CN118097036B - A point cloud reconstruction system and method suitable for objects containing light-transmitting materials - Google Patents

Abstract

The present invention discloses a point cloud reconstruction system and method for an object containing translucent materials. The system includes a turntable, an RGB-D camera, a camera bracket, a target object and an Apriltag calibration plate. The method is implemented based on the system, including starting the system, automatically collecting data of different viewing angles of the target object; aligning the RGB image and the depth image, extracting the Apriltag corner points in the RGB image, solving the rotation matrix and translation matrix from each corner point in the camera coordinate system to the corresponding point in the Apriltag coordinate system, and obtaining the external parameters of the RGB-D camera; merging the aligned RGB image and depth image data of different viewing angles, eliminating abnormal points caused by translucent materials; point cloud segmentation and target object extraction. The present invention can greatly reduce the error of depth information caused by strong light transmission ability, provides an important tool for three-dimensional modeling and analysis, and is suitable for application scenarios requiring high-quality point cloud data.

Description

A point cloud reconstruction system and method suitable for objects containing light-transmitting materials

Technical Field

The present invention belongs to the field of three-dimensional vision technology, and in particular relates to a point cloud reconstruction system and method suitable for objects containing light-transmitting materials.

Background technique

3D vision is an important branch of computer vision, focusing on understanding and processing images and scenes in three-dimensional space, involving the use of computers and digital image processing technology to acquire, analyze and present information about the three-dimensional world. In recent years, due to the rapid development of three-dimensional sensing technology and the explosive growth of three-dimensional geometric data, 3D vision research has broken through the traditional two-dimensional image space and achieved the analysis, understanding and interaction of three-dimensional space.

At present, the point cloud reconstruction systems for objects containing translucent materials mainly include point cloud reconstruction systems based on feature point matching, voxel reconstruction systems, and 3D scanner systems, but each system has various shortcomings; the main defects of point cloud reconstruction systems based on feature point matching are: 1) Complex calculation: Feature point extraction and matching usually require a lot of computing resources and time, especially when processing large-scale point clouds, which will limit the real-time performance of the system, and the area of translucent materials generally has fewer feature points, making it difficult to accurately match and reconstruct; 2) Noise and distortion: Feature point matching is easily affected by sensor noise, illumination changes, and distortion, which can lead to mismatching or instability in point cloud reconstruction; 3) Data dependence: Point cloud reconstruction using feature point matching will cause the light pulse emitted by the depth camera to be transmitted due to the translucent material on the surface of the object, which will cause a large number of abnormal points in the collected point cloud data. These abnormal points will affect the misjudgment of feature points by the point cloud feature extraction algorithm, and ultimately lead to reconstruction failure or low reconstruction accuracy. The main defect of voxel reconstruction is that the reconstruction accuracy is not ideal. The 3D scanner system is ideal in performance, but its main drawbacks are: complex installation and high price. In summary, there is an urgent need for a low-cost and easy-to-operate system that can reconstruct 3D point clouds of objects containing light-transmitting materials.

Summary of the invention

In order to solve the above technical problems, the present invention proposes a system for three-dimensional reconstruction of objects, which can significantly reduce the error of depth information of a target object caused by strong light transmittance, and realize low-cost and highly scalable object reconstruction, including a turntable, an RGB-D camera, a camera bracket, a target object and an Apriltag calibration plate.

The turntable is used to support and rotate the target object, presenting different angles of the object to the camera for data collection. The RGB-D camera has the ability to capture RGB color images and depth information, which is used to collect visual data of the object. The camera bracket is used to stably install the RGB-D camera to ensure that it maintains the appropriate position and angle during data collection. Part of the surface of the target object is composed of light-transmitting materials. The Apriltag calibration plate is placed on the turntable to provide positioning and orientation information for the image.

Based on the above point cloud reconstruction system, the present invention also provides a point cloud reconstruction method for an object containing light-transmitting materials, comprising the following steps:

Step 1: Start the point cloud reconstruction system to automatically collect RGB images and depth images of the target object from different perspectives;

Step 2: Align the RGB image and the depth image, extract the Apriltag corner points in the RGB image, solve the rotation matrix and translation matrix from each corner point in the camera coordinate system to the corresponding point in the Apriltag coordinate system, and obtain the external parameters of the RGB-D camera;

Step 3: Fuse the aligned RGB images and depth image data of different viewing angles, remove noise points whose average Euclidean distance to neighboring points is greater than the set threshold, determine the specific position of the light-transmitting area according to the brightness gradient, remove abnormal points caused by light-transmitting materials, and obtain a point cloud;

Step 4: Use a plane model segmentation algorithm to segment the point cloud after removing abnormal points and extract the target object.

Moreover, after starting the point cloud reconstruction system in step 1, the turntable starts to rotate at a constant speed, so that the target object placed on it also rotates at a constant speed. The RGB-D camera collects an RGB image and a depth image of the target object at regular intervals according to actual needs to obtain data of the target object at different perspectives.

Moreover, in step 2, the alignment coefficient of the RGB image and the depth image is obtained through the physical interval and angle of the actual placement of the RGB sensor and the TOF sensor to achieve alignment between the two. For each data automatically collected by the RGB-D camera, the alignment coefficient is used to align the RGB image and the depth image. The collected RGB image is grayed, and then edge extraction is performed to obtain the corner point of each April tag. The pixel coordinate system takes the upper left corner of the image as the origin, the u axis is horizontal to the right, and the v axis is vertically downward. Assuming that there are n Apriltag corner points in an RGB image, the coordinates in the RGB image are , corresponding to the depth value in the depth image is According to the ID value of the identified Apriltag corner point, its three-dimensional coordinate value in the Apriltag coordinate system is obtained as , the z coordinate is 0, the origin of the Apriltag coordinate system is located at the lower left corner of the Apriltag calibration plate, the x direction and the y direction coincide with the two sides of the initial position of the Apriltag calibration plate, and the z axis is perpendicular to the calibration plate plane and points upward. The coordinate system conforms to the right-hand coordinate system.

The origin of the camera coordinate system is located at the optical center of the camera. The X -axis points to the right side of the camera, perpendicular to the optical axis, and parallel to the image plane. The Y -axis points to the top of the camera, perpendicular to the optical axis, and parallel to the X- axis and the image plane. The Z -axis points to the viewing direction of the camera, that is, to the observed scene, and is parallel to the optical axis. The depth map aligned with the RGB image is used to obtain the corner depth value. Combined with the camera intrinsic parameters, the corner coordinates are converted from the pixel coordinate system to the camera coordinate system. The coordinate conversion formula is as follows:

(1)

Where u and v are the coordinate values of the pixel point in the pixel coordinate system. Respectively represent the focal length of the camera in the u and v directions in the pixel coordinate system, represents the pixel coordinates of the principal point, X , Y, and Z are the coordinate values of the pixel point in the camera coordinate system, and the Z value of the pixel point in the camera coordinate system is the depth value d .

Camera extrinsics refers to the camera's position and posture , including the rotation matrix R and the translation matrix t . Assume that there are multiple rays, each of which connects the camera optical center, the three-dimensional target point, and the projection of the target point on the camera plane. The point in the camera coordinate system is represented by Indicates that points in the Apriltag coordinate system are represented by In other words, solving the camera extrinsic parameters is to solve the rotation matrix and translation matrix of each point in the camera coordinate system to the corresponding point in the Apriltag coordinate system; the camera extrinsic parameters are adjusted by So that the expression The minimum value of is obtained, Represents the square of the 2-norm.

Moreover, in step 3, the aligned RGB image and depth image are first converted to the Apriltag coordinate system using the camera external parameters, and then the RGB image and depth image data collected from this perspective are converted to the Apriltag coordinate system at the initial moment according to the angle of rotation of the Apriltag at the current moment compared to the initial moment, so as to achieve multi-angle view fusion. Calculate the Euclidean distance between each point in the point cloud data after fusion of multiple RGB images and depth images from different perspectives and the points in its K neighborhood, and calculate the mean of all Euclidean distances and standard deviation , take the distance threshold ,in is a constant, i.e., the proportional coefficient. The point cloud is traversed again, and the points whose average Euclidean distance to the K neighboring points is greater than For abnormal points caused by translucent materials, the Harris feature point detection method is first used to extract feature points in the RGB image, and these feature points are evenly divided into N subsets. Then, the direction of intensity change of the area around each feature point in the subset is calculated. With each feature point as the origin, rays pointing to the direction of increased brightness are drawn. If a certain number of rays in the subset intersect at a common point, the closed area formed by these points is identified as the translucent material area, and the feature points in the area are deleted. This judgment and deletion operation is performed on the feature points in the subset one by one to eliminate abnormal data.

Moreover, in step 4, the RGB images and depth information of the target object collected continuously in steps 1-3 at various angles are unified to the Apriltag coordinate system at the initial moment, and after the data processing of the abnormal points caused by the light-transmitting material is eliminated, the multi-angle fused point cloud data is obtained. Each point cloud is screened, and in the Aptiltag coordinate system, the points with z coordinate values less than 0 and x and y greater than the size of the calibration plate are removed to obtain the point cloud information of the target object and the turntable. For the case where the turntable with a contact area with the target object is difficult to segment, the plane model segmentation algorithm is used to segment the fused point cloud data into voxels, and only the center point of each voxel is retained. The iteration number threshold, distance threshold and internal point number threshold are set, and the center points of M voxels are randomly selected. The plane equation is fitted using the Ransac algorithm to estimate the parameters of the plane model. The distance from the center point of all voxels to the estimated plane is calculated, and the point with a distance less than the threshold is regarded as the internal point. The number of internal points is counted, and the iteration stops when the specified number of iterations is reached or the number of internal points reaches the set threshold. Removing the internal points in the plane model means removing the turntable point cloud, and finally obtaining the three-dimensional target object model.

Compared with the prior art, the present invention has the following advantages:

1) Through the automated rotation of the turntable and regular image acquisition, data acquisition is automated, which reduces user intervention, makes data acquisition more efficient, and helps improve the real-time performance of the point cloud reconstruction system and the consistency of data acquisition;

2) By analyzing the Apriltag information in the captured image, the camera's external parameters, including position and orientation information, are calculated, providing a reliable basis for subsequent data processing and point cloud stitching, and improving the accuracy of point cloud reconstruction;

3) Obtain images at different angles by rotating the target object and fusing the multi-angle views to obtain a more comprehensive view of the object, which helps reduce occlusion and provide more geometric information, thereby improving the accuracy and completeness of point cloud reconstruction;

4) Considering the interference of the target object's translucent material on the depth information, the accuracy of the point cloud data is ensured by detecting and removing points with abnormal depth values caused by strong light transmittance, solving the noise and distortion problems of feature point matching;

5) Compared with other point cloud reconstruction systems on the market, the point cloud reconstruction system provided by the present invention has a relatively simple structure, uses relatively common equipment, and does not require a complicated installation process, thereby reducing the cost of the system and providing convenience in operation.

In summary, the present invention solves various problems existing in the current point cloud reconstruction system through technological innovation, and provides an efficient, accurate, low-cost and easy-to-operate point cloud reconstruction system and method suitable for objects containing translucent materials.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG. 1 is a structural device diagram of a point cloud reconstruction system for an object containing light-transmitting materials according to an embodiment of the present invention.

FIG. 2 is a flow chart of a method for reconstructing a point cloud of an object containing light-transmitting materials according to an embodiment of the present invention.

FIG. 3 is a flow chart of a camera extrinsic parameter estimation method according to an embodiment of the present invention.

FIG. 4 is a flow chart of data processing of light-transmitting materials in an embodiment of the present invention.

FIG5 is a flow chart of a point cloud segmentation algorithm in an embodiment of the present invention.

Detailed ways

In order to facilitate understanding of the present application, the present application will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present application are given in the drawings. However, the present application can be implemented in a variety of different forms and is not limited to the embodiments described herein. In addition, the embodiments provided are intended to provide a more thorough and comprehensive disclosure of the present application.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and/or" used herein includes any and all combinations of one or more of the related listed items.

Example 1

As shown in FIG1 , Embodiment 1 of the present invention provides a point cloud reconstruction system for an object containing light-transmitting materials, including a turntable, an RGB-D camera, a camera bracket, a car model, and an Apriltag calibration plate.

The turntable is used to support and rotate the target object, presenting different angles of it to the camera for data collection. The RGB-D camera has RGB (color image) and depth (distance information) capture capabilities, which are used to collect visual data of the object. The camera bracket is used to stably install the RGB-D camera to ensure that it maintains the appropriate position and angle during data collection. Part of the surface of the target object is composed of light-transmitting materials. In this embodiment, the target object is a car model, which includes a tempered glass window with light-transmitting properties. The Apriltag calibration plate is placed on the turntable to provide positioning and orientation information for the image. The entire point cloud reconstruction system needs to be placed in front of a solid color background.

Example 2

Based on the object point cloud reconstruction system provided in Example 1, Example 2 of the present invention further provides an object point cloud reconstruction method applicable to light-transmitting materials, as shown in FIG2 , comprising the following steps:

Step 1: Automatically collect data from different perspectives of the target object.

After starting the object point cloud reconstruction system, the turntable starts to rotate at a constant speed, causing the car model placed on it to rotate at a constant speed. The RGB-D camera collects an RGB image and a depth image of the car model at regular intervals according to actual needs to obtain data of the car model at different perspectives for subsequent processing.

Step 2: Align the RGB image and the depth image, extract the Apriltag corner points in the RGB image, solve the rotation matrix and translation matrix from each corner point in the camera coordinate system to the corresponding point in the Apriltag coordinate system, and obtain the external parameters of the RGB-D camera.

As shown in FIG3 , Embodiment 2 of the present invention further provides a camera extrinsic parameter estimation method, and the specific steps are as follows:

Since there is a certain physical distance between the RGB sensor and the TOF sensor in the RGB-D camera, it is necessary to obtain the alignment coefficient of the RGB image and the depth image through the actual physical interval and angle between the RGB sensor and the TOF sensor to achieve alignment between the two. For each data automatically collected by the RGB-D camera, the alignment coefficient needs to be used to align the RGB image and the depth image.

The collected RGB image is grayed out, and then edge extraction is performed to obtain the corner points of each April tag. The pixel coordinate system takes the upper left corner of the image as the origin, the u axis is horizontal to the right, and the v axis is vertically downward. Assuming that there are 15 complete April tags in an RGB image, and each April tag has 4 corner points, there are 60 April tag corner points, and the pixel coordinates in the RGB image are , corresponding to the depth value in the depth image is According to the ID value of the identified Apriltag corner point, its three-dimensional coordinate value in the Apriltag coordinate system is obtained as , the z coordinate is 0. The origin of the Apriltag coordinate system is located at the lower left corner of the Apriltag calibration plate. The x and y directions coincide with the lengths of the two sides of the initial position of the Apriltag calibration plate. The z axis is perpendicular to the calibration plate plane and points upward. The coordinate system conforms to the right-hand coordinate system.

The origin of the camera coordinate system is located at the optical center of the camera (the intersection of the optical axes). The X -axis points to the right of the camera, perpendicular to the optical axis, and parallel to the image plane. The Y -axis points to the top of the camera, perpendicular to the optical axis, and parallel to the X- axis and the image plane. The Z -axis points to the viewing direction of the camera, that is, to the observed scene, and is parallel to the optical axis. The depth map aligned with the RGB image is used to obtain the corner depth value. Combined with the camera intrinsic parameters, the corner coordinates are converted from the pixel coordinate system to the camera coordinate system. The coordinate conversion formula is as follows:

(1)

Where u and v are the coordinate values of the pixel point in the pixel coordinate system. Respectively represent the focal length of the camera in the u and v directions in the pixel coordinate system, represents the pixel coordinates of the principal point, X , Y, and Z are the coordinate values of the pixel point in the camera coordinate system, and the Z value of the pixel point in the camera coordinate system is the depth value d .

Camera extrinsics refers to the camera's position and posture , including the rotation matrix R and the translation matrix t . Assume that there are multiple rays, each of which connects the camera optical center, the three-dimensional target point, and the projection of the target point on the camera plane. Indicates that points in the Apriltag coordinate system are represented by In other words, solving the camera extrinsic parameters is to solve the rotation matrix and translation matrix of each point in the camera coordinate system to the corresponding point in the Apriltag coordinate system; the camera extrinsic parameters are adjusted by So that the expression The minimum value of is obtained, represents the square of the 2-norm. This embodiment uses the nonlinear optimization technology of Bundle Adjustment (BA) to solve the external parameters. BA improves the accuracy of estimation by minimizing the reprojection error and can achieve high-precision conversion.

Step 3: Fuse the aligned RGB images and depth image data of different perspectives, remove noise points whose average Euclidean distance to neighboring points is greater than the set threshold, determine the specific position of the translucent area based on the brightness gradient, and remove abnormal points caused by translucent materials.

As shown in FIG. 4 , Embodiment 2 of the present invention further provides a method for processing data of a light-transmitting material, which is specifically as follows:

First, the aligned RGB image and depth image are converted to the Apriltag coordinate system using the camera extrinsics. Then, according to the angle that the Apriltag at the current moment has rotated compared to the initial moment, the RGB image and depth image data collected from this perspective are converted to the Apriltag coordinate system at the initial moment, thereby realizing multi-angle view fusion. Since the light-transmitting windows of the car model may interfere with the depth information collected by the TOF camera, based on the continuous closure characteristics of the car model, the Euclidean distance between each point in the point cloud data after the fusion of multiple RGB images and depth images from different perspectives and the points in its K neighborhood is calculated (in this embodiment, K is 20, and the value can be taken according to the situation during specific implementation), and the mean of all Euclidean distances is calculated. and standard deviation , take the distance threshold ,in is a constant, i.e., the proportional coefficient. The point cloud is traversed again, and the points whose average Euclidean distance to the 20 neighboring points is greater than point.

For abnormal points caused by translucent materials, the Harris feature point detection method is first used to extract feature points in the RGB image, and these feature points are evenly divided into N subsets (N is 10 in this embodiment, and the value can be taken according to the situation during specific implementation). Then, the intensity change direction of the area around each feature point in the subset is calculated, and each feature point is used as the origin to make rays pointing to the direction of increasing brightness. If there are a certain number of rays in the subset that intersect at a common point, the closed area formed by these points is determined to be a translucent material area, and the feature points in the area are deleted. The above judgment and deletion operations are performed on the feature points in the subset one by one to eliminate abnormal data.

Step 4: point cloud segmentation and target object extraction.

Through steps 1-3, the RGB images and depth information of the target object at various angles are continuously collected and unified into the Apriltag coordinate system at the initial moment. After data processing of abnormal points caused by translucent materials is eliminated, multi-angle fused point cloud data is obtained.

In order to extract the car model from the scene, each point cloud is screened, and in the Aptiltag coordinate system, the points with z coordinate values less than 0 and x and y values greater than the size of the calibration plate are removed to obtain the point cloud information of the target object and the turntable. For the case where the turntable that has a contact area with the car model is difficult to segment, a plane model segmentation algorithm is used. The specific operation is shown in Figure 5. The fused point cloud data is segmented into voxels. For each voxel, only its center point is retained. The iteration number threshold, distance threshold and internal point number threshold are set (in this embodiment, the maximum number of iterations is set to 1000, the distance threshold is set to 0.01, and the internal point number threshold is set to 30% of the original point cloud number). The center points of M voxels are randomly selected (in this embodiment, M is 50, and the value can be taken according to the situation during specific implementation), and the plane equation is fitted using the Ransac algorithm to estimate the parameters of the plane model. Assume that is the set of M voxel centers, is any point in the point set P, It is a point set The center of , assuming that the fitted target plane passes through , and its normal vector is , using the least squares method to solve make The distance from all points to the plane is The value of is the smallest. Define the matrix ,but:

(2)

In the formula, It represents the norm of the matrix and the superscript T represents the matrix transpose.

Perform singular value decomposition on matrix A and get , substituting into formula (2) we get:

(3)

Where V is The unitary matrix of The unitary matrix of Is a In a matrix, each element on the main diagonal is a singular value, except for the elements on the main diagonal, which are all 0, that is, .

Will Use one The matrix W represents , then formula (3) is expressed as:

(4)

In the formula, , , are all singular values, and ; , , are the elements of the matrix W.

Finally, the fitted plane normal vector is obtained is the third column of matrix U , that is .

Calculate the distance from the center point of all voxels to the estimated plane, take the points with distance less than the threshold as inliers, count the number of inliers, and stop the iteration when the specified number of iterations is reached or the number of inliers reaches the set threshold. Remove the inliers in the plane model, that is, remove the turntable point cloud, and finally obtain the 3D target object model.

In specific implementation, the method proposed in the technical solution of the present invention can be implemented by technical personnel in this field using computer software technology to realize automatic operation process. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of the present invention and computer equipment running the corresponding computer program, should also be within the protection scope of the present invention.

The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific implementation cases described, or replace them with similar methods, without departing from the spirit of the present invention or exceeding the scope defined by the appended claims.

Claims (8)

1. A point cloud reconstruction method for an object containing light-transmitting materials, characterized in that: The object point cloud reconstruction process is implemented based on a point cloud reconstruction system, which includes a turntable, an RGB-D camera, a camera bracket, a target object and an Apriltag calibration plate; The turntable is used to support and rotate the target object, presenting different angles of the object to the camera for data collection; the RGB-D camera has the ability to capture RGB color images and depth information, and is used to collect visual data of the object; the camera bracket is used to stably install the RGB-D camera; part of the surface of the target object is made of light-transmitting materials; the Apriltag calibration plate is placed on the turntable to provide positioning and orientation information for the image; the entire point cloud reconstruction system needs to be placed in front of a solid color background; The object point cloud reconstruction process includes the following steps: Step 1: Start the point cloud reconstruction system to automatically collect RGB images and depth images of the target object from different perspectives; Step 2: Align the RGB image and the depth image, extract the Apriltag corner points in the RGB image, solve the rotation matrix and translation matrix from each corner point in the camera coordinate system to the corresponding point in the Apriltag coordinate system, and obtain the external parameters of the RGB-D camera; Step 3: Fuse the aligned RGB images and depth image data of different viewing angles, remove noise points whose average Euclidean distance to neighboring points is greater than the set threshold, determine the specific position of the light-transmitting area according to the brightness gradient, remove abnormal points caused by light-transmitting materials, and obtain a point cloud; First, the aligned RGB image and depth image are converted to the Apriltag coordinate system using the camera extrinsic parameters. Then, according to the angle that the Apriltag at the current moment has rotated compared to the initial moment, the RGB image and depth image data collected from this perspective are converted to the Apriltag coordinate system at the initial moment to achieve multi-angle view fusion. The Euclidean distance between each point in the point cloud data after fusion of multiple RGB images and depth images from different perspectives and the points in its K neighborhood is calculated, and the mean of all Euclidean distances is calculated. and standard deviation , take the distance threshold ,in is a constant, i.e., the proportional coefficient. The point cloud is traversed again, and the points whose average Euclidean distance to the K neighboring points is greater than For abnormal points caused by translucent materials, the Harris feature point detection method is first used to extract feature points in the RGB image, and these feature points are evenly divided into N subsets. Then, the intensity change direction of the area around each feature point in the subset is calculated. With each feature point as the origin, rays pointing to the direction where the brightness increases are drawn. If a certain number of rays in the subset intersect at a common point, the closed area formed by these points is identified as the translucent material area, and the feature points in the area are deleted. This judgment and deletion operation is performed on the feature points in the subset one by one to eliminate abnormal data. Step 4: Use a plane model segmentation algorithm to segment the point cloud after removing abnormal points and extract the target object. 2. A point cloud reconstruction method for an object containing a light-transmitting material as described in claim 1, characterized in that: after the point cloud reconstruction system is started in step 1, the turntable starts to rotate at a constant speed, so that the target object placed thereon also rotates at a constant speed; the RGB-D camera collects an RGB image of the target object and a depth image of the target object at regular intervals according to actual needs to obtain data of the target object at different viewing angles. 3. A point cloud reconstruction method for an object containing a light-transmitting material as described in claim 1, characterized in that: in step 2, the alignment coefficient of the RGB image and the depth image is obtained by the physical interval and angle of the actual placement of the RGB sensor and the TOF sensor to achieve alignment between the two, and for each data automatically collected by the RGB-D camera, the RGB image and the depth image are aligned using the alignment coefficient; the collected RGB image is grayed, and then edge extraction is performed to obtain each Apriltag corner point; the pixel coordinate system takes the upper left corner of the image as the origin, the u axis is horizontal to the right, and the v axis is vertical downward. Assuming that there are n Apriltag corner points in an RGB image, the pixel coordinates in the RGB image are , corresponding to the depth value in the depth image is According to the ID value of the identified Apriltag corner point, its three-dimensional coordinate value in the Apriltag coordinate system is obtained as , the z coordinate is 0, the origin of the Apriltag coordinate system is located at the lower left corner of the Apriltag calibration plate, the x direction and the y direction coincide with the two sides of the initial position of the Apriltag calibration plate respectively, the z axis is perpendicular to the calibration plate plane and points upward, and the coordinate system conforms to the right-hand coordinate system. 4. A point cloud reconstruction method for an object containing a light-transmitting material as claimed in claim 3, characterized in that: in step 2, the coordinate origin of the camera coordinate system is located at the optical center of the camera, the X -axis points to the right side of the camera, is perpendicular to the optical axis, and is parallel to the image plane, the Y -axis points to the top of the camera, is perpendicular to the optical axis, and is parallel to the X- axis and the image plane, and the Z -axis points to the observation direction of the camera, that is, points to the observed scene, and is parallel to the optical axis; the depth value of the corner point is obtained by using the depth map aligned with the RGB image, and the corner point coordinates are converted from the pixel coordinate system to the camera coordinate system in combination with the camera intrinsic parameter, and the coordinate conversion formula is as follows: (1) Where u and v are the coordinate values of the pixel point in the pixel coordinate system. Respectively represent the focal length of the camera in the u and v directions in the pixel coordinate system, represents the pixel coordinates of the principal point of the image, X , Y, Z are the coordinate values of the pixel point in the camera coordinate system, and the Z value of the pixel point in the camera coordinate system is the depth value d ; Camera extrinsics refers to the camera's position and posture , including the rotation matrix R and the translation matrix t . Assume that there are multiple rays, each of which connects the camera optical center, the three-dimensional target point, and the projection of the target point on the camera plane. The point in the camera coordinate system is represented by Indicates that points in the Apriltag coordinate system are represented by In other words, solving the camera extrinsic parameters is to solve the rotation matrix and translation matrix of each point in the camera coordinate system to the corresponding point in the Apriltag coordinate system; the camera extrinsic parameters are adjusted by So that the expression The minimum value of is obtained, Represents the square of the 2-norm. 5. A method for reconstructing a point cloud of an object containing a light-transmitting material as claimed in claim 1, characterized in that: in step 4, the RGB images and depth information of the target object continuously collected in step 1 at various angles are unified into the Apriltag coordinate system at the initial moment, and after removing the data of abnormal points caused by the light-transmitting material, the multi-angle fused point cloud data is obtained; each point cloud is screened, and in the Aptiltag coordinate system, the points with z coordinate values less than 0 and x , For the points where y is larger than the size of the calibration plate, the point cloud information of the target object and the turntable is obtained. In the case where the turntable with a contact area with the target object is difficult to segment, a plane model segmentation algorithm is used to segment the fused point cloud data into voxels. For each voxel, only its center point is retained. The iteration number threshold, distance threshold and internal point number threshold are set. The center points of M voxels are randomly selected, and the plane equation is fitted using the Ransac algorithm to estimate the parameters of the plane model. The distance from the center point of all voxels to the estimated plane is calculated, and the points with a distance less than the threshold are taken as internal points. The number of internal points is counted, and the iteration stops when the specified number of iterations is reached or the number of internal points reaches the set threshold. The internal points in the plane model are removed, that is, the turntable point cloud is removed, and finally the three-dimensional target object model is obtained. 6. A method for reconstructing a point cloud of an object containing a light-transmitting material as claimed in claim 5, characterized in that: the calculation process of estimating the parameters of the plane model by fitting the plane equation using the Ransac algorithm in step 4 is as follows: Assumptions is the set of M voxel centers, is any point in the point set P, It is a point set The center of , assuming that the fitted target plane passes through , and its normal vector is , using the least squares method to solve make The distance from all points to the plane is The value of is the smallest, defining the matrix ,but: (2) In the formula, represents the norm of the matrix, and the superscript T represents the matrix transpose; Perform singular value decomposition on matrix A and get , substituting into formula (2) we get: (3) Where V is The unitary matrix of The unitary matrix of Is a In a matrix, each element on the main diagonal is a singular value, except for the elements on the main diagonal, which are all 0, that is, ; Will Use one The matrix W represents , then formula (3) is expressed as: (4) In the formula, , , are all singular values, and ; , , are the elements of the matrix W; The final fitted plane normal vector is the third column of the matrix U. 7. A point cloud reconstruction device suitable for objects containing translucent materials, characterized in that it includes a processor and a memory, the memory is used to store program instructions, and the processor is used to call the program instructions in the memory to execute a point cloud reconstruction method suitable for objects containing translucent materials as described in any one of claims 1-6. 8. A readable storage medium, characterized in that a computer program is stored on the readable storage medium, and when the computer program is executed, a point cloud reconstruction method for an object containing a light-transmitting material as described in any one of claims 1 to 6 is implemented.