Palletizing loader and loading method thereof

By designing a palletizing loading machine, using laser point cloud stitching and rotating image feature descriptors to improve ICP algorithm, automated loading is achieved, solving the problems of low loading efficiency and high cost caused by manual dependence, and improving loading efficiency and security.

CN117864803BActive Publication Date: 2024-11-29杭州灵智科技数字化装备有限公司

Patent Information

Application Number
CN202410068025.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-17
Publication Date
2024-11-29
Estimated Expiration
2044-01-17

AI Technical Summary

Technical Problem

The existing loading methods rely on labor, resulting in low loading efficiency, high labor costs, and threats to human safety in harsh environments, making it difficult to achieve automated loading.

Method used

A palletizing loading machine is designed, including a loading platform, conveying device, a rotary table, a robotic arm and a vehicle-mounted platform. Through laser point cloud splicing and improved ICP algorithm based on the rotational image feature descriptor based on the intensity domain, it realizes automatic materials. Handling and palletizing trucks do not require manual participation.

Benefits of technology

Automatic palletizing and loading is realized, loading efficiency is improved, labor costs are reduced, and the safety and integrity of goods are ensured in harsh environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117864803B_ABST
    Figure CN117864803B_ABST
Patent Text Reader

Abstract

The present invention discloses a palletizing loader and a loading method thereof, which belong to the field of loading equipment. A palletizing loader comprises: a loading platform, having an upper layer and a lower layer, and the two ends of the lower layer are an entrance and an exit respectively; materials are placed on the upper layer; a truck is located on the lower layer; the palletizing loader also comprises: a conveying device, arranged on the loading platform, for transporting the materials on the upper layer away from the upper layer; a transfer table, arranged on the loading platform, located between the upper layer and the lower layer; a mechanical arm, arranged on the loading platform, for transporting the materials in the transfer table to the truck bucket; a vehicle-mounted platform, slidably arranged on the lower layer by an electric slide rail, for transporting the truck from the entrance to the exit. The beneficial effect of the present application is to provide a palletizing loader and a loading method thereof that can realize automated palletizing and loading, replace the manual loading method, realize accurate palletizing of materials layer by layer, and are highly automated, improve loading efficiency, and reduce labor costs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of loading equipment, and in particular to a palletizing loader and a loading method thereof. Background Art

[0002] With the continuous development of automation level, more and more industries have realized automated assembly, production and testing. The field of cargo storage and loading and unloading has gradually developed from traditional manual palletizing to automated palletizing and disassembly.

[0003] However, in the field of loading, pallet stacking, manual loading and semi-automatic loading are currently more commonly used. Therefore, improving the level of automation in the field of loading and replacing manual loading with automatic loading is an inevitable development trend in the field of loading in the future.

[0004] Compared with traditional manual loading, pallet stacking and semi-automatic loading have saved a lot of time and labor costs. However, this loading method increases the cost of pallets and cannot completely do without manual labor. Loading in harsh environments may even cause harm to the human body.

[0005] The existing loading method is highly dependent on manual labor and is difficult to do without manual labor, resulting in low loading efficiency and high labor costs, which is not conducive to the development of automated loading. Summary of the invention

[0006] The purpose of the present invention is to provide a palletizing loader and a loading method thereof. The present invention realizes automatic palletizing and loading, replaces the manual loading method, breaks the bottleneck in the existing loading field, realizes accurate palletizing of materials layer by layer, and is highly automated, does not require manual participation throughout the process, is intelligent and efficient, improves loading efficiency, and reduces labor costs.

[0007] The content of this application is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this application is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0008] In order to solve the technical problems mentioned in the above background technology section, some embodiments of the present application provide a palletizing loader, including:

[0009] The loading platform has an upper layer and a lower layer, and the two ends of the lower layer are an entrance and an exit respectively;

[0010] The materials are placed on the upper layer;

[0011] The vans are located on the lower level;

[0012] The pallet loader also includes:

[0013] A conveying device, provided on the loading platform, for transporting materials on the upper layer away from the upper layer;

[0014] The transfer platform is located on the loading platform, between the upper and lower levels;

[0015] The mechanical arm is installed on the loading platform and is used to move the materials in the transfer platform to the truck bed;

[0016] The vehicle-mounted platform is slidably mounted on the lower layer via an electric slide rail and is used to transport the truck from the entrance to the exit;

[0017] Among them, the materials transported by the conveying device are transported to the transfer platform; the robotic arm is located between the exit and the entrance; and a camera for monitoring the real-time status of the material woven bags is provided on the loading platform.

[0018] A loading method for a palletizing loader comprises the following steps:

[0019] S1, detect whether the truck arrives at the vehicle-mounted platform;

[0020] S2, collecting image information of cargo woven bags;

[0021] S3, collecting space information in the carriage;

[0022] S4. The robotic arm moves the materials on the transfer table to the carriage and stacks them according to the collected information.

[0023] Furthermore, in step S3, the car point cloud is first collected using a laser, and then the point cloud in the vertical direction is obtained, and the ICP algorithm is improved using a rotation image feature descriptor based on the intensity domain to solve the coordinate conversion parameters between point clouds in different frames, and finally the three-dimensional point cloud data collected by multiple scans are spliced ​​to convert the local point cloud data of multiple perspectives into the same reference coordinate system to obtain a complete three-dimensional point cloud of the spatial information in the car.

[0024] Furthermore, the ICP algorithm process is to filter the source point cloud and the target point cloud at the same time, then estimate the normal, then perform intensity domain rotation image features, then estimate the corresponding points of the two intensity domain rotation image features obtained from the source point cloud and the target point cloud, and finally ICP calculates the matching parameters.

[0025] Further, normal estimation, first set p i is the collected point cloud, a point in the data set, whose normal is n i , the k-neighborhood is N(p i )={q i,1 ,q i,2 ,…,q i,k}, for each point p i The corresponding covariance matrix C is as follows:

[0026]

[0027] Where t represents Q i The number of neighboring points of a point, Q represents Q i The total centroid of all points in the neighborhood of point is the jth eigenvalue of the covariance matrix, and V is the jth eigenvector.

[0028] Furthermore, the intensity domain rotation image features, assuming that point p represents a point in three-dimensional space, the normal vector of the three-dimensional surface at point p is represented by n, and P represents the tangent plane of the three-dimensional surface at the point; let x be any point other than point p on the three-dimensional surface, define a to represent the distance between point x and the normal vector n, and the distance from x to the tangent plane P is represented by B; let (α i , β i ) represents the projection of other points on the three-dimensional surface other than point p on P, i is the index of the two-dimensional mesh vertex, and statistics (α i , β i ) The two-dimensional histogram obtained by α and β represents the rotation image of point p. The value of the coordinate two-dimensional grid of the rotation image depends on the number and intensity value of the points (α, β) contained in the unified statistical interval.

[0029] Mapping function that projects a 3D point cloud onto a 2D mesh:

[0030] S P :R 3 →R 2 ;

[0031]

[0032] The 3D point cloud model is projected into a 2D grid through S projection, and the 2D model is rasterized using the following formula:

[0033]

[0034] In the above formula, b is the grid resolution, ω is the width of the two-dimensional grid, and (i, j) is the corresponding grid;

[0035] The normalized linear correlation coefficient estimates the similarity of two rotated images. The formula for calculating the linear correlation coefficient of two rotated images is as follows:

[0036]

[0037] Where P and Q represent two intensity domain transfer images, and x represents the i-th dimension feature of P and Q;

[0038] Where -1<R(P,Q)<1;

[0039] The following formula is used to describe the similarity of two intensity-rotated images:

[0040]

[0041] Where tanf represents the weight describing the predicted correlation coefficient value. For each point in the source point cloud P and the target point cloud 0, the rotation image descriptor in the intensity domain is calculated, and the corresponding points in the target point cloud 0 are found to obtain the corresponding point set.

[0042] Furthermore, the ICP calculates the concatenation parameters of the matching parameters including the following steps:

[0043] (1) Solving the initial transformation parameters. According to the basic principle of the ICP algorithm, after solving the corresponding point set, the least squares method is used to iteratively solve the rotation matrix and translation matrix by estimating the rigid body transformation to obtain the coordinate transformation parameters (x, y, z, θ, γ);

[0044] (2) Multi-sensor combination optimization splicing parameters, first construct a 10-dimensional state vector X, including three-directional position, velocity and attitude quaternion construction process equation and observation equation:

[0045] x k =f(x k-1 , μ k-1 ,ω k-1 );

[0046] z k =h(x k , v k );

[0047] The random variables k and vk represent process noise and observation noise respectively, which are assumed to be 0 and further linearized as follows:

[0048]

[0049]

[0050] Among them, A represents the Jacobian matrix of the partial derivative of f with respect to x, W represents the Jacobian matrix of the partial derivative of f with respect to ω, H represents the Jacobian matrix of the partial derivative of h, and V represents the Jacobian matrix of the partial derivative to calculate the prior estimation error covariance:

[0051]

[0052] Among them, W k Represents a Gaussian white noise model and satisfies a zero-mean distribution, that is:

[0053] p(ω)~N(0,Q);

[0054] p(v)~n(0,R);

[0055] Calculate the Kalman gain:

[0056]

[0057] Construct the observation equation:

[0058]

[0059] Update the error covariance:

[0060]

[0061] The intensity-domain-based rotation image descriptor is used to improve the ICP algorithm to solve the coordinate transformation parameters, high-precision IMU data and light intensity data for combined solution of x, y, and z.

[0062] Furthermore, the 3D laser point cloud stitching of the carriage is to solve the 6 transformation parameters (x, y, z, θ, γ), realize the splicing of two sets of point cloud data scanned under different viewing angles; take two frames of point cloud data sets to be spliced ​​under different scanning viewpoints as an example, set them as P and Q respectively, with overlapping areas, and their coordinate systems are OX n Y n Z n and OX b Y b Z b , let the positions of any points of the same name in the overlapping area 2 in the P and Q coordinate systems be p i ,q j , then any pair of points with the same name in the two point cloud datasets (p i ,q j ) satisfies the following transformation matrix:

[0063] q j =Mp i ;

[0064] in, M represents the transformation matrix;

[0065] The process of coordinate system transformation is to transform the initial coordinate system OX n Y n Z n After rotating around its three axes, the origin of the rotation system is translated to the next coordinate system OX b Y b Z b The transformation is described by the following formula:

[0066]

[0067] Where μ is the scale factor of the two coordinate systems, C and T are rotation matrix and translation matrix respectively, and the translation matrix T is:

[0068]

[0069] Follow the following rotation order:

[0070] OX n Y n Z n →OX1Y1Z1→OX2Y2Z2→OX b Y b Z b , the three rotations correspond to three transformation matrices:

[0071]

[0072]

[0073]

[0074] Multiplying these three equations together yields:

[0075]

[0076] Compute the rotation matrix:

[0077]

[0078] According to the calculated translation matrix T and rotation matrix C, the splicing between the two point cloud data P and Q is realized, and then all the point clouds are transformed into the global coordinate system to obtain the spliced ​​complete car point cloud.

[0079] Furthermore, step S2 collects image information of the cargo woven bag, including: using a camera to collect the image of the woven bag, repeatedly iterating the collected image of the woven bag based on the growth segmentation method of regional grayscale difference until all pixels of the image are traversed, and then using the dynamic region growing method to analyze and obtain the relationship model between the illumination change and the dynamic growth threshold, performing morphological processing on it, and then extracting the center and edge information of the woven bag to locate the woven bag.

[0080] Further, in step S4, through the matrix transformation formula B T O = B T C C T O To obtain the position relationship between the object to be operated and the base, the formula E T O = E T BB T O Obtain the position and posture relationship between the object to be operated and the end effector in real time for subsequent robot arm motion trajectory planning;

[0081] Among them, B represents the base coordinate system, C represents the camera coordinate system, O represents the object coordinate system, and E represents the gripper coordinate system, so for example: B T C It is the transformation matrix used to calculate the mutual transformation between the camera coordinate system and the base coordinate system; E T O It is the transformation matrix used to calculate the mutual transformation between the gripper coordinate system and the object coordinate system.

[0082] Compared with the prior art, the present invention has the following beneficial effects:

[0083] 1. Through the use of conveying devices, robotic arms and other devices, the materials on the upper layer are automatically transported to the carriage, reducing labor and improving work efficiency.

[0084] 2. By setting up the upper and lower layers, the truck and the materials are isolated, and automatic loading is carried out after the truck is in place, which can increase the loading space of the truck and prevent the truck from hitting the goods during movement and causing damage to the goods.

[0085] 3. The camera is used to capture images of the material packaging woven bags, and the images of the woven bags are processed and separated. At the same time, the spatial information in the carriage is collected and the data of the two are compared and processed, so that the robotic arm can stack and load the materials according to the actual situation in the carriage, which is convenient and fast. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The illustrative embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application.

[0087] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the components and elements are not necessarily drawn to scale.

[0088] In the attached picture:

[0089] Figure 1 is an overall schematic diagram according to an embodiment of the present application;

[0090] Figure 2 It is a structural schematic diagram of a part of the embodiment, mainly showing the structure of the transfer station;

[0091] Figure 3 It is an overall flow chart of the embodiment of the present application;

[0092] Figure 4 It is the framework of the Drost-PPF algorithm;

[0093] Figure 5 is to select the scene reference point S r Schematic diagram of

[0094] Figure 6 This is a schematic diagram of querying the model hash table and voting;

[0095] Figure 7 It is the flow chart of improved icp algorithm;

[0096] Figure 8 It is a schematic diagram of the characteristics of the intensity domain rotation image;

[0097] Fig. 9 It is the woven bag image processing process;

[0098] Fig.10 It is a flow chart of the dynamic region growing method;

[0099] Fig.11 It is the result graph of the (D, T) least squares fitting straight line;

[0100] Fig.12 It is a flowchart of automatic loading;

[0101] Fig.13 This is a calibration flow chart for clamping of a multi-adaptive clamping manipulator;

[0102] Fig.14 It is a schematic diagram of the center calibration of the woven bag;

[0103] Fig.15 This is a schematic diagram of the complete point cloud stitching result.

[0104] Reference numerals:

[0105] 100. Palletizing and loading machine;

[0106] 101, loading platform; 101a, upper level; 101b, lower level; 101c, entrance; 101d, exit;

[0107] 102. Conveying device;

[0108] 103. Relay station;

[0109] 104. Robotic arm;

[0110] 105. Car-mounted station;

[0111] 106. Electric slide. DETAILED DESCRIPTION

[0112] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0113] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0114] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0115] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0116] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0117] Embodiment 1

[0118] Reference Figure 1-2 The palletizing loader 100 includes: a loading platform 101, a conveying device 102, a transfer platform 103, a mechanical arm 104, and a vehicle-mounted platform 105; the loading platform 101 is divided into an upper and lower layer, and the two ends of the lower layer 101b are respectively an entrance 101c and an exit 101d for trucks; the conveying device 102 is a belt conveyor, which is arranged on the loading platform 101 and is used to transport the materials on the upper layer 101a to the transfer platform 103; the transfer platform 103 is arranged on the loading platform 101, between the upper layer 101a and the lower layer 101b The robot arm 104 is arranged on the loading platform 101, and is used to carry the materials on the transfer platform 103 to the compartment of the truck and stack the materials in the compartment. A plurality of cameras for monitoring the real-time status of woven bags filled with materials are arranged on the loading platform 101, and the cameras are CCD cameras; the vehicle-mounted platform 105 is slidably arranged on the lower layer 101b through an electronic slide rail, and is used to transport the truck from the entrance 101c to the bottom of the transfer platform 103 for loading, and then transport the truck to the exit 101d after the loading is completed.

[0119] When loading, the truck drives onto the vehicle platform 105 from the entrance 101c, and then the electric slide rail 106 drives the vehicle platform 105 and the truck to move to the bottom of the robotic arm 104 to load the materials. After the loading is completed, the electric slide rail 106 starts, driving the vehicle platform 105 and the truck to move to the exit 101d, and then the truck drives out from the exit 101d.

[0120] The automatic palletizing and loading by the robot arm 104 greatly saves manpower and improves efficiency.

[0121] Embodiment 2

[0122] Reference Figure 3-15 The loading method of the palletizing loader comprises the following steps:

[0123] S1, detect whether the truck arrives at the vehicle-mounted platform;

[0124] S2, collecting image information of cargo woven bags;

[0125] S3, collecting space information in the carriage;

[0126] S4. The robotic arm moves the materials on the transfer table to the carriage and stacks them according to the collected information.

[0127] In step S3, the car point cloud is first collected using laser, and then the point cloud in the vertical direction is obtained. The ICP algorithm is improved by using the intensity domain-based rotation image feature descriptor to solve the coordinate conversion parameters between point clouds in different frames. Finally, the three-dimensional point cloud data collected by multiple scans are spliced, so that the local point cloud data of multiple perspectives are converted to the same reference coordinate system to obtain a complete three-dimensional point cloud of the target object or scene.

[0128] Reference Figure 3-13 , the point cloud in the vertical direction is obtained, and the Drost-PPF algorithm is used to combine the global description method with the local description method through the generalized Hough voting idea. Compared with the traditional three-dimensional feature matching algorithm, the Drost-PPF voting algorithm has many advantages, especially for the interior of the car with baffle occlusion, sparse surface texture, and relatively sparse point cloud images. It has a good recognition effect that the traditional feature matching algorithm cannot obtain.

[0129] like Figure 3 As shown in the figure, the Drost-PPF algorithm can be generally divided into an offline training stage for the model and an online recognition stage for recognizing and estimating the pose of the target objects in the scene; the offline training stage requires constructing a global description of the model point cloud in the form of a hash table.

[0130] α m is the rotation angle of any pair of points in the model with respect to the intermediate reference coordinate system.

[0131] In the online recognition stage, the Drost-PPF algorithm uses the generalized Hough voting mechanism to obtain the reference point S r The corresponding LC parameters (m r ,α), which mainly includes the following links:

[0132] 1) Select scene reference point S r And calculate the discrete PPF descriptor F d (s r ,s j ) and the intermediate rotation parameter α s (Refer to Figure 5 );

[0133] 2) Query the model hash table and vote (refer to Figure 6 ).

[0134] Three-dimensional point cloud stitching. During the actual scanning and measurement process, due to the limitations of the scanning equipment, target objects or scenes, one scan can only obtain part of the three-dimensional point cloud data of the target object. The target object needs to be scanned multiple times, and the three-dimensional point cloud obtained each time is usually in a different coordinate system. Point cloud registration is to stitch the three-dimensional point cloud data collected by multiple scans, so that the local point cloud data of multiple perspectives are converted to the same reference coordinate system, and then the complete three-dimensional point cloud of the target object or scene is obtained.

[0135] The shape of the car is a relatively regular rectangular parallelepiped with a single three-dimensional geometric feature. However, there are influencing factors such as baffles inside the car. The intensity values ​​of the car point cloud collected only in the vertical direction at different positions have certain differences. Based on the research of existing three-dimensional laser point cloud filtering and registration algorithms, this method proposes to improve the ICP algorithm by using a rotation image feature descriptor based on the intensity domain to solve the coordinate transformation parameters between point clouds of different frames. This method first uses filtering to remove noise points, takes the initial frame in a static state as the reference, and then estimates the normal of each point in the point set, and solves the rotation image feature descriptor in the intensity domain for each point. The corresponding point pairs are estimated based on the rotation image feature descriptor in the intensity domain, and finally the transformation parameters are solved to match the remaining point clouds to the initial reference coordinate system. Please refer to the flowchart of the improved ICP algorithm. Figure 7 .

[0136] Filtering: When using a 3D laser scanner to scan the carriage, the traction rope, instrument bracket, robotic arm, etc. on the data acquisition device will inevitably block the acquisition process, causing the original point cloud data to contain a lot of noise. This method uses a straight-through filter and a statistical filter to pre-process the original point cloud. The straight-through filter sets a rectangular filter based on the data acquisition equipment to remove outliers in the specified area. Statistical filtering uses the statistical characteristics of the point cloud for filtering. The basic idea of ​​this method is to first perform filtering on any point q in the collected original point cloud data set Q. i Search the neighborhood and then calculate q i The distance t to all other points in the neighborhood i , distance from the mean value t0 and standard deviation σ0, if t i If the point exceeds a certain standard deviation σ0, it will be removed as a noise point.

[0137] Normal estimation: Surface normal is an important attribute of the geometric surface and plays a very important role in point cloud data processing. Let p i is the collected point cloud, a point in the data set, whose normal is n i ,k neighborhood is N(p i )={q i,1 ,q i,2 ,...,q i,k}, this method uses covariance analysis to estimate the normal of the point cloud surface. For a point on a smooth surface, its normal is the normal of the plane tangent to it. Since the point cloud is discrete, we can convert the problem of solving the surface normal into a least squares problem to perform plane fitting. Specifically, we use the KD tree to find the neighborhood and establish the covariance matrix by analyzing the eigenvectors and eigenvalues ​​of the covariance matrix. When estimating the normal of a point, the choice of the neighborhood is crucial to the accuracy of the normal solution. We need to select the part of the neighborhood that is close to the point being sought. If the selected neighborhood is too small, since the point cloud is discrete, more points need to be selected to ensure the accuracy of the normal solution. For each point p i The corresponding covariance matrix C is as follows:

[0138]

[0139] Where T represents Q i The number of neighboring points of a point, Q represents Q i The total centroid of all points in the neighborhood of a point is the jth eigenvalue of the covariance matrix, and V is the jth eigenvector. Since C is a positive definite symmetric matrix, the eigenvalues ​​are all real numbers and the eigenvectors are all orthogonal. By calculating the eigenvalues ​​and their corresponding eigenvectors, the normal of the neighborhood surface at point Q can be approximated. Finally, the minimum spanning tree method is used to adjust the normal direction of each point to obtain the final normal.

[0140] Reference Figure 8 , Intensity domain rotation image features: 3D point cloud data has higher degrees of freedom than 2D images, usually contains a large number of points, and has six degrees of freedom, so it is more complicated to extract rotation invariant features. In order to simplify the matching problem of 3D point cloud data, it is often converted into a 2D image for processing. In this way, mature algorithms for processing 2D images can be used to ultimately solve the matching problem in 3D point cloud data by matching relatively simple 2D images. Rotation image is a feature descriptor represented by a 2D grid image, which has the characteristics of rotation invariance and translation invariance. It plays an important role in fields such as target detection and image matching.

[0141] By converting 3D point cloud data into a rotated image, we can obtain a rotation-invariant feature representation in 2D space, which facilitates matching and recognition tasks. At the same time, the rotated image is also translation-invariant, that is, the translation of the image does not change the properties of its descriptor, which further enhances the superiority of the rotated image in dealing with matching problems. By using the rotated image as a feature descriptor, we are able to process complex 3D point cloud data more efficiently and achieve accurate matching and recognition. Assume that point p represents a point in 3D space, the normal vector of the 3D surface at point p is represented by n, and P represents the tangent plane of the 3D surface at that point. Let x be any point other than point p on the 3D surface, define a to represent the distance between point x and the normal vector n, and the distance from x to the tangent plane P is represented by B. Let (α i , β i ) represents the projection of other points on the three-dimensional surface other than point p on P, i is the index of the two-dimensional mesh vertex, and statistics (α i , β i )The two-dimensional histogram obtained represents the rotated image of point p, and α and β determine the coordinates of the rotated image. The value of the two-dimensional grid depends on the number and intensity values ​​of the points (α, β) contained in the unified statistical interval.

[0142] Mapping function that projects a 3D point cloud onto a 2D mesh:

[0143] S P :R 3 →R 2 ;

[0144]

[0145] The 3D point cloud model can be projected into a 2D grid through S projection. And the 2D model can be rasterized using the following formula:

[0146]

[0147] In the above formula, b is the grid resolution, ω is the width of the two-dimensional grid, and (i, j) is the corresponding grid. The rotated image has rotational invariance and translational invariance, that is, rigid transformation invariance, and is linearly correlated. The normalized linear correlation coefficient is used to estimate the similarity of two rotated images. The formula for calculating the linear correlation coefficient for two rotated images is as follows:

[0148]

[0149] In the formula, P and Q represent two intensity domain transfer images, and x represents the i-th dimension feature of P and Q. Where -1<R(P,Q)<1.

[0150] The following formula is used to describe the similarity of two intensity-rotated images:

[0151]

[0152] Where tanf represents the weight describing the predicted correlation coefficient value. For each point in the source point cloud P and the target point cloud 0, the rotation image descriptor in the intensity domain is calculated, and the corresponding points in the target point cloud 0 are found to obtain the corresponding point set.

[0153] The ICP algorithm calculates the matching splicing parameters:

[0154] (1) Solving the initial transformation parameters: According to the basic principle of the ICP algorithm, after solving the corresponding point set, the least squares method is used to iteratively solve the rotation matrix and translation matrix by estimating the rigid body transformation to obtain the coordinate transformation parameters (x, y, z, θ, γ).

[0155] The basic steps are as follows:

[0156] a. Remove the corresponding point set with incorrect estimation based on the estimated corresponding point set;

[0157] b. Estimate the rigid body transformation using the corresponding point set;

[0158] c. The structure of the optimization points;

[0159] d. Use rigid body transformation to transform the source point cloud to the coordinate system of the target point cloud;

[0160] e. Iterate until convergence.

[0161] (2) Multi-sensor combination optimization stitching parameters:

[0162] In order to solve the problem that the original point cloud data obtained by the 3D laser scanner is sparse in the vertical direction and to improve the accuracy of point cloud registration, this method proposes a technology using joint Kalman filtering. This technology combines the coordinate transformation parameters solved by the rotation image descriptor based on the intensity domain, the high-precision IMU (inertial measurement unit), and the light intensity data to optimize the point cloud stitching parameters. Kalman filtering is a method of realizing state estimation of discrete control systems through a series of mathematical equations. Its goal is to estimate the system state under the condition of minimizing the mean square error. The extended Kalman filter linearizes the nonlinear system and uses the Kalman gain to infinitely approximate the true value of the system state. In this method, the coordinate transformation parameters obtained by solving the rotation image descriptor based on the intensity domain are first used as the initial estimate, and then the parameters are fused with the high-precision IMU and light intensity data through joint Kalman filtering.

[0163] The joint Kalman filter uses the Kalman gain to correct the system state according to the current observation value, so as to obtain a more accurate state estimation result. By applying the joint Kalman filter method, the problem that the coordinate transformation parameters directly solved by ICP do not fully conform to the actual results can be overcome, and the accuracy of point cloud registration can be further improved. This method can effectively and comprehensively utilize data from different sources, provide more reliable and accurate point cloud stitching parameters, and thus improve the measurement results of 3D laser scanners.

[0164] First, construct a 10-dimensional state vector X, including the three-directional position, velocity and attitude quaternion construction process equations and observation equations:

[0165] x k =f(x k-1 ,μ k-1 ,ω k-1 );

[0166] z k =h(x k ,v k );

[0167] The random variables k and vk represent process noise and observation noise respectively, which are assumed to be 0 and further linearized as follows:

[0168]

[0169]

[0170] Among them, A represents the Jacobian matrix of the partial derivative of f with respect to x, W represents the Jacobian matrix of the partial derivative of f with respect to ω, H represents the Jacobian matrix of the partial derivative of h, and V represents the Jacobian matrix of the partial derivative to calculate the prior estimation error covariance: Where W kRepresents a Gaussian white noise model and satisfies a zero-mean distribution, that is:

[0171] p(ω)~N(0,Q);

[0172] p(v)~N(0,R);

[0173] Calculate the Kalman gain:

[0174]

[0175] Construct the observation equation:

[0176]

[0177] Update the error covariance:

[0178]

[0179] The intensity-domain-based rotation image descriptor is used to improve the ICP algorithm to solve the coordinate transformation parameters, high-precision IMU data and light intensity data for combined solution of x, y, and z.

[0180] Carriage 3D laser point cloud stitching: After the above steps, the six conversion parameters (x, y, z, θ, γ), the point cloud data obtained from two sets of scans at different viewing angles can be stitched together. Take two point cloud data sets to be stitched together from different scanning viewpoints as an example, set them as P and Q respectively, with overlapping areas, and their coordinate systems are OX n Y n Z n and OX b Y b Z b , let the positions of any points of the same name in the overlapping area 2 in the P and Q coordinate systems be p i ,q j , then any pair of points with the same name in the two point cloud datasets (p i ,q j ) satisfies the following transformation matrix:

[0181] q j =Mp i ;

[0182] in, M represents the transformation matrix;

[0183] The process of coordinate system transformation is to transform the initial coordinate system OX n Y n Z n After rotating around its three axes, the origin of the rotation system is translated to the next coordinate system OXb Y b Z b At the origin, the transformation can be described as the following formula:

[0184]

[0185] Where μ is the scale coefficient of the two coordinate systems, C and T are rotation matrix and translation matrix respectively, and the translation matrix T is:

[0186]

[0187] Follow the following rotation order:

[0188] OX n Y n Z n →OX1Y1Z1→OX2Y2Z2→OX b Y b Z b , the three rotations correspond to three transformation matrices:

[0189]

[0190]

[0191]

[0192] Multiplying these three equations together gives:

[0193]

[0194] Compute the rotation matrix:

[0195]

[0196] According to the calculated translation matrix T and rotation matrix C, the splicing between the two point cloud data P and Q can be realized, and then all the point clouds are transformed into the global coordinate system to obtain the spliced ​​complete compartment well point cloud. The complete splicing result of the point cloud is as follows: Fig.15 shown.

[0197] Reference Fig. 9 ,The woven bag image processing process includes the following steps:

[0198] a. Collect woven bag images;

[0199] b. Region-based threshold segmentation;

[0200] c. Dynamic region growing method;

[0201] d. Morphological processing;

[0202] e. Match and fuse with the original image;

[0203] f. Woven bag positioning.

[0204] Image acquisition is performed using a CCD camera.

[0205] Segmentation threshold: Use the growth segmentation method based on regional grayscale difference. In the algorithm based on regional grayscale difference, assume a set of seed points P (single pixel or partial area) and use it as the starting point of growth. The total number of pixels is N, the original grayscale value is f(x,y), and the pixel coordinates are (x,y). The grayscale mean m can be expressed as: Then, according to the growth criterion formula max|f(x,y)-m| (x,y)∈P <K, compare the grayscale values ​​of the seed point and its nearby pixels. If the absolute value of the difference is smaller than the threshold K, it is considered to have similar features and is included in the pixel area where the seed point is located. Then scan new pixels that have not yet been assigned and continue to grow as seed points. Repeat the iteration until all pixels in the image are traversed.

[0206] Reference Fig.10 ,The process of dynamic region growing method.

[0207] The optimal threshold relationship model for image region growth under different lighting conditions is shown in Table 1:

[0208]

[0209]

[0210] Table 1

[0211] Among them, M represents the average gray value of the image, S represents the gray value of the selected seed point, D is the absolute value of the difference between S and M, and T represents the optimal region growth threshold of the image. Matlab is used to draw the (D, T) discrete point distribution diagram of the experimental results of a total of 21 images under different illumination and fit them using the least squares method. The fitted straight line represents the linear relationship between D and T, as shown in Fig.11 shown.

[0212] The equation of the fitted straight line is:

[0213] T = 0.8593*D-13.4723;

[0214] That is, the relationship between the optimal growth value T and the image grayscale mean M and the seed point grayscale S is:

[0215] T=0.8593*(|MS|)-13.472335≤M≤150;

[0216] In the range of 35≤M<150, images can be collected and segmented normally, and the optimal growth threshold can be calculated by the above formula. When M<35, it means that the ambient light is too dark to capture the woven bag image normally, so image processing and woven bag positioning cannot be performed. In this case, the light source needs to be turned on to provide sufficient illumination. When M>150, it means that the ambient light is too strong and the collected image is overexposed, so conventional image processing cannot be performed. It should be noted that these two situations are extreme cases and generally do not occur during loading and stacking. The above analysis derives a relationship model between illumination changes and dynamic growth thresholds.

[0217] Morphological processing, close the image to fill the gaps between the woven bags: the closing operation is to first dilate and then erode the image, which is used to eliminate small pores in the object, connect narrow discontinuities and gullies, and smooth the object contour without changing the object area. The mathematical definition of the closing operation is as follows:

[0218]

[0219] Dilation operation: Its purpose is to attribute the background connected to the target object to the target area under the constraint of the structural element, so that the target boundary expands outward, and the area of ​​the target object area will also increase the corresponding number of points. Dilation will expand the originally unclear part into an obvious feature, which can be used to fill some tiny holes in the target area and remove the small noise points in the target area.

[0220] The mathematical definition of dilation is as follows:

[0221]

[0222] Among them, B ^ is the reflection set representing the symmetry of B about the origin.

[0223] Erosion operation: Erosion is the dual operation of dilation operation. It is an algorithm that shrinks the boundary inward. It is mainly used to remove certain specific pixels of the image, especially to eliminate small and meaningless noise points on the boundary. The formula is as follows:

[0224]

[0225] Edge feature extraction: Use Hough transform to extract the center and edge information of the woven bag.

[0226] For the Hough transform, the straight line function expression in the coordinate system is as follows:

[0227]

[0228] After simplification, we get:

[0229] r = xcosθ + ysinθ;

[0230] For any point (X0, Y0), all straight lines passing through the point can be represented as r θ =X0cosθ+Y0sinθ, that is, for each pair (r θ ,θ) represent a straight line passing through a fixed point (X0, Y0). Therefore, for a given point (X0, Y0), the (r θ ,θ) value, at θ-r θ It can be drawn as a sine curve in a plane.

[0231] The steps of the Hough transform method to implement the transformation are as follows:

[0232] 1). Set the parameter space (r θ ,θ) is divided into m×n cells and defines an accumulator matrix Q m×n , where m and n are equal parts of and respectively;

[0233] 2). Assign an accumulator Q(i,j) to each cell (i,j) in the parameter space and set its initial value to 0 to count the number of curves intersecting each cell;

[0234] 3). All points in the rectangular coordinate system (X i ,Y i ) and the calculated value of 0 are substituted into the formula r = xcosθ + ysinθ to obtain r θ , thus obtaining the parameter space point (r θ ,θ);

[0235] 4). Find the corresponding point in parameter space (r θ ,θ), add 1 to the accumulator value at that point, then Q(i,j)=Q(i,j)+1;

[0236] Repeat steps 3 and 4 to traverse all points in the rectangular coordinate system, and find the accumulator with the largest value in the accumulator matrix. The coordinates of its location are the parameter coordinates (r θ ,θ), from which we can get the corresponding straight line equation.

[0237] Improved PPHT algorithm:

[0238] PPHT: Cumulative Probability Hough Transform: During the operation, not all edge points are calculated, but only some edge points selected by the function are processed. Since the selection of these edge points is probabilistic, it is called Cumulative Probability Hough Transform.

[0239] Optimal three parameters:

[0240] In the PPHT function, the last three parameters intthreshold, doubleminLineLength and doublemaxLineGap are very important, and they directly affect the straight line extraction effect. Among them, threshold is the threshold of the number of points that determines whether a line exists, that is, if the number of points that make up a straight line is less than this value, it will not be judged as a straight line; minLineLength is the threshold of the minimum length of the line segment, that is, if the length of the detected straight line is less than this value, it will not be judged as a straight line;

[0241] maxLineGap (maximum line segment interval) is the threshold of the maximum distance between two points connected as the same straight line, that is, if the distance between two points in the same line detected is greater than this value, they will not be judged as a straight line. From the definition, it can be seen that the smaller the three values ​​are, the better the restoration of sample image details will be, and the recognition success rate will also increase. However, if the value is too small, it may also lead to excessive restoration of unnecessary details, resulting in recognition failure.

[0242] The specific steps to find the optimal three parameters are as follows:

[0243] 1) The initial value of the Maxp parameter is fixed to 50, and the range is changed from 50 to 350 with an interval of 50. The values ​​of the other two parameters are adjusted, and the first group of three parameter values ​​corresponding to the combination that is successfully identified are recorded.

[0244] 2) Then, with an interval of 50, the range is changed from 50 to 350, the value of the fixed parameter Max is changed, and the other two parameter values ​​are adjusted repeatedly, and the three parameter values ​​of all the second combinations that are successfully identified are recorded and counted.

[0245] 3) Determine the optimal Max parameter value of the variable Max based on the three parameter values ​​of the first combination and the three parameter values ​​of the second combination.

[0246] 4) Fix the value of the Th parameter and repeat step 3 to obtain the optimal Th parameter value with Th as a variable.

[0247] 5) According to the obtained optimal Max and Th parameter values, the optimal three parameter combinations can be comprehensively determined, as shown in Table 2:

[0248]

[0249]

[0250] Table 2

[0251] Woven bag center calibration: After extracting the outline of each woven bag, find the smallest circumscribed rectangle for each outline. The centroid of the rectangle is the center of each woven bag. Here we take one of the woven bags as an example. Fig.14 As shown, draw the minimum enclosing rectangle of its outline.

[0252] refer to Fig.12 , Automatic loading: When the operator confirms that the truck to be loaded has arrived in the loading area and the intelligent loading system for woven bags is in the initial position, the recognition system is started. The point cloud information of the carriage is obtained by laser, the carriage outline is extracted and the carriage size information is obtained, and combined with the image information of the woven bag, the stacking position of the first bag of woven bags is further calculated and the stacking strategy is formulated. Specifically, according to the current situation of the carriage, the number of woven bags that can be stacked in each row is determined to be Y bags, the number of rows that can be stacked in each layer is X, and the total number of layers that can be stacked is Z.

[0253] Then, the walking mechanism moves to the position of the first row of woven bags, and the stacking mechanism completes the stacking of the first bag of woven bags. After that, the control system determines whether the current row is full of Y bags of woven bags. If not, the outfeed belt conveyor of the stacking mechanism moves horizontally by one station to put down a bag of woven bags until the row is full. After completing the stacking operation of the row of woven bags, the walking mechanism continues to move to the position of the next row of woven bags and repeats the stacking operation of the row of woven bags until the stacking of all X rows of woven bags on this layer is completed.

[0254] Next, the lifting mechanism drives the entire system to lift one workstation, and the walking mechanism moves to the first row of woven bags on the previous layer. The stacking mechanism repeats the stacking of woven bags on this layer until all Z layers of woven bags are loaded.

[0255] During the loading process, the control system will determine whether all loading operations of the carriage have been completed. Once loading is completed, the bagged woven bag intelligent loading system will move to the initial position and prepare for the loading operation of the next truck.

[0256] Reference Fig.13 In order to meet the clamp's loading requirements for woven bags of various sizes and realize multi-stage loading of multiple woven bags on one vehicle, a clamp multi-adaptation mode is designed. The clamp calibration process is shown in the figure.

[0257] Through the matrix transformation formula B T O = B T C C T O To obtain the position relationship between the object to be operated and the base, the formula E T O =E T B B T O The position and posture relationship between the object to be operated and the end effector is obtained in real time for the subsequent robot arm motion trajectory planning. Among them, B represents the base coordinate system, C represents the camera coordinate system, O represents the object coordinate system, and E represents the gripper coordinate system, so for example: B T C It is the transformation matrix used to calculate the mutual transformation between the camera coordinate system and the base coordinate system; E T O It is the transformation matrix used to calculate the mutual transformation between the gripper coordinate system and the object coordinate system.

[0258] To sum up, the present application realizes automated palletizing and loading, replacing the manual loading method, breaking the bottleneck in the existing loading field, realizing precise palletizing of materials layer by layer, and is highly automated, requiring no human involvement throughout the process, which is intelligent and efficient, improving loading efficiency and reducing labor costs.

[0259] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A loading method of a palletizing loader, comprising: The loading platform (101) comprises an upper layer (101a) and a lower layer (101b), and two ends of the lower layer (101b) are an entrance (101c) and an exit (101d); Features: Materials are placed on the upper layer (101a); The truck is located on the lower level (101b); The palletizing and loading machine (100) further comprises: A conveying device (102) is disposed on the loading platform (101) and is used to transport the materials on the upper layer (101a) away from the upper layer (101a); A transfer platform (103) is provided on the loading platform (101) and is located between the upper layer (101a) and the lower layer (101b); A mechanical arm (104), disposed on the loading platform (101), for transporting materials in the transfer platform (103) to a truck bucket; A vehicle-mounted platform (105) is slidably disposed on the lower layer (101b) via an electric slide rail (106) and is used to transport a truck from the entrance (101c) to the exit (101d); The material transported by the conveying device (102) is transported to the transfer platform (103); the mechanical arm (104) is located between the exit (101d) and the entrance (101c); and a camera for monitoring the real-time status of the material woven bags is provided on the loading platform (101); The loading method of the palletizing loader comprises the following steps: S1, detect whether the truck arrives at the vehicle-mounted platform; S2, collecting image information of cargo woven bags; S3, collecting space information in the carriage; S4, the robotic arm moves the materials on the transfer table to the carriage and stacks them according to the collected information; In step S3, the point cloud of the carriage is first collected using laser, and then the point cloud in the vertical direction is obtained. The ICP algorithm is improved by using the rotation image feature descriptor based on the intensity domain to solve the coordinate conversion parameters between point clouds of different frames. Finally, the three-dimensional point cloud data collected by multiple scans are spliced ​​to convert the local point cloud data of multiple perspectives into the same reference coordinate system to obtain a complete three-dimensional point cloud of the spatial information in the carriage. The ICP algorithm process is to filter the source point cloud and the target point cloud at the same time, then estimate the normal, then perform intensity domain rotation image features, then estimate the corresponding points of the two intensity domain rotation image features obtained from the source point cloud and the target point cloud, and finally ICP calculates the matching parameters; ICP calculates the concatenation parameters of the matching parameters including the following steps: (1) Solving the initial transformation parameters. According to the basic principle of the ICP algorithm, after solving the corresponding point set, the least squares method is used to iteratively solve the rotation matrix and translation matrix by estimating the rigid body transformation to obtain the coordinate transformation parameters (x, y, z, θ, γ); (2) Multi-sensor combination optimization splicing parameters, first construct a 10-dimensional state vector X, including three-directional position, velocity and attitude quaternion construction process equation and observation equation: x k =f(x k-1 ,m k-1 Oh, oh k-1 ); z k =h(x k ,v k ); The random variables k and vk represent process noise and observation noise respectively, which are assumed to be 0 and further linearized as follows: Among them, A represents the Jacobian matrix of the partial derivative of f with respect to x, W represents the Jacobian matrix of the partial derivative of f with respect to ω, H represents the Jacobian matrix of the partial derivative of h, and V represents the Jacobian matrix of the partial derivative to calculate the prior estimation error covariance: Among them, W k Represents a Gaussian white noise model and satisfies a zero-mean distribution, that is: p(ω)~N(0,Q); p(v)~N(0,R); Calculate the Kalman gain: Construct the observation equation: Update the error covariance: The ICP algorithm is improved based on the intensity domain rotation image descriptor to solve the coordinate transformation parameters, high-precision IMU data and light intensity data to combine and solve x, y, z; Carriage 3D laser point cloud stitching: solve the 6 conversion parameters (x, y, z, θ, γ), realize the splicing of two sets of point cloud data scanned under different viewing angles; take two frames of point cloud data sets to be spliced ​​under different scanning viewpoints as an example, set them as P and Q respectively, with overlapping areas, and their coordinate systems are OX n Y n Z n and OX b Y b Z b , let the positions of any points of the same name in the overlapping area 2 in the P and Q coordinate systems be p i ,q j , then any pair of points with the same name in the two point cloud datasets (p i ,q j ) satisfies the following transformation matrix: q j =Mp i ; in, M represents the transformation matrix; The process of coordinate system transformation is to transform the initial coordinate system OX n Y n Z n After rotating around its three axes, the origin of the rotation system is translated to the next coordinate system OX b Y b Z b At the origin, the transformation is described as the following formula: Where μ is the scale factor of the two coordinate systems, C and T are rotation matrix and translation matrix respectively, and the translation matrix T is: Follow this rotation order: OX n Y n Z n →OX1Y1Z1→OX2Y2Z2→OX b Y b Z b , the three rotations correspond to three transformation matrices: Multiplying these three equations together yields: Compute the rotation matrix: According to the calculated translation matrix T and rotation matrix C, the splicing between the two point cloud data P and Q is realized, and then all the point clouds are transformed into the global coordinate system to obtain the spliced ​​complete car point cloud.

2. The loading method of the palletizing loader according to claim 1, characterized in that: Normal estimation, first assume that pi is the collected point cloud, a point in the data set, and its normal is n i , the k-neighborhood is N(p i )={q i,1 ,q i,2 , ..., q i,k }, for each point p i The corresponding covariance matrix C is as follows: Where t represents Q i The number of neighbors of a point, Indicates Q i The total centroid of all points in the point neighborhood is the jth eigenvalue of the covariance matrix. is the jth eigenvector, and λ is The characteristic value of .

3. The loading method of the palletizing loader according to claim 2, characterized in that: Intensity domain rotation image features, assuming that point p represents a point in three-dimensional space, the normal vector of the three-dimensional surface at point p is represented by n, and P represents the tangent plane of the three-dimensional surface at this point; let x be any point other than point p on the three-dimensional surface, define a to represent the distance between point x and the normal vector n, and the distance from x to the tangent plane P is represented by B; let (α i ,β i ) represents the projection of other points on the three-dimensional surface other than point p on P, i is the index of the two-dimensional mesh vertex, and statistics (α i ,β i ) The two-dimensional histogram obtained by α and β represents the rotation image of point p. The value of the coordinate two-dimensional grid of the rotation image depends on the number and intensity value of the points (α, β) contained in the unified statistical interval. Mapping function that projects a 3D point cloud onto a 2D mesh: S P :R 3 →R 2 ; The 3D point cloud model is projected into a 2D grid through S projection, and the 2D model is rasterized using the following formula: In the above formula, b is the grid resolution, ω is the width of the two-dimensional grid, and (i, j) is the corresponding grid; The normalized linear correlation coefficient estimates the similarity of two rotated images. The formula for calculating the linear correlation coefficient of two rotated images is as follows: In the formula, P and Q represent two intensity domain transfer images, x i ,y i Represent the i-th dimension features of P and Q respectively; N is the total number of pixels in the image; Where -1 <R(P,Q)<1; The following formula is used to describe the similarity of two intensity-rotated images: Where atanf represents the weight describing the predicted correlation coefficient value. For each point in the source point cloud P and the target point cloud 0, the rotation image descriptor in the intensity domain is calculated, and the corresponding points in the target point cloud 0 are found to obtain the corresponding point set.

4. The loading method of the palletizing loader according to claim 2, characterized in that: Step S2 collects image information of the cargo woven bag, including: using a camera to collect the image of the woven bag, repeatedly iterating the collected image of the woven bag based on the growth segmentation method of regional grayscale difference until all pixels of the image are traversed, and then using the dynamic region growing method to analyze and obtain the relationship model between the illumination change and the dynamic growth threshold, performing morphological processing on it, and then extracting the center and edge information of the woven bag to locate the woven bag.

5. The loading method of the palletizing loader according to claim 2, characterized in that:In step S4, the matrix transformation formula is B T O = B T C C T O To obtain the position relationship between the object to be operated and the base, the formula E T O = E T B B T O Obtain the position and posture relationship between the object to be operated and the end effector in real time for subsequent robot arm motion trajectory planning; Among them, B represents the base coordinate system, C represents the camera coordinate system, O represents the object coordinate system, and E represents the gripper coordinate system, so for example: B T C It is the transformation matrix used to calculate the mutual transformation between the camera coordinate system and the base coordinate system; E T O It is the transformation matrix used to calculate the mutual transformation between the gripper coordinate system and the object coordinate system.

Citation Information

Patent Citations

  • Intelligent identification robot loading system

    CN219807482U

Cited By

  • Full-automatic powdery material loading system

    CN122355073A

  • Full-automatic powder material loading control method

    CN122724991A