Vision-based cutting path generation system, cutting equipment, and cutting method

Abstract

This invention relates to a vision-based cutting path generation system, cutting equipment, and cutting method. The system includes a scanning camera control module, an image acquisition and recognition module, an image processing module, a cutting path planning module, and a cutting control module. The system addresses boundary recognition issues when patterns and backgrounds are of the same color or have low contrast by combining Laplacian algorithm, LBP, and GLCM texture extraction techniques. It employs contour optimization and curvature compensation algorithms to improve path smoothness and cutting quality, and optimizes the cutting sequence using a weighted function of area, complexity, and overlap to reduce material deformation. PID control dynamically adjusts the motor drive to achieve stable cutter speed control. This invention offers advantages such as high recognition accuracy, strong adaptability, intelligent path planning, and stable cutting process, significantly improving the automatic cutting efficiency and finished product quality of complex patterns or materials of the same color, with overall cutting accuracy fluctuations controlled within ±2%.

Description

Technical Field

[0001] This invention relates to the field of shoe material processing technology, specifically to a vision-based cutting path generation system, cutting equipment, and cutting method, which is particularly suitable for applications involving precise cutting of shoe materials using a cutting blade. Background Technology

[0002] In the footwear material processing industry, traditional pattern and color-matching material cutting techniques have significant drawbacks.

[0003] On the one hand, the lack of precise visual positioning methods makes it difficult to accurately identify the boundaries of complex patterns and materials of the same color with minimal color difference (≤5ΔE) on shoe materials, resulting in deviations in the cutting position and seriously affecting the quality and aesthetics of the finished shoe materials.

[0004] On the other hand, traditional methods rely solely on a single RGB color space for differentiation and lack sufficient texture feature extraction capabilities, making it difficult to effectively identify the boundaries of regions with similar colors but different textures, resulting in low cutting accuracy and efficiency.

[0005] In addition, traditional cutting systems cannot automatically adjust cutting parameters according to the characteristics of different materials such as leather and fabric (such as hardness and ductility), which can easily cause problems such as material tearing and accelerated tool wear, making it difficult to meet the diversified and high-precision production needs of modern shoe material processing. Summary of the Invention

[0006] To address the problems existing in current technologies, this invention aims to achieve precise cutting of shoe material patterns and materials of the same color through innovative technical means. Specifically, this invention overcomes the limitations of traditional single-color feature recognition by integrating HSV color space segmentation, LBP (Local Binary Pattern), and GLCM (Gray-Level Co-occurrence Matrix) texture feature extraction technologies. It achieves sub-pixel-level contour recognition with a matching accuracy of over 99%, accurately distinguishing boundaries even when colors are the same through texture differences. Simultaneously, a material property weighting model is constructed, combined with Bézier curve smoothing and curvature calculation algorithms, to automatically optimize the cutting path and parameters based on different materials. This reduces tool wear by 30% and cut edge burrs by up to 50%, comprehensively improving the quality and efficiency of shoe material cutting.

[0007] In a first aspect, the present invention provides a vision-based cutting path generation system, applied in a cutting device, comprising:

[0008] The scanning camera control module is used to receive speed information from the encoder set in the cutting equipment, adjust the scanning rhythm and exposure parameters of the object to be cut, ensure that the image acquisition ratio matches the movement speed and acquire the acquired image, and ensure the clarity and stability of the acquired image.

[0009] The image acquisition and recognition module is used to receive the acquired image and automatically adjust the acquisition parameters by analyzing the grayscale distribution and brightness characteristics of the acquired image to ensure that the details of the acquired image are clearly visible and to generate a clear image.

[0010] The image processing module is used to receive the clear image and obtain the pattern boundary contour data of the clear image through a threshold segmentation algorithm or a texture algorithm; wherein

[0011] When the pattern edge color of the clear image is similar to that of the background, the system first enhances the texture edge details by using the Laplacian algorithm, then uses the LBP algorithm and GLCM algorithm to jointly extract the local and global texture features of the clear image, and combines the weighted cross-entropy algorithm and Dice loss function to optimize the model training, so that the system can accurately identify the pattern boundary contour of the clear image based on texture and shape features even in the absence of obvious color contrast.

[0012] The cutting path planning module is used to receive the pattern boundary contour data and generate the optimal cutting path by calculating the contour optimization algorithm formula, the arc compensation algorithm formula, and the cutting sequence algorithm formula.

[0013] The cutting control module is used to control the cutting equipment to perform cutting operations according to the cutting path generated by the cutting path planning module; and

[0014] The data storage and management module is used to store the acquired image, the clear image, the pattern boundary contour data, and the cutting path information.

[0015] In the texture algorithm of the image processing module,

[0016] The Laplace algorithm formula is:

[0017]

[0018] in, I represents the Laplacian transform of the image at coordinates (x, y), and I represents the input image (grayscale matrix). This refers to the second-order partial derivative of the image in the x-direction, used to detect changes in horizontal edges. The second-order partial derivative of the image in the y-direction is used to detect changes in vertical edges;

[0019] The formula for the LBP algorithm is:

[0020]

[0021] in, The coordinates of the center pixel are given, P is the number of neighboring pixels (e.g., 8-neighborhood), and R is the neighborhood radius (pixel distance). Refers to the grayscale value of the center pixel. This refers to the grayscale value of the p-th pixel in the neighborhood. This is the sign function, used to compare the grayscale differences between the neighboring pixels and the center pixel;

[0022] The formula for the GLCM algorithm is:

[0023]

[0024] in, Number of gray levels This represents the co-occurrence probability of grayscale values ​​i and j at a specified direction and distance.

[0025] The formula for the weighted cross-entropy algorithm is:

[0026]

[0027] Where N is the total number of pixels in the image, and C is the number of categories (e.g., foreground / background). The weights for class c (used to balance imbalanced samples). This refers to the true label (0 or 1) of pixel i in category c. This refers to the probability that the model predicts pixel i belongs to category c;

[0028] The formula for the Dice loss function is:

[0029]

[0030] Where X refers to the set of foreground pixels predicted by the model, and Y refers to the set of actual foreground pixels. This refers to the number of pixels where the prediction intersects with the actual foreground.

[0031] The image processing module further includes a method for identifying and locating patterns in the clear image that are irregularly arranged, randomly oriented, or subject to occlusion or lighting changes. This method finds the best match between the template pattern and any pattern in the current view under any angle, size, and lighting conditions, and achieves highly robust recognition and positioning to ensure accurate generation of the cutting path.

[0032] Among them, the formula obtained through the FAST algorithm is: Corner points are detected, and local features of the pattern are constructed by combining SIFT and BRIEF descriptors, and then the Euclidean distance formula is used: Hamming distance formula: It determines the similarity between patterns to achieve accurate matching between templates and target patterns, maintaining robust recognition accuracy even when patterns vary in scale, angle, or lighting; where P is the coordinate of the center pixel. For the i-th pixel in the neighborhood, The grayscale value of the center pixel. Here, is the grayscale value of the neighboring pixels, and thresh is the grayscale difference threshold used to determine whether a pixel is a corner. Refers to the i-th dimension component of two 128-dimensional SIFT descriptors. Refers to the i-th bit of the two binary BRIEF descriptors.

[0033] Among them, the similarity transformation formula is: Affine transformation formula: Formula for normalized cross-correlation coefficient: By combining these methods, the system can accurately align and match the template with the pattern in the view, even in complex scenarios involving pattern rotation, scaling, offset, or lighting changes, ensuring the accuracy and stability of subsequent path generation; among these, The coordinates of the template point, Here are the coordinates of the target point after transformation, s is the scaling factor, and θ is the rotation angle. The translation is denoted by , which controls the position of the template in the image. (x, y) are the template pixel coordinates, (u, v) are the corresponding coordinates in the target image, T is the pixel value of the transformed template, and I is the pixel value of the corresponding region in the target image. and This represents the average pixel value of the template and target regions.

[0034] In the cutting path planning module, the contour optimization algorithm formula is used: And the formula for the circular arc compensation algorithm: Together, they ensure the smoothness and safety of the cutting path, avoiding quality issues caused by jagged edges and sharp corners. Used in conjunction, they achieve mathematical smoothing of the contour and optimization of physical cutting adaptability, ensuring the generation of high-quality and suitable cutting paths even for complex patterns; and through the cutting sequence algorithm formula: By combining material properties and regional information, a scientific cutting sequence is determined to reduce cutting risks and processing time; among other things, These are control points on the contour, which determine the shape of the curve. Let be a k-th order basis function that satisfies local support, t be a parameter variable, and after traversal, a continuous and smooth curve is generated, t∈[0,1]. For the area, The larger the ratio, the higher the priority, to avoid material deformation when cutting large areas; For the contour complexity, The larger the value, the more complex the shape, and it should be cut first to reduce the number of tool adjustments. For regional overlap, A higher value indicates a higher proportion of intersection with other areas, prioritizing cutting to avoid secondary damage; This is a weighting coefficient, which is adjusted according to the material.

[0035] The cutting control module also employs a PID control algorithm formula: The motor voltage adjustment is calculated to ensure the cutter head speed precisely follows the target speed, avoiding a decrease in cutting quality caused by speed fluctuations; among other things, For proportional, integral, and differential coefficients, the proportional term is... The output directly responds to the current error; the larger the error, the larger the output. (Integration) Accumulate historical errors, eliminate system static errors, calculus terms Predict the trend of error changes and suppress system oscillations.

[0036] Secondly, the present invention provides a cutting device, comprising:

[0037] The system comprises an object platform, a crossbeam, a line scan camera, a cutting blade device, a feeding device, a receiving device, and a computer. The computer is electrically connected to and controls the operation of the crossbeam, the line scan camera, and the cutting blade device. The computer integrates a cutting path generation system based on intelligent vision positioning, as described above.

[0038] The object platform is used to support the object to be cut, and a crossbeam is slidably connected to the object platform along the Y-axis.

[0039] A feeding device is provided behind the object platform for conveying the object to be cut, and the feeding device has a smoothing mechanism to smooth out the wrinkles of the material.

[0040] A material collection device is installed in front of the object platform to collect the material after cutting.

[0041] A wire scanning camera is fixedly connected to the lower end face of the crossbeam, and a cutting blade device is slidably connected to the upper X-axis of the crossbeam.

[0042] The line scan camera includes several acquisition elements arranged along the X-axis. Each acquisition element includes an image sensor, a cylindrical lens, a glass window, an LED light source, and a processing chip. The image sensor is electrically connected to the processing chip, and the processing chip is electrically connected to a computer.

[0043] The cutting blade device includes a connecting plate, a blade holder, and a cutting blade. The connecting plate is slidably connected to the crossbeam along the X-axis, and the blade holder is slidably connected to the connecting plate along the Z-axis. The cutting blade is mounted on the blade holder, and a blade head position sensor is provided on the blade holder. The blade head position sensor is used to monitor the state of the blade head and feed back the position information to the computer.

[0044] The blade holder is also equipped with an inkjet printing device, which includes a U-shaped bracket, a printing cylinder, a connecting block, and a printing component. The printing component is a spray valve or an inkjet cartridge, used for marking or assisting in positioning on the material surface.

[0045] Thirdly, the present invention provides a cutting method, wherein the cutting method includes the following steps:

[0046] Step 1: Fix the paper on the platform, and use the cutting control module to control the printing cylinder to drive the printing component to print calibration grid lines on the paper surface;

[0047] Step 2: The camera control module controls the line scan camera to scan the paper with the grid drawn on it. The image data is received and stitched by the image acquisition and recognition module to complete the horizontal alignment of the camera image.

[0048] Step 3: Perform camera calibration. The image processing module generates a camera grid and compares it with the printed grid in the acquired image. If there is a deviation, the X and Y offsets of the camera grid are automatically adjusted until the origin consistency calibration is completed. The calibration image is then saved to the data storage and management module.

[0049] Step 4: Turn on the power of each device, start the feeding device, and the object to be cut is conveyed to the object platform through the feeding device. During this process, the smoothing mechanism of the feeding device completes the surface wrinkle smoothing operation of the object to be cut. The object platform moves the object to be cut forward, and the line scan camera captures the image of the object to be cut.

[0050] Step 5: The image acquisition and recognition module receives image data and feeds it back to the camera control module to dynamically adjust the light source brightness, thereby improving image contrast and edge sharpness;

[0051] Step 6: The acquired image is transmitted to the image processing module, the design draft is imported, and after selecting the pattern, it is aligned with the real object image by scaling and rotating to generate a feature template. The FAST algorithm is used to extract image features and generate a stable template to achieve multi-angle matching and object positioning.

[0052] Step 7: The image processing module sends the template matching result to the cutting path planning module, which generates the corresponding optimal cutting path based on the template overlay image;

[0053] Step 8: The cutting path is converted into a control command and sent to the cutting control module. The cutting control module controls the cutting blade to execute the corresponding path cutting. The blade position sensor monitors and feeds back position information in real time to ensure accuracy.

[0054] Step 9: After the object to be cut on the object platform is cut, the object platform moves the material to the receiving device. The cutting device continues to cut subsequent objects. After the material is cut, the equipment is turned off and cleaned. When the equipment is used again, the operation can directly enter the scanning and cutting process without recalibrating the camera.

[0055] The beneficial effects of the technical solution of this invention are:

[0056] The cutting path generation system of this invention can complete the contour recognition of a single image within 500ms through image processing algorithms, and achieve sub-pixel alignment between the product contour and the design template through template matching technology, with a matching accuracy of >99%. Furthermore, based on a threshold segmentation algorithm of the RGB / HSV color space, it can identify areas of the same color on the material surface with a color difference ≤5ΔE, supporting color-coded segmentation and improving material utilization. Simultaneously, by extracting micro-texture differences (such as fabric weaving methods and leather pore distribution), it can distinguish pattern boundaries even when colors are the same. Texture features are less sensitive to changes in illumination than color features, reducing positioning errors caused by light fluctuations.

[0057] The cutting path generation system of this invention effectively combines threshold segmentation algorithms in the RGB / HSV color space with texture algorithms, breaking through the limitations of traditional single-feature segmentation and constructing a multimodal feature fusion framework that can adaptively adjust weights according to different cutting materials. Furthermore, by combining material properties (such as hardness and ductility) and cutting requirements (such as serrated edges and smooth edges), it generates the shortest path or cutting route with the fewest turning times through genetic algorithms or AI algorithms, improving cutting efficiency by more than 30% while reducing tool wear (extending tool life by 20%). After cubic B-spline fitting of the original contour (serrated), tool wear rate is reduced by 30%, and burrs on the cutting edge are reduced by 50%. Attached Figure Description

[0058] Figure 1 This is a structural block diagram of the computer in the path generation system of this invention.

[0059] Figure 2 This is a schematic diagram of the cutting device in an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the cutting blade device in an embodiment of the present invention;

[0061] Figure 4This is a schematic diagram of the transmission mechanism of the cutting blade in an embodiment of the present invention;

[0062] Figure 5 This is a schematic diagram of the acquisition element of the line scan camera in an embodiment of the present invention;

[0063] Figure 6 This is a schematic diagram of the inkjet device in an embodiment of the present invention;

[0064] Figure 7 This is a flowchart of the cutting method of the cutting device in an embodiment of the present invention;

[0065] Figure 8 This is a schematic diagram of the feeding device in an embodiment of the present invention. Detailed Implementation

[0066] To better understand the above technical solutions, the technical solutions of the embodiments of this specification will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. In the absence of conflict, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0067] The technical solution of this embodiment includes a vision-based positioning-based cutting path generation system, a cutting device, and a cutting method. The components of the cutting device are shown below:

[0068] I. Equipment Composition

[0069] See Figure 2 , Figure 3 and Figure 8 As shown, this embodiment provides a cutting device, comprising: an object platform 1, a crossbeam 2, a line scan camera 3, a cutting blade device 4, a computer 5, a receiving device 6, and a feeding device 7. The computer 5 is electrically connected to and controls the operation of the crossbeam 2, the line scan camera 3, and the cutting blade device 4. The computer integrates a cutting path generation system based on intelligent vision positioning.

[0070] The object platform 1 is used to support the object to be cut, and a crossbeam 2 is slidably connected to the object platform 1 along the Y-axis.

[0071] The material feeding device 7 is provided behind the object platform 1 for conveying the material to be cut, and the material feeding device has the function of smoothing out the wrinkles of the material.

[0072] A material collection device 6 is provided in front of the object platform 1 to collect the material after cutting.

[0073] The feeding device includes a frame and rollers rotatably mounted on the frame. The rollers rotate to convey the material. A smoothing mechanism is installed on the frame. The smoothing mechanism includes a fixed shaft rotatably connected to the frame. A striking wheel is mounted on the fixed shaft. The striking wheel is gear-shaped. The material extends from top to bottom. The rear of the striking wheel contacts the front face of the material. The fixed shaft is driven to rotate by a motor. The rotation of the fixed shaft drives the striking wheel to strike the material, thereby making the wrinkles in the material tend to be smooth. Subsequently, the material is squeezed by multiple extrusion shafts and then conveyed to the material platform in a smooth state.

[0074] A wire scanning camera 3 is fixedly connected to the lower end face of the crossbeam 2, and a cutting blade device 4 is slidably connected to the upper X-axis of the crossbeam 2.

[0075] like Figure 5 As shown, the line scan camera 3 includes several acquisition elements arranged along the X-axis. The acquisition elements include an image sensor 313, a cylindrical lens 312, a glass window 311, an LED light source 315, and a processing chip 314 integrated in a PCB board. The image sensor 313 is electrically connected to the processing chip 314, and the processing chip 314 is electrically connected to the computer 5.

[0076] The line scan camera includes a camera housing and several acquisition elements installed inside the camera housing. The acquisition elements include a processing chip, an image sensor, a cylindrical lens, glass, and an LED light source. The processing chip is electrically connected to a computer, and the image sensor is electrically connected to the processing chip. The cylindrical lens is placed between the glass and the image sensor. An LED light source facing the object is set on each of the left and right sides of the cylindrical lens. The light source refracted by the object passes through the glass and the cylindrical lens in sequence and reaches the image sensor, where it is converted into an electrical signal.

[0077] like Figure 3 , Figure 4 As shown, in this embodiment, a vibrating knife device is used as the cutting knife device. The cutting knife device 4 includes a connecting plate 41, a knife holder 42, and a cutting knife 43. The connecting plate 41 is slidably connected to the crossbeam 2 along the X-axis, and the knife holder 42 is slidably connected to the connecting plate 41 along the Z-axis. The cutting knife 43 is rotatably connected to the knife holder 42 through a bushing 45 and a bearing 47. A large gear 46 is fitted on the bushing 45 and is connected to the gear motor for transmission. The gear motor is electrically connected to the computer 5. A cutter head position sensor 49 is provided on the knife holder 42. The cutter head position sensor 49 is used to monitor the status of the cutter head and feed back the position information to the computer 5.

[0078] The cutting tool includes a connecting plate that slides along the X-axis on a crossbeam, a tool holder that slides along the Z-axis on the connecting plate, and a cutting head that is rotatably connected to the tool holder. The tool holder has a through hole for the cutting head to pass through, and a bushing is installed in the through hole. The cutting head is installed in the bushing, and a large gear and a bearing are fitted on the bushing. The bushing is rotatably connected to the tool holder through the bearing. The large gear is located above the tool holder and is connected to a gear motor. The gear motor is electrically connected to a computer. The bushing has a sensor recognition notch 48, and a cutting head position sensor facing the bushing is installed on the tool holder. The cutting head position sensor is electrically connected to the computer.

[0079] like Figure 6 As shown, an inkjet printing device 44 is also installed on the blade holder 42. The inkjet printing device 44 includes a U-shaped bracket 442 fixed to the blade holder. A printing cylinder 441 extending in the vertical direction is installed on the U-shaped bracket 442. A connecting block 444 in a limiting rod 443 that slides in the vertical direction within the U-shaped bracket 442 is fixed to the movable end of the printing cylinder 441. A printing component 445 is connected to the connecting block 444 by bolts. The printing component 445 can be a spray valve or an inkjet cartridge.

[0080] The technical solution of this embodiment includes a vision-based positioning-based cutting path generation system, a cutting device, and a cutting method. The computer system in the cutting path generation system is composed of the following:

[0081] II. Computer System Composition

[0082] like Figure 1 As shown, this embodiment provides a vision-based cutting path generation system, applied in a cutting device, comprising:

[0083] The scanning camera control module is used to receive speed information from the encoder set in the cutting equipment, adjust the scanning rhythm and exposure parameters of the object to be cut, ensure that the image acquisition ratio matches the movement speed and acquire the acquired image, and ensure the clarity and stability of the acquired image.

[0084] The image acquisition and recognition module is used to receive acquired images and automatically adjust the acquisition parameters by analyzing the grayscale distribution and brightness characteristics of the acquired images to ensure that the details of the acquired images are clearly visible and to generate clear images.

[0085] The image processing module receives a clear image and uses a thresholding segmentation algorithm or a texture algorithm to obtain the pattern boundary contour data of the clear image; wherein...

[0086] When the pattern edge color of a clear image is similar to that of the background, the system first enhances the texture edge details using the Laplacian algorithm, then uses the LBP and GLCM algorithms to jointly extract the local and global texture features of the clear image, and combines the weighted cross-entropy algorithm and Dice loss function to optimize model training, so that the system can accurately identify the pattern boundary contour of the clear image based on texture and shape features even in the absence of obvious color contrast.

[0087] The cutting path planning module receives pattern boundary contour data and generates the optimal cutting path through calculations using contour optimization algorithm formula, arc compensation algorithm formula, and cutting sequence algorithm formula.

[0088] The cutting control module is used to control the cutting equipment to perform cutting operations based on the cutting path generated by the cutting path planning module; and

[0089] The data storage and management module is used to store acquired images, clear images, pattern boundary contour data, and cutting path information.

[0090] The technical solution of this embodiment includes a vision-based positioning-based cutting path generation system, a cutting device, and a cutting method. The cutting method is as follows:

[0091] III. System Cutting Method

[0092] like Figure 7 As shown, this embodiment provides a cutting method, which includes the following steps:

[0093] Step 1: Fix the paper onto the platform. The cutting control module controls the printing cylinder to drive the printing component, printing calibration grid lines on the paper surface. Specifically, select paper large enough to cover the platform and fix it on the platform. The cutting control module controls the extension of the printing cylinder, causing the moving end of the cylinder to move the printing component downwards, making it contact the paper or create a suitable gap. The cutting control module then controls the movement of the blade holder, and the printing component prints grid lines on the paper.

[0094] Step 2: The camera control module controls the line scan camera to scan the paper with the grid drawn on it. The image data is received and stitched by the image acquisition and recognition module to complete the horizontal alignment of the camera images. Specifically, the camera control module starts the line scan camera to scan the paper, and the acquired image data is transmitted to the computer's image acquisition and recognition module. The image acquisition and recognition module adjusts the horizontally misaligned images, using the first camera head as a reference, and then adjusts the positions of the subsequent camera lenses in sequence to complete the stitching of the images.

[0095] Step 3: Camera calibration is performed. The image processing module generates a camera grid and compares it with the printed grid in the acquired image. If there is a discrepancy, the X and Y offsets of the camera grid are automatically adjusted until the origin consistency calibration is completed. The calibration image is then saved to the data storage and management module. Specifically, the camera is calibrated by generating a camera grid through the image processing module. The image acquisition and recognition module acquires the generated drawing image and transmits it to the image processing module. If the generated camera grid coincides with the acquired printed grid, it means that the camera and the cutting blade origin are consistent. If misalignment occurs, the image processing module will automatically adjust the camera grid, adjusting the X and Y offsets until it is aligned with the grid obtained from the scan. The image processing module then crops the image and saves it to the data storage and management module for subsequent automatic matching and retrieval.

[0096] Step 4: Turn on the power to all devices, start the feeding device, and the object to be cut is conveyed to the object platform through the feeding device. During this process, the smoothing mechanism of the feeding device completes the surface wrinkle smoothing operation of the object to be cut. The object platform moves the object to be cut forward, and the line scan camera captures the image of the object to be cut.

[0097] Step 5: Restart the line scan camera to acquire images of the object. The image acquisition and recognition module receives the image data and feeds it back to the camera control module to dynamically adjust the light source brightness, improving image contrast and edge sharpness. Specifically, the line scan camera is started via the computer's camera control module, which controls the light source of each camera to turn on. The acquired images are sent to the image acquisition and recognition module. Based on the image display, the image acquisition and recognition module feeds back signals to the camera control module. The camera control module controls the light source brightness in real time, thereby improving image contrast and ensuring that the acquired images are clear and have distinct boundaries. With the assistance of the image acquisition and recognition module, the line scan camera completes the image acquisition of the material surface.

[0098] Step 6: The acquired image is transmitted to the image processing module. The design draft is imported, and after selecting the pattern, it is aligned with the actual image through scaling and rotation to generate a feature template. The FAST algorithm is used to extract image features and generate a stable template to achieve multi-angle matching and object positioning. Specifically, the acquired image data is transmitted to the computer's image acquisition and recognition module. The image acquisition and recognition module synchronizes the acquired image to the image processing module, imports the material design draft, selects a single material graphic in the scanned image, and moves / rotates it to make the design draft and the material image coincide proportionally. Click "Generate Template," and use the SIFT algorithm to extract key features such as corner points, edges, and contours of the image, converting them into stable feature vectors to establish a visual template with green feature lines. The image processing module automatically optimizes the template quality to ensure that the feature lines are clear and free of redundant lines, improving the subsequent matching speed and similarity. Based on the SIFT feature vector, and using the template as a reference, multi-angle (0-360°) feature search is performed on the full-area irregular material image. Through high-speed fitting calculation by the computer's central processing unit (CPU) and graphics processing unit (GPU), the feature points of the template and the material image are matched, and the visual template is automatically overlapped to achieve accurate positioning of multiple images.

[0099] Step 7: The image processing module sends the template matching result to the cutting path planning module, which generates the corresponding optimal cutting path based on the template overlay image.

[0100] Step 8: The cutting path is converted into control commands and sent to the cutting control module. The cutting control module controls the cutting blade to execute the corresponding path cutting. The blade position sensor monitors and feeds back position information in real time to ensure accuracy. Specifically, the cutting path planning module converts the generated cutting path into control signals and sends them to the cutting control module. Based on the control signals, the cutting control module precisely controls the gear motor of the cutting blade, adjusting the blade's speed and direction to ensure the cutting blade cuts along the predetermined cutting path. Simultaneously, the blade position sensor monitors the blade's position information in real time and feeds it back to the computer.

[0101] Step 9: After the object to be cut on the object platform is cut, the object platform moves the material to the receiving device. The cutting device continues to cut subsequent objects. After the material is cut, the equipment is turned off and cleaned. When the equipment is used again, the operation can directly enter the scanning and cutting process without recalibrating the camera.

[0102] The following provides a more detailed technical description of the computer algorithm portion of the system in this embodiment:

[0103] In this embodiment, the vision-based positioning-based cutting path generation system, cutting device, and cutting method,

[0104] In the camera control module, through the formula The beam movement speed is calculated in real time using encoder pulse count, providing a basis for adjusting the camera scanning interval and ensuring that the image acquisition ratio matches the material movement speed. Here, v is the beam movement speed (mm / s), and p is the encoder pulse increment. The change in encoder pulse increment is given by L, which represents the linear displacement (mm / revolution) per motor revolution, T, which represents the sampling time interval (s), and R, which represents the reduction ratio.

[0105] According to the formula Adjust the camera scanning interval to ensure a constant physical distance between adjacent scan lines, avoiding image stretching or compression. The camera line scan interval (s) is given. The physical length (mm) corresponding to each line of scan.

[0106] According to the formula The camera scanning interval is controlled and adjusted to compensate for speed fluctuations during the acceleration and deceleration of the crossbeam, ensuring the stability of image acquisition. This is the amount to adjust the scan interval. This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For speed error, , For the target speed, For actual speed, The integral term of the error. Here, is the differential term of the error; de(t) is the minute change in the error, and dt is a differential symbol in mathematics. The meanings are the same, t is... The purpose is to distinguish the t in the two formulas.

[0107] In the image acquisition and recognition module, through the formula The probability of image grayscale distribution is calculated to analyze image brightness characteristics and determine whether overexposed or underexposed areas exist. for The probability of a pixel appearing. It represents the k-th gray level (0-255). grayscale value The number of pixels, where N is the total number of pixels in the image.

[0108] Through formula , Evaluating the overall brightness and contrast of the image provides quantitative indicators for parameter adjustment. The grayscale mean is... This represents the grayscale variance.

[0109] Through formula Iterative optimization of parameters such as exposure and gain concentrates the image grayscale distribution within the target region, improving the accuracy of feature extraction. The iteration stops when the change is less than a threshold or when the maximum number of iterations is reached. Let be the objective function. , For iteration variables, For the process The exposure parameter values ​​at the next time step after iterative adjustment. For learning rate, and For the objective function J, The partial derivatives of ...

[0110] Through formula , Quantify the degree of overexposure / underexposure and trigger parameter adjustment strategies. >5%, directly reduce exposure; if >5%, directly increasing gain to ensure image details are visible. For overexposure ratio, This represents the underexposure ratio.

[0111] In the image processing module, when the color contrast at the cutting boundary is significant, a threshold segmentation algorithm based on the RGB / HSV color space is used. This algorithm converts the RGB color space to HSV, separating hue, saturation, and brightness to reduce the impact of lighting changes on color segmentation. By setting the threshold range for each HSV channel, the target color region is extracted, and a binary mask is generated. The calculation formulas are as follows:

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] Where H is the hue, R, G, B are the pixel values ​​in the red, green, and blue channels, S is the saturation, C is the color difference, V is the brightness, max=max(R,G,B), min=min(R,G,B), C=max-min; Let H(x,y), S(x,y), V(x,y)) be a binary mask, where (H(x,y), S(x,y), V(x,y)) is the HSV value of pixel (x,y). The hue threshold range, This represents the saturation threshold range. This represents the brightness threshold range.

[0117] When cutting the boundary with the same color, a texture algorithm is used. The Gaussian filter (σ=1.0) is used to remove noise and preserve the edges. The Laplacian operator is used to enhance the texture details. Even if the pattern and the background color are the same, they can be distinguished by the texture difference, as shown in the Laplacian algorithm formula (1).

[0118]

[0119] in, I represents the Laplacian transform of the image at coordinates (x, y), and I represents the input image (grayscale matrix). This refers to the second-order partial derivative of the image in the x-direction, used to detect changes in horizontal edges. This refers to the second-order partial derivative of the image in the y-direction, used to detect changes in vertical edges.

[0120] Regarding texture feature extraction, the gray-level difference between the center point and its neighborhood is calculated based on LBP. The neighborhood texture of each pixel is converted into binary code to generate a texture histogram. LBP features with different radii (R=1,2,3) are fused to enhance robustness. Based on the gray-level difference distribution of pixel pairs in the GLCM quantization image, high contrast indicates clear texture edges, while low contrast indicates smooth textures. This distinguishes subtle texture differences against the same color background, as shown in formula (2) of the LBP algorithm and formula (3) of the GLCM algorithm.

[0121] ;

[0122] ;

[0123] in, This refers to the coordinates of the center pixel, where P is the number of neighboring pixels (e.g., 8-neighborhood), and R is the neighborhood radius (pixel distance). Refers to the grayscale value of the center pixel. This refers to the grayscale value of the p-th pixel in the neighborhood. This is a sign function used to compare the grayscale differences between the neighboring pixels and the center pixel. Number of gray levels This represents the co-occurrence probability of grayscale values ​​i and j at a specified direction and distance.

[0124] Regarding the calculation of weighted cross-entropy + Dice loss, the weight of edge pixels is strengthened. Even if the pattern and background colors are the same, the boundary can be accurately divided by features such as texture and shape, such as the formula (4) of the weighted cross-entropy algorithm and the formula (5) of the Dice loss function.

[0125] ;

[0126] ;

[0127] Where N is the total number of pixels in the image, and C is the number of categories (e.g., foreground / background). The weights for class c (used to balance imbalanced samples). This refers to the true label (0 or 1) of pixel i in category c. X refers to the probability that pixel i belongs to category c as predicted by the model; X refers to the set of foreground pixels predicted by the model, and Y refers to the set of actual foreground pixels. This refers to the number of pixels where the prediction intersects with the actual foreground. It refers to the number of pixels resulting from the product of the predicted foreground pixels and the actual foreground pixels.

[0128] During template generation and matching, scale-invariant feature points are detected using the Difference-of-Gaussian pyramid, the gradient histogram of the feature point's neighborhood is calculated, the principal direction is determined, and a SIFT feature descriptor, a 128-dimensional vector, is generated to encode the gradient direction distribution of the feature point's neighborhood. This is achieved through a formula... Calculate the sum of the absolute values ​​of the grayscale differences between the center pixel and its neighboring pixels. If the sum exceeds a threshold, then p is determined to be a corner point. For each pair of random pixels... Compare their grayscale values, if The corresponding bit is set to 1 if the bit is not specified, and to 0 otherwise, a BRIEF descriptor is generated. This descriptor is used to quickly describe the local appearance characteristics of image feature points, where P is the center pixel coordinate. For the i-th pixel in the neighborhood, The grayscale value of the center pixel. is the grayscale value of the neighboring pixels, and thresh is the grayscale difference threshold, used to determine whether it is a corner point.

[0129] Through formula Calculate the Euclidean distance between two floating-point descriptors; a smaller distance indicates higher similarity. SIFT descriptors encode local features using gradient direction histograms, and the Euclidean distance effectively measures the spatial difference of feature vectors. Refers to the i-th dimension component of two 128-dimensional SIFT descriptors.

[0130] Through formula Count the number of different bits in two binary strings. The smaller the Hamming distance, the more similar the descriptors are. Refers to the i-th bit of the two binary BRIEF descriptors.

[0131] Through the formula of similarity transformation matrix This method enables template rotation, scaling, and translation transformations, allowing template matching regardless of the object's orientation. It randomly selects four pairs of matching points, calculates a similarity transformation matrix, transforms all matching points using this matrix, counts the number of "interior points" satisfying the error threshold, and iterates until the matrix with the most interior points is found, suppressing the influence of out-of-place points. False matching points are eliminated through Random Sample Consensus (RANSAC). The optimal transformation matrix is ​​estimated using the interior points, improving the robustness of template matching to noise and false matching. The method utilizes the affine transformation matrix formula... Achieving precise matching under conditions of cutting transformation and transparency transformation of the object under inspection, using the normalized correlation coefficient formula. Eliminate the influence of lighting and confirm the final matching position.

[0132] in, The coordinates of the template point, Here are the coordinates of the target point after transformation, s is the scaling factor, and θ is the rotation angle. The translation is denoted by , which controls the position of the template in the image. (x, y) are the template pixel coordinates, (u, v) are the corresponding coordinates in the target image, T is the pixel value of the transformed template, and I is the pixel value of the corresponding region in the target image. and This represents the average pixel value of the template and target regions.

[0133] In the cutting path planning module, data such as pattern contour coordinates and color region masks output by the image processing module are received as the basis for path planning, and a contour optimization algorithm is used. Mathematical fitting is performed on the original contour to eliminate jagged edges caused by noise, improve cutting quality, and apply formulas at corners. Calculate the curvature k. When k is greater than the threshold, use a circular arc transition (the harder the material, the smaller the threshold) to avoid right-angle cutting that could cause the material to tear.

[0134] in, These are control points on the contour, which determine the shape of the curve. is a k-th order basis function that satisfies local support, t is a parameter variable, and after traversal, a continuous and smooth curve is generated, t∈[0,1]; k is the curvature at the corner of the pattern contour on the material to be cut, x is the abscissa at the corner of the pattern contour on the material to be cut, and y is the ordinate at the corner of the pattern contour on the material to be cut. Let x and y be the first derivatives. Let x and y be the second derivatives. and It is the mathematical derivative operator.

[0135] Simultaneously, based on factors such as material properties and regional complexity, the cutting order of each region is determined to avoid material deformation. The weighting function formula is as follows: ,in The cutting priority weight for the i-th region, and the region area. Prioritize areas with larger proportions to avoid material deformation during large-area cutting; contour complexity. A larger value indicates a more complex shape, which should be cut first to reduce the number of tool adjustments; area overlap. A higher value indicates a higher proportion of intersection with other areas, prioritizing cutting to avoid secondary damage; The weighting coefficients are adjusted based on the material. Based on graph theory and heuristic rules, the shortest path for tool movement is calculated to reduce idle travel time. The optimized path is then output as a set of instructions executable by the cutting tool.

[0136] In the cutting control module, the cutting blade is controlled according to the instruction set. The PID controller calculates the control quantity using three terms: proportional (P), integral (I), and derivative (D), and then uses the formula... The motor voltage u(t) is dynamically adjusted to ensure that the cutter head speed accurately tracks the target speed, ensuring that the fluctuation is ≤±2%. The actual moving speed is calculated based on adjacent position points and used as the feedback value of the PID controller to ensure that the speed is stable during the cutting process and avoid the decline in cutting quality due to speed fluctuation. The speed stability is monitored in real time, and the PID parameter adjustment is triggered when the fluctuation exceeds the threshold.

[0137] in, For proportional, integral, and differential coefficients, the proportional term is... The output directly responds to the current error; the larger the error, the larger the output. (Integral term) Accumulate historical errors, eliminate system static errors, calculus terms Predict the trend of error changes and suppress system oscillations. For speed error, , For the target speed, For actual speed, The integral term of the error. Let de(t) be the differential term of the error, and dt be the infinitesimal change in the error. In mathematics, dt is a differential symbol. The meanings of dt and t are the same, where t is... The purpose is to distinguish the t in the two formulas.

[0138] In summary, the technical solution of this embodiment can complete the contour recognition of a single image within 500ms through image processing algorithms, and achieve sub-pixel alignment between the product contour and the design template through template matching technology, with a matching accuracy of >99%. Furthermore, based on the threshold segmentation algorithm of the RGB / HSV color space, it can identify areas of the same color on the material surface with a color difference ≤5ΔE, supporting color-based segmentation and improving material utilization. Simultaneously, by extracting micro-texture differences (such as fabric weaving methods and leather pore distribution), even with the same color, pattern boundaries can be distinguished. Texture features are less sensitive to changes in illumination than color features, reducing positioning errors caused by light fluctuations.

[0139] The technical solution in this embodiment effectively combines threshold segmentation algorithms in the RGB / HSV color space with texture algorithms, breaking through the limitations of traditional single-feature segmentation and constructing a multimodal feature fusion framework that can adaptively adjust weights according to the different materials being cut. Furthermore, by combining material properties (such as hardness and ductility) and cutting requirements (such as serrated edges and smooth edges), it generates the shortest path or cutting route with the fewest turning times through genetic algorithms or AI algorithms, improving cutting efficiency by more than 30% while reducing tool wear (extending tool life by 20%). After cubic B-spline fitting of the original contour (serrated), tool wear rate is reduced by 30%, and burrs on the cutting edge are reduced by 50%.

[0140] In addition, the technical solution of this embodiment also has the following beneficial effects:

[0141] In this embodiment, the line scan camera has multiple sets of acquisition elements arranged along the X-axis, which effectively expands the lateral field of view. Through the multi-lens collaborative acquisition mechanism, it significantly reduces single-lens distortion error and achieves high resolution. This is to distinguish t in the two formulas.

[0142] In summary, the technical solution of this embodiment can complete the contour recognition of a single image within 500ms through image processing algorithms, and achieve sub-pixel alignment between the product contour and the design template through template matching technology, with a matching accuracy of >99%. Furthermore, based on the threshold segmentation algorithm of the RGB / HSV color space, it can identify areas of the same color on the material surface with a color difference ≤5ΔE, supporting color-based segmentation and improving material utilization. Simultaneously, by extracting micro-texture differences (such as fabric weaving methods and leather pore distribution), even with the same color, pattern boundaries can be distinguished. Texture features are less sensitive to changes in illumination than color features, reducing positioning errors caused by light fluctuations.

[0143] The technical solution in this embodiment effectively combines threshold segmentation algorithms in the RGB / HSV color space with texture algorithms, breaking through the limitations of traditional single-feature segmentation and constructing a multimodal feature fusion framework that can adaptively adjust weights according to the different materials being cut. Furthermore, by combining material properties (such as hardness and ductility) and cutting requirements (such as serrated edges and smooth edges), it generates the shortest path or cutting route with the fewest turning times through genetic algorithms or AI algorithms, improving cutting efficiency by more than 30% while reducing tool wear (extending tool life by 20%). After cubic B-spline fitting of the original contour (serrated), tool wear rate is reduced by 30%, and burrs on the cutting edge are reduced by 50%.

[0144] In addition, the technical solution of this embodiment also has the following beneficial effects:

[0145] The technical solution of this embodiment uses a line scan camera with multiple sets of acquisition elements arranged along the X-axis, which effectively expands the lateral field of view. Through the multi-lens collaborative acquisition mechanism, it significantly reduces the distortion error of a single lens and achieves high-resolution, seamless image acquisition, which is especially suitable for panoramic scanning of large-format materials.

[0146] The technical solution of this embodiment features a three-dimensional sliding design with the crossbeam along the Y-axis, the connecting plate along the X-axis, and the cutter holder along the Z-axis. Combined with the rotational vibration function of the cutter head, this allows the cutting blade to move freely in three-dimensional space, precisely adapting to complex cutting paths such as curves and oblique lines, and fully meeting the high-precision cutting requirements of diverse patterns.

[0147] The inkjet printing device in this embodiment is equipped with various printing components such as a spray valve or inkjet cartridge. It can complete operations such as contour pre-drawing, color area marking, or number printing before cutting, realizing an integrated "visual recognition-marking-cutting" workflow, effectively improving process continuity and production efficiency. It automatically adjusts the positions of subsequent lenses based on the first camera lens, eliminating parallax between lenses through software algorithms to form a complete wide-format image with a stitching error of less than 0.5 pixels, reducing manual calibration time. Grid line calibration ensures that the camera and cutting blade share the same origin, guaranteeing a one-to-one correspondence between image coordinates and physical cutting positions, avoiding cutting offsets caused by coordinate deviations.

[0148] In this embodiment, the cutter head position sensor provides real-time feedback of rotational speed and position data. The computer dynamically adjusts the motor parameters through a PID control algorithm, strictly controlling the cutting speed fluctuation within ±2%, effectively improving the flatness of the cut surface and ensuring the stability of the cutting quality. After the initial calibration, the system memorizes the coordinate relationship between the camera and the cutting origin, eliminating the need for repeated calibration for subsequent cuts. The preparation time for a single cut is reduced from 30 minutes to 5 minutes, making it suitable for small-batch, multi-production scenarios.

[0149] The technical solution of this embodiment is not limited to using a vibrating knife; other cutting tools can be used. It is also not limited to using a feeding device. When cutting non-flexible materials, the material can be placed directly on the object platform.

[0150] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A vision-based cutting path generation system, applied in cutting equipment, characterized in that, include: The scanning camera control module is used to receive speed information from the encoder set in the cutting equipment, adjust the scanning rhythm and exposure parameters of the object to be cut, ensure that the image acquisition ratio matches the movement speed and acquire the acquired image, and ensure the clarity and stability of the acquired image. The image acquisition and recognition module is used to receive the acquired image and automatically adjust the acquisition parameters by analyzing the grayscale distribution and brightness characteristics of the acquired image to ensure that the details of the acquired image are clearly visible and to generate a clear image. An image processing module is used to receive the clear image and obtain the pattern boundary contour data of the clear image through a threshold segmentation algorithm or a texture algorithm; in When the pattern edge color of the clear image is similar to that of the background, the system first enhances the texture edge details by using the Laplacian algorithm, then uses the LBP algorithm and GLCM algorithm to jointly extract the local and global texture features of the clear image, and combines the weighted cross-entropy algorithm and Dice loss function to optimize the model training, so that the system can accurately identify the pattern boundary contour of the clear image based on texture and shape features even in the absence of obvious color contrast. The cutting path planning module is used to receive the pattern boundary contour data and generate the optimal cutting path information by calculating the contour optimization algorithm formula, the arc compensation algorithm formula, and the cutting sequence algorithm formula. The cutting control module is used to control the cutting equipment to perform cutting operations according to the cutting path generated by the cutting path planning module. as well as The data storage and management module is used to store the acquired image, the clear image, the pattern boundary contour data, and the cutting path information; In the cutting path planning module, the contour optimization algorithm formula is used: And the formula for the circular arc compensation algorithm: Together, they ensure the smoothness and safety of the cutting path, avoiding quality problems caused by jagged edges and sharp corners. When used together, they achieve mathematical smoothing of the contour and optimization of physical cutting adaptability, ensuring that even complex patterns can generate high-quality and applicable cutting paths. And through the cutting order algorithm formula: By combining material properties and regional information, a scientific cutting sequence can be determined to reduce cutting risks and processing time. in, The coordinates of the curve points under parameter t. These are control points on the contour, which determine the shape of the curve. Let be a k-th order basis function that satisfies local support, t be a parameter variable, and after traversal, a continuous and smooth curve is generated, t∈[0,1]. k is the curvature at the corner of the pattern outline on the material to be cut, used to describe the sharpness of the outline corner; x is the horizontal coordinate of the corner of the pattern outline on the material to be cut; and y is the vertical coordinate of the corner of the pattern outline on the material to be cut. Let x and y be the first derivatives. Let x and y be the second derivatives. and It is the mathematical derivative operator; For the region The priority cutting weight value, The larger the value, the more preferentially the corresponding area is cut to avoid material deformation in large areas due to subsequent cutting; For the contour complexity, The larger the value, the more complex the shape, and it should be cut first to reduce the number of tool adjustments. For regional overlap, A higher value indicates a higher proportion of intersection with other areas, prioritizing cutting to avoid secondary damage; This is a weighting coefficient, which is adjusted according to the material.

2. The system according to claim 1, characterized in that, In the texture algorithm of the image processing module, The Laplace algorithm formula is: ; in, Let I be the Laplacian transform of the image at coordinates (x, y), and let I be the input image. This refers to the second-order partial derivative of the image in the x-direction, used to detect changes in horizontal edges. The second-order partial derivative of the image in the y-direction is used to detect changes in vertical edges; The formula for the LBP algorithm is: ; in, coordinates The local binary pattern encoding result at that location, The coordinates of the center pixel are given, P is the number of neighboring pixels, and R is the neighborhood radius. Refers to the grayscale value of the center pixel. Let be the gray value of the p-th pixel in the neighborhood. This is the sign function, used to compare the grayscale differences between the neighboring pixels and the center pixel; The formula for the GLCM algorithm is: ; in, For image contrast, Number of gray levels This represents the co-occurrence probability of grayscale values ​​i and j at a specified direction and distance. The formula for the weighted cross-entropy algorithm is: ; in, This is the weighted cross-entropy loss value, where N is the total number of pixels in the image and C is the number of classes. The weight of category c, This refers to the true label of pixel i in category c. This refers to the probability that the model predicts pixel i belongs to category c; The formula for the Dice loss function is: ; in, Here, X represents the set of foreground pixels predicted by the model, and Y represents the set of true foreground pixels. This refers to the number of pixels where the prediction intersects with the actual foreground. It refers to the number of pixels resulting from the product of the predicted foreground pixels and the actual foreground pixels.

3. The system according to claim 2, characterized in that, The image processing module also includes a method for identifying and locating patterns in the clear image that are irregularly arranged, randomly oriented, or subject to occlusion or lighting changes. This method finds the best match between the template pattern and the pattern in the current view under any angle, size, and lighting conditions, and achieves highly robust recognition and positioning to ensure accurate generation of the cutting path.

4. The system according to claim 3, characterized in that, Based on the FAST algorithm formula: Corner points are detected, and local features of the pattern are constructed by combining SIFT and BRIEF descriptors, and then the Euclidean distance formula is used: Hamming distance formula: It determines the similarity between patterns to achieve accurate matching between templates and target patterns, maintaining robust recognition accuracy even when patterns vary in scale, angle, and lighting; among other things, To determine whether pixel P is a corner response value, P is the coordinate of the center pixel. For the i-th pixel in the neighborhood, The grayscale value of the center pixel. Here, is the grayscale value of the neighboring pixels, and thresh is the grayscale difference threshold used to determine whether a point is a corner. This is related to the Euclidean distance formula. Let Euclidean distance be the distance between two image feature points. and The i-th component of two 128-dimensional SIFT descriptors, in the Hamming distance formula. The Hamming distance values ​​for the two patterns under the BRIEF descriptor. and Refers to the i-th bit of the two binary BRIEF descriptors.

5. The system according to claim 3, characterized in that, Using the similarity transformation formula: Affine transformation formula: Formula for normalized cross-correlation coefficient: By combining these methods, the system can accurately align and match the template with the pattern in the view, even in complex scenarios involving pattern rotation, scaling, offset, or changes in lighting, ensuring the accuracy and stability of subsequent path generation; among these, The coordinates of the template point, Here are the coordinates of the target point after transformation, s is the scaling factor, and θ is the rotation angle. The translation amount controls the position of the template in the image, where (x, y) are the template pixel coordinates and (u, v) are the corresponding coordinates in the target image. The normalized cross-correlation coefficient is used to evaluate the similarity between template T and image I, where T is the pixel value of the transformed template and I is the pixel value of the corresponding region in the target image. and This represents the average pixel value of the template and target regions.

6. The system according to claim 1, characterized in that, The cutting control module also employs a PID control algorithm formula: The motor voltage adjustment is calculated to ensure the cutter head speed precisely follows the target speed, avoiding a decrease in cutting quality caused by speed fluctuations; among other things, For proportional, integral, and differential coefficients, the proportional term is... The output directly responds to the current error; the larger the error, the larger the output. (Integration) Accumulate historical errors, eliminate system static errors, calculus terms Predict the trend of error changes and suppress system oscillations.

7. A cutting device, characterized in that, include: The system comprises an object platform, a crossbeam, a line scan camera, a cutting blade device, a feeding device, a receiving device, and a computer. The computer is electrically connected to and controls the operation of the crossbeam, the line scan camera, and the cutting blade device. The computer integrates a vision-based positioning-based cutting path generation system according to claim 1. The object platform is used to support the object to be cut, and a crossbeam is slidably connected to the object platform along the Y-axis. A feeding device is provided behind the object platform for conveying the object to be cut, and the feeding device has a smoothing mechanism to smooth out the wrinkles of the material. A material collection device is installed in front of the object platform to collect the material after cutting. A wire scanning camera is fixedly connected to the lower end face of the crossbeam, and a cutting blade device is slidably connected to the upper X-axis of the crossbeam. The cutting blade device includes a connecting plate, a blade holder, and a cutting blade. The connecting plate is slidably connected to the crossbeam along the X-axis, and the blade holder is slidably connected to the connecting plate along the Z-axis. The cutting blade is mounted on the blade holder, and a blade head position sensor is provided on the blade holder. The blade head position sensor is used to monitor the state of the blade head and feed back the position information to the computer. The blade holder is also equipped with an inkjet printing device, which includes a U-shaped bracket, a printing cylinder, a connecting block, and a printing component. The printing component is a spray valve or an inkjet cartridge, used for marking or assisting in positioning on the material surface.

8. A cutting method for the cutting equipment according to claim 7, characterized in that, The cutting method includes the following steps: Step 1: Fix the paper on the platform, and use the cutting control module to control the printing cylinder to drive the printing component to print calibration grid lines on the paper surface; Step 2: The camera control module controls the line scan camera to scan the paper with the grid printed on it. The image data is received and stitched by the image acquisition and recognition module to complete the horizontal alignment of the camera image. Step 3: Perform camera calibration. The image processing module generates a camera grid and compares it with the printed grid in the acquired image. If there is a deviation, the X and Y offsets of the camera grid are automatically adjusted until the origin consistency calibration is completed. The calibration image is then saved to the data storage and management module. Step 4: Turn on the power of each device, start the feeding device, and the object to be cut is conveyed to the object platform through the feeding device. During this process, the smoothing mechanism of the feeding device completes the surface wrinkle smoothing operation of the object to be cut. The object platform moves the object to be cut forward, and the line scan camera captures the image of the object to be cut. Step 5: The image acquisition and recognition module receives image data and feeds it back to the camera control module to dynamically adjust the light source brightness, thereby improving image contrast and edge sharpness; Step 6: The acquired image is transmitted to the image processing module, the design draft is imported, and after selecting the pattern, it is aligned with the real object image by scaling and rotating to generate a feature template. The FAST algorithm is used to extract image features and generate a stable template to achieve multi-angle matching and object positioning. Step 7: The image processing module sends the template matching result to the cutting path planning module, which generates the corresponding optimal cutting path based on the template overlay image; Step 8: The cutting path is converted into a control command and sent to the cutting control module. The cutting control module controls the cutting blade to execute the corresponding path cutting. The blade position sensor monitors and feeds back position information in real time to ensure accuracy. Step 9: After the object to be cut on the object platform is cut, the object platform moves the material to the receiving device. The cutting device continues to cut subsequent objects. After the material is cut, the equipment is turned off and cleaned. When the equipment is used again, the operation can directly enter the scanning and cutting process without recalibrating the camera.

Images