A method and device for workpiece point cloud fusion in laser cladding processing

Abstract

The present invention discloses a method and device for fusing workpiece point clouds during laser cladding processing. The method comprises: obtaining an infrared image and a point cloud depth image of the workpiece to be measured; fusing the point cloud depth image with the infrared image; building and training an image segmentation model based on a convolutional neural network; identifying high-temperature areas in the fused image using the image segmentation model, and cropping and removing the high-temperature areas from the point cloud data; analyzing the obtained point cloud data and preprocessing it in the order of outlier removal, RANSAC plane segmentation, data denoising, and voxel downsampling; and stitching the preprocessed point cloud data in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching, and ICP fine stitching to obtain a stitched workpiece point cloud. The present invention can effectively reduce the impact of high-temperature thermal radiation on data collected by an active optical camera, achieve real-time acquisition of workpiece shape data during laser cladding processing, and improve acquisition speed and accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing, and specifically relates to a method and device for performing workpiece point cloud fusion in laser cladding processing. Background Art

[0002] Laser cladding, a key additive manufacturing technology, can be used in a variety of applications, including directly shaping metal materials, depositing functional coatings, and repairing precision components. Real-time monitoring of the workpiece surface shape is crucial during the laser cladding process, as it allows for real-time perception of the workpiece surface shape and timely adjustment of process path parameters, thereby avoiding machining deviations and improving machining accuracy and efficiency.

[0003] Real-time detection of the three-dimensional shape of the workpiece is a core technology in the laser cladding process. However, there is a local high-temperature molten pool in the laser cladding process, and the radiation energy is concentrated in the visible light and infrared bands, which affects the active optical camera's capture of the reflection pattern. In the high-temperature area of ​​the molten pool, point cloud data may be missing or have reduced accuracy. Although some existing methods can be used to capture the surface shape of the workpiece during the laser cladding process, they cannot eliminate the interference of local high-temperature thermal radiation on the acquisition of the workpiece surface shape, thereby affecting the acquisition accuracy and speed. Therefore, in the context of the rapid development of intelligent manufacturing technology, there is an urgent need for a new method that can capture the surface shape of the workpiece in real time during laser cladding processing. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a method and device for performing workpiece point cloud fusion in laser cladding processing, so as to effectively capture the surface shape of the workpiece in real time.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A method for performing workpiece point cloud fusion in laser cladding processing comprises the following steps:

[0007] S1: Acquire the infrared image and point cloud depth image of the workpiece to be measured;

[0008] S2: Fusion processing of point cloud depth image and infrared image;

[0009] S3: Build and train a convolutional neural network-based image segmentation model;

[0010] S4: identifying the high temperature area in the fused image obtained in S2 by using the image segmentation model, and cropping and removing the high temperature area on the point cloud data;

[0011] S5: Analyze the point cloud data obtained in S4 and preprocess it in the order of outlier removal, RANSAC plane segmentation, data denoising and voxel downsampling;

[0012] S6: The point cloud data preprocessed in S5 is stitched in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching, and ICP fine stitching to obtain the stitched workpiece point cloud.

[0013] Furthermore, the step S2 specifically includes the following sub-steps:

[0014] The infrared image and the point cloud depth image are matched pixel by pixel, and then the pixels of the infrared image and the point cloud depth image are superimposed. The superimposed pixel value X c The calculation is as follows:

[0015] X c =pX a +qX b

[0016] Among them, a represents the pixel of the point cloud depth image, b represents the pixel of the infrared image, and X a Represents the pixel value of the point cloud depth image, X b represents the pixel value of the infrared image, and p and q represent the weights of the pixels.

[0017] Furthermore, the fusion processing also includes: before performing the fusion operation, processing the infrared image and the point cloud depth image into the same size.

[0018] Furthermore, in step S3:

[0019] The image segmentation model includes an encoder and a decoder. The encoder is based on the DeeplabV3+ framework and uses GhostNet as the backbone network for feature extraction. The shallow features of the backbone network are obtained when the output step size is 4, and the deep features of the backbone network are obtained when the output step size is 16. The multi-rate hollow convolution ASPP module is used on the deep features. Then, in the decoder, the feature map is processed by 1x1 convolution and upsampling and then merged with the shallow features. Finally, the segmentation result is output through 3x3 convolution and upsampling.

[0020] The training process of the image segmentation model includes: collecting infrared images and point cloud depth images during the laser cladding process; fusing the point cloud depth images with the corresponding infrared images; annotating the areas with excessively high temperatures pixel by pixel on the fused image containing the molten pool area to obtain a one-hot encoded label; and training the image segmentation model using the annotated labels and the fused image.

[0021] Furthermore, the S4 specifically includes the following sub-steps:

[0022] Multiple point cloud depth images are acquired in the order in which they were taken by the camera. The resulting fused image is then used with a trained image segmentation model to identify high-temperature areas, which represent the molten pool areas in the laser cladding process. If a high-temperature area is identified, the corresponding point cloud data is cropped using that area.

[0023] Furthermore, step S5 includes the following sub-steps:

[0024] S5.1: Calculate the statistical features of the points in the neighborhood around each point of the multiple cropped point clouds in the order of camera shooting, select sparse outliers and remove them;

[0025] S5.2: For unnecessary planes in the original point cloud, use the random sampling consistency RANSAC algorithm to perform plane segmentation and remove irrelevant parts;

[0026] S5.3: Use radius filtering to remove surrounding noise from the point cloud data after plane segmentation;

[0027] S5.4: Use the same ratio of voxel downsampling for all point clouds obtained in S5.3 to improve algorithm efficiency.

[0028] Furthermore, step S6 includes the following sub-steps:

[0029] S6.1: Calculate the point cloud normals using the PCA principal component analysis algorithm for each of the multiple point clouds to be spliced ​​after preprocessing in S5;

[0030] S6.2: Take the first point cloud in the camera shooting order as the reference point cloud, calculate the fast point feature histogram (FPFH) descriptor of the first subsequent point cloud and the reference point cloud, and iterate the sampling consistency initial alignment (SAC-IA) algorithm to obtain the initial transformation matrix;

[0031] S6.3: Using the initial transformation matrix, transform the first subsequent point cloud into the reference point cloud coordinate system to obtain the initially transformed point cloud.

[0032] S6.4: Find corresponding point pairs between the reference point cloud and the initially transformed point cloud, remove incorrect point pairs using the RANSAC algorithm, and then calculate the rotation matrix and translation vector to minimize the root mean square error between the corresponding matching point pairs to achieve fine stitching. Also, set a threshold and a maximum number of iterations to control the accuracy and time consumption of fine stitching.

[0033] S6.5: Repeat S6.2 to S6.5, stitching the subsequent point clouds onto the stitched point cloud in sequence until all point clouds are stitched together.

[0034] A device for performing workpiece point cloud fusion in laser cladding processing, comprising:

[0035] Image acquisition unit: used to obtain infrared image data and point cloud depth image of the workpiece to be measured, and respectively obtain the workpiece surface thermal distribution information and workpiece surface three-dimensional shape information;

[0036] Image fusion unit: used to perform pixel-by-pixel correspondence between the infrared image and the point cloud depth image, and perform weighted summation of the corresponding pixels to obtain a fused image; the fused image has both the thermal distribution information of the infrared image and the three-dimensional structure information of the point cloud depth image;

[0037] Image segmentation unit: used for segmenting the high temperature area in the fused image through an image segmentation model;

[0038] Point cloud processing unit: used to determine whether each point cloud data point falls within the high-temperature area, and perform preprocessing of outlier removal, RANSAC plane segmentation, data denoising and voxel downsampling on the point cloud data after removing the high-temperature area. The point cloud data preprocessed by S5 is stitched in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching and ICP fine stitching to obtain the fused workpiece point cloud.

[0039] An electronic device has a storage module, which includes instructions loaded and executed by a processor; the instructions enable the processor to process a method for performing workpiece point cloud fusion in laser cladding processing when executed.

[0040] A computer-readable storage medium stores one or more programs, which, when executed by a processor, implement a method for performing workpiece point cloud fusion in laser cladding processing.

[0041] The beneficial effects of the present invention are as follows:

[0042] The method of the present invention for performing workpiece point cloud fusion in laser cladding processing can achieve real-time perception of the workpiece surface shape and improve acquisition accuracy and speed by fusing the collected infrared image and point cloud depth image and effectively eliminating the interference of high-temperature molten pool radiation on the active optical camera acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a method for performing workpiece point cloud fusion in laser cladding processing provided by an embodiment of the present invention.

[0044] Figure 2 This is a framework diagram of the image segmentation model provided by an embodiment of the present invention.

[0045] Figure 3 This is a flowchart of point cloud processing provided by an embodiment of the present invention.

[0046] Figure 4 It is a schematic diagram of a device for performing workpiece point cloud fusion in laser cladding processing provided by an embodiment of the present invention.

[0047] Figure 5 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0049] like Figure 1 As shown, the method for performing workpiece point cloud fusion in laser cladding processing according to an embodiment of the present invention includes the following steps:

[0050] S1: Obtain the infrared image and point cloud depth image of the workpiece to be measured.

[0051] In this embodiment, step S1 is specifically as follows:

[0052] Acquire an infrared image and a point cloud depth image of the workpiece to be measured. The infrared image can be obtained by a thermal imager, while the point cloud depth image can be obtained by an active optical camera, such as a structured light camera or a TOF camera.

[0053] S2: Fusion processing of point cloud depth image and infrared image.

[0054] In this embodiment, step S2 is specifically as follows:

[0055] a. Process the infrared image and point cloud depth image into the same size;

[0056] b. Match the infrared image and the point cloud depth image pixel by pixel;

[0057] c. Superimpose the pixels of the infrared image and the point cloud depth image, and the superimposed pixel value X c The calculation is as follows:

[0058] X c =pX a +qX b

[0059] Among them, a represents the pixel of the point cloud depth image, b represents the pixel of the infrared image, and X a Represents the pixel value of the point cloud depth image, X b represents the pixel value of the infrared image, and p and q represent the weights of the pixel points. In this embodiment, p is set to 0.4 and q is set to 0.6.

[0060] S3: Build and train an image segmentation model based on convolutional neural networks.

[0061] In this embodiment, in step S3:

[0062] like Figure 2 As shown in the figure, the image segmentation model includes an encoder and a decoder. The encoder is based on the DeeplabV3+ framework and uses GhostNet as the backbone network for feature extraction. When the output step size is 4, the shallow features of the backbone network are obtained, and when the output step size is 16, the deep features of the backbone network are obtained. The multi-rate void convolution ASPP module is used on the deep features, and then the feature map is merged with the shallow features after 1x1 convolution and upsampling in the decoder. Finally, the segmentation result is output through 3x3 convolution and upsampling.

[0063] The training process of the image segmentation model includes: collecting infrared images and point cloud depth images during the laser cladding process; fusing the point cloud depth image with the corresponding infrared image; annotating the areas with excessively high temperatures pixel by pixel on the fused image containing the molten pool area to obtain one-hot encoded labels; and training the image segmentation model using the annotated labels and the fused image.

[0064] In this embodiment, the output of the network is trained using focal loss.

[0065]

[0066] where (1-p i,j ) γ is a dynamic adjustment factor. W and H represent the width and height of the image respectively, p i,j Indicates the probability of the defect class at position (i, j) estimated by the network, and the dynamic adjustment factor in the focus loss function is set to 2. The model parameters are updated using the stochastic gradient descent method for backpropagation, with the optimizer momentum of 0.9, the weight decay parameter of 0.0001, the batch size of 8, and the learning rate of 0.005.

[0067] S4: Identify the high-temperature area in the fused image obtained in S2 through the image segmentation model, and crop and remove the high-temperature area on the point cloud data.

[0068] In this embodiment, step S4 is specifically as follows:

[0069] Multiple point cloud depth images are acquired in the order captured by the cameras. The resulting fused image is then used with a trained image segmentation model to identify high-temperature regions, representing the melt pool during laser cladding. If a high-temperature region is identified, the corresponding point cloud data is cropped using that region. Specifically, for each point cloud data point, if its location falls within the high-temperature region, the point is removed from the point cloud data to ensure that subsequent point cloud data processing is not affected by the high-temperature region.

[0070] S5: Analyze the point cloud data obtained in S4 and preprocess it in the order of outlier removal, RANSAC plane segmentation, data denoising and voxel downsampling.

[0071] In this embodiment, step S5 is specifically as follows:

[0072] S5.1: Calculate the statistical features of the points in the neighborhood around each point of the multiple cropped point clouds in the order of camera shooting, select sparse outliers and remove them;

[0073] S5.2: For unnecessary planes in the original point cloud, use the random sampling consistency RANSAC algorithm to perform plane segmentation and remove irrelevant parts;

[0074] S5.3: Use radius filtering to remove surrounding noise from the point cloud data after plane segmentation;

[0075] S5.4: Use the same ratio of voxel downsampling for all point clouds obtained in S5.3 to improve algorithm efficiency.

[0076] S6: The point cloud data preprocessed in S5 is stitched in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching, and ICP fine stitching to obtain the stitched workpiece point cloud.

[0077] In this embodiment, step S6 is specifically as follows:

[0078] S6.1: Calculate the point cloud normals using the PCA principal component analysis algorithm for each of the multiple point clouds to be spliced ​​after preprocessing in S5;

[0079] S6.2: Take the first point cloud in the camera shooting order as the reference point cloud, calculate the fast point feature histogram (FPFH) descriptor of the first subsequent point cloud and the reference point cloud, and iterate the sampling consistency initial alignment (SAC-IA) algorithm to obtain the initial transformation matrix;

[0080] S6.3: Using the initial transformation matrix, transform the first subsequent point cloud into the reference point cloud coordinate system to obtain the initially transformed point cloud.

[0081] S6.4: Find corresponding point pairs between the reference point cloud and the initially transformed point cloud, remove incorrect point pairs using the RANSAC algorithm, and then calculate the rotation matrix and translation vector to minimize the root mean square error between the corresponding matching point pairs to achieve fine stitching. Also, set a threshold and a maximum number of iterations to control the accuracy and time consumption of fine stitching.

[0082] S6.5: Repeat S6.2 to S6.5, stitching the subsequent point clouds onto the stitched point cloud in sequence until all point clouds are stitched together.

[0083] On the other hand, Figure 3 As shown, an embodiment of the present invention provides a device for performing workpiece point cloud fusion in laser cladding processing, including an image acquisition unit 401, an image fusion unit 402, an image segmentation unit 403 and a point cloud processing unit 404.

[0084] The image acquisition unit 401 is used to acquire infrared image data and point cloud depth images of the workpiece to be measured, respectively obtaining thermal distribution information and three-dimensional shape information of the workpiece surface. The infrared image can be obtained by a thermal imager, and the point cloud depth image can be obtained by an active optical camera, such as a structured light camera or a time-of-flight camera.

[0085] The image fusion unit 402 is used to perform pixel-by-pixel correspondence between the infrared image and the point cloud depth image, and perform weighted summation on the corresponding pixels to obtain a fused image; the fused image has both the thermal distribution information of the infrared image and the three-dimensional structure information of the point cloud depth image.

[0086] Furthermore, the image fusion unit 402 includes:

[0087] The first computing subunit 4021 is used to process the infrared image and the point cloud depth image into the same size;

[0088] The second calculation sub-unit 4022 is used to match the infrared image and the point cloud depth image pixel by pixel and perform pixel superposition. The superposed pixel value X c The calculation is as follows:

[0089] X c =pX a +qX b

[0090] Among them, a represents the pixel of the point cloud depth image, b represents the pixel of the infrared image, and X a Represents the pixel value of the point cloud depth image, X b represents the pixel value of the infrared image, and p and q represent the weights of the pixels.

[0091] The image segmentation unit 403 is configured to segment the high temperature area in the fused image using an image segmentation model.

[0092] In this embodiment, the image segmentation model adopts a lightweight deep learning image segmentation model. For a description of the model, please refer to the introduction of the model in the previous method.

[0093] The flow chart of the point cloud processing unit 404 is as follows: Figure 4 As shown, specifically including:

[0094] The cropping subunit 4041 is used to obtain multiple point cloud data in the order of camera shooting, and at the same time determine whether there is a to-be-cropped area in the image output by the image segmentation model. If so, the point cloud data is cropped using the area.

[0095] The pre-processing sub-unit 4042 is used to pre-process the plurality of point cloud data after being cropped in the order of camera shooting in the order of outlier removal, RANSAC plane segmentation, data denoising and voxel down-sampling.

[0096] The coarse registration subunit 4043 is used to perform coarse registration on the multiple point cloud data after preprocessing in the order of camera shooting according to the order of normal vector calculation, FPFH feature descriptor calculation and SAC-IA algorithm to realize point cloud transformation.

[0097] The fine registration subunit 4044 is used to finely align the multiple point cloud data in the order of camera shooting after the rough registration in the order of finding corresponding points, removing erroneous point pairs using the RANSAC algorithm, solving the minimization objective function and judging whether the termination condition is met, and at the same time judge whether all point clouds are spliced.

[0098] Same as the above concept, Figure 4 The device for performing workpiece point cloud fusion in laser cladding processing shown can be implemented as an electronic device, including one or more storage devices 501 and a processor 502; Figure 5 This figure shows a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. A storage device 501 is used to store one or more programs, and there can be one or more processors 502. The one or more programs are executed by the processor 502, thereby enabling the processor 502 to implement any of the methods for workpiece point cloud fusion in laser cladding processing described in the present invention.

[0099] The electronic device uses a storage device 501 to store one or more programs, which may be software programs, computer executable programs and modules, including program instructions / modules for the method for performing workpiece point cloud fusion in laser cladding processing provided in an embodiment of the present invention, for implementing the method for performing workpiece point cloud fusion in laser cladding processing in the above method embodiment. The processor 502 can implement various functional applications and data processing by running the software programs, instructions and modules stored in the storage device 501. The storage device 501 can be composed of one or more different types of storage devices, for example, a flash memory device, a disk storage device or other non-volatile solid-state storage device. In addition, the storage device 501 may also include a remote storage connected to the device via a network, examples of which include but are not limited to a local area network, the Internet, a mobile communication network and a combination thereof.

[0100] When one or more programs included in the electronic device are executed by the one or more processors 502 , the program performs the above steps S1 to S6 .

[0101] In an embodiment of the present invention, a computer-readable storage medium is further provided, the storage medium being configured to store one or more programs. When executed by a processor, the programs implement the steps according to various exemplary embodiments of the present invention described in the aforementioned method for performing workpiece point cloud fusion in laser cladding processing.

[0102] In the present invention, if the module / unit for integrating the workpiece point cloud fusion device in laser cladding processing is implemented in the form of a software functional unit, it can be stored in a computer-readable storage medium. The module / unit can be implemented by hardware related to computer program instructions. The computer program includes computer program code and can be stored in a computer-readable storage medium in the form of source code, object code, executable file, etc. Computer-readable storage media may include any entity or device capable of carrying the computer program code, such as a portable disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM) or other optical or magnetic storage media. When the processor executes the program, the steps of the method of each embodiment of the present invention can be implemented.

[0103] The electronic devices and computer-readable storage media provided in the aforementioned embodiments acquire infrared images and point cloud depth images of the workpiece under test and fuse them together to achieve the fusion of depth features and temperature information. By segmenting the fused image, areas affected by high-temperature thermal radiation are detected and removed, completing the stitching and fusion of the point cloud slices. This effectively reduces the impact of high-temperature thermal radiation on the data collected by the active optical camera.

[0104] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.

Claims

1. A method for workpiece point cloud fusion in laser cladding processing, characterized in that: The following steps are involved: S1: Acquire the infrared image and point cloud depth image of the workpiece to be measured; S2: Fusion processing of point cloud depth image and infrared image; S3: Build and train a convolutional neural network-based image segmentation model; S4: identifying the high temperature area in the fused image obtained in S2 by using the image segmentation model, and cropping and removing the high temperature area on the point cloud data; S5: Analyze the point cloud data obtained in S4 and preprocess it in the order of outlier removal, RANSAC plane segmentation, data denoising and voxel downsampling; S6: The point cloud data preprocessed in S5 is stitched in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching, and ICP fine stitching to obtain the stitched workpiece point cloud.

2. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: The infrared image and the point cloud depth image are matched pixel by pixel, and then the pixels of the infrared image and the point cloud depth image are superimposed. The superimposed pixel value X c The calculation is as follows: X c =pX a +qX b Among them, a represents the pixel of the point cloud depth image, b represents the pixel of the infrared image, and X a Represents the pixel value of the point cloud depth image, X b represents the pixel value of the infrared image, and p and q represent the weights of the pixels.

3. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 2, characterized in that: The fusion processing also includes: before performing the fusion operation, processing the infrared image and the point cloud depth image into the same size.

4. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 2, characterized in that: In the step S3: The image segmentation model includes an encoder and a decoder. The encoder is based on the DeeplabV3+ framework and uses GhostNet as the backbone network for feature extraction. The shallow features of the backbone network are obtained when the output step size is 4, and the deep features of the backbone network are obtained when the output step size is 16. The multi-rate hollow convolution ASPP module is used on the deep features, and then the feature map is merged with the shallow features in the decoder after 1x1 convolution and upsampling. Finally, the segmentation result is output through 3x3 convolution and upsampling; The training process of the image segmentation model includes: collecting infrared images and point cloud depth images during the laser cladding process; fusing the point cloud depth images with the corresponding infrared images; annotating the areas with excessively high temperatures pixel by pixel on the fused image containing the molten pool area to obtain a one-hot encoded label; and training the image segmentation model using the annotated labels and the fused image.

5. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 1, characterized in that: The S4 specifically includes the following sub-steps: Multiple point cloud depth images are acquired in the order in which they were taken by the camera. The resulting fused image is then used with a trained image segmentation model to identify high-temperature areas, which represent the molten pool areas in the laser cladding process. If a high-temperature area is identified, the corresponding point cloud data is cropped using that area.

6. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 1, characterized in that: The step S5 includes the following sub-steps: S5.1: Calculate the statistical features of the points in the neighborhood around each point of the multiple cropped point clouds in the order of camera shooting, select sparse outliers and remove them; S5.2: For unnecessary planes in the original point cloud, use the random sampling consistency RANSAC algorithm to perform plane segmentation and remove irrelevant parts; S5.3: Use radius filtering to remove surrounding noise from the point cloud data after plane segmentation; S5.4: Use the same ratio of voxel downsampling for all point clouds obtained in S5.3 to improve algorithm efficiency.

7. The method for performing workpiece point cloud fusion in laser cladding processing according to claim 1, characterized in that: The step S6 includes the following sub-steps: S6.1: Calculate the point cloud normals using the PCA principal component analysis algorithm for each of the multiple point clouds to be spliced ​​after preprocessing in S5; S6.2: Take the first point cloud in the camera shooting order as the reference point cloud, calculate the fast point feature histogram (FPFH) descriptor of the first subsequent point cloud and the reference point cloud, and iterate the sampling consistency initial alignment (SAC-IA) algorithm to obtain the initial transformation matrix; S6.3: Using the initial transformation matrix, transform the first subsequent point cloud into the reference point cloud coordinate system to obtain the initially transformed point cloud. S6.4: Find corresponding point pairs between the reference point cloud and the initially transformed point cloud, remove incorrect point pairs using the RANSAC algorithm, and then calculate the rotation matrix and translation vector to minimize the root mean square error between the corresponding matching point pairs to achieve fine stitching. Also, set a threshold and a maximum number of iterations to control the accuracy and time consumption of fine stitching. S6.5: Repeat S6.2 to S6.5, stitching the subsequent point clouds onto the stitched point cloud in sequence until all point clouds are stitched together.

8. A device for performing workpiece point cloud fusion in laser cladding processing, characterized in that: include: Image acquisition unit: used to obtain infrared image data and point cloud depth image of the workpiece to be measured, and respectively obtain the workpiece surface thermal distribution information and workpiece surface three-dimensional shape information; Image fusion unit: used to perform pixel-by-pixel correspondence between the infrared image and the point cloud depth image, and perform weighted summation of the corresponding pixels to obtain a fused image; the fused image has both the thermal distribution information of the infrared image and the three-dimensional structure information of the point cloud depth image; Image segmentation unit: used for segmenting the high temperature area in the fused image through an image segmentation model; Point cloud processing unit: used to determine whether each point cloud data point falls within the high-temperature area, and perform preprocessing of outlier removal, RANSAC plane segmentation, data denoising and voxel downsampling on the point cloud data after removing the high-temperature area. The point cloud data preprocessed by S5 is stitched in the order of normal vector calculation, FPFH feature descriptor calculation, SAC-IA coarse stitching and ICP fine stitching to obtain the fused workpiece point cloud.

9. An electronic device comprising a storage module comprising instructions loaded and executed by a processor; the instructions causing the processor to process the method for performing workpiece point cloud fusion in laser cladding processing as described in any one of claims 1 to 7 when executed.

10. A computer-readable storage medium storing one or more programs, characterized in that: When the program is executed by a processor, the method for performing workpiece point cloud fusion in laser cladding processing as described in any one of claims 1 to 7 is implemented.

Images