Image type labor protection glove dipping positioning device

Abstract

The application provides an image type labor protection glove dipping positioning device, which comprises an image recognition module for detecting a hand mold, and a dipping angle servo control module for adjusting the position of a laser emitting device. When the glove is dipped on the hand mold, the hand mold is vertically rotated to make the characteristic position of the hand mold turn to a laser detection limit point, so that the posture of the hand mold is determined as a dipping posture by a flow line. The laser detection limit point is formed by a laser spot emitted by a laser sensor. According to the recognition result of the hand mold of the image recognition module, the dipping angle servo control module is used to move the position of the laser emitting device to compensate for the size deviation of the hand mold, and the error of a chain sprocket is compensated by the dipping angle servo control module, so that the dipping postures of the hand molds are consistent when the hand molds turn to the laser detection limit point. The application can timely determine the size of the labor protection glove dipping hand mold, and automatically adjust the dipping angle of the hand mold according to the corresponding size of the hand mold, so that the best dipping forming effect is achieved.

Description

Technical Field

[0001] This invention relates to the field of dip coating technology, and in particular to an image-based positioning device for dip coating of work gloves. Background Technology

[0002] In existing technologies, the hand mold dipping process is as follows: Figure 1 As shown. 1. The production line uses a chain drive to transport the hand molds to the top of the impregnation tank. 2. Each hand mold first rotates 180° counterclockwise, then slowly rotates clockwise to the laser detection limit angle. 3. Once the laser spot detects the hand mold's limit position, the slow clockwise rotation stops. 4. The impregnation tank rises, completing the hand mold impregnation process.

[0003] The limitations of existing technologies are analyzed in the following table:

[0004]

[0005] In existing technologies, due to the cumulative error caused by the varying lengths of each chain link in the limited chain transmission mechanism, the stop position of the mold bar in the laser inspection area has a left / right variation range, which in turn causes the limiting angle deviation of the laser inspection mold, seriously affecting the consistency quality of the nitrile latex molding area of ​​each mold bar.

[0006] In summary, the defects in the current existing technology of the dip-coating process for work gloves include (such as...) Figure 2 ):

[0007] 1. Error caused by cumulative clearance of the transmission mechanism: After the hand mold enters the laser inspection area of ​​the dipped part, there is a cumulative clearance error range of the transmission mechanism, which causes the limiting angle of the laser inspection hand mold to have a left / right deviation.

[0008] 2. Deviation in height / height caused by differences in hand mold number: Currently, the hand mold impregnation process requires manual identification of the hand mold size and model, and manual adjustment of the laser sensor position according to the hand mold size in order to achieve the optimal impregnation angle. Summary of the Invention

[0009] This invention proposes an image-based positioning device for dipped work gloves, which can promptly determine the size of the dipped work glove mold and automatically adjust the dipping angle of the mold according to the corresponding mold size to achieve the best dipping molding effect.

[0010] The present invention adopts the following technical solution.

[0011] An image-based positioning device for dipping and positioning work gloves includes an image recognition module for detecting hand molds located at the chain drive mechanism of the dipping production line, and a dipping angle servo control module for adjusting the position of a laser emitter. When dipping the glove onto the hand mold fitted onto the mold bar of the chain drive mechanism, the mold bar drives the hand mold to rotate vertically around the wrist of the hand mold so that the hand mold's feature position is rotated to the laser detection limit point, allowing the production line to determine that the hand mold's posture is the dipping posture.

[0012] The laser detection limit point is formed by the laser spot emitted by the laser sensor. The positioning device moves the position of the laser emitting device according to the hand mold recognition result of the image recognition module through the glue dipping angle servo control module to compensate for the hand mold specification deviation, and compensates for the chain sprocket error through the glue dipping angle servo control module, so that the glue dipping posture of the hand mold tends to be consistent when the hand mold turns to the laser detection limit point.

[0013] A dipping tank, which is a nitrile rubber tank, is provided at the end of the chain drive mechanism. During dipping, multiple gloved hand molds are fixed in the same posture at the mold bar of the chain drive mechanism to form each mold bar. The chain drive mechanism first sends each mold bar to the top of the dipping tank in chain drive mode. Then, the mold bar rotates counterclockwise vertically by a preset angle of 180 degrees. Then, the mold bar slowly rotates clockwise vertically to the laser detection limit point, so that the laser spot of the laser emitting device falls on the preset position of the hand mold. After the production line detects the laser spot at the hand mold, it determines that the hand mold is in the dipping posture. The dipping tank rises so that the hand mold enters the dipping tank in the dipping posture and is immersed for a preset time of 4-8 seconds to ensure that the nitrile rubber latex is evenly adhered to the surface of the hand mold.

[0014] The image recognition module includes a webcam and a Raspberry Pi controller with a built-in Raspberry Pi.

[0015] A webcam is installed in front of the nitrile tank and connected to the Raspberry Pi controller. The webcam is connected to the Raspberry Pi controller via a network cable. The webcam takes images of each hand mold before impregnation according to the impregnation production line and transmits each hand mold image to the Raspberry Pi controller.

[0016] The Raspberry Pi controller first uses AI machine vision technology to automatically identify the size of the hand mold based on the image of the hand mold before impregnation captured by the camera. Then, it sends the identified hand mold size to the PLC controller of the impregnation angle servo control module.

[0017] The dip angle servo control module includes a PLC controller and a dip angle servo control mechanism. The PLC controller adjusts the dip angle servo mechanism to adjust the position of the laser spot according to the hand mold size transmitted by the image recognition module, so that when the hand mold rotates vertically to reach the laser spot, the posture of each hand mold reaches the optimal dip angle.

[0018] The positioning device also includes a cloud platform module; after each hand mold is dipped in glue, the PLC controller transmits the processing parameters of each hand mold back to the cloud platform module via the Raspberry Pi controller; the cloud platform module records the processing parameters of the hand mold in the glue dipping process (production date, time, work gloves model, glue dipping angle, glue dipping time).

[0019] The controller is a Raspberry Pi controller with a built-in Raspberry Pi, which uses a hand mold size discrimination algorithm to identify the hand mold size. The hand mold size discrimination algorithm includes a hand mold area discrimination algorithm and a palm key point discrimination algorithm.

[0020] The hand mold area discrimination algorithm uses an AI neural network (e.g., Yolov8-seg and Mask R-CNN) to estimate the mask region of the hand mold; the AI ​​neural network further calculates the mask area index and the maximum mask width index as discrimination indicators to determine the size of the hand mold;

[0021] YOLOv8-seg consists of a backbone, a feature fusion neck network, and a segmentation head. The backbone employs a CSP dark network structure, with the core component being the C2f module for extracting detailed image features. The neck network uses a PAN+FPN structure. PAN+FPN is an optimization and improvement of the classic feature pyramid FPN. It fuses feature maps at levels P3, P4, and P5 through upsampling and downsampling.

[0022] The difference between YOLOv8-seg and the YOLOv8 architecture lies in the subdivision. In addition to the detection branch responsible for predicting object boundaries and categories, YOLOv8-seg also generates a set of prototype masks and corresponding mask coefficients from the P3 features. After cropping and thresholding, the final pixel-level segmentation mask is obtained.

[0023] Mask R-CNN introduces a RoI Align module based on Faster R-CNN to address the pixel-level alignment problem. The backbone network (ResNet FPN) is responsible for extracting multi-scale feature maps. The Region Proposal Network (RPN) generates candidate regions that may contain objects. RoI Align ensures precise alignment between the extracted features and input pixels, effectively improving the accuracy of mask segmentation. The detection branch of the decoupled head outputs classification and bounding boxes. Furthermore, a segmentation branch parallel to the detection branch is added to output the target mask.

[0024] The palm keypoint discrimination algorithm uses AI neural networks (e.g., YOLOv8-pose and MediaPipeHands) to estimate the position of palm keypoints; the AI ​​neural network further calculates the length of specific keypoints (e.g., ...). Figure 4The length indicators of key points on the palm include the middle finger length indicator LM and the palm length indicator LH.

[0025] The length indicator for the middle finger of a hand mold (LM): the length from key point 9 to 12;

[0026] Hand mold palm length index LH: the length from key point 0 to 12;

[0027] YOLOv8-pose and YOLOv8-seg share the core backbone and neck network. The main difference between the two lies in the final decoupled head output design. The pose branch of YOLOv8-pose calculates the coordinates of hand keypoints through regression.

[0028] The MediaPipe Hands model employs a typical two-stage process: First, a lightweight detector locates the hand, then a keypoint model regresses 21 keypoints to the hand region. The hand detector provides the bounding box and rotation angle of the hand. Additionally, the hand landmark model outputs the 3D coordinates of the 21 keypoints. To conserve computational power, MediaPipe Hands only executes the detector in the first frame or when the target is missing. At other times, keypoints from the previous frame are used to calculate the current RoI, enabling real-time detection.

[0029] The hand mold size discrimination algorithm includes a method for setting the hand mold size discrimination threshold, which includes a training phase and a validation phase.

[0030] The training phase is as follows: First, 200 un-immersion images of each size of hand mold are collected. Based on the hand mold size discrimination algorithm, the average and standard deviation of four discrimination indicators (mask area, maximum mask width, LM index, and LH index) are calculated for each size. The range of the average ± 3 standard deviations is then used as the upper and lower bounds of the candidate discrimination thresholds for hand mold size. Next, candidate discrimination thresholds with overlapping ranges are checked and eliminated. Finally, the candidate discrimination threshold with the smallest standard deviation is selected as the upper and lower bounds of the hand mold size discrimination indicator thresholds.

[0031] The verification phase specifically involves testing the upper and lower bounds of the hand mold size discrimination threshold using 100 images of the hand mold before immersion for each size, and verifying whether the hand mold size discrimination threshold achieves a 100% accuracy rate.

[0032] The controller is a Raspberry Pi controller with a built-in Raspberry Pi, and the impregnation angle servo control module includes a PLC controller and an impregnation angle servo control mechanism.

[0033] The PLC controller receives the hand mold identification dimensions transmitted by the Raspberry Pi and controls the dip angle servo control mechanism to adjust the dip angle of the hand mold according to the hand mold dimensions. The PLC controller also sends the dip processing parameters of each hand mold back to the Raspberry Pi controller.

[0034] The dipped angle servo control mechanism is used to correct the cumulative error caused by the varying lengths of each chain link in the chain drive mechanism and the detection error of the laser sensor. The dipped angle servo control mechanism includes a left and right positioning mechanism (2) and a vertical positioning mechanism (3) connected to the laser sensor (1), such as Figure 5 .

[0035] The left and right positioning mechanism is formed by installing the existing fixed laser sensor on the positioning mechanism that can be adjusted left and right. It is used to adjust the horizontal position of the end point of the hand mold conveyed by the chain drive mechanism to the impregnation tank. The left and right positioning mechanism is mainly composed of a cylinder (21), an error adjustment wheel (22) and a left and right adjustment spring slider (23). When the hand mold is conveyed to the impregnation station according to the chain drive mode of the chain drive mechanism, the stopping position of the chain roller (24) of the chain drive mechanism is affected by the cumulative error of the different lengths of each chain link, and there will be a variation range of leading and lagging. The left and right positioning mechanism is driven by the cylinder at the top to move the error adjustment wheel downward. When the error adjustment wheel moves downward, it will naturally contact the outer edge of the two sets of chain rollers (24) of the chain drive mechanism below, and move downward along the outer edge of the chain rollers to ensure that each impregnation can stop at the midpoint of the two sets of chain rollers. The error adjustment wheel is used to automatically compensate for the leading and lagging error of the chain rollers to ensure that the final horizontal position of each hand mold is consistent with the horizontal position of the laser sensor.

[0036] When the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide the reset thrust for the left and right positioning mechanism, such as... Figure 6 As shown;

[0037] The vertical positioning mechanism is used to adjust the vertical position of the laser sensor. The vertical positioning mechanism mainly consists of a stepper motor (31), a linear slide rail (32), and a laser sensor bracket (33) for fixing the laser sensor. The PLC controller controls the stepper motor and linear slide rail according to the size of the hand mold, automatically adapting the height of the laser sensor to the size of the dipped hand mold, achieving the optimal dipped limit angle measurement position for the laser sensor. Figure 7 As shown.

[0038] In the glove dipping process, the positioning device first uses an AI discrimination algorithm and discrimination threshold parameters to automatically identify the glove mold number and transmits the mold size to the servo control module PLC for controlling the dipping angle. Specifically, the servo control module PLC controls the stepper motor and linear guide rail of the vertical positioning mechanism based on the glove mold size, automatically adapting to the height of the laser sensor for the dipping glove mold size to achieve the optimal dipping limit angle measurement position. Figure 11 As shown, Figure 11 The upper area 1 is a schematic diagram of the height of the laser sensor for a small hand mold. Figure 11 The diagram in section 2 illustrates the height of a fixed laser sensor. When the size of the hand mold increases, using a fixed laser sensor height will result in insufficient rotation angle during clockwise rotation, causing errors in the mold's adhesive dipping angle. Figure 11 The lower three areas are schematic diagrams of the best embodiment. After the AI ​​discrimination algorithm identifies the large-sized hand mold, it controls the stepper motor of the vertical positioning mechanism to raise the height of the laser sensor, so that the laser spot rises to achieve the optimal dip-resin limit angle measurement position.

[0039] In the glove dipping process, the positioning device automatically adjusts the cumulative clearance error of the transmission mechanism through the dipping angle servo control module to achieve the optimal dipping angle. Specifically, if the error in the current work gloves dipping process includes the error caused by the cumulative clearance of the transmission mechanism, the left and right positioning mechanisms of the dipping angle servo control module can automatically adjust the cumulative clearance error of the transmission mechanism. The steps are as follows: When the hand mold is conveyed to the dipping station according to the chain drive mode, the left and right positioning mechanisms push the error adjustment wheel downward through the cylinder above it. When the error adjustment wheel moves downward, it will naturally contact the outer edge of the two sets of chain rollers below and move downward along the outer edge of the chain rollers, ensuring that each hand mold can stop at the midpoint of the two sets of chain rollers during each dipping process; the error adjustment wheel automatically compensates for the cumulative clearance error of the chain rollers being ahead or behind, ensuring the consistency between the final position of each hand mold and the position of the laser sensor; when the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide the reset thrust of the left and right positioning mechanisms. Figure 12 As shown. Figure 12 The upper region is a schematic diagram of the error caused by the cumulative backlash of the transmission mechanism. Figure 12 The lower region is a schematic diagram of the error caused by the cumulative backlash of the transmission mechanism.

[0040] In the prior art, the dip angle of the hand mold is fixed and cannot be precisely adjusted according to the size of the hand mold; or it requires manual setting of adjustment components to adjust the dip angle of the glove template. However, the present invention improves accuracy and production efficiency by using image-based automatic recognition of hand mold size, and the PLC can accurately adjust the dip angle of the hand mold according to the automatically recognized hand mold size.

[0041] This invention proposes an image-based positioning device for dipped work gloves. This method combines AI machine vision recognition technology with a servo control mechanism for the dipped angle of the work gloves, achieving a fast and efficient positioning device. This method can promptly determine the size of the hand mold for dipped work gloves and automatically adjust the dipped angle based on the corresponding hand mold size to achieve the best dipping and molding effect.

[0042] This invention is portable and easy to operate, and can be adapted to different work gloves dipping production line equipment.

[0043] The advantages of this invention also include:

[0044] 1. Compared with existing fixed single dip angle technology, the present invention uses AI image recognition technology, which can automatically identify the size of the hand mold and adjust the optimal dip angle.

[0045] 2. The dip angle servo mechanism designed in this invention can automatically adjust the dip angle according to the size of the hand mold, without requiring manual setting.

[0046] 3. The cloud platform module of this invention can record all impregnation processing parameters to ensure processing quality.

[0047] 4. By designing a left and right positioning mechanism, this invention ensures the consistency of the limit angle measurement of each hand mold by the laser sensor, thus ensuring the accuracy of angle positioning. Attached Figure Description

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0049] Appendix Figure 1 This is a schematic diagram of the dip-coating process for work gloves in an embodiment of the present invention (in the figure, area 1 shows: the production line uses a chain drive mode to transport the hand mold to the top of the dip-coating tank; area 2 shows: each hand mold first rotates counterclockwise 180° and then slowly rotates clockwise to the laser detection limit angle; area 3 shows: when the laser spot detects the limit position of the hand mold, the slow clockwise rotation stops; area 4 shows: the dip-coating tank rises, completing the dip-coating of the hand mold).

[0050] Appendix Figure 2 This is a schematic diagram of the error in the positioning of the dipped work gloves in the traditional technical solution (the left area shows the error caused by the cumulative clearance of the transmission mechanism, and the right area shows the error caused by the difference in the hand model number).

[0051] Appendix Figure 3 This is a schematic diagram of the architecture of the image-based work gloves dip-coating positioning system in an embodiment of the present invention;

[0052] Appendix Figure 4This is a schematic diagram of the palm key point location based on the palm key point discrimination algorithm in an embodiment of the present invention (the numbers in the figure are palm key point numbers).

[0053] Appendix Figure 5 This is a schematic diagram of the impregnation angle servo control mechanism in an embodiment of the present invention;

[0054] Appendix Figure 6 This is a schematic diagram of the left and right positioning mechanism of the dip-in angle servo control mechanism in an embodiment of the present invention;

[0055] Appendix Figure 7 This is a schematic diagram of the vertical positioning mechanism of the dip-in angle servo control mechanism in an embodiment of the present invention;

[0056] Appendix Figure 8 This is a schematic diagram of the recognition results of the Mask R-CNN algorithm in an embodiment of the present invention;

[0057] Appendix Figure 9 This is a schematic diagram of the recognition results of the MediaPipe Hands algorithm in an embodiment of the present invention;

[0058] Appendix Figure 10 This is a schematic diagram of the recognition results of the YOLOv8-pose algorithm in an embodiment of the present invention;

[0059] Appendix Figure 11 This is a schematic diagram of how the AI ​​algorithm, in conjunction with the servo mechanism, adjusts the hand mold to the optimal impregnation angle in an embodiment of the present invention.

[0060] Appendix Figure 12 This is a schematic diagram of the automatic adjustment of the cumulative clearance error of the transmission mechanism by the dip angle servo control module in an embodiment of the present invention;

[0061] In the figure: laser sensor (1), left and right positioning mechanism (2), vertical positioning mechanism (3), cylinder (21), error adjustment wheel (22), left and right adjustment spring slider (23), chain roller (24), stepper motor (31), linear slide rail (32), laser sensor bracket (33). Detailed Implementation

[0062] As shown in the figure, an image-based positioning device for dipped gloves includes an image recognition module for detecting hand molds located at the chain drive mechanism of the dipping production line, and a dipping angle servo control module for adjusting the position of a laser emitter. When the glove is dipped at the hand mold fitted onto the mold bar of the chain drive mechanism, the mold bar drives the hand mold to rotate vertically around the wrist so that the hand mold's feature position is rotated to the laser detection limit point, allowing the production line to determine that the hand mold's posture is the dipping posture.

[0063] The laser detection limit point is formed by the laser spot emitted by the laser sensor. The positioning device moves the position of the laser emitting device according to the hand mold recognition result of the image recognition module through the glue dipping angle servo control module to compensate for the hand mold specification deviation, and compensates for the chain sprocket error through the glue dipping angle servo control module, so that the glue dipping posture of the hand mold tends to be consistent when the hand mold turns to the laser detection limit point.

[0064] A dipping tank, which is a nitrile rubber tank, is provided at the end of the chain drive mechanism. During dipping, multiple gloved hand molds are fixed in the same posture at the mold bar of the chain drive mechanism to form each mold bar. The chain drive mechanism first sends each mold bar to the top of the dipping tank in chain drive mode. Then, the mold bar rotates counterclockwise vertically by a preset angle of 180 degrees. Then, the mold bar slowly rotates clockwise vertically to the laser detection limit point, so that the laser spot of the laser emitting device falls on the preset position of the hand mold. After the production line detects the laser spot at the hand mold, it determines that the hand mold is in the dipping posture. The dipping tank rises so that the hand mold enters the dipping tank in the dipping posture and is immersed for a preset time of 4-8 seconds to ensure that the nitrile rubber latex is evenly adhered to the surface of the hand mold.

[0065] The image recognition module includes a webcam and a Raspberry Pi controller with a built-in Raspberry Pi.

[0066] A webcam is installed in front of the nitrile tank and connected to the Raspberry Pi controller. The webcam is connected to the Raspberry Pi controller via a network cable. The webcam takes images of each hand mold before impregnation according to the impregnation production line and transmits each hand mold image to the Raspberry Pi controller.

[0067] The Raspberry Pi controller first uses AI machine vision technology to automatically identify the size of the hand mold based on the image of the hand mold before impregnation captured by the camera. Then, it sends the identified hand mold size to the PLC controller of the impregnation angle servo control module.

[0068] The dip angle servo control module includes a PLC controller and a dip angle servo control mechanism. The PLC controller adjusts the dip angle servo mechanism to adjust the position of the laser spot according to the hand mold size transmitted by the image recognition module, so that when the hand mold rotates vertically to reach the laser spot, the posture of each hand mold reaches the optimal dip angle.

[0069] The positioning device also includes a cloud platform module; after each hand mold is dipped in glue, the PLC controller transmits the processing parameters of each hand mold back to the cloud platform module via the Raspberry Pi controller; the cloud platform module records the processing parameters of the hand mold in the glue dipping process (production date, time, work gloves model, glue dipping angle, glue dipping time).

[0070] The controller is a Raspberry Pi controller with a built-in Raspberry Pi, which uses a hand mold size discrimination algorithm to identify the hand mold size. The hand mold size discrimination algorithm includes a hand mold area discrimination algorithm and a palm key point discrimination algorithm.

[0071] The hand mold area discrimination algorithm uses an AI neural network (e.g., Yolov8-seg and Mask R-CNN) to estimate the mask region of the hand mold; the AI ​​neural network further calculates the mask area index and the maximum mask width index as discrimination indicators to determine the size of the hand mold;

[0072] YOLOv8-seg consists of a backbone, a feature fusion neck network, and a segmentation head. The backbone employs a CSP dark network structure, with the core component being the C2f module for extracting detailed image features. The neck network uses a PAN+FPN structure. PAN+FPN is an optimization and improvement of the classic feature pyramid FPN. It fuses feature maps at levels P3, P4, and P5 through upsampling and downsampling.

[0073] The difference between YOLOv8-seg and the YOLOv8 architecture lies in the subdivision. In addition to the detection branch responsible for predicting object boundaries and categories, YOLOv8-seg also generates a set of prototype masks and corresponding mask coefficients from the P3 features. After cropping and thresholding, the final pixel-level segmentation mask is obtained.

[0074] Mask R-CNN introduces a RoI Align module based on Faster R-CNN to address the pixel-level alignment problem. The backbone network (ResNet FPN) is responsible for extracting multi-scale feature maps. The Region Proposal Network (RPN) generates candidate regions that may contain objects. RoI Align ensures precise alignment between the extracted features and input pixels, effectively improving the accuracy of mask segmentation. The detection branch of the decoupled head outputs classification and bounding boxes. Furthermore, a segmentation branch parallel to the detection branch is added to output the target mask.

[0075] The palm keypoint discrimination algorithm uses AI neural networks (e.g., YOLOv8-pose and MediaPipeHands) to estimate the position of palm keypoints; the AI ​​neural network further calculates the length of specific keypoints (e.g., ...). Figure 4 The length indicators of key points on the palm include the middle finger length indicator LM and the palm length indicator LH.

[0076] The length indicator for the middle finger of a hand mold (LM): the length from key point 9 to 12;

[0077] Hand mold palm length index LH: the length from key point 0 to 12;

[0078] YOLOv8-pose and YOLOv8-seg share the core backbone and neck network. The main difference between the two lies in the final decoupled head output design. The pose branch of YOLOv8-pose calculates the coordinates of hand keypoints through regression.

[0079] The MediaPipe Hands model employs a typical two-stage process: First, a lightweight detector locates the hand, then a keypoint model regresses 21 keypoints to the hand region. The hand detector provides the bounding box and rotation angle of the hand. Additionally, the hand landmark model outputs the 3D coordinates of the 21 keypoints. To conserve computational power, MediaPipe Hands only executes the detector in the first frame or when the target is missing. At other times, keypoints from the previous frame are used to calculate the current RoI, enabling real-time detection.

[0080] The hand mold size discrimination algorithm includes a method for setting the hand mold size discrimination threshold, which includes a training phase and a validation phase.

[0081] The training phase is as follows: First, 200 un-immersion images of each size of hand mold are collected. Based on the hand mold size discrimination algorithm, the average and standard deviation of four discrimination indicators (mask area, maximum mask width, LM index, and LH index) are calculated for each size. The range of the average ± 3 standard deviations is then used as the upper and lower bounds of the candidate discrimination thresholds for hand mold size. Next, candidate discrimination thresholds with overlapping ranges are checked and eliminated. Finally, the candidate discrimination threshold with the smallest standard deviation is selected as the upper and lower bounds of the hand mold size discrimination indicator thresholds.

[0082] The verification phase specifically involves testing the upper and lower bounds of the hand mold size discrimination threshold using 100 images of the hand mold before immersion for each size, and verifying whether the hand mold size discrimination threshold achieves a 100% accuracy rate.

[0083] The controller is a Raspberry Pi controller with a built-in Raspberry Pi, and the impregnation angle servo control module includes a PLC controller and an impregnation angle servo control mechanism.

[0084] The PLC controller receives the hand mold identification dimensions transmitted by the Raspberry Pi and controls the dip angle servo control mechanism to adjust the dip angle of the hand mold according to the hand mold dimensions. The PLC controller also sends the dip processing parameters of each hand mold back to the Raspberry Pi controller.

[0085] The dip-coating angle servo control mechanism is used to correct the cumulative error caused by the varying lengths of each chain link in the chain drive mechanism and the detection error of the laser sensor. The dip-coating angle servo control mechanism includes a left-right positioning mechanism 2 and a vertical positioning mechanism 3 connected to the laser sensor 1. Figure 5 .

[0086] The left and right positioning mechanism is formed by mounting the existing fixed laser sensor on a positioning mechanism that can be adjusted left and right. It is used to adjust the horizontal position of the end point of the hand mold conveyed to the impregnation tank by the chain drive mechanism. The left and right positioning mechanism is mainly composed of a cylinder 21, an error adjustment wheel 22, and a left and right adjustment spring slider 23. When the hand mold is conveyed to the impregnation station according to the chain drive mode of the chain drive mechanism, the stopping position of the chain roller 24 of the chain drive mechanism is affected by the cumulative error of the different lengths of each chain link, and there will be a variation range of leading and lagging. The left and right positioning mechanism is driven by the cylinder at the top to move the error adjustment wheel downward. When the error adjustment wheel moves downward, it will naturally contact the outer edge of the two sets of chain rollers 24 of the chain drive mechanism below, and move downward along the outer edge of the chain rollers to ensure that each impregnation stops at the midpoint of the two sets of chain rollers. The error adjustment wheel is used to automatically compensate for the leading and lagging errors of the chain rollers to ensure that the final horizontal position of each hand mold is consistent with the horizontal position of the laser sensor.

[0087] When the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide the reset thrust for the left and right positioning mechanism, such as... Figure 6 As shown;

[0088] The vertical positioning mechanism is used to adjust the vertical position of the laser sensor. The vertical positioning mechanism mainly consists of a stepper motor 31, a linear slide rail 32, and a laser sensor bracket 33 for fixing the laser sensor. The PLC controller, based on the size of the hand mold, controls the stepper motor and linear slide rail to automatically adapt to the height of the laser sensor of the dipped hand mold, achieving the optimal dipped limit angle measurement position for the laser sensor. Figure 7 As shown.

[0089] In the glove dipping process, the positioning device first uses an AI discrimination algorithm and discrimination threshold parameters to automatically identify the glove mold number and transmits the mold size to the servo control module PLC for controlling the dipping angle. Specifically, the servo control module PLC controls the stepper motor and linear guide rail of the vertical positioning mechanism based on the glove mold size, automatically adapting to the height of the laser sensor for the dipping glove mold size to achieve the optimal dipping limit angle measurement position. Figure 11 As shown, Figure 11 The upper area 1 is a schematic diagram of the height of the laser sensor for a small hand mold. Figure 11 The diagram in section 2 illustrates the height of a fixed laser sensor. When the size of the hand mold increases, using a fixed laser sensor height will result in insufficient rotation angle during clockwise rotation, causing errors in the mold's adhesive dipping angle. Figure 11The lower three areas are schematic diagrams of the best embodiment. After the AI ​​discrimination algorithm identifies the large-sized hand mold, it controls the stepper motor of the vertical positioning mechanism to raise the height of the laser sensor, so that the laser spot rises to achieve the optimal dip-resin limit angle measurement position.

[0090] In the glove dipping process, the positioning device automatically adjusts the cumulative clearance error of the transmission mechanism through the dipping angle servo control module to achieve the optimal dipping angle. Specifically, if the error in the current work gloves dipping process includes the error caused by the cumulative clearance of the transmission mechanism, the left and right positioning mechanisms of the dipping angle servo control module can automatically adjust the cumulative clearance error of the transmission mechanism. The steps are as follows: When the hand mold is conveyed to the dipping station according to the chain drive mode, the left and right positioning mechanisms push the error adjustment wheel downward through the cylinder above it. When the error adjustment wheel moves downward, it will naturally contact the outer edge of the two sets of chain rollers below and move downward along the outer edge of the chain rollers, ensuring that each hand mold can stop at the midpoint of the two sets of chain rollers during each dipping process; the error adjustment wheel automatically compensates for the cumulative clearance error of the chain rollers being ahead or behind, ensuring the consistency between the final position of each hand mold and the position of the laser sensor; when the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide the reset thrust of the left and right positioning mechanisms. Figure 12 As shown. Figure 12 The upper region is a schematic diagram of the error caused by the cumulative backlash of the transmission mechanism. Figure 12 The lower region is a schematic diagram of the error caused by the cumulative backlash of the transmission mechanism.

[0091] Example:

[0092] Example 1 includes: 1. The algorithm for determining hand mold size and the calculation process for the determination threshold. 2. A verification experiment of the hand mold size determination threshold.

[0093] 1. Hand mold size discrimination algorithm: The following two AI discrimination algorithms are used: one based on the area of ​​the hand mold and the other based on key points of the palm.

[0094] (1) Based on the hand mold area discrimination algorithm and discrimination threshold calculation process:

[0095] The image recognition module uses Yolov8-seg and Mask R-CNN algorithms to estimate the sizes of three different hand molds (M, L, XL). The image recognition module automatically calculates the mask area and maximum mask width of the hand molds.

[0096] The recognition results of the Mask R-CNN algorithm in the example are as follows: Figure 8 As shown, the image recognition module correctly estimated the area of ​​the hand mold and successfully identified three sizes of hand molds (M size: 33070 px2; L size: 37548 px2; XL size: 38296 px2).

[0097] The average and standard deviation of the hand mold discrimination index for each size model during the statistical training phase are shown in Table 1.

[0098]

[0099] (2) Hand key point discrimination algorithm and discrimination threshold calculation process:

[0100] The image recognition module uses YOLOv8-pose and MediaPipe Hands algorithms to estimate the sizes of three different hand molds (M, L, XL). Based on the automatically identified key points of the hand mold, the image recognition module automatically calculates the LM length and LH length indices for specific key points.

[0101] The recognition results of the MediaPipe Hands algorithm in the example are as follows: Figure 9 As shown, the image recognition module correctly estimated the length of the key points of the hand mold and successfully identified three sizes of hand molds (M size: 307.4 px; L size: 327.1 px; XL size: 333.1 px).

[0102] The average and standard deviation of the hand mold discrimination index for each size and model during the statistical training phase are shown in Table 2.

[0103]

[0104] (3) Calculation process of the discrimination threshold in the example:

[0105] This method first uses the range of the mean ± 3 standard deviations as the candidate discrimination threshold for hand mold size. Then, it excludes candidate discrimination thresholds that have overlapping ranges among different sizes.

[0106] Because the LM index of the YOLOv8-pose algorithm in the embodiment has the smallest standard deviation and there are no overlapping intervals in the discrimination thresholds of each size model, the LM index can be used as the hand mold size discrimination index in the embodiment.

[0107] 2. Verification experiment on the threshold for judging hand mold size:

[0108] Based on Table 2 of the embodiments, the range of the average value of the LM index ± 3 standard deviations is used as the upper and lower bounds of the discrimination threshold. A verification experiment was conducted using 100 hand mold images of each size before immersion, and the results are shown in Table 3.

[0109]

[0110] The recognition results of the YOLOv8-pose algorithm in the example are as follows: Figure 10As shown, the image recognition module correctly estimated the length of the key points of the hand mold and successfully identified hand molds of three sizes (M size: 112.7 px; L size: 120.7 px; XL size: 124 px).

[0111] Example 2: This method automatically adjusts the position of the laser sensor to achieve the optimal impregnation angle.

[0112] Current pain points in the glove impregnation process include impregnation errors caused by differences in hand mold sizes. Currently, manual identification of the hand mold size and manual adjustment of the laser sensor position are required to achieve the optimal impregnation angle.

[0113] This method eliminates the need for manual setting of the hand mold size and model, automatically adjusting the laser sensor position to achieve the optimal impregnation angle. The AI ​​algorithm, in conjunction with the servo mechanism, adjusts the impregnation angle in the following steps:

[0114] 1. This method uses the AI ​​discrimination algorithm and discrimination threshold parameters of Example 1 to automatically identify the hand mold number and transmit the hand mold size to the impregnation angle servo control module PLC.

[0115] 2. The servo control module PLC controls the stepper motor and linear guide rail of the vertical positioning mechanism according to the size and model of the hand mold, automatically adapting to the height of the laser sensor of the dipped hand mold size to achieve the optimal dipped limit angle measurement position (e.g., Figure 11 ). Figure 11-1 (Above) is a schematic diagram of the height of the laser sensor for a small hand mold. Figure 11-2 (middle) is a schematic diagram of the height of a fixed laser sensor. When the size of the hand mold increases, if a fixed laser sensor is used, insufficient rotation angle will occur during clockwise rotation, resulting in an error in the mold's adhesive dipping angle. Figure 11-3 (Below) is a schematic diagram of the preferred embodiment. After the AI ​​discrimination algorithm identifies the large-sized hand mold, it controls the stepper motor of the vertical positioning mechanism to raise the height of the laser sensor to achieve the optimal dip-resin limit angle measurement position.

[0116] Example 3: The dip angle servo control module of this method automatically adjusts the cumulative clearance error of the transmission mechanism to achieve the optimal dip angle.

[0117] The current pain point in the dip-coating process of work gloves includes errors caused by the accumulated clearance of the transmission mechanism. This method utilizes a dip-coating angle servo control module with a left and right positioning mechanism that can automatically adjust for the accumulated clearance error of the transmission mechanism. The implementation steps are as follows:

[0118] When the hand mold is conveyed to the impregnation station via chain drive, the left and right positioning mechanism of this method is driven downwards by the upper cylinder to move the error adjustment wheel. As the error adjustment wheel moves downwards, it naturally contacts the outer edges of the two sets of chain rollers below and moves downwards along the outer edges of the chain rollers, ensuring that each impregnation stops at the midpoint of the two sets of chain rollers. The error adjustment wheel automatically compensates for the cumulative gap error of the chain rollers, ensuring the consistency between the final position of each hand mold and the position of the laser sensor. When the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide a reset thrust for the left and right positioning mechanism (e.g., ...). Figure 12 ). Figure 12 (Above) is a schematic diagram of the error caused by the lack of accumulated clearance in the transmission mechanism. Figure 12 (Below) is a schematic diagram of the error caused by the cumulative backlash in the transmission mechanism.

Claims

1. A video-based positioning device for dipping and positioning work gloves, characterized in that: The system includes an image recognition module for detecting hand molds located at the chain drive mechanism of the glue-dipping production line, and a glue-dipping angle servo control module for adjusting the position of the laser emitting device. When the glove fitted onto the hand mold is being glued, the hand mold rotates vertically around the wrist to bring the hand mold feature position to the laser detection limit point, allowing the production line to determine that the hand mold posture is the glue-dipping posture. The laser detection limit point is formed by the laser spot emitted by the laser sensor. The positioning device moves the position of the laser emitting device according to the hand mold recognition result of the image recognition module through the glue dipping angle servo control module to compensate for the hand mold specification deviation, and compensates for the chain sprocket error through the glue dipping angle servo control module, so that the glue dipping posture of the hand mold tends to be consistent when the hand mold turns to the laser detection limit point.

2. The imaging-based positioning device for dipped and positioned work gloves according to claim 1, characterized in that: A dipping tank, which is a nitrile rubber tank, is provided at the end of the chain drive mechanism. During dipping, multiple gloved hand molds are fixed in the same posture at the mold bar of the chain drive mechanism. The chain drive mechanism first sends each mold to the top of the dipping tank in chain drive mode. Then, the mold bar rotates counterclockwise vertically by a preset angle, and then slowly rotates clockwise vertically to the laser detection limit point, so that the laser spot of the laser emitting device falls on the preset position of the hand mold. After the production line detects the laser spot at the hand mold, it determines that the hand mold is in the dipping posture. The dipping tank rises so that the hand mold enters the dipping tank in the dipping posture and is immersed for a preset time to ensure that the nitrile rubber latex is evenly adhered to the surface of the hand mold.

3. The imaging-based positioning device for dipped and positioned work gloves according to claim 2, characterized in that: The image recognition module includes a webcam and a controller; A webcam is installed in front of the nitrile tank and connected to the Raspberry Pi controller; the webcam takes images of each hand mold before immersion according to the impregnation production line and transmits each hand mold image to the controller; The controller first uses AI machine vision technology to automatically identify the size of the hand mold based on the image of the hand mold before impregnation captured by the camera, and then sends the identified hand mold size to the PLC controller of the impregnation angle servo control module. The dip angle servo control module includes a PLC controller and a dip angle servo control mechanism. The PLC controller adjusts the dip angle servo mechanism to adjust the position of the laser spot according to the hand mold size transmitted by the image recognition module, so that when the hand mold rotates vertically to reach the laser spot, the posture of each hand mold reaches the optimal dip angle.

4. The imaging-based positioning device for dipped and positioned work gloves according to claim 3, characterized in that: The positioning device also includes a cloud platform module; after each hand mold is dipped in glue, the PLC controller transmits the processing parameters of each hand mold back to the cloud platform module via the Raspberry Pi controller; the cloud platform module records the processing parameters of the hand mold in the glue dipping process.

5. The imaging-based positioning device for dipped and positioned work gloves according to claim 3, characterized in that: The controller uses a hand mold size discrimination algorithm to identify the hand mold size. The hand mold size discrimination algorithm includes a hand mold area discrimination algorithm and a palm key point discrimination algorithm. The hand mold area discrimination algorithm uses an AI neural network to estimate the mask area of ​​the hand mold; the AI ​​neural network further calculates the mask area index and the maximum mask width index as discrimination indicators to determine the size of the hand mold. The palm key point discrimination algorithm uses an AI neural network to estimate the position of palm key points; the AI ​​neural network further calculates the length of specific key points; the length indicators of palm key points include the middle finger length indicator LM and the palm length indicator LH.

6. The imaging-based positioning device for dipped and positioned work gloves according to claim 5, characterized in that: The hand mold size discrimination algorithm includes a method for setting the hand mold size discrimination threshold, which includes a training phase and a validation phase. The training phase is as follows: First, during the training phase, images of hand molds of various sizes before immersion are collected. Based on the hand mold size discrimination algorithm, the average and standard deviation of the mask area index, maximum mask width index, LM index, and LH index are calculated for each size model. Then, the range of the average value ± 3 standard deviations is used as the upper and lower bounds of the candidate discrimination thresholds for hand mold size. Next, candidate discrimination thresholds with overlapping ranges are checked and eliminated. Finally, the candidate discrimination threshold with the smallest standard deviation is selected as the upper and lower bounds of the thresholds for the hand mold size discrimination index. The verification phase specifically involves testing the upper and lower bounds of the hand mold size discrimination threshold using multiple unimmersed hand mold images of different sizes, and verifying whether the hand mold size discrimination threshold achieves a 100% accuracy rate.

7. The imaging-based positioning device for dipped and positioned work gloves according to claim 3, characterized in that: The controller is a Raspberry Pi controller with a built-in Raspberry Pi, and the impregnation angle servo control module includes a PLC controller and an impregnation angle servo control mechanism. The PLC controller receives the hand mold identification dimensions transmitted by the Raspberry Pi and controls the dip angle servo control mechanism to adjust the dip angle of the hand mold according to the hand mold dimensions. The PLC controller also sends the dip processing parameters of each hand mold back to the Raspberry Pi controller. The dipped angle servo control mechanism is used to correct the cumulative error caused by the different lengths of each chain link in the chain drive mechanism and the detection error of the laser sensor. The dipped angle servo control mechanism includes a left and right positioning mechanism (2) and a vertical positioning mechanism (3) connected to the laser sensor (1).

8. The imaging-based positioning device for dipped and positioned work gloves according to claim 7, characterized in that: The left and right positioning mechanism is formed by mounting the laser sensor on the positioning mechanism that can be adjusted left and right. It is used to adjust the horizontal position of the end point of the chain drive mechanism conveying the hand mold to the impregnation tank. The left and right positioning mechanism consists of a cylinder (21), an error adjustment wheel (22), and a left and right adjustment spring slider (23). The vertical positioning mechanism is used to adjust the vertical position of the laser sensor; the vertical positioning mechanism consists of a stepper motor (31), a linear slide rail (32), and a laser sensor bracket (33) for fixing the laser sensor. The PLC controller controls the stepper motor and the linear slide rail according to the size of the hand mold, automatically adapting the height of the laser sensor to the size of the dipped hand mold, so as to achieve the optimal dipped limit angle measurement position of the laser sensor.

9. The imaging-based positioning device for dipped and positioned work gloves according to claim 8, characterized in that: In the glove impregnation process, the positioning device first uses an AI discrimination algorithm and discrimination threshold parameters to automatically identify the glove mold number and transmits the glove mold size to the servo control module PLC for controlling the impregnation angle. Specifically, the servo control module PLC controls the stepper motor and linear slide rail of the vertical positioning mechanism according to the glove mold size and model, automatically adapting the height of the laser sensor to the size of the impregnated glove mold to achieve the optimal impregnation limit angle measurement position. When the glove mold size increases, the AI ​​discrimination algorithm identifies the large-sized glove mold and controls the stepper motor of the vertical positioning mechanism to raise the height of the laser sensor, thereby increasing the laser spot height to achieve the optimal impregnation limit angle measurement position.

10. The imaging-based positioning device for dipped and positioned work gloves according to claim 8, characterized in that: In the glove dipping process, the positioning device automatically adjusts the cumulative clearance error of the transmission mechanism through the dipping angle servo control module to achieve the optimal dipping angle. Specifically, if the error in the current work gloves dipping process includes the error caused by the cumulative clearance of the transmission mechanism, the left and right positioning mechanisms of the dipping angle servo control module can automatically adjust the cumulative clearance error of the transmission mechanism. The steps are as follows: When the hand mold is conveyed to the dipping station according to the chain drive mode, the left and right positioning mechanisms push the error adjustment wheel downward through the cylinder above it. When the error adjustment wheel moves downward, it will naturally contact the outer edge of the two sets of chain rollers below and move downward along the outer edge of the chain rollers, ensuring that each hand mold can stop at the midpoint of the two sets of chain rollers during each dipping process. The error adjustment wheel automatically compensates for the cumulative clearance error of the chain rollers being ahead or behind, ensuring the consistency between the final position of each hand mold and the position of the laser sensor. When the cylinder piston rises and returns, the adjusting springs installed at both ends of the left and right adjusting sliders provide the reset thrust of the left and right positioning mechanisms.

Images