Intelligent license plate manufacturing machine control system

Abstract

The application relates to the technical field of instruction control, in particular to a license plate intelligent manufacturing machine control system, and the following scheme is provided: after a machine arm acquires manufacturing instructions and a process formula, the machine arm completes font assembly, press forming and heat transfer printing processing, and a narrow camera is integrated at the end to realize synchronous quality detection. A rasterized scanning path aligned with a license plate version is generated, the machine arm is controlled to collect multiple local images in movement, the images are projected to a license plate plane coordinate system in combination with end posture and camera calibration information, splicing and fine adjustment are carried out based on character anchor point constraints, a complete license plate image is obtained, and quality determination is carried out. The application does not require an additional detection station, avoids secondary grabbing and transmission, improves the integration and detection precision of the license plate manufacturing process, and is suitable for an automatic production environment of a compact license plate manufacturing cabin.

Description

Technical Field

[0001] This application relates to the field of instruction control technology, and in particular to a control system for an intelligent license plate making machine. Background Technology

[0002] Existing license plate production lines typically rely on manual handling and independent workstations, resulting in low efficiency in the production process and difficulty in improving production cycle time. Independently set-up visual inspection workstations require additional hardware and space, resulting in complex structures, large footprints, and difficulty in deploying in compact environments. Panoramic cameras are susceptible to the effects of reflective film and raised characters during imaging, leading to distortion and misjudgments in the inspection results, and unstable quality control. Furthermore, traditional gripping methods often require secondary placement or re-attachment when license plate positioning accuracy is insufficient, extending the operation cycle, increasing the risk of film damage, and impacting license plate consistency and yield.

[0003] To address the above issues, this application presents a control system for an intelligent license plate production machine. Summary of the Invention

[0004] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing a smart license plate manufacturing machine control system. After acquiring manufacturing instructions and process formulas, the robotic arm completes character mold assembly, pressing, and heat transfer processing, and integrates a narrow-field camera at its end for simultaneous quality inspection. By generating a rasterized scanning path aligned with the license plate layout, the robotic arm is controlled to acquire multiple local images during movement. These images are then projected onto the license plate's planar coordinate system based on the end-effector posture and camera calibration information. Finally, based on character anchor point constraints, the images are stitched together and fine-tuned to obtain a complete license plate image, which is then used for quality assessment.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A smart license plate production machine control system is applied to a license plate production control device. The license plate production control device includes a controller and a correction mechanism. The correction mechanism is disposed at the gripping end of a robotic arm and is used to correct the position and posture of the license plate when the robotic arm grips the license plate. The robotic arm also includes a narrow-field camera, which is disposed at the end of the robotic arm. The system includes:

[0007] The instruction acquisition module is used to acquire manufacturing instructions and process formulas, and control the robotic arm to grab letter molds and license plate blanks that match the process formulas according to the manufacturing instructions.

[0008] The license plate manufacturing module is used to complete the pressing and heat transfer process under the control of the controller to obtain the heat-transferred license plate.

[0009] The license plate restoration module is used to acquire multiple local images during the movement of the robotic arm using the narrow-field camera, combine the posture parameters of the robotic arm end with the camera calibration information, project the local images onto the license plate plane coordinate system, and stitch them together according to the rasterized scanning path to obtain the license plate image.

[0010] The license plate detection module is used to determine the quality of the license plate image. If the image is qualified, the license plate is transferred to the license plate collection exit and a corresponding production number is generated and synchronized to the background. If the image is unqualified, the license plate is put into the defective product collection box and the unqualified information is recorded.

[0011] The card-making and processing module includes:

[0012] The letter mold assembly unit is used to control the robotic arm to pick up the letter mold and place it in the positioning position of the preset mold, pick up the license plate blank and feed it to the predetermined position directly above the lower letter mold, and then pick up the upper letter mold and place it directly above the license plate blank to form a mold assembly.

[0013] The pressing and forming unit is used to control the robotic arm to move the mold assembly to the pressing station, and to press and form the license plate blank according to the pressure and holding time parameters of the process formula to obtain a license plate semi-finished product.

[0014] The heat transfer unit is used to control the robotic arm to transfer the semi-finished license plate to the heat transfer station, and to perform heat transfer processing according to the temperature and transfer time parameters of the process formula to obtain the heat-transferred license plate.

[0015] The character mold returning unit is used to control the robotic arm to return the upper and lower character molds to the character mold slots and exit the card making station after the heat transfer process is completed.

[0016] The system also includes a correction module, which is configured with correction logic. This correction logic is used to correct the position and orientation of the license plate when the robotic arm grasps it, including:

[0017] After the license plate is attracted, negative pressure feedback is collected at the grasping end, wherein the negative pressure feedback includes:

[0018] The adjustment direction is determined based on the changing trend of the negative pressure feedback, and the grasping posture and planar position of the grasping end are corrected until the feedback signal stabilizes.

[0019] After the feedback signal stabilizes, the gripping end maintains its adsorption posture, and the license plate is moved to the reference edge of the mold. The correction mechanism, in conjunction with the controller's mode switching, enters the virtual fixture mode to obtain a compensation signal. The gripping end is then subjected to posture compensation based on the compensation signal.

[0020] By switching the mode of the controller in conjunction with the correction mechanism, the virtual fixture mode is entered, and a compensation signal is obtained, including:

[0021] While maintaining the license plate adsorption, the robotic arm is controlled to move the license plate along the normal direction to the reference edge of the mold, and the contact state is determined according to the change of joint torque.

[0022] When the contact state is determined, the normal advance of the robotic arm is stopped, and it is switched to sliding along the tangential direction of the reference edge, while maintaining the license plate adsorption state.

[0023] During the sliding process, the contact feedback signal is monitored. When the contact feedback signal is a symmetrical signal, it is determined that the license plate is aligned with the reference edge.

[0024] The robotic arm is controlled to slide to the positioning angle and perform an attitude correction operation to make the long side of the license plate parallel to the reference side and the short side perpendicular to the reference side, thereby obtaining the corresponding compensation signal.

[0025] The license plate restoration module includes:

[0026] The scanning path generation unit is used to generate a rasterized scanning path based on the planar range of the license plate;

[0027] An image acquisition unit is used to control the movement of the robotic arm along the rasterized scanning path and to acquire multiple local images using the equidistant distance of the end on the license plate plane as a trigger condition.

[0028] An initial image generation unit is used to combine the end-effector's pose and preset narrow-field camera calibration information to project the multiple local images onto the license plate plane coordinate system, and stitch them together according to the grid position to obtain an initial license plate image.

[0029] An image fine-tuning unit is used to extract character anchor points from the initial license plate image using an edge extraction algorithm, wherein the character anchor points include at least one of the character outer frame corner points, stroke intersection points, and character baselines; and to fine-tune the initial license plate image using the character anchor points as constraints to generate a license plate image.

[0030] The license plate restoration module is equipped with an image restoration strategy. The image restoration strategy reconstructs the entire license plate image based on multiple local images captured by the narrow-span camera and outputs a license plate image for quality judgment. The image restoration strategy includes path planning logic, image projection logic, and image fine-tuning logic. The path planning logic is configured in the scanning path generation unit, the image projection logic is configured in the initial image generation unit, and the image fine-tuning logic is configured in the image fine-tuning unit.

[0031] The path planning logic includes:

[0032] The main scanning direction is determined by the long side of the license plate and the baseline of the characters, and the vertical direction of the main scanning direction is determined by the line connecting the center line of the license plate outer frame and the mounting hole. A license plate plane coordinate system aligned with the license plate layout is established and the corresponding coverage boundary is determined.

[0033] Based on the license plate layout, the coverage boundary is divided into an outer frame edge area and multiple character areas, and grid parameters are set for each area. The grid parameters include the scan line spacing, the overlap of adjacent scan lines, and the boundary margin. The scan line spacing is determined based on the effective field of view of the narrow-angle camera and the preset overlap.

[0034] Intersecting calibration lines perpendicular to the main scanning direction are inserted into the coverage boundary at preset intervals, and the intersection of the intersecting calibration lines and the main scanning lines serves as alignment nodes.

[0035] A set of mutually parallel main scan lines is generated in each character area according to the main scan direction;

[0036] For each main scan line, calculate the start and end points, where the start and end points are located on the lines corresponding to the alignment nodes;

[0037] The main scan lines are arranged in a serpentine sequence so that the end point of an adjacent main scan line is adjacent to the start point of the next main scan line, thus obtaining an initial grid.

[0038] The initial grid is translated and rotated for fine adjustment based on the layout offset of the license plate, so that the main scanning direction is parallel to the character baseline and the grid origin of the initial grid is anchored to the positioning mark in the upper left corner of the license plate.

[0039] The output includes the main scan line sequence, the start and end coordinates of each main scan line, and the rasterized scan path with the coordinates of the alignment nodes.

[0040] The image projection logic includes:

[0041] The attitude parameters of the robotic arm end effector at each trigger position are collected. The attitude parameters include the spatial position of the end effector and the attitude angle relative to the license plate plane.

[0042] The preset narrow-field camera calibration information is invoked, which includes the camera's intrinsic parameters and corresponding extrinsic parameters;

[0043] The attitude parameters and the extrinsic parameters are combined to obtain the projection matrix of the camera in the license plate plane coordinate system corresponding to each local image;

[0044] Based on the intrinsic parameters, the pixel coordinates of each local image are converted into a point set in the license plate plane coordinate system through the projection matrix;

[0045] The point set is normalized and mapped according to the rasterized scanning path to obtain the initial license plate image.

[0046] The image fine-tuning logic includes:

[0047] Anchor point constraints are established based on the license plate layout. These constraints include constraints on the straightness and parallelism of character baselines, spacing and column alignment of adjacent characters, and parallelism and verticality of character outlines and license plate outlines.

[0048] For each local image at its stitching position in the grid, the character anchor point deviation within the corresponding coverage area is detected. Without changing the grid index, the compensation parameters for the corresponding local image are calculated based on the character anchor point deviation, so that the character anchor points within the coverage area of ​​the local image satisfy the anchor point constraint relationship.

[0049] The local image is compensated according to the compensation parameters, and a license plate image is generated by combining the initial license plate image and the compensated local image.

[0050] The license plate detection module includes:

[0051] The image analysis unit is used to receive the license plate image output by the license plate restoration module, segment the character region and the outer frame region, and extract image features for detection.

[0052] The quality detection unit performs quality determination based on the image features. The quality determination includes at least the detection of character edge sharpness, the detection of character stroke breakage, the detection of character outline straightness, the detection of color density consistency of characters and outlines, and the detection of character position deviation.

[0053] The judgment output unit is used to generate a qualified signal when the quality inspection is qualified, control the robotic arm to transfer the license plate to the license plate collection exit and generate the corresponding production number to be synchronized to the background; when the quality inspection is unqualified, it generates an unqualified signal, controls the robotic arm to put the license plate into the defective product collection box and records the unqualified information.

[0054] Compared with the prior art, the beneficial effects of this application are:

[0055] This application integrates a narrow-field camera at the end effector of a robotic arm with an image reconstruction strategy combining rasterization scanning and character anchor point constraints. This achieves the integration of license plate production and inspection processes, eliminating reliance on a separate inspection station and reducing additional hardware investment and space requirements. Introducing posture parameters and calibration information during image stitching effectively eliminates distortion and misalignment caused by end-effector movement, improving the accuracy of license plate image reconstruction. Furthermore, the fine-tuning process using character anchor point constraints binds the image to the geometric features of the license plate characters, ensuring more reliable quality assessment results. The overall solution not only shortens the production cycle and avoids errors caused by secondary handling, but also enables stable and automated license plate production and quality control throughout a compact space. Attached Figure Description

[0056] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0057] Figure 1 A structural diagram of the license plate production control device provided in this application embodiment;

[0058] Figure 2 A block diagram of a smart license plate making machine control system provided in this application embodiment;

[0059] Figure 3 A flowchart illustrating the intelligent license plate manufacturing robot arm control method provided in this application embodiment;

[0060] Figure 4 This is a schematic diagram of the license plate plane coordinate system provided in the embodiments of this application;

[0061] Figure 5 A schematic diagram of the coverage boundary division provided in the embodiments of this application;

[0062] Figure 6 This is a schematic diagram of the cross calibration lines provided in an embodiment of this application.

[0063] Reference numerals: 100, robotic arm body; 101, controller; 102, gripping end; 103, calibration mechanism; 104, narrow-field camera. Detailed Implementation

[0064] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0065] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0066] This embodiment is designed for the real-world production environment of an integrated license plate manufacturing cabin: pressing, heat transfer, picking and placing, and quality inspection are all integrated into a compact cabin. The robotic arm serves as both the executor of material handling and the mobile viewpoint for inspecting the quality of license plate production.

[0067] In this highly integrated, compact layout, traditional visual inspection stations often require additional fixed cameras and independent optical spaces within the license plate manufacturing compartment to acquire a complete panoramic image of the license plate in one go. However, this approach inevitably leads to problems such as limited installation space, complex optical path reflections, and glare from the reflective film on the license plate surface, making the inspection results prone to distortion, missed detections, or misjudgments. At the same time, fixed stations also force the production process to add extra handling actions between license plate manufacturing, transfer, and inspection, which not only prolongs the production time but also increases the risk of secondary damage to the license plate film.

[0068] Therefore, this embodiment proposes a narrow-field camera configuration scheme that works in conjunction with the end effector of a robotic arm:

[0069] A narrow-field camera moves with the arm, simultaneously capturing multiple local images while performing the license plate handling action. Through specific path planning, projection stitching, and character anchor point constraints, a complete license plate detection image is reconstructed. This effectively avoids the spatial and optical conflicts of traditional vision workstations, integrating the detection process into the handling trajectory. A separate panoramic detection channel is no longer needed, thus shortening the process and reducing hardware layout complexity while ensuring detection accuracy.

[0070] In terms of process collaboration, the geometric transition of license plates from flat to raised characters, the semi-cured window after heat transfer, and the strict requirement for no indentation on the surface determine that the gripping strategy cannot rely on drop-and-grab for secondary correction.

[0071] Based on this, this embodiment completes the correction while the object is held:

[0072] First, the negative pressure trend at the gripping end is used as a fit signal to drive a slight wrist movement and a light sweeping motion on the plane, quickly converging to a better center and normal direction. Then, without loosening the suction, a virtual fixture slides along the edge, using the mold reference edge and positioning angle for light touch and return to center, finally writing the compensation amount back into the subsequent loading and pressing trajectory. The correction is naturally embedded in the same set of handling actions, without interrupting the rhythm or introducing the risk to the film surface caused by secondary clamping.

[0073] This application is particularly suitable for several common and challenging scenarios: the confined space within a mobile license plate making vehicle, where installation distance is limited and external lighting is uncontrollable; the need for staggered hot-end control during multi-camera parallel operation to reduce peak loads, leaving tighter time windows for imaging and handling; and the instability of traditional fixed-threshold quality inspection due to microscopic differences in aluminum plates and reflective films from different batches, while multi-angle local images acquired by the arm provide redundancy for anti-reflection and detail verification. Even after slight vibration, weak mesh, or temporary shutdown and restart, the system can restore a consistent reconstruction coordinate system using the grid path and alignment nodes, ensuring the comparability and traceability of images from different batches.

[0074] In the specific application scenario of this application, the enclosed license plate manufacturing chamber has an overall box-like structure, used to integrate functional modules such as pressing, heat transfer, picking and placing, and quality inspection within a limited space. Inside the chamber is a robotic arm that, driven by a controller, grasps and transports the letter molds, license plate blanks, and finished products. The end of the robotic arm is equipped with a gripping end and an adjacent correction mechanism for automatic position and posture correction during license plate adsorption. A follow-up narrow-field camera is also configured at the end of the robotic arm to simultaneously acquire local images of the license plate surface during handling. An operation display interface is located at the front of the chamber for inputting process parameters and displaying test results. Through this integrated arrangement, the license plate manufacturing chamber can achieve automated license plate preparation and quality inspection within a compact space.

[0075] refer to Figure 1 , Figure 1 This is a structural diagram of the license plate production control device provided in an embodiment of this application.

[0076] Figure 1 The license plate production control device shown includes a robotic arm body 100, a controller 101, a gripping end 102, a correction mechanism 103, and a narrow-field camera 104, wherein:

[0077] The robotic arm body 100 is used to perform the grasping, carrying and positioning actions required for license plate production under the drive of the controller 101;

[0078] Controller 101 is used to control the movement of the robotic arm body 100 according to the manufacturing instructions and process formula;

[0079] The gripping end 102 is located at the end of the robotic arm body 100 and is used to pick up or hold the license plate blank, letter mold and finished license plate.

[0080] The correction mechanism 103 is set above or to the side of the gripping end 102 and is used to correct the position and posture of the license plate during the gripping process, so as to ensure that the license plate is accurately aligned with the reference edge of the mold.

[0081] A narrow-field camera 104 is positioned near the grasping end 102 and captures multiple local images of the license plate surface as the robotic arm body 100 moves, so as to reconstruct a complete license plate image through image stitching and character anchor point constraints.

[0082] refer to Figure 2 This application provides a schematic diagram of a control system module for an intelligent license plate making machine.

[0083] In one example, this application embodiment provides a smart license plate making machine control system, applied to a license plate making control device. The license plate making control device includes a controller and a correction mechanism. The correction mechanism is disposed at the gripping end of the robotic arm and is used to correct the position and posture of the license plate when the robotic arm grips the license plate. The robotic arm also includes a narrow-field camera, which is disposed at the end of the robotic arm. The system includes:

[0084] The instruction acquisition module is used to acquire manufacturing instructions and process formulas, and control the robotic arm to grab letter molds and license plate blanks that match the process formulas according to the manufacturing instructions.

[0085] The license plate manufacturing module is used to complete the pressing and heat transfer process under the control of the controller to obtain the heat-transferred license plate.

[0086] The license plate restoration module is used to acquire multiple local images during the movement of the robotic arm using the narrow-field camera, combine the posture parameters of the robotic arm end with the camera calibration information, project the local images onto the license plate plane coordinate system, and stitch them together according to the rasterized scanning path to obtain the license plate image.

[0087] The license plate detection module is used to determine the quality of the license plate image. If the image is qualified, the license plate is transferred to the license plate collection exit and a corresponding production number is generated and synchronized to the background. If the image is unqualified, the license plate is put into the defective product collection box and the unqualified information is recorded.

[0088] The card-making and processing module includes:

[0089] The letter mold assembly unit is used to control the robotic arm to pick up the letter mold and place it in the positioning position of the preset mold, pick up the license plate blank and feed it to the predetermined position directly above the lower letter mold, and then pick up the upper letter mold and place it directly above the license plate blank to form a mold assembly.

[0090] The pressing and forming unit is used to control the robotic arm to move the mold assembly to the pressing station, and to press and form the license plate blank according to the pressure and holding time parameters of the process formula to obtain a license plate semi-finished product.

[0091] The heat transfer unit is used to control the robotic arm to transfer the semi-finished license plate to the heat transfer station, and to perform heat transfer processing according to the temperature and transfer time parameters of the process formula to obtain the heat-transferred license plate.

[0092] The character mold returning unit is used to control the robotic arm to return the upper and lower character molds to the character mold slots and exit the card making station after the heat transfer process is completed.

[0093] The system also includes a correction module for correcting the position and orientation of the license plate when the robotic arm grasps it.

[0094] The license plate restoration module includes:

[0095] The scanning path generation unit is used to generate a rasterized scanning path based on the planar range of the license plate;

[0096] An image acquisition unit is used to control the movement of the robotic arm along the rasterized scanning path and to acquire multiple local images using the equidistant distance of the end on the license plate plane as a trigger condition.

[0097] An initial image generation unit is used to combine the end-effector's pose and preset narrow-field camera calibration information to project the multiple local images onto the license plate plane coordinate system, and stitch them together according to the grid position to obtain an initial license plate image.

[0098] An image fine-tuning unit is used to extract character anchor points from the initial license plate image using an edge extraction algorithm, wherein the character anchor points include at least one of the character outer frame corner points, stroke intersection points, and character baselines; and to fine-tune the initial license plate image using the character anchor points as constraints to generate a license plate image.

[0099] The license plate detection module includes:

[0100] The image analysis unit is used to receive the license plate image output by the license plate restoration module, segment the character region and the outer frame region, and extract image features for detection.

[0101] The quality detection unit performs quality determination based on the image features. The quality determination includes at least the detection of character edge sharpness, the detection of character stroke breakage, the detection of character outline straightness, the detection of color density consistency of characters and outlines, and the detection of character position deviation.

[0102] The judgment output unit is used to generate a qualified signal when the quality inspection is qualified, control the robotic arm to transfer the license plate to the license plate collection exit and generate the corresponding production number to be synchronized to the background; when the quality inspection is unqualified, it generates an unqualified signal, controls the robotic arm to put the license plate into the defective product collection box and records the unqualified information.

[0103] In one example, this application provides a method for controlling a smart license plate making robotic arm.

[0104] Next, with reference to the accompanying drawings, the intelligent license plate manufacturing robot arm control method provided in the embodiments of this application will be described. Figure 3 The method shown is applied to a license plate production control device, which includes a controller and a correction mechanism. The correction mechanism is disposed at the gripping end of the robotic arm and is used to correct the position and posture of the license plate when the robotic arm grips it. The robotic arm also includes a narrow-field camera disposed at the end of the robotic arm. The method includes:

[0105] S1: Obtain the manufacturing instructions and process formula, and control the robotic arm to grab the letter mold and license plate blank that match the process formula according to the manufacturing instructions;

[0106] In this embodiment, the production instruction is triggered by the front-end terminal scanning the business receipt barcode. Based on this, the controller reads the formula parameters corresponding to the size, background color, and font specifications from the local process library and sends them to the motion control and process execution channel under the same task token.

[0107] Furthermore, the robotic arm first removes the letter mold from the letter mold slot, then picks up the license plate blank at the blank loading position; during the gripping action, the correction mechanism at the gripping end intervenes:

[0108] Using negative pressure changes as a fit signal, small adjustments are made to the end-effector's pitch, yaw, and micro-displacement to align the adsorption center with the geometric center of the blank. The grip compensation amount is then recorded for subsequent placement and molding trajectory correction. This approach aims to incorporate the alignment process into the initial gripping step, avoiding secondary corrections during placement and re-adsorption, shortening cycle time, and reducing the risk of film surface indentation. Simultaneously, task tokens are individually linked to process formulations to ensure consistency and traceability of formulations in subsequent processes.

[0109] Those skilled in the art will understand that the source of the production instructions can be an external business system or a local task list, and the formula fields can include pressure, holding time, transfer temperature and duration, hot window limit, etc. It is only necessary to meet the minimum requirement of retrieving the corresponding process parameters by token, and this application does not impose any further limitations.

[0110] S2: The controller controls the robotic arm to complete the license plate making process, and obtains the heat-transfer license plate;

[0111] In this embodiment, the card-making process is performed along a fixed process path: lower mold positioning - blank loading - upper mold fitting - pressing - heat transfer.

[0112] The pressing stage is executed according to the pressure and holding time curves in the formula, while the heat transfer stage is controlled by temperature and transfer time. Before being fed into the pressing and transfer stations, the robotic arm incorporates the hold compensation amount obtained from S1 to ensure that the mold stacking and transfer contact surfaces are aligned under the same reference. Considering the semi-curing window after heat transfer, the motion is planned as a serpentine path with a time window, prohibiting prolonged pauses near the hot end to reduce the risk of thermal warping and localized adhesion. After completion, the upper and lower molds are reset to their mold slots by a mold return action to ensure fixture consistency for the next cycle.

[0113] S3: The narrow-angle camera captures images and performs quality judgment; if the judgment is qualified, the license plate is transferred to the license plate collection exit and a corresponding production number is generated and synchronized to the background; if the judgment is unqualified, the license plate is put into the defective product collection box and the unqualified information is recorded.

[0114] In this embodiment, a narrow-field camera is mounted near the gripping end of the robotic arm to capture partial images of the license plate. The acquisition strategy employs a rasterized path aligned with the license plate layout, triggering exposure based on the equidistant travel distance of the end effector on the license plate plane, thereby eliminating sampling unevenness caused by speed fluctuations. Subsequently, combining the end effector posture and camera calibration, the partial image is projected onto the license plate plane and stitched together according to the raster positions to form an initial complete image. To suppress reflections and accumulated errors, character anchor points are extracted from the initial complete image. These anchor points constrain the local mosaic blocks to perform minor translation and rotation corrections, generating the license plate image. The image is then evaluated for character edge sharpness, stroke continuity, outer frame straightness, color density consistency, and character position deviation.

[0115] Those skilled in the art will understand that the grid step size, overlap, and judgment threshold can be set according to the imaging resolution and target accuracy, and the data processing algorithm can adopt edge operators, connected component analysis, or learning models. It is only necessary to support the evaluation of the geometric and appearance indicators of the license plate image to a minimum. This application does not impose any further limitations.

[0116] Before delving into the specific technical details of the steps, the embodiments of this application need to be emphasized again.

[0117] This embodiment first defines the mutual constraints between imaging and transportation within the compact card-making cabin:

[0118] Because the process chain is arranged in a closed loop within a limited cavity, the installation distance and incident angle of the camera device are restricted. If a single full-frame acquisition is used, edge distortion and strong reflective stripes are likely to occur, making it difficult to guarantee the geometric consistency between the character strokes and the outer frame lines. Therefore, this embodiment binds the acquisition process with the transport action, allowing the narrow-span imaging unit to travel within the card coordinate system along the end trajectory. Local images are accumulated under a stable geometric reference through equidistant triggering, thus replacing the stringent requirements of large field-of-view optical conditions with kinematic constraints.

[0119] Regarding parameter settings, those skilled in the art will understand that the step size, overlap, and trigger threshold can be adjusted according to the target resolution and rhythm requirements, as long as they can be minimally satisfied to allow for stable projection and subsequent splicing within the card face plane. This application does not impose any further limitations.

[0120] During the reconstruction phase, instead of proceeding in the traditional order of first assembling and then checking, a geometric-semantic joint framework is introduced that is isomorphic to the license plate layout:

[0121] The local image is first projected onto the card face plane through end pose and camera calibration to form an initial stitching that is aligned with the layout; then, the anchor point set consisting of the corner points of the character outline, the intersection of strokes and the baseline alignment points is used as a constraint to make minor pose corrections to each local block, so that the character column spacing, baseline straightness and the orthogonality of the outline converge consistently throughout the entire image.

[0122] This change in order is not simply an algorithm replacement; it aims to use layout knowledge as the dominant clue for error convergence, avoiding drift caused by insufficient pure geometric features in reflective or sparsely textured areas.

[0123] To ensure that the handling and alignment actions do not interrupt each other, the alignment process is folded into the same holding cycle:

[0124] After adsorption is established, the trend of negative pressure change is used as an indicator of fit. Small-scale probes are made for end pitch, yaw and planar light sweep, and the system quickly converges to a better center and normal. Then, without loosening the adsorption, the system completes the light touch and sliding along the reference edge. The system returns to the center at the positioning angle through the symmetric feedback criterion. The compensation amount is immediately written back to the pose parameters of the subsequent molding and transfer.

[0125] Understandably, the correction behavior is an implicit action under the gripping end during a single hold, the beat is not significantly lengthened, and the membrane surface avoids the risk of indentation and stretching caused by secondary clamping.

[0126] Regarding anomalies and recovery, considering the potential for slight vibrations, temporary pauses, or time window changes due to hot-end peak shifts within the card-making chamber, trajectory planning uses a serpentine coverage and node alignment framework. After any interruption, data acquisition can resume from the nearest node without disrupting the consistency of the overall map coordinates. If anomalous feedback is continuously detected during correction or reconstruction, the process will degrade to retaining only the geometrically necessary minimum action set to ensure neither misjudgment nor misplacement, returning to the complete path once external conditions recover. Maintaining steady-state coupling between imaging, handling, and process without relying on high-bandwidth transmission or external computing clusters, it is suitable for reuse in various deployment configurations such as front-end windows, mobile chambers, or parallel production lines.

[0127] Next, the technical details of the card-making process in this application will be elaborated.

[0128] The principle of this embodiment in the license plate manufacturing process is that by sequentially and systematically converging the mold assembly, pressing and forming and heat transfer processing into the same robotic arm's execution chain, the forming and pattern transfer of the license plate no longer rely on manual repositioning or multi-process handling, but are directly driven by the same control logic.

[0129] Specifically, during the alternating assembly of the lower character mold, license plate blank, and upper character mold, the robotic arm utilizes preset positioning positions to ensure the consistency of the mold assembly, avoiding the accumulation of deviations caused by multiple loading and unloading. In the pressing stage, precise control of pressure and holding time parameters in the process formula ensures uniform and consistent character forming. In the heat transfer stage, dual control of temperature and time guarantees stable transfer and adhesion of ink or film layers on the surface of the semi-finished product. Throughout the entire process, the return of the character mold and the exit from the workstation are both completed by the robotic arm, forming a closed loop. This not only shortens the production cycle of a single piece but also reduces mold wear and the probability of deviation.

[0130] Those skilled in the art will understand that parameters such as pressure, temperature, and time can be set according to the actual material and batch, as long as the minimum quality requirements of the license plate forming and transfer process are met. This application does not impose any further limitations.

[0131] In one example, the step of controlling the robotic arm to complete the card-making process via a controller includes:

[0132] Control the robotic arm to gripper the character mold and place it at the preset mold positioning position;

[0133] The robotic arm is controlled to pick up the license plate blank and load it to a predetermined position directly above the lower letter mold;

[0134] The robotic arm is controlled to grip the upper letter mold and place it directly above the license plate blank, forming a pressing mold assembly together with the lower letter mold.

[0135] The robotic arm is controlled to move the mold assembly to the pressing station, where it is pressed and shaped according to the pressure and holding time parameters of the process formula to obtain a semi-finished license plate.

[0136] The robotic arm is controlled to transfer the semi-finished license plate to the heat transfer station, where heat transfer processing is performed according to the temperature and transfer time parameters of the process formula.

[0137] After the heat transfer process is completed, the robotic arm is controlled to return the upper and lower letter molds to the letter mold slots and exit the license plate making station to obtain the heat-transferred license plate.

[0138] Next, we will elaborate on the technical aspects of the license plate correction method in this application.

[0139] In the automated process of license plate manufacturing, there is always an unavoidable uncertainty in the handling and placement of license plate blanks. For example, during the adsorption and gripping stage, the blank may experience slight skew or displacement due to its own warping, differences in surface friction, or uneven distribution of vacuum adsorption. During the handling process, inertia or vibration may cause further deviations in posture. If these minor cumulative deviations are not corrected, they will be amplified during the mold pressing or heat transfer stages, manifesting as blurred character edges, skewed outer frames, or even localized demolding, thus directly affecting the quality of the finished product.

[0140] Traditional methods typically rely on secondary gripping or additional positioning stations for compensation, but this not only prolongs the cycle time but also increases the complexity of the robotic arm's path planning and energy consumption. This embodiment integrates a correction mechanism at the gripping end and directly introduces feedback and a virtual fixture mode after adsorption, achieving rapid correction of the license plate's position and posture, thereby avoiding additional secondary gripping actions.

[0141] Understandably, in the processing logic of this application, the correction process is naturally embedded into the existing action chain, which not only improves the real-time performance of the correction, but also ensures a high degree of consistency between the license plate and the mold reference.

[0142] In one example, when the robotic arm grasps the license plate, the method further includes:

[0143] After the license plate is attracted, negative pressure feedback is collected at the grasping end;

[0144] Specifically, the change in negative pressure at the moment of adsorption directly reflects the sealing quality between the suction cup and the card surface, the deviation between the adsorption center and the geometric center, and the gap leakage caused by local warping. If reliable feedback is not obtained during this period, subsequent posture correction will lack quantitative basis and may easily lead to cascading errors as the card enters the mold with misalignment.

[0145] In this embodiment, corresponding sampling points are set at the grasping end. Negative pressure is sampled at medium to high frequencies, and combined with short-window median and moving average composite filtering to form two sets of statistics for the early and stable adsorption stages. At the same time, the pump load current or valve group pressure drop is read as auxiliary quantities to identify micro-leakage and local warping. The controller immediately records the baseline window after adsorption and calculates the mean, fluctuation amplitude, recovery time, and high-frequency ripple amplitude within the stable window as the fit index. If the ripple and the negative pressure mean are inversely coupled and the recovery time exceeds the set range, an edge leakage mark is added to the current cycle, and the step size of subsequent attitude correction is limited to avoid large attitude changes before the seal is closed.

[0146] The adjustment direction is determined based on the changing trend of the negative pressure feedback, and the grasping posture and planar position of the grasping end are corrected until the feedback signal stabilizes.

[0147] Specifically, the negative pressure exhibits a monotonically increasing exploitable region as attitude and planar displacement change: when the adsorption center moves closer to the geometric center and the suction cup normal is closer to the card surface normal, the sealing area increases, leakage decreases, the average negative pressure increases, and ripple decreases. Rather than making a large adjustment all at once, it is better to probe and evaluate within a safe, small range using micro-oscillations and micro-movements, gradually converging based on feedback trends, thereby completing the initial correction without disrupting the established adsorption.

[0148] In this embodiment, after the adsorption stabilizes, the controller performs small micro-swings and micro-movements along the wrist pitch, yaw, and planar axes of the gripping end in a preset sequence. The amplitude, duration, and pause interval are all configurable. Immediately after each micro-movement, the adhesion difference is calculated. If the improvement is significant, an offset of the same magnitude is accumulated along that direction, and the pause for the next attempt is shortened. If the improvement is limited, the step size is halved or the direction is reversed. To prevent slippage, all micro-movements are performed within the adsorption safety zone. Speed ​​and acceleration are limited by material friction and suction cup specifications. A brief pause is inserted between movements to allow negative pressure to recover before proceeding to the next round of judgment. When the adhesion improvement after multiple consecutive attempts is below a threshold and the ripple remains low, convergence is determined. At this point, the total offset of the wrist posture and planar position is written into the grip compensation amount and locked to the subsequent handling and placement trajectory.

[0149] After the feedback signal stabilizes, the gripping end maintains its adsorption posture, and the license plate is moved to the reference edge of the mold. The correction mechanism, in conjunction with the controller's mode switching, enters the virtual fixture mode to obtain a compensation signal. The gripping end is then subjected to posture compensation based on the compensation signal.

[0150] Specifically, negative pressure-driven gripping can only achieve relatively optimized sealing and centering consistency, while pressing and transfer rely on absolute alignment with the mold reference edge and positioning angle. Direct hard contact positioning can easily leave marks on reflective film or spray coating. Therefore, while maintaining adsorption, it is necessary to gently touch along the edge and to the corner using a restricted normal force and a sliding tangential approach. By utilizing the geometric relationship between the edge and the corner, a write-backable attitude offset can be obtained, so that the gripping coordinate system and the mold reference can be established in a consistent reference.

[0151] In one example, the correction mechanism, in conjunction with the controller's mode switching, enters a virtual fixture mode to obtain a compensation signal, including:

[0152] While maintaining the license plate adsorption, the robotic arm is controlled to move the license plate along the normal direction to the reference edge of the mold, and the contact state is determined according to the change of joint torque.

[0153] When the contact state is determined, the normal advance of the robotic arm is stopped, and it is switched to sliding along the tangential direction of the reference edge, while maintaining the license plate adsorption state.

[0154] During the sliding process, the contact feedback signal is monitored. When the contact feedback signal is a symmetrical signal, it is determined that the license plate is aligned with the reference edge.

[0155] The robotic arm is controlled to slide to the positioning angle and perform an attitude correction operation to make the long side of the license plate parallel to the reference side and the short side perpendicular to the reference side, thereby obtaining the corresponding compensation signal.

[0156] Next, we will elaborate on the technical aspects of the method related to license plate images in this application.

[0157] Understandably, existing license plate production lines typically have a dedicated visual inspection station after pressing and heat transfer. This station often requires additional equipment such as panoramic industrial cameras, independent lighting systems, and fixed inspection fixtures, increasing both equipment procurement and maintenance costs. Furthermore, it occupies extra space in a compact license plate manufacturing environment, limiting the overall layout. In addition, there is a material handling connection between the independent inspection station and the license plate manufacturing process. This connection not only delays the production cycle but may also introduce new uncertainties due to secondary transport or positional errors.

[0158] In this embodiment, the narrow-field camera is directly integrated into the end effector of the robotic arm. The robotic arm can simultaneously acquire partial images of the license plate during its natural loading and unloading trajectory, thus eliminating the need for a separate inspection station. This approach avoids additional hardware, but it also introduces new technical challenges:

[0159] Due to the limited field of view of narrow-field cameras, a single capture cannot cover the entire license plate plane, resulting only in a partial image. Furthermore, the camera's posture and position constantly change during spatial movement, and direct stitching often leads to cumulative distortion and misalignment. Without processing, the final stitched image will neither fully cover the license plate nor accurately support quality assessment indicators such as character edge sharpness and stroke breakage.

[0160] It is important to emphasize that the narrow-field camera is used in this application because it is small in size, low in cost, and easy to install. It can be directly embedded into the end effector of the robotic arm without significant modifications to the overall card-making chamber structure. Compared to wide-field panoramic cameras, narrow-field cameras significantly reduce reliance on light sources and installation space, while also covering any workstation as the robotic arm moves, enabling an integrated process of on-demand inspection. This avoids the complexity of reserving additional inspection stations and light source arrangements in a compact card-making device, and also reduces the risk of downtime due to equipment upgrades or maintenance, thereby improving the overall continuity and flexibility of production.

[0161] In this embodiment, the narrow-field camera is combined with the robotic arm's motion trajectory. Through gridded path planning and local image stitching strategies, the limitation of limited coverage in a single acquisition is effectively overcome. Using real-time posture data from the robotic arm's end effector and camera calibration parameters, different local images can be uniformly projected onto the same license plate coordinate system. Furthermore, error correction is achieved through geometric constraints on character anchor points, ensuring that the final stitched image achieves comparable results to independent inspection stations in terms of size ratio, character spacing, and outer frame straightness. This approach retains the cost and placement advantages of narrow-field cameras while, with the cooperation of algorithms and path control, achieving near-panoramic inspection quality assessment capabilities, thus completing the entire integrated process of license plate production and inspection within a limited space.

[0162] In one example, the process of acquiring images and determining their quality using the narrow-field camera includes:

[0163] S3.1: Generate a rasterized scanning path based on the planar range of the license plate, control the robotic arm to move along the rasterized scanning path, and acquire multiple local images with the equidistant distance of the end on the license plate plane as the trigger condition;

[0164] Specifically, the rasterized scanning path establishes the license plate coordinates primarily along the long side of the license plate and the character baseline. A boundary margin is set by indenting the coverage boundary along the outer frame to compensate for edge distortion and mechanical overshoot. To avoid uneven sampling spacing caused by speed fluctuations, the trigger condition is not time but the travel distance of the end point projected onto the license plate. The scan lines use a serpentine sequence to reduce idle travel, and cross-calibration lines are inserted at fixed intervals to form alignment nodes, facilitating subsequent stitching constraints. Raster parameters include scan line spacing, overlap width, trigger travel step size, and boundary retraction.

[0165] In one example, generating a rasterized scan path based on the planar extent of the license plate includes:

[0166] S3.1.1: Determine the main scanning direction using the long side of the license plate and the baseline of the characters, and determine the vertical direction of the main scanning direction using the center line of the license plate outer frame and the line connecting the mounting holes. Establish a license plate plane coordinate system aligned with the license plate layout and determine the corresponding coverage boundary.

[0167] Specifically, the main scanning direction is parallel to the long side of the license plate and uses the position of the character baseline as a reference. This is because the character information has prior knowledge of column alignment and equal spacing in this direction, allowing subsequent overlap planning and character anchor point constraints to share the same geometric reference. The vertical direction is determined by connecting the outer frame centerline and the mounting holes. This ensures consistency with the overall layout even with minor cutting deviations or clamping offsets in the incoming material, thus avoiding the cumulative skew introduced by using only one edge as a vertical reference. The coverage boundary is set with a boundary allowance within the outer frame to absorb instantaneous errors in projection and trajectory, reducing the impact of field of view overshoot and edge distortion on splicing.

[0168] In this embodiment, the geometric parameters of the license plate's long side, outer frame centerline, and mounting holes are derived from the process formula, and the character baseline direction is directly given through the template. When establishing the license plate's planar coordinates, the upper left corner positioning mark is taken as the origin, the long side direction is the x-axis, the vertical direction is the y-axis, and the outer frame indentation is calculated based on the lens's effective field of view and calibration residual.

[0169] refer to Figure 4 , Figure 4 This is a schematic diagram of the license plate plane coordinate system provided in the embodiments of this application.

[0170] Figure 4 This demonstrates a method for establishing a planar coordinate system based on the license plate layout. The long side of the license plate is used as the primary scanning direction, and the character baseline serves as a horizontal reference. The line connecting the center line of the license plate's outer frame and the mounting holes defines the vertical direction of the primary scanning direction, thus forming a stable two-dimensional coordinate system. Using this coordinate system, the grid origin can be set at the positioning mark in the upper left corner of the license plate, thereby dividing the character area and the side strip area into a unified geometric framework, providing a reference for subsequent rasterized scanning path generation and image stitching.

[0171] S3.1.2: Divide the coverage boundary into an outer frame edge area and multiple character areas according to the license plate layout, and set grid parameters for each area. The grid parameters include the scan line spacing, the overlap of adjacent scan lines, and the boundary margin. The scan line spacing is determined based on the effective field of view of the narrow-angle camera and the preset overlap.

[0172] Specifically, the character area has higher requirements for resolution and anti-reflection than the side strip area. If a uniform spacing is used, it will result in invalid sampling or insufficient detail. Therefore, different spacing and overlap are allocated to each area to achieve a balance between pace and quality. The boundary margin is used to offset small-range motion errors and calibration residuals, so that the edge of the field of view does not become the main seam of the stitching.

[0173] In this embodiment, the grid parameters can be calculated using a field-of-view model. Specifically, based on the effective field-of-view width of the narrow-field camera, the lens focal length, and the calibrated projection scale, combined with the target overlap ratio, the spacing of each scan line in the license plate plane coordinate system is calculated. Simultaneously, the boundary margin can be obtained by contour detection or reference point identification of the license plate's outer frame edge to ensure that the scan coverage area is slightly larger than the actual area. This compensates for minor offsets or calibration residuals during end-effector movement, preventing edges from being misjudged as gaps during image stitching.

[0174] Furthermore, the division between the outer frame border area and the character area can be preset through layout rules, such as logical partitioning according to the character row and column layout, outer frame thickness, and mounting hole positions specified in industry standards; alternatively, it can be dynamically determined during operation through image edge extraction and character region segmentation algorithms. Both are conventional methods that can be understood and implemented by those skilled in the art. This application is not limited to a specific partitioning method, as long as it ensures that differentiated raster parameters are used between the character area and the border area during the scanning path generation process.

[0175] refer to Figure 5 , Figure 5 This is a schematic diagram of the coverage boundary division provided for an embodiment of this application.

[0176] Figure 5 The diagram shows that within the overall plane of the license plate, the coverage boundary is divided into multiple character zones. It is easy to understand that license plates typically include eight characters.

[0177] exist Figure 6 In embodiments not shown, the outer frame edge area specifically refers to the strip-shaped region between the edge of the license plate's outer frame and the character area, typically including the border area around the license plate and the annular buffer area around the mounting holes. Since this area does not carry character information, its geometric features are mainly manifested as the continuity of straight lines or rectangular frames; therefore, the requirements for resolution and overlap during image acquisition are lower than those for the character area. By setting a larger scan line spacing and a lower overlap within the outer frame edge area, the scanning path can be effectively shortened, reducing the number of local images, thereby reducing the computational load for image stitching and detection latency.

[0178] Understandably, regardless of the method used, the purpose of defining the outer frame border area is to achieve regional parameter allocation, that is, to allocate more computing and detection resources to the character area while maintaining overall coverage, so as to improve the quality of the final image stitching and the accuracy of detection.

[0179] refer to Figure 6 , Figure 6 This is a schematic diagram of the cross calibration lines provided in an embodiment of this application.

[0180] S3.1.3: Insert cross calibration lines perpendicular to the main scanning direction into the coverage boundary at preset intervals, and the intersection of the cross calibration lines and the main scanning lines shall be used as alignment nodes;

[0181] Specifically, cross calibration lines are used to interrupt cross-row error propagation and provide rigid anchor points between rows and columns during the stitching stage; the intersection point is used as an alignment node, which can be restarted from the nearest node in the event of a subsequent pause or abnormal recovery without destroying the consistency of coordinates across the entire map.

[0182] In this embodiment, the spacing of the cross calibration lines is set according to the distribution of the character columns, and they are preferentially placed in the gaps between the character columns to reduce the occlusion of the character area. Each cross calibration line is required to be configured with at least one alignment node, and the node is forced to be exposed and the timestamp and end pose are recorded so as to serve as a priority reference during projection and unification.

[0183] S3.1.4: Generate a set of mutually parallel main scan lines in each character area according to the main scan direction;

[0184] Specifically, the main scan lines must be strictly parallel to ensure a constant overlap bandwidth and be aligned with the character baseline so that the row direction of subsequent character anchor points is consistent; when crossing areas, the scan line numbers must be kept continuous to facilitate splicing indexes and quality traceability.

[0185] S3.1.5: For each main scan line, calculate the start point and end point, where the start point and end point are located on the line corresponding to the alignment node;

[0186] Specifically, having the start and end points fall on the alignment node lines allows each scan line to start and end on a rigid reference, thereby reducing drift accumulation at serpentine reversals; combined with the lead-in and lead-out distances, it can prevent the start and end frames from becoming blurred due to acceleration and deceleration.

[0187] In this embodiment, the start and end points are initially calculated using the intersection of the scan line and the coverage boundary. Then, the intersection point is pushed along the scan direction to the nearest aligned node line and a fixed lead-in / lead-out distance is added.

[0188] S3.1.6: Arrange the main scan lines in a serpentine sequence so that the end point of an adjacent main scan line is adjacent to the start point of the next main scan line, thus obtaining an initial grid;

[0189] Specifically, the serpentine sequence reduces unproductive return trips and keeps the turning positions of adjacent rows fixed, making it easier to form stable start and end references near the cross calibration line; the end point being adjacent to the next start point can shorten the reversing path.

[0190] S3.1.7: Adjust the initial grid by translation and rotation according to the layout offset of the license plate, so that the main scanning direction is parallel to the character baseline and the grid origin of the initial grid is anchored to the positioning mark at the upper left corner of the license plate.

[0191] Specifically, license plates experience slight translational and rotational offsets during clamping. If the initial grid is not fully verified, the scan lines and character columns will slowly deviate, affecting the alignment of subsequent character anchor points. Anchoring the grid origin to the upper left corner positioning mark establishes a stable spatial starting point for splicing and judgment, facilitating cross-batch traceability.

[0192] In this embodiment, the layout offset is given by the clamping datum and template parameters; if a stable in-situ offset has been recorded in the previous batch of sampling, it is applied directly before generating the grid; when fast pre-check is enabled, the actual placement rotation angle can be calculated using the geometric features of the short side of the outer frame and the hole position, and the initial grid is rotated by this small angle and translated along the two axes to the positioning mark; the fine-tuned grid is checked for parallelism with the character baseline direction, and if the deviation exceeds the limit, it is rolled back and prompted to reread the offset.

[0193] S3.1.8: Output a rasterized scan path containing the main scan line sequence, the start and end coordinates of each main scan line, and the coordinates of the alignment nodes;

[0194] Specifically, the path output must simultaneously carry geometric information and operational constraints to ensure consistent reference to the same coordinate system in the execution and reconstruction stages; the main scan line sequence is used to determine the serpentine execution order, the start and end coordinates are used for trigger management and splicing positioning, and the alignment node coordinates are used for cross-line consistency, anomaly recovery, and anchor point weight allocation.

[0195] In this embodiment, the output includes: a unique number for each scan line, its corresponding area number, running direction, start and end coordinates, a list of aligned nodes, a range of prohibited sub-segments, a trigger stroke step distance, suggested speed and acceleration, turning transition radius, thermal window limits, and boundary retraction amount, etc. It also saves the template version used during path generation, boundary margins, and overlapping targets for traceability. During execution, the execution end attaches the actual trigger timestamp and attitude record to the same number, which is used to accurately position the frames during the projection and stitching stages.

[0196] S3.2: Combining the end-effector's posture and the preset narrow-field camera calibration information, the multiple local images are projected onto the license plate plane coordinate system and stitched together according to the grid position to obtain the initial license plate image;

[0197] Specifically, unifying local images requires mapping frames with different poses and perspectives to the same card face plane coordinates. To reduce the impact of lens distortion and installation errors, image plane correction based on calibration files is performed before projection, and the card face projection relationship is calculated based on the end pose and hand-eye extrinsic parameters. To ensure time consistency, the exposure timestamp and end pose are aligned using the same source clock or time synchronization mechanism. During the stitching stage, no free deformation is performed; frame blocks are simply placed in their corresponding positions according to the grid index to avoid geometric drift caused by early overfitting.

[0198] In one example, obtaining the initial license plate image includes:

[0199] S3.2.1: Collect the attitude parameters of the end effector at each trigger position, the attitude parameters including the spatial position of the end effector and the attitude angle relative to the license plate plane;

[0200] S3.2.2: Call the preset narrow-field camera calibration information, which includes the camera's intrinsic parameters and corresponding extrinsic parameters;

[0201] S3.2.3: Combine the attitude parameters with the extrinsic parameters to obtain the projection matrix of the camera in the license plate plane coordinate system corresponding to each local image;

[0202] S3.2.4: Based on the intrinsic parameters, the pixel coordinates of each local image are converted into a point set in the license plate plane coordinate system through the projection matrix;

[0203] S3.2.5: Normalize and map the point set according to the rasterized scanning path to obtain the initial license plate image.

[0204] Specifically, the spatial position and attitude angle of the robotic arm's end effector at each trigger point are acquired in real time through joint encoders and force control feedback. These attitude parameters are combined with pre-calibrated extrinsic parameters of the narrow-span camera to establish a projection matrix of the camera in the license plate plane coordinate system, thereby associating the camera's imaging plane with the physical plane of the license plate. Combined with the camera's intrinsic parameters (including focal length, principal point offset, and distortion coefficients), the pixel coordinates of each local image are projected one by one onto the license plate plane coordinate system, forming a point set data corresponding to the actual geometry. Subsequently, the point set data is normalized and mapped according to the rasterized scanning path, ensuring seamless stitching of each local image within a unified reference system, thus obtaining the initial license plate image. Because this process combines multiple attitude acquisitions with coordinate transformations, it effectively offsets the cumulative distortion caused by attitude shifts resulting from the robotic arm's movement, ensuring that the stitched image is geometrically consistent with the actual license plate surface.

[0205] Those skilled in the art will understand that the acquisition of intrinsic and extrinsic parameters can be achieved through conventional planar target calibration, and the acquisition of posture parameters can be obtained by relying on the conventional joint encoders and end-effector pose calculations of industrial robots. Therefore, this mapping and conversion method does not depend on specific hardware. It only needs to provide the minimum pose information and calibration data to complete the coordinate unification. This application does not impose any further limitations here.

[0206] S3.3: Extract character anchor points from the initial license plate image using an edge extraction algorithm, wherein the character anchor points include at least one of the character outline corner points, stroke intersection points, and character baselines;

[0207] Specifically, character anchors provide stable semantic constraints for subsequent geometric fine-tuning. Features that are crucial to the overall layout are prioritized, including straight lines on the four sides of the outline, corner points of the character outline, stroke intersections, and alignment points on the character baseline. To avoid interference from reflective stripes and noisy edges, the initial image is contrast-stretched and reflective highlight areas are suppressed before anchor extraction. In overlapping areas, segments with higher clarity are prioritized for feature detection.

[0208] In this embodiment, the outer frame straight lines are extracted from the global edge map using a straight line detection method, and the aspect ratio of the layout is used as a consistency check. The character candidate area is coarsely partitioned based on the column spacing and top and bottom margins of the layout template. Corner points and stroke intersections are extracted within the partitions to form an anchor point candidate set. To improve reliability, the sharpness index, local contrast, and deviation from the template position are calculated for each anchor point, and weights are assigned accordingly. High-brightness edges that may be caused by reflections are eliminated or downweighted based on brightness distribution and the continuity of neighborhood texture.

[0209] S3.4: Fine-tune the initial license plate image using the character anchor points as constraints to generate a license plate image;

[0210] Specifically, although the initial image already satisfies overall coverage, slight misalignment may still occur in some areas due to triggering and pose errors. The fine-tuning stage does not deform the entire image; instead, it compensates for frame blocks with minute translations and small-angle rotations on a grid-by-grid basis. The compensation range is limited by mechanical and imaging tolerances to avoid excessive deformation that could damage the true geometry. Constraints are derived from the set of anchor points, with the goal of ensuring that character baselines converge to a straight line globally, that the spacing between adjacent characters conforms to the layout, and that the parallel and perpendicular relationships between the outer frame and the character outer frame are maintained.

[0211] In one example, the initial license plate image is fine-tuned using the character anchor points as constraints to generate a license plate image, including:

[0212] Anchor point constraints are established based on the license plate layout. These constraints include constraints on the straightness and parallelism of character baselines, spacing and column alignment of adjacent characters, and parallelism and verticality of character outlines and license plate outlines.

[0213] For each local image at its stitching position in the grid, the character anchor point deviation within the corresponding coverage area is detected. Without changing the grid index, the compensation parameters for the corresponding local image are calculated based on the character anchor point deviation, so that the character anchor points within the coverage area of ​​the local image satisfy the anchor point constraint relationship.

[0214] The local image is compensated according to the compensation parameters, and a license plate image is generated by combining the initial license plate image and the compensated local image.

[0215] S3.5: Perform quality assessment on the license plate image and output the assessment result, wherein the quality assessment includes the detection of character edge clarity, stroke breakage, outer frame straightness, color density consistency and character position deviation;

[0216] Specifically, quality assessment uses license plate images as a unified reference, employing a combined geometric and appearance index to evaluate characters and outlines, avoiding threshold drift caused by multiple camera positions or inconsistent coordinates. The evaluation covers key aspects such as character edge sharpness, stroke continuity, outline straightness, color density consistency, and character position deviation. The judgment criteria are derived from process formulations or experience templates and can be adaptively fine-tuned batch by batch; these details are not elaborated upon here.

[0217] In this embodiment, the sharpness of character edges is evaluated by local gradient and edge stability measurement, and continuity is determined by the break count and minimum continuity length after the skeleton is refined; the straightness of the outer frame is measured by the deviation of the four-sided contour, and the degree of corner closure is checked; the color density consistency is achieved by statistically analyzing the brightness and color distribution in the character area and the background area respectively, and detecting uneven coating or transfer voids; the character position deviation is measured by center deviation and spacing deviation under a unified coordinate system based on the layout template.

[0218] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A smart license plate making machine control system, applied to a license plate making control device, characterized in that, The license plate production control device includes a controller and a correction mechanism. The correction mechanism is located at the gripping end of the robotic arm of the intelligent license plate production machine and is used to correct the position and posture of the license plate when the robotic arm grips it. The robotic arm also includes a narrow-field camera, which is disposed at the end of the robotic arm. The system includes: The instruction acquisition module is used to acquire manufacturing instructions and process formulas, and control the robotic arm to grab letter molds and license plate blanks that match the process formulas according to the manufacturing instructions. The license plate manufacturing module is used to complete the pressing and heat transfer process under the control of the controller to obtain the heat-transferred license plate. The license plate restoration module is used to acquire multiple local images during the movement of the robotic arm using the narrow-field camera, combine the posture parameters of the robotic arm end with the camera calibration information, project the local images onto the license plate plane coordinate system, and stitch them together according to the rasterized scanning path to obtain the license plate image. The license plate detection module is used to determine the quality of the license plate image. If the quality is qualified, the license plate is transferred to the license plate collection exit and a corresponding production number is generated and synchronized to the background. If the quality is unqualified, the license plate is put into the defective product collection box and the unqualified information is recorded. The system also includes a correction module, which is configured with correction logic. This correction logic is used to correct the position and orientation of the license plate when the robotic arm grasps it, including: After the license plate is attracted, negative pressure feedback is collected at the grasping end, wherein the negative pressure feedback includes: The adjustment direction is determined based on the changing trend of the negative pressure feedback, and the grasping posture and planar position of the grasping end are corrected until the feedback signal stabilizes. After the feedback signal stabilizes, maintaining the gripping posture of the gripping end, the license plate is moved to the reference edge of the mold. The correction mechanism, in conjunction with the controller's mode switching, enters the virtual fixture mode to obtain a compensation signal. Based on this compensation signal, posture compensation is performed on the gripping end, including: While maintaining the license plate adsorption, the robotic arm is controlled to move the license plate along the normal direction to the reference edge of the mold, and the contact state is determined according to the change of joint torque. When the contact state is determined, the normal advance of the robotic arm is stopped, and it is switched to sliding along the tangential direction of the reference edge, while maintaining the license plate adsorption state. During the sliding process, the contact feedback signal is monitored. When the contact feedback signal is a symmetrical signal, it is determined that the license plate is aligned with the reference edge. The robotic arm is controlled to slide to the positioning angle and perform an attitude correction operation to make the long side of the license plate parallel to the reference side and the short side perpendicular to the reference side, thereby obtaining the corresponding compensation signal.

2. The intelligent license plate production machine control system according to claim 1, characterized in that, The card-making and processing module includes: The letter mold assembly unit is used to control the robotic arm to pick up the letter mold and place it in the positioning position of the preset mold, pick up the license plate blank and feed it to the predetermined position directly above the lower letter mold, and then pick up the upper letter mold and place it directly above the license plate blank to form a mold assembly. The pressing and forming unit is used to control the robotic arm to move the mold assembly to the pressing station, and to press and form the license plate blank according to the pressure and holding time parameters of the process formula to obtain a license plate semi-finished product. The heat transfer unit is used to control the robotic arm to transfer the semi-finished license plate to the heat transfer station, and to perform heat transfer processing according to the temperature and transfer time parameters of the process formula to obtain the heat-transferred license plate. The character mold returning unit is used to control the robotic arm to return the upper and lower character molds to the character mold slots and exit the card making station after the heat transfer process is completed.

3. The intelligent license plate production machine control system according to claim 1, characterized in that, The license plate restoration module includes: The scanning path generation unit is used to generate a rasterized scanning path based on the planar range of the license plate; An image acquisition unit is used to control the movement of the robotic arm along the rasterized scanning path and to acquire multiple local images using the equidistant distance of the end on the license plate plane as a trigger condition. An initial image generation unit is used to combine the end-effector's pose and preset narrow-field camera calibration information to project the multiple local images onto the license plate plane coordinate system, and stitch them together according to the grid position to obtain an initial license plate image. An image fine-tuning unit is used to extract character anchor points from the initial license plate image using an edge extraction algorithm, wherein the character anchor points include at least one of the character outer frame corner points, stroke intersection points, and character baselines; and to fine-tune the initial license plate image using the character anchor points as constraints to generate a license plate image.

4. The intelligent license plate production machine control system according to claim 3, characterized in that, The license plate restoration module is equipped with an image restoration strategy. The image restoration strategy reconstructs the entire license plate image based on multiple local images captured by the narrow-span camera and outputs a license plate image for quality judgment. The image restoration strategy includes path planning logic, image projection logic, and image fine-tuning logic. The path planning logic is configured in the scanning path generation unit, the image projection logic is configured in the initial image generation unit, and the image fine-tuning logic is configured in the image fine-tuning unit.

5. The intelligent license plate production machine control system according to claim 4, characterized in that, The path planning logic includes: The main scanning direction is determined by the long side of the license plate and the baseline of the characters, and the vertical direction of the main scanning direction is determined by the line connecting the center line of the license plate outer frame and the mounting hole. A license plate plane coordinate system aligned with the license plate layout is established and the corresponding coverage boundary is determined. Based on the license plate layout, the coverage boundary is divided into an outer frame edge area and multiple character areas, and grid parameters are set for each area. The grid parameters include the scan line spacing, the overlap of adjacent scan lines, and the boundary margin. The scan line spacing is determined based on the effective field of view of the narrow-angle camera and the preset overlap. Intersecting calibration lines perpendicular to the main scanning direction are inserted into the coverage boundary at preset intervals, and the intersection of the intersecting calibration lines and the main scanning lines serves as alignment nodes. A set of mutually parallel main scan lines is generated in each character area according to the main scan direction; For each main scan line, calculate the start and end points, where the start and end points are located on the lines corresponding to the alignment nodes; The main scan lines are arranged in a serpentine sequence so that the end point of an adjacent main scan line is adjacent to the start point of the next main scan line, thus obtaining an initial grid. The initial grid is translated and rotated for fine adjustment based on the layout offset of the license plate, so that the main scanning direction is parallel to the character baseline and the grid origin of the initial grid is anchored to the positioning mark in the upper left corner of the license plate. The output includes the main scan line sequence, the start and end coordinates of each main scan line, and the rasterized scan path with the coordinates of the alignment nodes.

6. The intelligent license plate production machine control system according to claim 4, characterized in that, The image projection logic includes: The attitude parameters of the robotic arm end effector at each trigger position are collected. The attitude parameters include the spatial position of the end effector and the attitude angle relative to the license plate plane. The preset narrow-field camera calibration information is invoked, which includes the camera's intrinsic parameters and corresponding extrinsic parameters; The attitude parameters and the extrinsic parameters are combined to obtain the projection matrix of the camera in the license plate plane coordinate system corresponding to each local image; Based on the intrinsic parameters, the pixel coordinates of each local image are converted into a point set in the license plate plane coordinate system through the projection matrix; The point set is normalized and mapped according to the rasterized scanning path to obtain the initial license plate image.

7. The intelligent license plate production machine control system according to claim 4, characterized in that, The image fine-tuning logic includes: Anchor point constraints are established based on the license plate layout. These constraints include constraints on the straightness and parallelism of character baselines, spacing and column alignment of adjacent characters, and parallelism and verticality of character outlines and license plate outlines. For each local image at its stitching position in the grid, the character anchor point deviation within the corresponding coverage area is detected. Without changing the grid index, the compensation parameters for the corresponding local image are calculated based on the character anchor point deviation, so that the character anchor points within the coverage area of ​​the local image satisfy the anchor point constraint relationship. The local image is compensated according to the compensation parameters, and a license plate image is generated by combining the initial license plate image and the compensated local image.

8. The intelligent license plate production machine control system according to claim 1, characterized in that, The license plate detection module includes: The image analysis unit is used to receive the license plate image output by the license plate restoration module, segment the character region and the outer frame region, and extract image features for detection. The quality detection unit performs quality determination based on the image features. The quality determination includes at least the detection of character edge sharpness, the detection of character stroke breakage, the detection of character outline straightness, the detection of color density consistency of characters and outlines, and the detection of character position deviation. The judgment output unit is used to generate a qualified signal when the quality inspection is qualified, control the robotic arm to transfer the license plate to the license plate collection exit and generate the corresponding production number to be synchronized to the background; when the quality inspection is unqualified, it generates an unqualified signal, controls the robotic arm to put the license plate into the defective product collection box and records the unqualified information.

Images