Multi-color FDM printing real-time visual pose correction method and 3D printer

Abstract

The present application relates to the technical field of 3D printer calibration, and specifically provides a multi-color FDM printing real-time visual pose correction method and a 3D printer.The pose correction method comprises the following steps: obtaining an image of a printing layer based on a camera above a printing platform; obtaining image information at a nozzle based on a camera of a printing head; extracting physical space coordinates of reference feature points based on the collected image; solving pose parameters by an improved PnP algorithm based on the physical parameter coordinates of the reference feature points; and adjusting printing parameters based on the pose solving result and real-time monitoring data.The present application uses the color boundary of the printing material itself as a natural visual reference, effectively avoiding the cumbersome operation of manually pasting physical markers and the risk of marker failure caused by high temperature.

Description

Technical Field

[0001] This invention relates to the field of 3D printer calibration technology, and in particular to a real-time visual pose correction method for multi-color FDM printing and a 3D printer. Background Technology

[0002] In the FDM (Fused Deposition Modeling) printing process, the pose accuracy of the printed parts directly affects the molding quality, and high-precision control is required through real-time visual correction.

[0003] In existing technologies, pose correction methods generally rely on manually affixing physical markers (such as QR codes, metal reference points, etc.) to the surface of the printed part or construction platform. These markers serve as visual positioning references, and visual sensors acquire images of the markers and calculate pose deviations to complete the correction. However, the above methods have significant drawbacks: on the one hand, manually affixing physical markers requires additional labor costs, and the placement and angle of the markers need to be manually calibrated, increasing operational complexity and time costs, especially in mass production or complex structure printing scenarios, resulting in low efficiency; on the other hand, the nozzle and printed part surface temperatures are high during FDM printing (typically exceeding 100°C, and for some materials reaching over 200°C), and the physical markers are prone to adhesive failure, deformation, or detachment due to high temperatures. This causes the visual sensor to fail to recognize the positioning reference, leading to pose correction interruption, which in turn causes problems such as accumulated printing deviations and printing failures, seriously affecting printing stability and reliability. To solve the above problems, this application proposes a real-time visual pose correction method for multi-color FDM printing and a 3D printer. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time visual pose correction method and a 3D printer for multi-color FDM printing, so as to solve the problem that current FDM printing visual pose correction methods rely on manual marking.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A real-time visual pose correction method for multi-color FDM printing, the pose correction method comprising the following steps: Images of the printed layer are acquired using a camera above the printing platform; image information at the nozzles is acquired using a camera on the print head. Extract the physical space coordinates of reference feature points from the acquired images; Pose parameters are calculated using an improved PnP algorithm based on the physical parameter coordinates of reference feature points. Based on the pose calculation results and real-time monitoring data, the printing parameters are adjusted.

[0006] Furthermore, the image processing method for the camera above the printing platform includes the following steps: Correcting lens distortion based on camera calibration parameters; Extract the feature channels of the target color, generate a single-color mask image, synchronize with timestamps and coordinates, and remove background pixels outside the printing area; Enhance the edge color of the target; The pixel coordinates of the effective boundary features are converted into physical space coordinates through the extrinsic parameter matrix.

[0007] Furthermore, when enhancing the edge color of the target edge, the line segment length, straightness error, and consecutive frame offset are set to filter valid boundary features that meet the conditions.

[0008] Furthermore, when removing background pixels, isolated noise points with an area smaller than a preset value are also removed.

[0009] Furthermore, the image processing method for the camera captured by the print head includes the following steps: Distortion correction is performed on the acquired images based on camera intrinsic parameters, distortion coefficients, and extrinsic parameter matrices. Based on the preset installation position of the nozzle, a dedicated ROI for the nozzle feature points is defined in the image; The nozzle marker points are separated by threshold segmentation; Accurate contours of marker points are located through edge detection; For the filtered circular contours, the pixel coordinates of the marker points are obtained by Hough circle detection; The pixel coordinates of the marker points are converted into normalized image coordinates.

[0010] Furthermore, the method for separating nozzle marker points includes the following steps: Perform weighted grayscale on the ROI region; The optimal threshold T is automatically calculated using the Otsu method, and the grayscale values ​​are then... Figure 2 Values ​​are converted, and pixels with grayscale values ​​less than a preset value are set as foreground pixels, while the remaining pixels are set as background pixels. The background pixels are nozzle pixels.

[0011] Furthermore, when acquiring the marker points, the confidence level is calculated based on the edge response value of the marker points according to formula (4). : ;Formula (4) in, is the edge response value of the i-th nozzle feature point.

[0012] Furthermore, the calculation of pose parameters includes the following steps: Weights are assigned to each feature point; Construct a weighted projection error objective function; Iterative optimization and pose parameter calculation are performed based on the Levenberg-Marquardt (LM) algorithm; Convert the pose parameters calculated by the LM algorithm into actual values ​​in the printing coordinate system.

[0013] Furthermore, the weighted projection error objective function is shown in formula (5): ;Formula (5) in, For projection function, Let be a rotation matrix. It is a translation vector. This is the camera intrinsic parameter matrix at the printhead. To print the intrinsic parameter matrix of the camera above the printing platform, Here are the normalized image coordinates of the i-th nozzle feature point. Normalized image coordinates of the j-th color boundary feature point The weights of nozzle feature points The weights of color boundary feature points 3D coordinates of the i-th nozzle feature point in the printing reference coordinate system The 3D coordinates of the j-th color boundary feature point in the printing reference coordinate system.

[0014] The present invention also discloses a 3D printer, wherein the pose correction method of the printer during operation includes the above-mentioned real-time visual pose correction method for multi-color FDM printing.

[0015] In summary, the present invention has the following advantages compared with the prior art: The real-time visual pose correction method for multi-color FDM printing disclosed in this invention uses the color boundary of the printing material itself as a natural visual reference, effectively avoiding the tedious operation of manually pasting physical marks and the risk of mark failure caused by high temperature; by using dual cameras to collect multi-source visual data in real time and combining it with an improved weighted PnP algorithm for fusion processing, the accuracy and robustness of pose calculation are significantly improved, and the geometric accuracy and interlayer color alignment quality of multi-color FDM printed parts are significantly improved. Attached Figure Description

[0016] Figure 1 The system architecture diagram for implementing the real-time visual pose correction method for multi-color FDM printing disclosed in the embodiments of the present invention is shown.

[0017] Figure 2 This is a flowchart of the real-time visual pose correction method for multi-color FDM printing disclosed in an embodiment of the present invention.

[0018] Figure 3This is a flowchart of the image processing method for the camera above the printing platform in the real-time visual pose correction method for multi-color FDM printing disclosed in an embodiment of the present invention.

[0019] Figure 4 This is a flowchart of the image processing method for the camera capturing images from the print head in the real-time visual pose correction method for multi-color FDM printing disclosed in an embodiment of the present invention.

[0020] Figure 5 This is a flowchart of the method for calculating pose parameters in the real-time visual pose correction method for multi-color FDM printing disclosed in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Example 1 like Figure 1 As shown, an embodiment of the present invention provides a real-time visual pose correction method for multi-color FDM printing. The system architecture on which the correction method relies includes a dual-view image acquisition module, an edge computing unit, a task mapping module, and a parameter storage module.

[0023] The multi-dimensional information acquisition module includes two cameras located at different positions on the printer, as well as a linear laser emitter, an infrared thermometer, and a gloss meter. One camera is mounted above the printing platform, and this position integrates the linear laser emitter, infrared thermometer, and gloss meter to acquire information about the printed layer. The other camera is located on the side of the print head, with its lens facing the nozzle, to acquire information about the spatial orientation of the nozzle and the extrusion status of the consumables. The data from the two cameras, the linear laser emitter, the infrared thermometer, and the gloss meter are transmitted wirelessly via WiFi or via wired connection, and a clock module is configured to synchronize the timestamps of the acquired information.

[0024] The edge computing unit uses a computing chip, such as the RK3588 chip, which integrates image preprocessing, feature extraction, pose calculation, correction decision, anomaly handling and parameter adaptation algorithms to achieve rapid data processing. The edge computing unit can also use the data MCU of the 3D printer to achieve the merging and processing of print data.

[0025] The task mapping module is used to map position and posture correction information (printer operating parameters, such as filament extrusion speed, filament melting temperature, printhead position, etc.). After the edge computing unit calculates the position and posture correction information, it is sent to the printer's actuators, such as the printhead fine-tuning structure and platform heating mechanism, through the task mapping module to realize position and posture adjustment and parameter regulation.

[0026] The parameter storage module is used to store consumable parameter data, camera calibration parameters (intrinsic parameter matrix, distortion coefficient, extrinsic parameter matrix), platform deformation calibration data (initial deformation matrix, thermal expansion coefficient), laser extraction material adaptation parameter table, color feature extraction material and lighting adaptation parameter table (including complete Lab threshold), and other data required and generated during pose correction. It can be set separately or use the storage structure of the 3D printer.

[0027] like Figure 2 As shown, the pose correction method includes the following steps: Step S100: Acquire an image of the printed layer using a camera above the printing platform; acquire image information of the nozzle using a camera on the print head; In this embodiment, the camera, linear laser emitter, infrared thermometer, and gloss meter need to be initialized and calibrated before operation, including loading the camera's intrinsic and extrinsic parameter calibration data, constructing the laser plane equation, and synchronizing the timestamps of each sensor.

[0028] The camera calibration method is as follows: based on a preset reference board, the intrinsic parameter matrix of the two cameras (such as the camera intrinsic parameter matrix at the printhead) is determined using the Zhang Zhengyou calibration method. The image pixel coordinates and the printing physical space coordinates are established by using distortion coefficients (radial distortion k1=-0.04, k2=0.005; tangential distortion p1=0.001, p2=-0.0005) and extrinsic parameter matrices (rotation matrix R, translation vector T).

[0029] The linear laser emitter calibration method is as follows: Adjust the linear laser emitter so that the laser stripe is parallel to the printing platform, and calibrate the mapping relationship between the laser stripe pixel offset and the physical height (e.g., 1 pixel = 0.002 mm) using a standard gauge block (accuracy ±0.001 mm). Orthogonal experiment parameter library loading: Based on the user's selected multi-color consumable combination (including special filling consumables), load the corresponding orthogonal experiment data matrix, fitting model (including special consumable adjustment coefficients), error analysis results, significance test data, and new consumable access verification standards (including qualification judgment thresholds and parameter library update rules). Illumination baseline calibration: Collect the mean value of the background V channel of the printing area under the initial ambient light (e.g., ), serving as the baseline value for adaptive lighting adjustment; Lab color space calibration: Define the range of Lab space values ​​(e.g., L∈[0,100], a∈[-128,127], b∈[-128,127]), and calibrate the channel conversion accuracy using a standard color chart (e.g., X-RiteColorCheckerPassport) to ensure that the measurement error of Δa / Δb is ≤1.

[0030] In this embodiment, the timestamp synchronization error between the two cameras is controlled within 1ms.

[0031] Step S200: Extract the physical space coordinates of the reference feature points based on the acquired image; The edge computing unit performs the following processing on the images captured by the camera at the printing platform: S111. Correct lens distortion based on camera calibration parameters; normalize the image grayscale (map pixel values ​​to [0,255]), and use a bilateral filtering algorithm (window 5×5, σs=5, σr=20) to filter noise; S112, Color Channel Separation and Multi-Dimensional Adaptation: Convert the RGB image to HSV color space, extract the feature channels of the target color based on the current printing color combination, and generate a single-color mask image; through timestamp synchronization and coordinate registration, remove background pixels outside the printing area (e.g., in this case, coordinate range: X∈[5mm,295mm], Y∈[5mm,195mm], adapted to a 300×200mm printing platform). Preferably, in this embodiment, when removing background pixels, isolated noise points with an area of ​​less than 3 pixels are also removed, and color boundary segments with a continuous number of pixels of ≥5 are retained; S113, Enhance the edge color of the target edge; To enhance the edge color of target objects: The Sobel operator (3×3 convolution kernel) is used to calculate the X / Y axis gradient, and a gradient threshold is set (in this embodiment, the gradient threshold = 50), retaining edge pixels with gradients greater than the threshold; a Gaussian weighted Laplacian operator (σ = 1.2, 3×3 convolution kernel) is superimposed to enhance edge details and suppress noise; pixel difference operations are performed on adjacent color mask images (difference > 50) to mark candidate boundary points; one dilation operation (filling edge gaps) and one erosion operation (removing burrs) are performed using 3×3 rectangular structuring elements to optimize boundary continuity; Preferably, in this embodiment, feature purification is also included: Set the line segment length, straightness error (using least squares to fit the line and calculate the deviation value), and consecutive frame offset, then filter valid boundary features that meet the conditions: In this embodiment, the line segment length is ≥ 5 pixels (physical length ≥ 0.01 mm); the straightness error is ≤ 0.05 (the least squares method is used to fit the straight line, and the deviation value / line segment length is ≤ 0.05); the position offset of 3 consecutive frames is ≤ 1 pixel (physical offset ≤ 0.002 mm). It should be noted that in this embodiment, when the number of effective feature points is less than 8, feature point re-extraction is triggered, that is, the sampling density is increased (from taking 1 point every 5 mm to taking 1 point every 3 mm), the number of nozzle feature points is increased (from 5 to 8), and the maximum number of re-extraction iterations is 3. S114. Convert the pixel coordinates of the effective boundary features into physical space coordinates (unit: mm) using the extrinsic parameter matrix. The conversion formulas are shown in formulas (2) and (3): Formula (2) , formula (3) in, The x-coordinate is the pixel coordinate. Principal point x-coordinate offset Horizontal equivalent focal length This represents the X-axis value of the physical space coordinates. The vertical coordinate is the pixel coordinate. The offset of the ordinate of the main point. The equivalent focal length in the vertical direction. This represents the Y-axis value in physical space coordinates. Provided by laser ranging data.

[0032] It should be noted that in this embodiment, if there is a problem of laser stripe occlusion in the image, that is, the laser stripes are incomplete, the laser stripes are supplemented by the following steps: When the laser stripe is missing ≥3 consecutive sampling points, cubic spline interpolation is used to complete it. The formula for solving the nodes is as follows: Let the interpolation nodes be (3 valid points before and after occlusion), interpolation function ; By the continuity condition of the first derivative: ,have to Equation (1) By the continuity condition of the second derivative: ,have to Equation (2) Boundary conditions: Solving Equation (3) Solving for coefficients based on equations (1), (2), and (3) .

[0033] The camera image processing steps for the printing end are as follows: S211. Image preprocessing: Based on camera intrinsic parameters, distortion coefficients, and extrinsic parameter matrices, distortion correction is performed on the acquired images; The image grayscale values ​​of the nozzle area are mapped from [0, 255] to enhance the contrast between the marker points and the nozzle's metal surface. A 5×5 filter window is used. ), filtering thermal noise and camera noise while preserving edge details to avoid blurring of marker edges; S212. Based on the preset installation position of the nozzle, define the nozzle feature point-specific ROI in the image.

[0034] S213. Separate the nozzle marker points from the metallic background using threshold segmentation, as follows: Perform weighted grayscale conversion on the ROI region, using the following formula: The color image is converted to a grayscale image, and the difference in brightness of the focus markers is used. Grayscale value , , These are the RGB values ​​of the image; The optimal threshold T is automatically calculated using the Otsu method, and the grayscale values ​​are then... Figure 2 Pixels with a grayscale value less than T (black ceramic markers) are set as foreground, while pixels with a grayscale value greater than or equal to T (nozzle metal surface) are set as background.

[0035] S214. Locate the precise contour of the marker points through edge detection, as follows: Perform Canny operator edge detection on the binarized image, with the parameters set as follows: Gaussian filter. (Smooth noise); Low threshold = 50, High threshold = 150; Output complete edge contours of marker points, removing stray edges from the nozzle's metal surface; Contour filtering calls contour extraction functions (such as OpenCV's findContours) to obtain all edge contours. The contours of the marker points are then filtered according to preset ranges for contour area and contour roundness, as shown in this embodiment. Pixels; Outline roundness: (Roundness formula (Roundness)) ,in, Let be the area of ​​the shape's outline. (Total length of the shape's boundary), ensure that the marker points are circular, and remove strip-shaped edges.

[0036] S215. For the filtered circular contours, obtain the pixel coordinates of the marker points using Hough circle detection: Hough circle detection parameter settings, as in this embodiment, are adapted to the pixel size of a 1mm diameter marker point, with the following parameters: minimum radius = 3 pixels, maximum radius = 8 pixels; accumulator threshold = 100 (to suppress pseudo-circle interference); center-to-center distance ≥ 6 pixels; For the detected circular contour, extract the coordinates of its center pixel. 3 or 8, corresponding to the preset number of markers), and record the edge response value of each marker (for subsequent confidence determination).

[0037] In this embodiment, this step also includes feature point verification and post-processing: Quantity and position verification: Quantity verification: If the number of detected marker points is less than the preset number (e.g., less than 3), feature point re-extraction is triggered—expanding the ROI range (X∈[480,800], Y∈[280,440]) and lowering the Canny threshold to 40, with a maximum of 3 iterations; if it fails 3 times, the valid coordinates of the previous frame are used for padding; Position verification: Calculate the pixel spacing between marker points. If the deviation from the preset spacing (6 pixels, corresponding to 2mm physical spacing) is ≤1 pixel, it is judged as a false feature point and removed.

[0038] This embodiment also includes edge response value confidence calculation: Calculate the confidence level of the edge response value of the marker point based on formula (4). (Range 0-1): Formula (4), where, The edge response value of the i-th nozzle feature point is used. Points with a confidence level < 0.7 are marked as "low confidence points" and their weight is reduced (from 0.6 to 0.3) during subsequent pose calculations.

[0039] S216, Coordinate Normalization: Convert the pixel coordinates of the marker points to normalized image coordinates to eliminate the influence of camera intrinsic parameters. The formula is as follows: ,in Let these be the coordinates of the camera's principal point. It is the equivalent focal length.

[0040] The output of this step is the pixel coordinates and normalized coordinates of the nozzle feature points, and the confidence score of each feature point.

[0041] Step S300: Based on the physical parameter coordinates of the reference feature points, the pose parameters are calculated using the improved PnP algorithm. The edge computing unit performs pose calculation based on the extracted valid feature points. The specific process is as follows: S310. Assign weights to each feature point. In this embodiment, the nozzle feature point has a weight of 0.6 (the weight is reduced to 0.3 for low confidence points), the color boundary feature point has a weight of 0.4 (the weight is increased to 0.5 for curve boundary feature points). S320. Construct the weighted projection error objective function, as shown in formula (5). ;Formula (5) in, For projection function, Let be a rotation matrix. It is a translation vector. This is the camera intrinsic parameter matrix at the printhead. To print the intrinsic parameter matrix of the camera above the printing platform, Here are the normalized image coordinates of the i-th nozzle feature point. Normalized image coordinates of the j-th color boundary feature point The weights of nozzle feature points The weights of color boundary feature points 3D coordinates of the i-th nozzle feature point in the printing reference coordinate system The 3D coordinates of the j-th color boundary feature point in the printing reference coordinate system; S330. Iterative optimization and pose parameter calculation based on the Levenberg-Marquardt (LM) algorithm: The rotation matrix... Translation vector Using the result from the previous frame as the initial value, substitute it into the objective function to calculate the initial error. ; For projection function Regarding pose parameters (Euler angles of rotation matrix R) The translation vector t Taking the partial derivative, we obtain the Jacobian matrix J; then we construct an approximation of the Hessian matrix. (W is a weighted diagonal matrix, with diagonal elements as follows) Simultaneously, the error residual vector (r = (observed coordinates - projected coordinates)) is calculated; Through damping factor Balanced iteration step size: If the current iteration error Less than the previous error This indicates that the step size is reasonable and should be reduced. (Multiplied by 0.1) enhances the characteristics of the Gauss-Newton method to accelerate convergence; if ( > This indicates that the step size is too large; increase it. (Multiply by 10) and switch to gradient descent mode to ensure stability; Solve the regularized equation (I is the identity matrix, (For parameter increments), update pose parameters: Rotation matrix update: via Euler angle increments ,Will Updated to Translation vector update: ; If the preset conditions are met, the iteration stops and the current pose parameters are output. , : Number of iterations ≤ 1200 (can be increased to 1500 for high reflectivity conditions); Parameter change between two adjacent iterations Weighted projection error .

[0042] Step S330: Convert the pose parameters calculated by the LM algorithm into actual values ​​in the printing coordinate system using formula (6). That is, the coordinates of the reference feature point of the printhead / nozzle in the printing coordinate system: ;Formula (6) in, The coordinates of the center of the nozzle end face are: .

[0043] Step S400: Based on the pose calculation results and real-time monitoring data, adjust the printing parameters. The specific steps are as follows: The difference between the actual and theoretical values ​​of the reference feature points of the nozzle camera is calculated to obtain the nozzle spatial offset. The difference between the actual and theoretical values ​​of the reference feature points of the platform camera is calculated to obtain the color boundary deviation. The difference between the measured actual floor height and the theoretical floor height is calculated to obtain the floor height deviation. Printing parameters are controlled based on the printer's built-in adjustment methods and calculated deviations.

[0044] Example 2 A 3D printer, wherein the pose correction method during printer operation includes the real-time visual pose correction method for multi-color FDM printing described in Example 1.

[0045] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0046] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time visual pose correction method for multi-color FDM printing, characterized in that, The pose correction method includes the following steps: Images of the printed layer are acquired using a camera above the printing platform; image information at the nozzles is acquired using a camera on the print head. Extract the physical space coordinates of reference feature points from the acquired images; Pose parameters are calculated using an improved PnP algorithm based on the physical parameter coordinates of reference feature points. Based on the pose calculation results and real-time monitoring data, the printing parameters are adjusted.

2. The real-time visual pose correction method for multi-color FDM printing according to claim 1, characterized in that, The image processing method for images captured by a camera above the printing platform includes the following steps: Correcting lens distortion based on camera calibration parameters; Extract the feature channels of the target color, generate a single-color mask image, synchronize with timestamps and coordinates, and remove background pixels outside the printing area; Enhance the edge color of the target; The pixel coordinates of the effective boundary features are converted into physical space coordinates through the extrinsic parameter matrix.

3. The real-time visual pose correction method for multi-color FDM printing according to claim 2, characterized in that, When enhancing the color of edge targets, set the line segment length, straightness error, and consecutive frame offset to filter valid boundary features that meet the conditions.

4. The real-time visual pose correction method for multi-color FDM printing according to claim 2, characterized in that, When removing background pixels, isolated noise points with an area smaller than a preset value are also removed.

5. The real-time visual pose correction method for multi-color FDM printing according to claim 1, characterized in that, The image processing method for the camera capturing the printhead includes the following steps: Distortion correction is performed on the acquired images based on camera intrinsic parameters, distortion coefficients, and extrinsic parameter matrices. Based on the preset installation position of the nozzle, a dedicated ROI for the nozzle feature points is defined in the image; The nozzle marker points are separated by threshold segmentation; Accurate contours of marker points are located through edge detection; For the filtered circular contours, the pixel coordinates of the marker points are obtained by Hough circle detection; The pixel coordinates of the marker points are converted into normalized image coordinates.

6. The real-time visual pose correction method for multi-color FDM printing according to claim 5, characterized in that, The method for separating nozzle marker points includes the following steps: Perform weighted grayscale on the ROI region; The optimal threshold T is automatically calculated using the Otsu method. The grayscale image is then binarized, and pixels with grayscale values ​​less than a preset value are set as foreground pixels, while the remaining pixels are set as background pixels. The background pixels are nozzle pixels.

7. The real-time visual pose correction method for multi-color FDM printing according to claim 5, characterized in that, When acquiring the marker points, the confidence level is also calculated based on the edge response value of the marker points according to formula (4). : Official (4) in, is the edge response value of the i-th nozzle feature point.

8. The real-time visual pose correction method for multi-color FDM printing according to claim 1, characterized in that, Solving the pose parameters includes the following steps: Weights are assigned to each feature point; Construct a weighted projection error objective function; Iterative optimization and pose parameter calculation are performed based on the Levenberg-Marquardt (LM) algorithm; Convert the pose parameters calculated by the LM algorithm into actual values ​​in the printing coordinate system.

9. The real-time visual pose correction method for multi-color FDM printing according to claim 8, characterized in that, The objective function for the weighted projection error is shown in formula (5): Official (5) in, For projection function, For rotation matrix, It is a translation vector. This is the camera intrinsic parameter matrix at the printhead. To print the intrinsic parameter matrix of the camera above the platform, Here are the normalized image coordinates of the i-th nozzle feature point. Normalized image coordinates of the j-th color boundary feature point The weights of nozzle feature points The weights of color boundary feature points 3D coordinates of the i-th nozzle feature point in the printing reference coordinate system The 3D coordinates of the j-th color boundary feature point in the printing reference coordinate system.

10. A 3D printer, characterized in that, The posture correction method for the printer during operation includes the real-time visual posture correction method for multi-color FDM printing as described in any one of claims 1-9.

Images