A method and system for detecting production behavior

Abstract

This invention relates to a production behavior detection method and system, relating to the field of production detection. The method includes: acquiring workstation image information; analyzing and determining production action frames based on the workstation image information; annotating and extracting action features from the production action frames; determining whether the action features are consistent with qualified features in a preset historical qualified action database; if consistent, marking the product as a processed product; if inconsistent, acquiring the corresponding product image and issuing a prompt; determining product quality information based on the product image, and determining whether it is qualified based on the product quality information and preset quality parameters; if the product quality information is qualified, marking the product as a processed product; if the product quality information is unqualified, marking the product as a product awaiting processing; and re-inspecting the processed product using magnetic field detection to complete the action detection. This application has the effect of ensuring production quality and efficiency.

Description

Technical Field

[0001] This invention relates to the field of production detection, and in particular to a method and system for detecting production behavior. Background Technology

[0002] Production behavior detection is a technical means of determining whether products are qualified and whether behaviors are compliant by analyzing the operational behavior and work processes of employees.

[0003] When workers in factories are assembling or installing semi-finished products from the production line, production behavior detection technology is often used to ensure the efficiency and quality of production activities. In existing technologies, managers are generally on-site to supervise the workers' work and judge whether the production behavior is qualified, thereby ensuring product quality.

[0004] Because on-site inspections rely on human judgment, managers need to supervise a large number of workers, making supervision difficult. When the actual product is highly complex or the errors caused by human processing are subtle, it is difficult to detect the problems. Misconnection and misinsertion of parts occur frequently, leading to a decrease in yield and subsequent product rework, which seriously affects quality and efficiency. Summary of the Invention

[0005] To ensure production quality and efficiency, this invention provides a production behavior detection method and system.

[0006] In a first aspect, the present invention provides a method for detecting production behavior, which adopts the following technical solution:

[0007] A method for detecting production behavior, comprising:

[0008] Collect workstation image information;

[0009] The production action frame is determined based on the analysis of the workstation image information;

[0010] The action features are annotated and extracted from the production action frames;

[0011] Determine whether the action feature is consistent with the qualified features in a preset historical qualified action database;

[0012] If they match, the product will be marked as a processed product;

[0013] If there is a discrepancy, the corresponding product image will be captured and a prompt will be issued;

[0014] Based on the product image, determine the product quality information, and based on the product quality information and preset quality parameters, determine whether it is qualified;

[0015] If the product quality information is qualified, the product will be marked as a processed product;

[0016] If the product quality information is unqualified, the product will be marked as a product awaiting processing.

[0017] The processed product is re-inspected by magnetic field detection to complete motion detection.

[0018] By adopting the above technical solution, the production action frames are determined by collecting workstation image information, and then the fixed-point action features are extracted. The action is used to determine whether the processed product does not need to be reworked. Finally, the processed product is re-inspected by magnetic field detection, thus completing the action detection process. This enables real-time monitoring of production behavior and accurate judgment of product quality, improves the controllability of the production process and the stability of product quality, thereby ensuring the quality and efficiency of production.

[0019] Optionally, the annotation and extraction of motion features from the production action frames include:

[0020] In the production action frame, mark and determine the action marker points;

[0021] The limb shape is determined based on the production action frame and the preset clothing image;

[0022] Based on the motion markers and the limb shape, key motion points are marked and determined in the production motion frame;

[0023] Based on the marked key action points, a motion trajectory is formed by connecting them;

[0024] The movement characteristics of limb swinging are determined based on the movement trajectory and the limb shape.

[0025] By adopting the above technical solution, the motion trajectory can be obtained through the linkage of point and shape markings, which can more accurately capture and analyze the motion characteristics in the production process, providing a precise basis for subsequent motion comparison and quality judgment.

[0026] Optionally, marking and determining action marker points in the production action frame includes:

[0027] Based on the production action frame, extract the contour features of all targets within a single frame;

[0028] Based on the contour features, human candidate regions and non-human regions are distinguished;

[0029] The shape integrity rate and limb movement trend are determined based on the candidate human body regions and the preset human body shape.

[0030] The human candidate regions whose shape integrity rate is less than the preset shape judgment rate are classified as non-human regions;

[0031] Based on the contour features and limb movement trends of the redefined human candidate regions, the worker's limb features are determined;

[0032] Action markers are determined based on the worker's limb characteristics.

[0033] By adopting the above technical solution, the contour features of all targets in a single frame are extracted based on the production action frame, and the human candidate area and non-human area are distinguished. Then, the shape integrity is used for screening, and the worker's limb features are determined in the human candidate area to obtain action marker points, thereby improving the accuracy of action marker points and laying the foundation for subsequent action feature extraction.

[0034] Optionally, determining motion markers based on the worker's limb characteristics includes:

[0035] The static and dynamic parts of the limbs are determined based on the worker's limb characteristics;

[0036] Extract the dynamic shape based on the dynamic portion;

[0037] Joint positioning is determined based on the static and dynamic components;

[0038] The joint trajectory is determined based on the dynamic shape and the joint positioning.

[0039] The joint movement point is determined based on the joint trajectory;

[0040] Static points are determined based on the static components and the preset human body shape;

[0041] The static fixed point, the joint fixed point, and the joint moving point are used as motion marker points.

[0042] By employing the above technical solution, the static and dynamic parts of a limb are determined based on the worker's limb characteristics. The dynamic shape is extracted, joint anchor points and trajectories are determined, and then the moving and static anchor points of the joints are identified. These points are then used as motion markers. This approach allows for a more comprehensive capture of the limb's motion characteristics, improving the representativeness of the motion markers and the accuracy of the motion features.

[0043] Optionally, determining whether the action feature is consistent with the qualified features in a preset historical qualified action database includes:

[0044] Determine the trajectory shape based on the described action trajectory;

[0045] Determine the motion angle based on the trajectory shape;

[0046] The motion displacement is determined based on the motion characteristics and the preset human body shape;

[0047] The amplitude of the movement is determined based on the movement angle and the movement displacement;

[0048] The degree of overlap of actions is determined based on the passability range of the passability features in the preset historical passability database and the action range.

[0049] When the overlap is greater than the preset judgment overlap, the action feature is consistent with the qualified features in the preset historical qualified action database.

[0050] When the overlap is not greater than the preset judgment overlap, the action feature is inconsistent with the qualified features in the preset historical qualified action database.

[0051] By adopting the above technical solution, the motion angle and motion displacement are determined based on the motion trajectory, and the motion amplitude is further determined. By comparing with the qualified features in the historical qualified motion database, it is determined whether the motion is consistent. Through multi-dimensional motion feature analysis, the accuracy and reliability of motion comparison are improved.

[0052] Optionally, determining the limb shape based on the production action frame and the preset clothing image includes:

[0053] In the image of the production action frame, a region matching the preset clothing features is selected, and the clothing image within that region is extracted;

[0054] Based on the image of the clothing, extract its overall outline to determine the outline of the clothing;

[0055] The garment curve is extracted from the garment outline, and the curvature and fit of the curve are recorded.

[0056] The loft is determined based on the degree of curvature and the outline of the garment;

[0057] The limb shape is determined based on the fit and the fluffiness.

[0058] By adopting the above technical solution, clothing images are extracted from the images of the generated motion frames, thereby determining the clothing outline and curves, recording the curvature and fit of the curves, further determining the fluffiness, and determining the limb shape based on the fit and fluffiness. Through the extraction and analysis of clothing features, the shape and movement state of the limbs are more accurately reflected, providing a reference for the extraction of motion features.

[0059] Optionally, re-inspection via magnetic field testing includes:

[0060] Collect overall magnetic field signal data of product parts and mark abnormal magnetic field areas with signal fluctuations;

[0061] The magnetic field anomaly point is determined by fitting the changes in signal data in the magnetic field anomaly region.

[0062] The actual installation position of the part is determined based on the magnetic field anomaly point and the preset theoretical installation point;

[0063] The position deviation is obtained by calculating the difference between the actual installation coordinates of the part and the preset theoretical installation point.

[0064] When the position deviation exceeds the preset position tolerance threshold, a prompt is given at the preset theoretical installation point;

[0065] When the position deviation is not greater than the preset position tolerance threshold, the re-inspection is completed.

[0066] By adopting the above technical solution, the overall magnetic field signal data of the product parts is collected to determine the magnetic field anomaly points. Based on the magnetic field anomaly points and the preset theoretical installation points, the actual installation position of the parts is determined. The difference is calculated to obtain the position deviation, which is compared with the preset position tolerance threshold to determine whether there is an error. The product is re-inspected through magnetic field detection, thereby improving the accuracy and reliability of product quality inspection.

[0067] Optionally, when determining the production action frame based on the workstation image information, the method further includes:

[0068] Collect the magnetic field characteristics of the parts to be selected;

[0069] The magnetic field characteristics of the component are compared with the characteristics of a reference component in a preset standard component magnetic field database to determine whether the magnetic field strengths of the two components match.

[0070] If a match is found, the magnetic field distribution of the part is determined based on the magnetic field characteristics of the part.

[0071] The difference between the magnetic field distribution of the part and the preset required magnetic field distribution of the part is calculated to obtain the magnetic field deviation value;

[0072] If the magnetic field deviation value is not less than the preset deviation threshold, the selected part is inaccurate, and a part selection prompt is given based on the magnetic field distribution of the part.

[0073] If the magnetic field deviation value is less than the preset deviation threshold, then the selected part is determined to be accurate.

[0074] By adopting the above technical solution, the magnetic field characteristics of the parts to be selected are collected and compared with the benchmark parts characteristics in the preset standard parts magnetic field database to determine whether the magnetic field strength matches. The magnetic field distribution and magnetic field deviation of the parts are further determined, and the parts selection prompts are given. By comparing the magnetic field characteristics, the accuracy of parts selection is improved.

[0075] Optionally, determining product quality information based on the product image includes:

[0076] In the product image, the part features to be processed at the corresponding workstation are selected by using a preset part image frame, and the position of the selected part feature is defined as the part position point;

[0077] Determine the depth of field, shape, and holes of the part at the specified part location point;

[0078] The tightness of the part is determined based on the depth-of-field value of the part;

[0079] The deviation value of the part is determined based on the shape of the part and the holes in the part;

[0080] The flatness of the part is determined based on the shape of the part and the preset product shape.

[0081] The tightness, the deviation value of the part, and the flatness of the part are used as product quality information.

[0082] By adopting the above technical solution, the part features to be processed at the corresponding workstation are selected in the product image, the part position points are determined, and the part depth value, part shape and part holes are further determined, thereby determining the tightness, part deviation value and part flatness of the part as product quality information. Through multi-parameter part analysis, the comprehensiveness and accuracy of product quality inspection are ensured.

[0083] Secondly, this application provides a production behavior detection system, which adopts the following technical solution:

[0084] A production behavior detection system, comprising:

[0085] The acquisition module is used to acquire workstation image information and product images;

[0086] Memory, used to store programs that implement any production behavior detection method;

[0087] The processor loads and executes programs from memory.

[0088] In summary, this application includes at least one of the following beneficial technical effects:

[0089] 1. Collect workstation image information to determine production action frames, then extract fixed-point action features, determine whether the processed product does not need rework based on the action, and finally use magnetic field detection to re-inspect the processed product to complete the action detection process. This enables real-time monitoring of production behavior and accurate judgment of product quality, improves the controllability of the production process and the stability of product quality, thereby ensuring the quality and efficiency of production.

[0090] 2. Extract clothing images from the images of the production motion frames to determine the clothing outline and curves, record the curvature and fit of the curves, further determine the fluffiness, and determine the limb shape based on the fit and fluffiness. Through the extraction and analysis of clothing features, the shape and movement state of the limbs are more accurately reflected, providing a reference for the extraction of motion features.

[0091] 3. Collect overall magnetic field signal data of product parts to determine magnetic field anomalies. Based on the magnetic field anomalies and the preset theoretical installation points, determine the actual installation position of the parts. Calculate the difference to obtain the position deviation, compare it with the preset position tolerance threshold to determine if there is an error, and re-inspect the product through magnetic field detection to improve the accuracy and reliability of product quality inspection. Attached Figure Description

[0092] Figure 1 This is a flowchart of a production behavior detection method according to an embodiment of the present invention;

[0093] Figure 2 This is a flowchart of a method for determining limb shape according to an embodiment of the present invention;

[0094] Figure 3 This is a flowchart of a re-inspection method according to an embodiment of the present invention. Detailed Implementation

[0095] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0096] This application discloses a method for detecting production behavior.

[0097] Reference Figure 1 A method for detecting production behavior includes the following steps:

[0098] Step S100: Collect workstation image information.

[0099] Workstation image information refers to image data of the work area acquired through cameras, including continuously collected video and corresponding multi-frame images. The cameras are pre-set by technicians at the workstations on the production line according to the actual situation, which will not be elaborated here.

[0100] Step S101: Analyze and determine the production action frame based on the workstation image information.

[0101] Production action frames refer to keyframes with production actions obtained from workstation images.

[0102] After selecting multiple frames from the workstation image information at preset intervals, the production action frames are obtained.

[0103] Step S102: Annotate and extract action features in the production action frames.

[0104] Motion characteristics refer to the oscillating trajectory characteristics of a worker's production movements, which are used to determine whether the operation is qualified.

[0105] The specific methods are described in steps S200 to S204, and will not be repeated here.

[0106] The process of annotating and extracting motion features from production motion frames includes the following steps:

[0107] Step S200: Mark and determine action marker points in the production action frame.

[0108] Action markers refer to the joints of the worker's skeleton marked in the production action frame.

[0109] The specific methods are described in steps S300 to S305, and will not be repeated here.

[0110] The steps involved in annotating and determining action marker points in a production action frame are as follows:

[0111] Step S300: Extract the contour features of all targets within a single frame based on the production action frame.

[0112] Contour features refer to the contour shape of all target objects within a production action frame.

[0113] First, a foundation for contour extraction is established. Then, edge detection is used to locate the target boundary and extract the contours of all targets within a single frame. Next, target overlap is distinguished and resolved. By combining image morphology processing and contour analysis algorithms, the contour features of all targets within a single frame are extracted. This is common knowledge known to those skilled in the art and will not be elaborated here.

[0114] Step S301: Distinguish between human candidate regions and non-human regions based on contour features.

[0115] Human body candidate region refers to the region that may contain a human body.

[0116] Non-human areas refer to areas that do not contain human bodies.

[0117] The core of the human candidate region refers to the body parts involved in the worker's operation (such as hands), while the non-human region includes semi-finished products (such as products to be processed), parts to be assembled (such as screws or connectors), tools, and background clutter. The contour features of the two differ significantly (the human region has an irregular and dynamic contour, while the non-human region has a relatively regular and fixed shape). The core features are first selected, then verified through multi-dimensional algorithms, and finally dynamically assisted in the judgment to achieve accurate differentiation, ensuring that subsequent analysis focuses on human operational behavior. This is common knowledge known to those skilled in the art and will not be elaborated upon here.

[0118] Step S302: Determine the shape integrity rate and limb movement trend based on the candidate human body region and the preset human body shape.

[0119] The shape of the human body refers to the outline of the external structure and overall form of the human body, which is preset by technicians according to the actual situation, and will not be elaborated here.

[0120] Shape integrity rate refers to the spatial fit index between the candidate human body region and the human body shape, expressed as a percentage.

[0121] The tendency of a worker's limb movement refers to the direction of the next movement of the limb.

[0122] First, the candidate human body region is aligned with the human body shape to make their feature spaces consistent. Then, the contour overlap and key feature missing rate are calculated separately. The calculation results are substituted into the weighted summation formula to calculate the shape integrity rate. The weights are preset by the technicians according to the actual situation and will not be elaborated here.

[0123] Limb movement trends are used to determine whether the movement direction and speed of the candidate human body area (such as the hand) conform to the standard process of production operation (such as moving towards the semi-finished product interface during the alignment stage, rather than moving away). By combining the dynamic changes of limbs in the candidate human body area in continuous frames, the limb movement trends are analyzed (predicting the intention of the action, such as whether to move towards the interface or whether to prepare to complete the assembly), thereby obtaining the limb movement trends. This is common knowledge known to those skilled in the art and will not be elaborated here.

[0124] Step S303: Divide the human candidate regions with a shape integrity rate less than the preset shape judgment rate into non-human regions.

[0125] The shape judgment rate refers to the threshold used to determine whether the shape integrity rate is that of a human body. It is preset by technicians according to the actual situation and will not be elaborated here.

[0126] Human candidate regions with a shape integrity rate lower than the shape judgment rate are classified as non-human regions, thus completing the division of regions containing human bodies.

[0127] Step S304: Determine the worker's limb characteristics based on the contour features and limb movement trends of the redefined human candidate regions.

[0128] Worker limb features refer to the shape characteristics of a worker's limbs, which are image data used to determine motion markers.

[0129] By matching static contour features with dynamic limb movement trends, complete worker limb features can be identified and determined.

[0130] Step S305: Determine the action marker points based on the worker's limb characteristics.

[0131] The specific method for determining the action marker point is described in steps S400 to S406, and will not be repeated here.

[0132] Determining motion markers based on worker limb characteristics includes the following steps:

[0133] Step S400: Determine the static and dynamic parts of the limbs based on the worker's limb characteristics.

[0134] The static part refers to the clear, static shape and features of a limb.

[0135] The dynamic part refers to the ambiguous part of the shape characteristics and movement of the limbs.

[0136] The core feature of a statically clear part is that the details within the frame are clear and the grayscale is stable. The dynamically blurred part is that the motion causes the edge diffusion within the frame, the gradual change of grayscale, and the obvious displacement between frames. The determination method needs to combine the visual feature analysis of a single frame with the temporal correlation analysis of consecutive frames. Accurate differentiation can be achieved by quantifying the differences. This is common knowledge known to those skilled in the art and will not be elaborated here.

[0137] Step S401: Extract the dynamic shape based on the dynamic part.

[0138] Dynamic shape refers to the shape characteristics of dynamic parts.

[0139] First, the clear edges of the dynamic shape are restored using a deblurring algorithm. Then, the contour extraction range is precisely located and locked. Based on the restored dynamic region, the dynamic contour line is initially obtained through edge enhancement and contour extraction algorithms. The initial contour may have local breaks or redundant points, which are then optimized to extract the dynamic shape.

[0140] Step S402: Determine the joint positioning points based on the static and dynamic components.

[0141] Joint pinpoints refer to the location points of joints.

[0142] The fixed point connecting the static and dynamic parts is called the joint fixed point.

[0143] Step S403: Determine the joint trajectory based on the dynamic shape and joint positioning.

[0144] Joint trajectory refers to the movement path of a joint.

[0145] Once the joint fixed point is determined in the dynamic shape, the motion path traversed by the motion around the joint fixed point is called the joint trajectory.

[0146] For example, the dynamic shape is a fan shape that the arm slides across, the position of the joint is the center of the fan, and the arc length of the fan is the joint trajectory.

[0147] Step S404: Determine the joint movement point based on the joint trajectory.

[0148] Joint movement points refer to the key points of joint movement.

[0149] During movement, the joint slides continuously in space along with the limb's movements. These points together constitute the joint's movement trajectory. The basic unit that reflects the dynamic positional changes of the joint is the joint's moving point.

[0150] Step S405: Determine the static fixed point based on the static part and the preset human body shape.

[0151] Static fixed point refers to the location point of the static part.

[0152] The joints that are relatively fixed and do not change with limb movement, and are determined one by one according to the shape of the human body, are called static fixed points.

[0153] For example, when working, the humerus (arm) is in a static support state, and the joint at the shoulder of the scapula is a static fixed point.

[0154] Step S406: Use the static fixed point, joint fixed point, and joint moving point as motion marker points.

[0155] Static fixed points, joint fixed points, and joint moving points are used as motion markers to determine motion characteristics.

[0156] Step S201: Determine the limb shape based on the production action frame and the preset clothing image.

[0157] Clothing images refer to images of various workers' clothing, used to identify the shape of workers' limbs from the characteristics of production movements.

[0158] Limb shape refers to the external shape of a worker's limbs in a production motion frame.

[0159] The specific methods are described in steps S600 to S604, and will not be repeated here.

[0160] Reference Figure 2 Determining limb shapes based on production motion frames and preset clothing images includes the following steps:

[0161] Step S600: Select the region that matches the preset clothing features in the image of the production action frame, and extract the clothing image within that region.

[0162] Clothing characteristics refer to the appearance features of clothing, such as shape, color, and location. These are preset by technicians based on the actual situation and will not be elaborated here.

[0163] Clothing images refer to images with clothing features extracted from images of production action frames.

[0164] By annotating clothing features in a large number of images, the labeled images are input into the YOLO large model. The clothing features are extracted by repeatedly stacking images using a PyTorch network architecture. The error between the results and the data is calculated. When the error is less than 1%, the model is put into use.

[0165] The working image is input into the YOLO large model. When clothing features are recognized, the clothing features are marked and selected from the image of the production action frame. The selected image is the clothing image, which is common knowledge known to those skilled in the art and will not be elaborated here.

[0166] Step S601: Based on the clothing image, extract its overall outline to determine the clothing outline.

[0167] Clothing outline refers to the overall shape of a clothing image.

[0168] The method for determining the outline of clothing is the same as step S300, and will not be repeated here.

[0169] Step S602: Extract the garment curve from the garment outline and record the curvature and fit of the curve.

[0170] Clothing curves refer to the curved portion of the garment's silhouette. Degree of curvature refers to the extent to which the garment's curves bend. Fit refers to how well the garment's curves fit the body.

[0171] The specific methods are described in steps S6020 to S6025, and will not be repeated here.

[0172] Step S6020: Based on the clothing curve, extract and statistically determine the number of curve inflection points and the number of folds of the clothing, and calculate the curvature value at each inflection point based on the clothing curve.

[0173] The number of curve inflection points refers to the number of key nodes where the direction of the curve changes. Curve inflection points are like the connection points where the curved edge of a collar turns into a straight edge.

[0174] The number of pleats refers to the number of continuous undulations formed by the stacking of fabric, such as the stacking caused by the looseness of the garment.

[0175] The curvature value refers to the numerical value that quantifies the degree of curvature at the inflection point.

[0176] The number of inflection points and the number of folds in the statistical curve are extracted from the clothing curve and obtained through the corresponding counting algorithm. This is common knowledge known to those skilled in the art and will not be elaborated here.

[0177] The curvature value at each inflection point is calculated by substituting the data of the clothing curve into the preset curvature formula. The curvature formula is as follows:

[0178]

[0179] The inflection point is P(xP,yP) (the point where curvature needs to be calculated), the one adjacent sampling point before the inflection point is A(xA,yA), and the one adjacent sampling point after the inflection point is B(xB,yB). Then the curvature value at the inflection point P is k.

[0180] Step S6021: Determine the degree of curvature based on the number of curve inflection points and the curvature value matching.

[0181] The more inflection points a curve has and the greater its curvature value, the greater the degree of bending. The degree of bending is obtained by inputting the number of inflection points and the curvature value into a preset bending degree database. The bending degree database is a database that is preset by technicians according to the actual situation. The bending degree database contains the relationship between the number of inflection points and the curvature value and the degree of bending. The actual parameters are preset by technicians according to the actual situation, which will not be elaborated here.

[0182] Step S6022: Determine the fold spacing based on the number of folds and the degree of curvature.

[0183] The fold spacing refers to the average distance between the centers of adjacent effective folds on the same continuous curved curve segment of clothing.

[0184] First, the target curve segment of the fold concentration is identified by the degree of curvature, and the actual physical length of the segment is calculated. Then, the number of intervals in most fold concentrations is determined by combining the number of folds. Finally, the fold spacing is obtained by dividing the length of the target curve segment by the number of intervals.

[0185] Step S6023: Determine the fold density based on the fold spacing and curvature value.

[0186] Pleat density refers to the density of the effective pleat distribution within a specific area of ​​clothing.

[0187] The fold spacing and curvature values ​​are substituted into a preset fold model. The fold model first calibrates the parameter units and precision and removes outliers, while matching the garment type and process standards to achieve data consistency and scenario adaptation. Then, the fold spacing, curvature values ​​and other parameters are accurately mapped according to the model input port requirements, and single or batch input is performed according to different interaction methods such as visualization software and code interfaces. Finally, the model outputs compliance judgment, grade classification and other results according to preset logic, and is directly applied to actual scenarios such as production quality inspection, pattern optimization and virtual fitting, forming a complete closed-loop process. The specific parameters are preset by technical personnel according to the actual situation, and will not be elaborated here.

[0188] Step S6024: Determine the interval distance difference of the corresponding positions based on the clothing curve and the human body shape.

[0189] The interval distance difference refers to the spatial distance difference between each point on the clothing curve and the corresponding point on the human body curve.

[0190] Based on the different structural partitions of the clothing curves, corresponding matching areas are determined on the baseline contour of the human body. Then, within each matching area, sampling points corresponding to the clothing curves and the baseline contour of the human body are selected: sampling points are taken at evenly spaced positions on each key structural curve, and corresponding baseline sampling points are taken at the same horizontal / vertical positions on the baseline contour of the human body. Next, the straight-line distance between each pair of corresponding sampling points is calculated to obtain the interval distance value at a single position, and the distribution of all interval distance values ​​within each key structural area is recorded. Then, based on the interval distance values ​​within each key structural area, outliers caused by fabric wrinkles or coordinate acquisition errors are removed. Subsequently, the corrected interval distance values ​​within each key structural area are statistically analyzed to obtain the interval distance difference at corresponding positions. Finally, the interval distance difference results for corresponding positions of each key structure are output.

[0191] Step S6025: Determine the fit based on the gap distance difference and fold density.

[0192] After assigning corresponding weights to the interval distance difference and fold density, the fit is obtained through weighted calculation.

[0193] The specific weights are preset by technical personnel based on the actual situation, and will not be elaborated here.

[0194] Step S603: Determine the fluffiness based on the degree of curvature and the outline of the garment.

[0195] Loft refers to the degree to which clothing is fluffy.

[0196] The specific methods are described in steps S6030 to S6034, and will not be repeated here.

[0197] Step S6030: Calculate the actual coverage area of ​​the clothing in the production action frame based on the clothing outline.

[0198] Actual coverage area refers to the physical area of ​​the clothing itself, including wrinkles, fluff, and excess areas beyond the human body.

[0199] First, image preprocessing is used to eliminate interference and enhance the outline of clothing. Then, clothing outline extraction and screening are performed to eliminate background interference. The extracted outlines are repaired to reduce occlusion and breaks. Finally, the projected area of ​​clothing is calculated and statistically analyzed based on pixel area to determine the actual coverage area.

[0200] Step S6031: Calculate the clothing coverage area corresponding to the clothing within the human body shape based on the actual coverage area mapping.

[0201] Clothing coverage area refers to the surface area of ​​the human body covered by clothing.

[0202] First, establish a baseline coverage area for the human body, divide the human body parts according to the clothing coverage logic, and calculate the baseline area; then, map the coverage areas of the clothing and human body parts through image registration or grid alignment; then, subtract redundant areas and non-fitting areas from the initial mapped area; finally, sum the effective coverage areas of each part to obtain the clothing coverage area.

[0203] Step S6032: Calculate the area difference between the area covered by clothing and the actual area covered.

[0204] The area difference refers to the difference between the area covered by clothing and the actual area covered.

[0205] The difference between the actual covered area and the area covered by clothing is the area difference.

[0206] Step S6033: Determine the degree of unevenness on the surface of the clothing based on the outline of the clothing and the shape of the human body.

[0207] The degree of unevenness refers to the difference in three-dimensional undulation between the surface shape enclosed by the outline of clothing and the corresponding reference shape of the human body.

[0208] Align the clothing outline with the human body shape to ensure that every point on the clothing surface can find a reference point corresponding to the human body shape, avoiding misjudgment of concavity and convexity due to coordinate misalignment. For each point on the clothing surface after accurate matching, calculate its vertical distance to the human body shape, distinguish between convex and concave, obtain the data of the corresponding point, and then obtain the degree of concavity and convexity through weighted calculation.

[0209] Step S6034: Calculate the fluffiness based on the weighted average of the degree of curvature, the degree of unevenness, and the area difference.

[0210] The specific weights for curvature, unevenness, and area difference are preset by technicians based on actual conditions and will not be elaborated here.

[0211] Step S604: Determine the limb shape based on fit and fluffiness.

[0212] Limb shape refers to the shape and characteristics of a limb.

[0213] If the fit is high, it means that the curves of the clothing are close to the shape of the limb. Based directly on the curves of the clothing, we can eliminate the small abnormal protrusions caused by fabric wrinkles (judged by the stability of the continuous frame curves) and obtain the preliminary outline of the limb.

[0214] If the fit is low, further analysis is needed based on the loft: if the loft is high, it indicates that there is a lot of redundant space in the clothing. Key support points in the clothing curve (such as the curve inflection points corresponding to the acromion, elbow, and hip) are extracted. These points are less affected by the loft. Based on the key support points, the area difference between the clothing outline and the human body baseline outline is reduced (refer to the area redundancy ratio corresponding to the loft) to obtain the basic outline of the limb.

[0215] If the fit and loft are low, it indicates a structural deviation between the garment's pattern and the body shape. Based on the direction of the straight / curved structures such as the side seams and armholes of the garment, adjust the corresponding dimensions of the human body's baseline contour (e.g., the side seam inclination angle corresponds to the limb width, and the armhole curvature corresponds to the arm thickness) to improve the fit between the adjusted baseline contour and the garment's curve, thus forming a preliminary limb contour.

[0216] Based on the basic outline of the limb, and combined with the curvature of the clothing curves (the curvature direction corresponds to the trend of limb movement, such as the armhole curve curving inward when the elbow bends), the outline details are corrected: for parts with a high degree of curvature (such as cuffs and trouser hems), the outline curvature is adjusted according to the curvature direction to match the limb joint activity state.

[0217] Verify the connection between the corrected limb contour and adjacent parts (such as whether the transition between the shoulder contour and the arm contour is natural). If there are gaps in the connection, further optimize by referring to the proportional relationship of the human body shape, so as to determine the final limb shape.

[0218] Step S202: Mark and determine key motion points in the production motion frame based on motion marker points and limb shapes.

[0219] Critical action points refer to the key locations in production processes.

[0220] In each production action frame, the action markers are connected by the shape of the limbs. The center point of the line segment is the key action point of that limb shape. By marking all the key action points of the limb shape in the single frame image, all the key action points in the single frame are obtained, and thus the key action points of the production action frame are obtained.

[0221] For example, the knee and the groin are two action markers. Connecting these two points with the shape of the limbs creates a line segment on the thigh, and the center point of this line segment is the key action point.

[0222] Step S203: Based on the marked key action points, connect them to form an action trajectory.

[0223] The motion trajectory refers to the movement path formed by connecting key motion points.

[0224] Based on the order of the production action frames, key action points at the same location are linked together to obtain the corresponding action trajectory.

[0225] For example, by connecting the key movement points of the thigh, the movement path can be obtained, which is the movement trajectory of the thigh.

[0226] Step S204: Determine the movement characteristics of limb swinging based on the movement trajectory and limb shape.

[0227] Movement characteristics refer to the movements made by limb swinging, which are used to judge whether the movement is qualified.

[0228] The movement characteristics are the actions performed by controlling the shape of the limbs along the movement trajectory using key movement points of the limb shape.

[0229] Step S103: Determine whether the action features are consistent with the qualified features in the preset historical qualified action database.

[0230] The historical qualified action database refers to a database that stores the characteristics of qualified actions. It is pre-set by technicians according to the actual situation and will not be elaborated on here.

[0231] Qualified features refer to characteristic actions in the historical qualified action database that meet the requirements for qualified production.

[0232] For details, please refer to steps S500 to S506, which will not be repeated here.

[0233] Determining whether the action characteristics match the qualified characteristics in the preset historical qualified action database includes the following steps:

[0234] Step S500: Determine the trajectory shape based on the motion trajectory.

[0235] Trajectory shape refers to the shape characteristics of a motion trajectory.

[0236] The shape that a limb moves along a movement trajectory is called the trajectory shape.

[0237] For example, the shape of a limb is the arm. The fan-shaped arc formed by the arm's movement trajectory is the trajectory shape.

[0238] Step S501: Determine the motion angle based on the trajectory shape.

[0239] The motion angle refers to the angle through which the trajectory shape slides.

[0240] For example, the trajectory shape of a fan is determined by first determining the position of the fan's center, and then the arc angle of the fan-shaped circle formed by connecting the fan shape and the fan's center is the motion angle.

[0241] Step S502: Determine the motion displacement based on the motion characteristics and the preset human body shape.

[0242] Action displacement refers to the change in displacement within the trajectory of an action.

[0243] After the human body shape is controlled to move according to the motion feature, the displacement between the starting key motion point and the last key motion point in the motion feature is the motion displacement.

[0244] Step S503: Determine the amplitude of the movement based on the movement angle and movement displacement.

[0245] Amplitude of motion refers to the magnitude of the production motion.

[0246] The range of motion = angle of motion × displacement of motion / standard reference value. The standard reference value is set in advance by technicians according to the actual situation, and will not be elaborated here.

[0247] Step S504: Determine the degree of overlap of actions based on the qualified range and action range of qualified features in the preset historical qualified action database.

[0248] The acceptable range refers to the standard range of acceptable production actions.

[0249] Movement overlap refers to the degree of matching between the range of motion and the acceptable range of motion.

[0250] The degree of overlap of movements is calculated based on the acceptable range and the range of motion. The calculation method is common knowledge known to those skilled in the art and will not be elaborated here.

[0251] Step S505: When the overlap is greater than the preset judgment overlap, the action feature is consistent with the qualified features in the preset historical qualified action database.

[0252] The degree of overlap refers to the preset threshold for the degree of overlap of actions, which is set in advance by technicians according to the actual situation, and will not be elaborated here.

[0253] When the overlap is greater than the judgment overlap, it indicates that the action characteristics are consistent with the qualified characteristics.

[0254] Step S506: When the overlap is not greater than the preset judgment overlap, the action feature is inconsistent with the qualified features in the preset historical qualified action database.

[0255] When the degree of coincidence is not greater than the judged degree of coincidence, it indicates that the action feature is inconsistent with the qualified feature.

[0256] Step S104: If they are consistent, mark the product as a processed product.

[0257] A processed product refers to a product to be re-inspected produced through qualified production actions.

[0258] If the action feature is consistent with the qualified feature in the historical qualified action database, it indicates that the production action meets the requirements, and wait for re-inspection.

[0259] Step S105: If they are inconsistent, collect the corresponding product image and issue a prompt.

[0260] A product image refers to the image data of the corresponding product.

[0261] A prompt refers to a signal reminding that the production action is unqualified.

[0262] The prompting method is common knowledge known to those skilled in the art and will not be elaborated here.

[0263] If the action feature is inconsistent with the qualified feature in the historical qualified action database, it indicates that the production action does not meet the requirements and rework is needed.

[0264] Step S106: Determine the product quality information based on the product image, and determine whether it is qualified based on the product quality information and the preset quality parameters.

[0265] Product quality information refers to the data sizes of the tightness, part deviation value, and part flatness related to quality extracted from the product image.

[0266] Quality parameters refer to the parameters of the tightness, part deviation value, and part flatness when the product quality meets the standard, which are preset by technicians according to the actual situation and will not be elaborated here.

[0267] For the specific method of determining the product quality information, refer to Steps S900 to S906, and it will not be elaborated here.

[0268] When the product quality information is greater than the quality parameter, it is qualified; otherwise, it is unqualified.

[0269] Determining the product quality information based on the product image includes the following steps:

[0270] Step S900: In the product image, frame the part features required for processing at the corresponding workstations with a preset part image frame, and define the position where the part features are framed as the part position point.

[0271] A product image refers to the image data of the product to be inspected.

[0272] Part images refer to part images that meet production requirements. They are pre-set by technicians based on actual conditions and will not be elaborated on here.

[0273] The part location point refers to the position of the part in the product image.

[0274] The specific method for selecting part features from the product image is described in step S600, and will not be repeated here.

[0275] The part location point is obtained by selecting the part features by means of a bounding box. This is common knowledge known to those skilled in the art and will not be elaborated here.

[0276] Step S901: Determine the depth of field, shape, and holes of the part at the part location point.

[0277] The depth of field value of a part refers to the depth of field value of the part image.

[0278] Part shape refers to the shape characteristics of a part.

[0279] Holes in a part refer to the hole features on a part.

[0280] Use a camera to take multiple images of the same scene at different focal lengths, or use a lens with a variable focal length to continuously change the focal length during the shooting process to obtain a series of images with different focus states.

[0281] For each pixel or region in the image, its focus metric is calculated using a gradient-based method that evaluates pixel sharpness by calculating the high-frequency components or edge information of the image.

[0282] For example, a search-based method can be used to find the image that maximizes the focus metric of a particular pixel among multiple images; the focal length of this image is then related to the depth of field of that pixel. Alternatively, a mathematical model can be established between focus metric and depth of field, and the depth of field can be estimated by fitting the model. Based on the focus metric of the part image at different focal lengths, the focal length at which each pixel or region is most sharp can be analyzed, thereby inferring the depth of field value of the part at that location.

[0283] The method of determining the shape and holes of a part based on images of its location points is common knowledge known to those skilled in the art and will not be elaborated upon here.

[0284] Step S902: Determine the tightness of the part based on the part's depth of field value.

[0285] Tightness refers to the degree of tightness or looseness of a part.

[0286] The greater the depth of field value of a part, the greater its tightness. The tightness is obtained by inputting the depth of field value of the part into a preset tightness database. The tightness database is a database that is preset by technicians according to the actual situation. The tightness database contains the relationship between the depth of field value of the part and the tightness. The actual parameters are preset by technicians according to the actual situation, which will not be elaborated here.

[0287] Step S903: Determine the part deviation value based on the part shape and part holes.

[0288] Part deviation value refers to the magnitude of the deviation between the part shape and the part hole.

[0289] When the shape of a part fails to completely cover the hole in the part, the deviation value of the part is calculated based on the difference between the outer boundary of the part shape and the outer boundary of the hole. The specific method is common knowledge known to those skilled in the art and will not be elaborated here.

[0290] Step S904: Determine the flatness of the part based on its shape and the preset product outline.

[0291] Product appearance refers to the standard shape of the product, which is set in advance by technicians according to the actual situation, and will not be elaborated here.

[0292] Flatness refers to the degree of flatness of a part. The specific methods are common knowledge known to those skilled in the art and will not be elaborated here.

[0293] Step S905: Use tightness, part deviation value and part flatness as product quality information.

[0294] By using tightness, part deviation, and part flatness as product quality information, multiple parameters are determined to ensure comprehensive inspection.

[0295] Step S107: If the product quality information is qualified, then mark the product as a processed product.

[0296] If the product quality information is qualified, it means that the product needs to be re-inspected.

[0297] Step S108: If the product quality information is unqualified, then mark the product as a product to be processed.

[0298] Products awaiting processing refer to products that require reprocessing.

[0299] If the product quality information is unqualified, it means that the production was defective and rework is required.

[0300] Step S109: Re-inspect the processed product through magnetic field detection to complete motion detection.

[0301] Magnetic field testing refers to the method of detecting whether a product is qualified by using magnetic field signals.

[0302] The specific method for re-inspection is described in steps S700 to S705, and will not be repeated here.

[0303] The processed products are re-inspected by magnetic field detection to determine if there are any errors in part selection or part placement, and then motion detection is completed.

[0304] Reference Figure 3 The re-inspection via magnetic field detection includes the following steps:

[0305] Step S700: Collect overall magnetic field signal data of the product parts and mark the abnormal magnetic field areas with signal fluctuations.

[0306] Magnetic field signal data refers to the magnetic field signal data generated by the product parts, which is acquired by a preset Hall effect sensor. The Hall effect sensor is preset by technicians according to the actual situation, and will not be described in detail here.

[0307] An abnormal magnetic field area refers to a region in which the magnetic field signal fluctuates abnormally due to incorrect assembly of product parts.

[0308] When the overall magnetic field signal data is inconsistent with the preset magnetic field data, the location of the signal fluctuation is determined, which is the magnetic field anomaly area.

[0309] The magnetic field data is preset by technicians based on the actual situation, and will not be elaborated here.

[0310] Step S701: Determine the magnetic field anomaly point by fitting the changes in signal data in the magnetic field anomaly region.

[0311] A magnetic field anomaly point refers to the specific location where the magnetic field signal is abnormal, that is, the specific location where a part is misassembled, such as a hole or a joint.

[0312] The location with the greatest change in signal data within the magnetic field anomaly region is designated as the magnetic field anomaly point.

[0313] Step S702: Determine the actual installation position of the part based on the magnetic field anomaly point and the preset theoretical installation point.

[0314] The theoretical mounting point refers to the standard mounting position of the part, which is set in advance by technicians according to the actual situation, and will not be elaborated here.

[0315] The actual installation location of a part refers to its actual installation position.

[0316] The magnetic field model of the theoretical installation point is compared with the actual detected magnetic field anomaly point. The spatial offset of the part is inferred by mathematical algorithm or physical model. The specific method is common knowledge known to those skilled in the art and will not be elaborated here.

[0317] Step S703: Calculate the difference between the actual installation coordinates of the part and the preset theoretical installation point to obtain the position deviation.

[0318] Position deviation refers to the distance between the actual installation position of a part and its theoretical installation point.

[0319] The difference between the actual installation position of a part and its theoretical installation point is called the position deviation.

[0320] Step S704: When the position deviation is greater than the preset position tolerance threshold, a prompt is given at the preset theoretical installation point.

[0321] The position tolerance threshold refers to the allowable range of positional deviation, which is preset by technicians according to the actual situation and will not be elaborated here.

[0322] The prompt indicates a message reminding the user that the part's installation position is significantly off-center.

[0323] If the positional deviation is not greater than the positional tolerance threshold, it indicates that there is a problem with the installation and the installation should be carried out according to the instructions of the theoretical installation point.

[0324] Step S705: When the position deviation is not greater than the preset position tolerance threshold, the re-inspection is completed.

[0325] When the positional deviation is not greater than the positional tolerance threshold, it indicates that the installation is correct and the re-inspection is complete.

[0326] Determining production action frames based on workstation image information includes the following steps:

[0327] Step S800: Collect the magnetic field characteristics of the part to be selected.

[0328] The magnetic field characteristics of a component refer to the magnetic field characteristics generated by the component.

[0329] The data collection process is the same as step S700, and will not be repeated here.

[0330] Step S801: Compare the magnetic field characteristics of the part with the reference part characteristics in the preset standard part magnetic field database to determine whether the magnetic field strengths of the two are matched.

[0331] The standard parts magnetic field database refers to a database that stores the magnetic field characteristics of standard parts. It is pre-set by technicians according to actual conditions and will not be elaborated here.

[0332] Reference part characteristics refer to the magnetic field characteristics of standard parts.

[0333] Magnetic field strength refers to the intensity of a magnetic field.

[0334] If the magnetic field strength of the magnetic field feature of a part is consistent with the magnetic field strength of the feature of the reference part, then they are matched; otherwise, they are not matched.

[0335] Step S802: If a match is found, the magnetic field distribution of the part is determined based on the magnetic field characteristics of the part.

[0336] The magnetic field distribution of a component refers to the distribution of the magnetic field of the component.

[0337] If the magnetic field characteristics of the part match those of the reference part, it means that the selected part is consistent with the required part.

[0338] If the magnetic field characteristics of the part do not match those of the reference part, it indicates that the selected part and the required part are inconsistent.

[0339] Step S803: Calculate the difference between the magnetic field distribution of the part and the preset required magnetic field distribution of the part to obtain the magnetic field deviation value.

[0340] The required magnetic field distribution for the parts refers to the magnetic field distribution that is set in advance by technicians based on the actual situation, and will not be elaborated here.

[0341] The magnetic field deviation value refers to the deviation between the magnetic field distribution of the part and the required magnetic field distribution.

[0342] The magnetic field deviation value is obtained by subtracting the required magnetic field distribution from the magnetic field distribution of the component.

[0343] The specific methods are common knowledge known to those skilled in the art and will not be elaborated here.

[0344] Step S804: If the magnetic field deviation value is not less than the preset deviation threshold, the selected part is inaccurate, and a part selection prompt is given based on the magnetic field distribution of the part.

[0345] The deviation threshold refers to the allowable range of magnetic field deviation, which is preset by technicians according to the actual situation and will not be elaborated here.

[0346] The prompt indicates that the selected part is inaccurate.

[0347] If the magnetic field deviation value is not less than the deviation threshold, it indicates that the selected part is inaccurate, and the correct part will be selected based on the magnetic field distribution of the part.

[0348] Step S805: If the magnetic field deviation value is less than the preset deviation threshold, then the selected part is determined to be accurate.

[0349] If the magnetic field deviation is less than the deviation threshold, it indicates that the selected part is accurate, and the selection of the part is confirmed.

[0350] Based on the same inventive concept, embodiments of the present invention provide a production behavior detection system, including:

[0351] The acquisition module is used to acquire workstation image information, product images, overall magnetic field signal data, and part magnetic field characteristics.

[0352] The memory is used to store programs that implement any production behavior detection method.

[0353] The processor loads and executes programs from memory.

[0354] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0355] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting production behavior, characterized in that, include: Collect workstation image information; The production action frame is determined based on the analysis of the workstation image information; The action features are annotated and extracted from the production action frames; Determine whether the action feature is consistent with the qualified features in a preset historical qualified action database; If they match, the product will be marked as a processed product; If there is a discrepancy, the corresponding product image will be captured and a prompt will be issued; Based on the product image, determine the product quality information, and based on the product quality information and preset quality parameters, determine whether it is qualified; If the product quality information is qualified, the product will be marked as a processed product; If the product quality information is unqualified, the product will be marked as a product awaiting processing. The processed product is re-inspected by magnetic field detection to complete motion detection; The action features obtained by annotating and extracting the production action frames include: In the production action frame, mark and determine the action marker points; The limb shape is determined based on the production action frame and the preset clothing image; Based on the motion markers and the limb shape, key motion points are marked and determined in the production motion frame; Based on the marked key action points, a motion trajectory is formed by connecting them; The movement characteristics of limb swinging are determined based on the movement trajectory and the limb shape; Determining the limb shape based on the production action frame and the preset clothing image includes: In the image of the production action frame, a region matching the preset clothing features is selected, and the clothing image within that region is extracted; Based on the image of the clothing, extract its overall outline to determine the outline of the clothing; The garment curve is extracted from the garment outline, and the curvature and fit of the curve are recorded. The loft is determined based on the degree of curvature and the outline of the garment; The limb shape is determined based on the fit and the fluffiness.

2. The production behavior detection method according to claim 1, characterized in that, In the production action frame, the annotation and determination of action marker points includes: Based on the production action frame, extract the contour features of all targets within a single frame; Based on the contour features, human candidate regions and non-human regions are distinguished; The shape integrity rate and limb movement trend are determined based on the candidate human body regions and the preset human body shape. The human candidate regions whose shape integrity rate is less than the preset shape judgment rate are classified as non-human regions; Based on the contour features and limb movement trends of the redefined human candidate regions, the worker's limb features are determined; Action markers are determined based on the worker's limb characteristics.

3. The production behavior detection method according to claim 2, characterized in that, Determining motion markers based on the worker's limb characteristics includes: The static and dynamic parts of the limbs are determined based on the worker's limb characteristics; Extract the dynamic shape based on the dynamic portion; Joint positioning is determined based on the static and dynamic components; The joint trajectory is determined based on the dynamic shape and the joint positioning. The joint movement point is determined based on the joint trajectory; Static points are determined based on the static components and the preset human body shape; The static fixed point, the joint fixed point, and the joint moving point are used as motion marker points.

4. The production behavior detection method according to claim 1, characterized in that, Determining whether the action feature matches the qualified features in the preset historical qualified action database includes: Determine the trajectory shape based on the described action trajectory; Determine the motion angle based on the trajectory shape; The motion displacement is determined based on the motion characteristics and the preset human body shape; The amplitude of the movement is determined based on the movement angle and the movement displacement; The degree of overlap of actions is determined based on the passability range of the passability features in the preset historical passability database and the action range. When the overlap is greater than the preset judgment overlap, the action feature is consistent with the qualified features in the preset historical qualified action database. When the overlap is not greater than the preset judgment overlap, the action feature is inconsistent with the qualified features in the preset historical qualified action database.

5. The production behavior detection method according to claim 1, characterized in that, Re-inspection via magnetic field detection includes: Collect overall magnetic field signal data of product parts and mark abnormal magnetic field areas with signal fluctuations; The magnetic field anomaly point is determined by fitting the changes in signal data in the magnetic field anomaly region. The actual installation position of the part is determined based on the magnetic field anomaly point and the preset theoretical installation point; The position deviation is obtained by calculating the difference between the actual installation coordinates of the part and the preset theoretical installation point. When the position deviation exceeds the preset position tolerance threshold, a prompt is given at the preset theoretical installation point; When the position deviation is not greater than the preset position tolerance threshold, the re-inspection is completed.

6. The production behavior detection method according to claim 1, characterized in that, When determining the production action frame based on the workstation image information, the method further includes: Collect the magnetic field characteristics of the parts to be selected; The magnetic field characteristics of the component are compared with the characteristics of a reference component in a preset standard component magnetic field database to determine whether the magnetic field strengths of the two components match. If a match is found, the magnetic field distribution of the part is determined based on the magnetic field characteristics of the part. The difference between the magnetic field distribution of the part and the preset required magnetic field distribution of the part is calculated to obtain the magnetic field deviation value; If the magnetic field deviation value is not less than the preset deviation threshold, the selected part is inaccurate, and a part selection prompt is given based on the magnetic field distribution of the part. If the magnetic field deviation value is less than the preset deviation threshold, then the selected part is determined to be accurate.

7. The production behavior detection method according to claim 1, characterized in that, Determining product quality information based on the product image includes: In the product image, the part features to be processed at the corresponding workstation are selected by using a preset part image frame, and the position of the selected part feature is defined as the part position point; Determine the depth of field, shape, and holes of the part at the specified part location point; The tightness of the part is determined based on the depth-of-field value of the part; The deviation value of the part is determined based on the shape of the part and the holes in the part; The flatness of the part is determined based on the shape of the part and the preset product shape. The tightness, the deviation value of the part, and the flatness of the part are used as product quality information.

8. A production behavior detection system, characterized in that, include: The acquisition module is used to acquire workstation image information and product images; A memory for storing a program that implements the production behavior detection method according to any one of claims 1 to 7; The processor loads and executes programs from memory.

Images