to anomalies helps to reduce the downtime of production equipment and maintain production continuity. In addition, by adjusting the operating parameters in a timely manner, product quality problems caused by abnormal production equipment can be avoided, and preventive maintenance and timely troubleshooting also help to reduce long-term maintenance costs and prevent potential safety accidents.

[0177] For example, assume that on a playing card production line, there is a device responsible for printing, and its temperature and speed are key operating parameters. Sensors continuously monitor the temperature and speed of the printing device and send the data to the central control system. The central control system analyzes the data, finds that the temperature is 10 degrees higher than the normal value, determines that the equipment is abnormal, and the system automatically sends a control instruction to the PLC of the printing device, requesting it to reduce the temperature. After receiving the instruction, the PLC adjusts the operation of the cooling system to lower the temperature to the normal range. The central control system continues to monitor the operating parameters until the temperature stabilizes within the normal range, ensuring that the equipment resumes normal operation, thereby enabling real-time monitoring of the equipment status, timely discovery and handling of abnormal situations, and improving production efficiency and product quality, while reducing maintenance costs and safety risks.

[0178] In another embodiment, before inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the playing card production line, it may further include:

[0179] Obtain multiple video samples of the playing card production line, where the video samples include normal video samples and abnormal video samples;

[0180] Segment each of the video samples respectively, and based on the image feature matching algorithm, screen out multiple training images from each of the video samples respectively;

[0181] For each of the video samples, adjust the size of each of the training images to be consistent with the size of a preset standard image, obtaining multiple standard training images for each of the video samples;

[0182] For each of the video samples, a sequence of standard training images for each video sample is generated by splicing multiple standard training images in the order of the time nodes corresponding to each standard training image.

[0183] The YOLO algorithm is used to train the sequence of standard training images for each video sample, and when the training result meets the requirements, a YOLO model is obtained.

[0184] This embodiment can collect videos of a poker production line that include normal operations and abnormal situations. This video will be used to train a YOLO model capable of identifying abnormalities on the production line. Each video sample is split into individual frames as potential training images. Key frames containing feature information for training the model are extracted from each video sample using an image feature matching algorithm (such as SIFT, SURF, or ORB) as training images.

[0185] To ensure the consistency of the training data, the selected training images are adjusted to the same size as the preset standard images, which helps the YOLO model to learn and identify better. The standard training images after size adjustment are spliced into a sequence in chronological order to obtain a sequence of standard training images for each video sample. This sequence of standard training images will be used to simulate a video stream so that the YOLO model can learn the features of the video changing over time.

[0186] Then, the YOLO (You Only Look Once) algorithm is used to train the spliced sequence of standard training images. YOLO is a real-time object detection method that can identify objects and abnormalities in images. Continuous training is performed until the model reaches a satisfactory accuracy and recall rate. After meeting this requirement, the model can be used for real-time monitoring of the poker production line to detect abnormal situations.

[0187] This embodiment can train using normal and abnormal samples, enabling the model to more accurately identify abnormalities. At the same time, the trained YOLO model can analyze the image sequence in real time, quickly respond to abnormalities on the production line, which also helps to prevent accidents and improve product quality. In addition, automated anomaly detection reduces the need for manual monitoring, lowering costs and increasing efficiency.

[0188] Please refer to Figure 5 , an embodiment of the present invention also provides a monitoring system for a poker production line, which may include:

[0189] A receiving module 21 for receiving the original video of the poker production line collected by the monitoring device;

[0190] A splitting module 22, configured to split the original video into multiple original video frames, and screen out at least two consecutive original video frames as target images based on an image feature matching algorithm, where the target images include at least one original video frame that matches a preset image feature;

[0191] An adjustment module 23, configured to adjust the size of each of the target images to be consistent with the size of a preset standard image, to obtain multiple standard target images;

[0192] A splicing module 24, configured to splice the multiple standard target images in the order of the time nodes corresponding to each of the standard target images, to generate a standard target image sequence;

[0193] An analysis module 25, configured to input the standard target image sequence into a pre-trained YOLO model for analysis, to obtain the production status of the poker card production line;

[0194] An output module 26, configured to output an alarm message when it is detected that the production status of the poker card production line is abnormal.

[0195] The monitoring system of the poker card production line provided by the present invention reduces the workload of manual monitoring through automated original video frame splitting and target image screening, and also reduces the amount of image processing during subsequent computer analysis, improving the monitoring efficiency; by using an image feature matching algorithm, frames that match the preset image feature can be more accurately screened out from consecutive original video frames, improving the accuracy of target detection; by adjusting the size of the target images to be consistent with the preset standard image, it helps to unify the data format and facilitates the processing and analysis of the YOLO model; by splicing multiple standard target images into a standard target image sequence in the order of the time nodes and inputting them into a pre-trained YOLO model for analysis, the YOLO model can comprehensively consider the previous and subsequent standard target images during the image analysis process to improve the accuracy of production status analysis, and output an alarm message when an abnormality is detected, improving the real-time performance of monitoring.

[0196] Regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0197] In one embodiment, the present invention also proposes a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to execute the above-mentioned monitoring method for the poker card production line. Wherein, the storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0198] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. This computer program can be stored in a storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0199] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0200] The above embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent of the present invention should be subject to the appended claims.

Claims

1. A monitoring method for a poker card production line, characterized in that, Including: Receiving the original video of the poker card production line collected by the monitoring device; Segmenting the original video into multiple original video frames, and screening out at least two consecutive original video frames as target images based on an image feature matching algorithm, where the target images include at least one original video frame that matches the preset image features; Resizing the size of each of the target images to be consistent with the size of the preset standard image to obtain multiple standard target images; Stitching the multiple standard target images in the order of the time nodes corresponding to each of the standard target images to generate a standard target image sequence; Inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker card production line; When it is detected that the production status of the poker card production line is abnormal, outputting an alarm message; Among them, the inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker card production line includes: Using the pre-trained YOLO model to evenly divide each of the standard target images in the standard target image sequence into multiple image blocks; For each of the standard target images, respectively predicting the bounding box and probability value of each corresponding image block, where the bounding box contains the center point coordinates and size information of the image block, and the probability value is the probability that the target to be monitored falls into the bounding box, and the targets to be monitored include poker cards and production equipment on the poker card production line; After removing the overlapping bounding boxes using the non-maximum suppression algorithm, removing the remaining bounding boxes with probability values lower than the preset probability value according to the threshold filtering method to obtain the target bounding box of each of the standard target images; Analyzing the targets to be monitored corresponding to the target bounding boxes of each of the standard target images respectively to obtain the production status of the poker card production line.

2. The monitoring method of the poker card production line according to claim 1, characterized in that, Respectively predicting the bounding box of each corresponding image block includes: Identifying all edge segments of each image block corresponding to each of the standard target images according to the edge detection algorithm; Starting from a preset starting point, traversing each edge segment of each image block in the counterclockwise direction and assigning a chain code value to each edge segment; Calculating the difference between the chain code values of every two adjacent edge segments in the counterclockwise direction to obtain all the chain code differences corresponding to each image block; Calculating the mean square error of all the chain code differences corresponding to each image block to obtain the edge smoothness of each image block, and the edge smoothness is used to characterize the smoothness of the edge formed by all the edge segments of the corresponding image block; When there are abnormal image blocks with edge smoothness less than the preset edge smoothness, performing edge smoothing processing on the abnormal image blocks according to the edge smoothing algorithm until the edge smoothness of the abnormal image blocks is greater than or equal to the preset edge smoothness; When the edge smoothness of each image block is greater than or equal to the preset edge smoothness, connecting all the edge segments of each image block respectively as the corresponding bounding box.

3. The monitoring method of the poker card production line according to claim 1, characterized in that, When it is detected that the production status of the poker card production line is abnormal, outputting an alarm message, including: When it is detected that there is an abnormality in the production status of the playing card production line, abnormal information and warning information are generated, and the abnormal information includes the position and type of the abnormal point on the playing card production line; Generate an adjustment strategy for the playing card production line based on the abnormal information; Compress and output the adjustment strategy and the warning information.

4. The monitoring method of the poker card production line according to claim 1, wherein After analyzing the standard target image sequence by inputting it into a pre-trained YOLO model to obtain the production status of the playing card production line, it further includes: When it is detected that there is no abnormality in the production status of the playing card production line, receive a playing card simulation image obtained by a monitoring device taking a picture of the playing cards produced by the playing card production line; Perform analog-to-digital conversion processing on the playing card simulation image to generate a playing card digital image; Identify the playing card digital image, and extract the image features of the playing card digital image; Match the image features of the playing card digital image with the template image features pre-stored in the database to obtain the matching result of the playing cards; Classify and package the playing cards according to the matching result.

5. The monitoring method of the poker card production line according to claim 1, characterized in that, After analyzing the standard target image sequence by inputting it into a pre-trained YOLO model to obtain the production status of the playing card production line, it further includes: When it is detected that there is no abnormality in the production status of the playing card production line, receive a playing card production line image obtained by a monitoring device taking a picture of the playing card production line; Identify the posture features of the staff in the playing card production line image; Compare the posture features of the staff with the abnormal posture features pre-stored in the database to obtain a comparison result; Judge whether there is an abnormality in the posture of the staff according to the comparison result; When it is determined that there is an abnormality in the posture of the staff, send a prompt message of abnormal posture to the staff.

6. The monitoring method of the poker card production line according to claim 5, characterized in that, Identifying the posture features of the staff in the playing card production line image includes: Input the playing card production line image into a pre-constructed convolutional neural network model to obtain a target playing card production line image. The convolutional neural network model includes a plurality of residual blocks, and each residual block is used to mark at least one local feature of the playing card production line image. Between the residual blocks, the output result of the previous residual block is input to the input of the next adjacent residual block through a skip connection; Extract target local features from all the marked local features of the target playing card production line image to obtain the posture features of the staff; Comparing the posture features of the staff with the abnormal posture features pre-stored in the database to obtain a comparison result, including: Align the format and dimension of the posture features of the staff with the abnormal posture features pre-stored in the database; According to the Pr-VIPE algorithm, map the posture key points of the posture features and the posture key points of the abnormal posture features to the same embedding space with view invariance; In the embedding space, calculate the Euclidean distance between the posture features and the abnormal posture features; Compare the Euclidean distance with a preset threshold to generate a comparison result between the posture feature of the staff member and the abnormal posture feature.

7. The monitoring method of the poker card production line according to claim 1, wherein When it is detected that there is an abnormality in the production status of the poker production line, after outputting an alarm message, it further includes: Receive the operation data of each production device in the poker production line collected by the acquisition device; When it is determined that the target production device is abnormal according to the operation data, send a control instruction to the control device corresponding to the target production device, so that the control device corrects the operation parameters of the target production device according to the control instruction.

8. The monitoring method of the poker card production line according to claim 1, wherein Before inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line, it further includes: Obtain multiple video samples of the poker production line, where the video samples include normal video samples and abnormal video samples; Segment each of the video samples, and based on the image feature matching algorithm, select multiple training images from each of the video samples respectively; For each of the video samples, adjust the size of each of the training images to be consistent with the size of the preset standard image respectively, to obtain multiple standard training images for each of the video samples; For each of the video samples, splice the multiple standard training images in the order of the time nodes corresponding to each of the standard training images respectively, to generate a standard training image sequence for each of the video samples; Use the YOLO algorithm to train the standard training image sequence of each of the video samples, and when the training result meets the requirements, obtain the YOLO model.

9. A monitoring system for a poker card production line, characterized in that, It includes: A receiving module, configured to receive the original video of the poker production line collected by the monitoring device; A segmentation module, configured to segment the original video into multiple original video frames, and based on the image feature matching algorithm, select at least two consecutive original video frames as target images, where the target images include at least one original video frame that matches the preset image feature; An adjustment module, configured to adjust the size of each of the target images to be consistent with the size of the preset standard image respectively, to obtain multiple standard target images; A splicing module, configured to splice the multiple standard target images in the order of the time nodes corresponding to each of the standard target images respectively, to generate a standard target image sequence; An analysis module, configured to input the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line; An output module, configured to output an alarm message when it is detected that there is an abnormality in the production status of the poker production line; Wherein, the inputting the standard target image sequence into a pre-trained YOLO model for analysis to obtain the production status of the poker production line includes: Using the pre-trained YOLO model, evenly divide each of the standard target images in the standard target image sequence into multiple image blocks; For each of the standard target images, the bounding boxes and probability values of the corresponding image patches are predicted respectively. The bounding boxes contain the central point coordinates and size information of the image patches, and the probability values are the probabilities that the targets to be monitored fall into the bounding boxes. The targets to be monitored include playing cards and the production equipment on the playing card production line; After removing the overlapping bounding boxes using the non-maximum suppression algorithm, the remaining bounding boxes with probability values lower than the preset probability value are removed according to the threshold filtering method to obtain the target bounding boxes of each standard target image; The production status of the playing card production line is obtained by analyzing the targets to be monitored corresponding to the target bounding boxes of each standard target image respectively.

Images