Complex surface parts reconstruction method based on robotic arm

Abstract

The present invention provides a method for reconstructing complex curved surface parts based on a robotic arm, comprising the following steps: S1. setting a laser scanner at the end of a six-axis mechanical tool hand to form a hand-eye system, and calibrating the hand-eye system; S2. determining the scanning order and optimal scanning trajectory of the laser scanner for the scanning target; S3. controlling the laser scanning to obtain point cloud data of the scanning target according to the scanning order and scanning trajectory determined in step S2, and fitting and reconstructing the point cloud data. Through the above method, it is possible to effectively avoid missed scans or repeated scans when scanning the scanning target, thereby avoiding the point cloud of a partial area of ​​the scanning object being inaccurate, and it is also possible to improve the scanning efficiency and more effectively ensure the accuracy of the final scanning result.

Description

Technical Field

[0001] The invention relates to a part model reconstruction method, in particular to a complex curved surface part reconstruction method based on a robot arm. Background Art

[0002] With the development of material science and equipment manufacturing industry, high-performance materials and complex thin-walled structural parts (blades, propellers, etc.) account for a growing proportion of metal parts. Such parts generally have the characteristics of long production cycle and high manufacturing cost. When they are damaged, direct replacement will greatly increase the cost of parts and extend the production cycle. Therefore, it is of great practical significance to repair the damage of high-performance material parts and complex thin-walled structural parts.

[0003] In order to repair parts, it is necessary to accurately and completely obtain the surface point cloud information of the parts. In the existing technology, most of them use handheld laser scanners to obtain the surface point cloud of complex parts, and use commercial software to reconstruct the surface. Such a solution has two obvious defects: the handheld scanner needs to be attached with a considerable number of label points, and the scanning result does not have the location information of the parts. Because it is impossible to determine which areas on the parts have been scanned, it often causes many areas on the parts to be scanned repeatedly, resulting in too dense scanning point clouds and too large generated files, which increases the burden of post-processing. In addition, it is easy for some areas to not be scanned, resulting in the missing of some areas of the obtained part model file; later calibration is required in the robot coordinate system, and the accuracy is difficult to guarantee during the later calibration. Based on the above, the existing technology cannot accurately obtain the surface point cloud information of the parts.

[0004] Therefore, in order to solve the above technical problems, it is urgent to propose a new technical means. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a complex curved surface part reconstruction method based on a robotic arm, which can effectively avoid missed scans or repeated scans when scanning the scanning target, thereby avoiding the point cloud of part of the scanned object area being inaccurate, and can also improve the scanning efficiency and more effectively ensure the accuracy of the final scanning results.

[0006] The present invention provides a complex curved surface part reconstruction method based on a robotic arm, comprising the following steps:

[0007] S1. Setting the laser scanner at the end of the six-axis mechanical tool hand to form a hand-eye system, and calibrating the hand-eye system;

[0008] S2. Determine the scanning order and optimal scanning trajectory of the laser scanner for the scanning target;

[0009] S3. Control the laser scanning to obtain the point cloud data of the scanned target according to the scanning sequence and scanning trajectory determined in step S2, and perform fitting and reconstruction on the point cloud data.

[0010] Further, in step S1, calibrating the laser scanner specifically includes:

[0011] S11. Using a plane chessboard as a calibration plate, and acquiring an image of the calibration plate by a laser scanner;

[0012] S12. Establish hand-eye calibration model:

[0013] Build the calibration model:

[0014] AX = XB, where A and B represent transformation matrices and X is the desired pose matrix;

[0015]

[0016] in, b T g is the pose matrix of the robot tool hand end in the robot coordinate system, c T t is the pose matrix of the camera in the calibration board coordinate system, b T g and c T t The superscripts 1 and 2 in the matrix represent two different viewing angles of the camera and the two matrices are determined based on the initial relative position matrix T;

[0017] Express matrix A, matrix B, and matrix X as pose matrices:

[0018]

[0019] R A is the rotation matrix of matrix A, T A are all displacement matrices of matrix A;

[0020] R B is the rotation matrix of matrix B, T B are all displacement matrices of matrix B;

[0021] R X is the rotation matrix of matrix X, T X are all displacement matrices of matrix X;

[0022] The calibration model is expressed as a system of equations:

[0023]

[0024] Convert a rotation matrix to a rotation vector:

[0025]

[0026] Normalize the rotation vector:

[0027] Among them, r a is the rotation vector of matrix A, r b is the rotation vector of matrix B, N a is the rotation vector r a The module, N b is the rotation vector r b Model;

[0028] Compute the modified Rodrigues vector:

[0029]

[0030] Calculate the initial rotation vector p x ':

[0031] skew(p a +p b ) x '=p a -p b ;

[0032] Calculate the rotation vector p x :

[0033]

[0034] Determine the rotation and translation matrices for the hand-eye system:

[0035]

[0036] (R A -1)T X =R X T B -T A ; Where: I represents the identity matrix, skew(p(x)) represents the rotation vector p x The skewness of .

[0037] Further, in step S2, the optimal scanning order of the laser scanner is determined by the following method:

[0038] The part as the scanning target is partitioned, and the center point of each partition is translated along the normal of the partition by an optimal working distance to obtain the scanning target point P. in；projected ;

[0039] The target point P will be scanned in；projectedAnd the normal information corresponding to each scanned target point constitutes the pose matrix T in；projected :T in；projected ={T1,T2,T3,…,T i ,…,T m}, T i represents the pose matrix of the i-th partition, m represents the number of partitions, i = 1, 2, ..., m;

[0040] The classic TSP algorithm is used to determine the optimal scanning sequence of the laser scanner.

[0041] Further, in step S2, the optimal scanning trajectory is determined by the following method:

[0042] S201. The pose matrix T of the i-th scan partition i Substitute it into the IKFast robot inverse kinematics solver to calculate the trajectory point configuration solution set of the laser scanner. m :

[0043] Solutions m ={Solutions1,Solutions2,…,Solutions n}

[0044] Where m represents the inverse solution set of the mth partition, n represents the n configuration solutions of the mth inverse solution set, to They respectively represent the axis angles of the six axes of the mechanical tool hand;

[0045] S202. The hand-eye system performs collision detection on the scanned target and deletes the configuration solution in which the collision occurs;

[0046] S203. Repeat steps S201 and S202 to find the configuration solution of each partition, generate a configuration solution set, and substitute all the configuration solutions into the graph generator to generate a weighted Q-layer robot configuration path graph Graph, and use the n solutions in Solution1 as the n nodes of the first layer of the path graph Graph; and so on, there are a total of m×n nodes in the path graph, and connect each node of the two adjacent layers to form an edge and form an edge set Edges: Edges={edge 11,21 ,…edge mn,(m+1)n}, where edge mn,(m-1)n represents the edge formed by connecting the nth node of the mth layer with the nth node of the m+1th layer,

[0047] S204. Calculate the weight of each edge:

[0048]

[0049]

[0050] Among them, Solution (m)n represents the nth solution of the mth configuration solution set, represents the i-th axis angle of the n-th solution of the m-th configuration solution set;

[0051] S205. Traverse the robot configuration path graph Graph, sum the weights of all edges on each path, and select the path with the smallest sum of weights as the optimal path.

[0052] Furthermore, collision detection specifically includes:

[0053] Establish a three-dimensional coordinate system.

[0054] Construct the smallest cube that can protect the scanned target as a bounding box;

[0055] Determine the position coordinates Q of any point on the scanning target i (x i ,y i ,z i ), then at any position of the scan target:

[0056] x min ≤x i ≤x max

[0057] y min ≤y i ≤y max

[0058] z min ≤z i ≤z max

[0059] Then: The size of the bounding box is:

[0060] a=x max -y min

[0061] b=y max -y min

[0062] c=z max -z min

[0063] Among them: a, b, c represent the length, width and height of the bounding box respectively;

[0064] Therefore, the robot tool hand and the scanning target do not collide if any of the following conditions are met:

[0065] (1) On the X-axis, the maximum X-value of the machine tool hand is less than the minimum X-value of the scanned object;

[0066] (2) On the X-axis, the maximum X-value of the scanned object is less than the minimum X-value of the machine tool hand;

[0067] (3) On the Y axis, the minimum Y value of the machine tool hand is greater than the maximum Y value of the scanned object;

[0068] (4) On the Y axis, the minimum Y value of the scanned object is greater than the maximum Y value of the machine tool hand;

[0069] (5) On the Z axis, the minimum Z value of the machine tool hand is greater than the maximum Z value of the scanned object;

[0070] (6) On the Z axis, the minimum Z value of the scanned object is greater than the maximum Z value of the machine tool hand;

[0071] If the robot tool hand and the scanning target do not meet all the above conditions, the robot tool hand and the scanning target will collide and the corresponding configuration solution will be deleted.

[0072] Further, step S3 specifically includes:

[0073] Convert the point cloud data scanned at different times into the same coordinate system;

[0074] Project the point cloud data onto the same plane;

[0075] The two-dimensional Delaunay growth method is used to connect the triangular meshes and establish the topological relationship of the two-dimensional space;

[0076] According to the coordinate projection relationship of the point cloud data, the topology of the two-dimensional space is mapped to the three-dimensional space to form the topological relationship reconstruction of the potential space.

[0077] The beneficial effects of the present invention are as follows: Through the present invention, it is possible to effectively avoid missed scans or repeated scans when scanning a scanning target, thereby avoiding the point cloud of a partial area of ​​the scanning object being inaccurate, and it is also possible to improve the scanning efficiency and more effectively ensure the accuracy of the final scanning result. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0079] Figure 1 It is a flow chart of the present invention.

[0080] Figure 2 This is a schematic diagram of scanning order optimization in the present invention.

[0081] Figure 3 Schematic diagram of the configuration path of the present invention.

[0082] Figure 4 It is the local projection diagram of the three-dimensional k domain.

[0083] Figure 5 It is a schematic diagram of greedy projection triangulation of the present invention.

[0084] Figure 6 Schematic diagram of the MLS projection process of the present invention. DETAILED DESCRIPTION

[0085] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0086] The present invention provides a complex curved surface part reconstruction method based on a robotic arm, comprising the following steps:

[0087] S1. Setting the laser scanner at the end of the six-axis mechanical tool hand to form a hand-eye system, and calibrating the hand-eye system;

[0088] S2. Determine the scanning order and optimal scanning trajectory of the laser scanner for the scanning target;

[0089] S3. Control the laser scanning to obtain the point cloud data of the scanning target according to the scanning order and scanning trajectory determined in step S2, and fit and reconstruct the point cloud data. Through the above method, it is possible to effectively avoid missed scans or repeated scans when scanning the scanning target, thereby avoiding the point cloud of part of the scanning object area being inaccurate, and it is also possible to improve the scanning efficiency and more effectively ensure the accuracy of the final scanning result.

[0090] In this embodiment, in step S1, calibrating the laser scanner specifically includes:

[0091] S11. Using a plane chessboard as a calibration plate, and acquiring an image of the calibration plate by a laser scanner;

[0092] S12. Establish hand-eye calibration model:

[0093] Build the calibration model:

[0094] AX = XB, where A and B represent transformation matrices and X is the desired pose matrix;

[0095]

[0096] in, b T g is the pose matrix of the robot tool hand end in the robot coordinate system, c T t is the pose matrix of the camera in the calibration board coordinate system, b T g andc T t The superscripts 1 and 2 in the matrix represent two different viewing angles of the camera and the two matrices are determined based on the initial relative position matrix T;

[0097] Express matrix A, matrix B, and matrix X as pose matrices:

[0098]

[0099] R A is the rotation matrix of matrix A, T A are all displacement matrices of matrix A;

[0100] R B is the rotation matrix of matrix B, T B are all displacement matrices of matrix B;

[0101] R X is the rotation matrix of matrix X, T X are all displacement matrices of matrix X;

[0102] The calibration model is expressed as a system of equations:

[0103]

[0104] Convert a rotation matrix to a rotation vector:

[0105]

[0106] Normalize the rotation vector:

[0107] Among them, r a is the rotation vector of matrix A, r b is the rotation vector of matrix B, N a is the rotation vector r a The module, N b is the rotation vector r b Model;

[0108] Compute the modified Rodrigues vector:

[0109]

[0110] Calculate the initial rotation vector p x ':

[0111] skew(p a +p b ) x '=p a -p b ;

[0112] Calculate the rotation vector p x :

[0113]

[0114] Determine the rotation and translation matrices for the hand-eye system:

[0115]

[0116] (R A -1)T X =R X T B -T A ; Where: I represents the identity matrix, skew(p(x)) represents the rotation vector p x After determining the skewness of the rotation matrix and the displacement matrix, the position and posture of the laser scanner and the manipulator can be obtained, thereby completing the calibration and ensuring the accuracy of subsequent scanning; in the above-mentioned, the specific dimensions of each matrix and the value of each element are determined by existing methods.

[0117] In this embodiment, in step S2, the optimal scanning order of the laser scanner is determined by the following method:

[0118] The part as the scanning target is partitioned, and the center point of each partition is translated along the normal of the partition by an optimal working distance to obtain the scanning target point P. in；projected ;

[0119] The target point P will be scanned in；projected And the normal information corresponding to each scanned target point constitutes the pose matrix T in；projected :T in；projected ={T1,T2,T3,…,T i ,…,T m}, T i represents the pose matrix of the i-th partition, m represents the number of partitions, i = 1, 2, ..., m;

[0120] The classic TSP algorithm is used to determine the optimal scanning order of the laser scanner. Through the above method, the accuracy of the final result can be ensured. Figure 2 As shown, Figure 2 (a) is the scanning order before optimization. Figure 2 (b) is the optimized scanning order.

[0121] In this embodiment, in step S2, the optimal scanning trajectory is determined by the following method:

[0122] S201. The pose matrix T of the i-th scan partition iSubstitute it into the IKFast robot inverse kinematics solver to calculate the trajectory point configuration solution set of the laser scanner. m :

[0123] Solutions m ={Solutions1,Solutions2,…,Solutions n}

[0124] Where m represents the inverse solution set of the mth partition, n represents the n configuration solutions of the mth inverse solution set, to They respectively represent the axis angles of the six axes of the mechanical tool hand;

[0125] S202. The hand-eye system performs collision detection on the scanned target and deletes the configuration solution in which the collision occurs;

[0126] S203. Repeat steps S201 and S202 to find the configuration solution of each partition, generate a configuration solution set, and substitute all the configuration solutions into the graph generator to generate a weighted Q-layer robot configuration path graph Graph, and use the n solutions in Solution1 as the n nodes of the first layer of the path graph Graph; and so on, there are a total of m×n nodes in the path graph, and connect each node of the two adjacent layers to form an edge and form an edge set Edges: Edges={edge 11,21 ,…edge mn,(m+1)n}, where edge mn,(m-1)n represents the edge formed by connecting the nth node of the mth layer with the nth node of the m+1th layer,

[0127] S204. Calculate the weight of each edge:

[0128]

[0129]

[0130] Among them, Solution (m)n represents the nth solution of the mth configuration solution set, represents the i-th axis angle of the n-th solution of the m-th configuration solution set;

[0131] S205. Traverse the robot configuration path graph Graph, sum the weights of all edges on each path, and select the path with the smallest sum of weights as the optimal path.

[0132] Among them, collision detection specifically includes:

[0133] Establish a three-dimensional coordinate system.

[0134] Construct the smallest cube that can protect the scanned target as a bounding box;

[0135] Determine the position coordinates Q of any point on the scanning target i (x i ,y i ,z i ), then at any position of the scan target:

[0136] x min ≤x i ≤x max

[0137] y min ≤y i ≤y max

[0138] z min ≤z i ≤z max

[0139] Then: The size of the bounding box is:

[0140] a=x max -y min

[0141] b=y max -y min

[0142] c=z max -z min

[0143] Among them: a, b, c represent the length, width and height of the bounding box respectively;

[0144] Therefore, the robot tool hand and the scanning target do not collide if any of the following conditions are met:

[0145] (1) On the X-axis, the maximum X-value of the machine tool hand is less than the minimum X-value of the scanned object;

[0146] (2) On the X-axis, the maximum X-value of the scanned object is less than the minimum X-value of the machine tool hand;

[0147] (3) On the Y axis, the minimum Y value of the machine tool hand is greater than the maximum Y value of the scanned object;

[0148] (4) On the Y axis, the minimum Y value of the scanned object is greater than the maximum Y value of the machine tool hand;

[0149] (5) On the Z axis, the minimum Z value of the machine tool hand is greater than the maximum Z value of the scanned object;

[0150] (6) On the Z axis, the minimum Z value of the scanned object is greater than the maximum Z value of the machine tool hand;

[0151] If the robot hand and the scanning target do not meet all the above conditions, the robot hand and the scanning target collide, and the corresponding configuration solution is deleted. Through the above method, the accuracy of the final result can be guaranteed.

[0152] In this embodiment, step S3 specifically includes:

[0153] Convert the point cloud data scanned at different times into the same coordinate system;

[0154] Project the point cloud data onto the same plane;

[0155] The two-dimensional Delaunay growth method is used to connect the triangular meshes and establish the topological relationship of the two-dimensional space;

[0156] According to the coordinate projection relationship of the point cloud data, the topology of the two-dimensional space is mapped to the three-dimensional space to form the topological relationship reconstruction of the potential space.

[0157] Specifically: Figure 4-Figure 6 As shown: The two-dimensional Delaunay growth method is used to connect triangular meshes, including:

[0158] For any point in the point cloud, the kd-tree neighborhood search algorithm is used to determine its neighborhood containing k neighboring points. During the triangulation algorithm, the points in the point cloud are divided into four types: free points, completed points, boundary points, and edge points;

[0159] Determine the projection tangent plane of a point and its neighboring points, and project all points in the neighborhood onto the two-dimensional plane. For the neighborhood of any point, calculate the approximate normal vector through the normal estimation algorithm, and then use the plane equation to solve the tangent plane of the point. That is, if the normal vector n of point M0 (x0, y0, z0) is n = (A, B, C), point M (x, y, z) is a point on the tangent plane passing through M0, so the tangent plane can be expressed as:

[0160] A(x-x0)+B(y-y0)+C(z-z0)=0

[0161] Transform the point in space to the two-dimensional tangent plane Π through the projection matrix (translation and rotation transformation), the projection transformation matrix T M The definition of is as follows

[0162] T M =T c R x R y

[0163] Based on the above formula, calculate any point p(x i ,yi ,z i ) on its tangent plane is as follows:

[0164] (x i ′,y i ′,z i ′,1)=T M (x i ′,y i ′,z i ′,1) T ;

[0165] A greedy algorithm is used for the points on the projection plane. Each time, only the local optimal point is considered as the growth point. The plane is triangulated to determine the topological relationship. Then, based on the projection relationship, the topological relationship of the three-dimensional space is constructed. This process is repeated continuously to finally achieve the complete surface reconstruction of the object model. The specific process is as follows:

[0166] 1) Any point p i As the initial point, the kd-tree algorithm is used to search for its neighboring points, and the distance point p is selected in the neighborhood. i The shortest point p j And make a straight line connection;

[0167] 2) Calculate the edge p i p j The nearest point p k , make triangular connections and construct triangle Δp i p j p k ;

[0168] 3) Through step 2, select the triangle Δp i p j p k The closest point on one side builds a new triangle;

[0169] 4) Repeat step 3 until the complete topological relationship of the object surface is constructed and the algorithm ends.

[0170] The moving least squares (MLS) method is used to smooth the point cloud, so that the reconstructed surface is smooth and delicate, and a better reconstruction effect is obtained. Figure 6 : On the local fitting surface g, project a point cloud surface onto the surface, which is the MLS projection smoothing of the point cloud. The algorithm is completed by the following two steps:

[0171] (1) For a smooth point r, solve its local reference plane H

[0172] (2) Local mapping: A polynomial is fitted to the neighborhood of point r through the local plane H defined in step 1. The MLS algorithm is a prior art and will not be described in detail here.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A complex surface part reconstruction method based on a robotic arm, characterized by: The following steps are involved: S1. Setting the laser scanner at the end of the six-axis mechanical tool hand to form a hand-eye system, and calibrating the hand-eye system; S2. Determine the scanning order and optimal scanning trajectory of the laser scanner for the scanning target; S3. Control the laser scanning to obtain the point cloud data of the scanned target according to the scanning order and scanning trajectory determined in step S2, and perform fitting and reconstruction on the point cloud data; In step S2, the optimal scanning order of the laser scanner is determined by the following method: The part as the scanning target is partitioned, and the center point of each partition is translated along the normal of the partition by an optimal working distance to obtain the scanning target point P. in；projected ; The target point P will be scanned in；projected And the normal information corresponding to each scanned target point constitutes the pose matrix T in；projected :T in；projected ={T1,T2,T3,…,T i ,…,T m }, Ti represents the pose matrix of the i-th partition, m represents the number of partitions, i = 1, 2, ..., m; The optimal scanning sequence of the laser scanner is determined using the classic TSP algorithm; The optimal scanning trajectory is determined by the following method: S201. The pose matrix T of the i-th scan partition i Substitute it into the IKFast robot inverse kinematics solver to calculate the trajectory point configuration solution set of the laser scanner. m : Solutions m ={Solutions1,Solutions2,…,Solutions n } Where m represents the inverse solution set of the mth partition, n represents the n configuration solutions of the mth inverse solution set, to They respectively represent the axis angles of the six axes of the mechanical tool hand; S202. The hand-eye system performs collision detection on the scanned target and deletes the configuration solution in which the collision occurs; S203. Repeat steps S201 and S202 to find the configuration solution of each partition, generate a configuration solution set, and substitute all the configuration solutions into the graph generator to generate a weighted Q-layer robot configuration path graph Graph, and use the n solutions in Solution1 as the n nodes of the first layer of the path graph Graph; and so on, there are a total of m×n nodes in the path graph, and connect each node of the two adjacent layers to form an edge and form an edge set Edges: Edges={edge 11,21 ,…edge mn,(m+1)n }, where edge mn,(m-1)n represents the edge formed by connecting the nth node of the mth layer with the nth node of the m+1th layer, S204. Calculate the weight of each edge: Among them, Solution (m)n represents the nth solution of the mth configuration solution set, represents the i-th axis angle of the n-th solution of the m-th configuration solution set; S205. Traverse the robot configuration path graph Graph, sum the weights of all edges on each path, and select the path with the smallest sum of weights as the optimal path.

2. The complex curved surface part reconstruction method based on a robot arm according to claim 1, characterized in that: In step S1, calibrating the laser scanner specifically includes: S11. Using a plane chessboard as a calibration plate, and acquiring an image of the calibration plate by a laser scanner; S12. Establish hand-eye calibration model: Build the calibration model: AX = XB, where A and B represent transformation matrices and X is the desired pose matrix; in, b T g is the pose matrix of the robot tool hand end in the robot coordinate system, c T t is the pose matrix of the camera in the calibration board coordinate system, b T g and c T t The superscripts 1 and 2 in the matrix represent two different viewing angles of the camera and the two matrices are determined based on the initial relative position matrix T; Express matrix A, matrix B, and matrix X as pose matrices: R A is the rotation matrix of matrix A, T A are all displacement matrices of matrix A; R B is the rotation matrix of matrix B, T B are all displacement matrices of matrix B; R X is the rotation matrix of matrix X, T X are all displacement matrices of matrix X; The calibration model is expressed as a system of equations: Convert a rotation matrix to a rotation vector: Normalize the rotation vector: Among them, r a is the rotation vector of matrix A, r b is the rotation vector of matrix B, N a is the rotation vector r a The module, N b is the rotation vector r b Model; Compute the modified Rodrigues vector: Calculate the initial rotation vector p x ': skew(p a +p b )p x '=p a -p b ; Calculate the rotation vector p x : Determine the rotation and translation matrices for the hand-eye system: (R A -1)T X =R X T B -T A ; Where: I represents the identity matrix, skew(p(x)) represents the rotation vector p x skewness.

3. The complex curved surface part reconstruction method based on a robot arm according to claim 1, characterized in that: Collision detection specifically includes: Establish a three-dimensional coordinate system, Construct the smallest cube that can protect the scanned target as a bounding box; Determine the position coordinates Q of any point on the scanning target i (x i ,y i ,z i ), then at any position of the scan target: x min ≤x i ≤x max and min ≤y i ≤y max With min ≤of i ≤of max Then: The size of the bounding box is: a=x max -and min by max -and min wh=z max -With min Among them: a, b, c represent the length, width and height of the bounding box respectively; Therefore, the robot tool hand and the scanning target do not collide if any of the following conditions are met: (1) On the X-axis, the maximum X-value of the machine tool hand is less than the minimum X-value of the scanned object; (2) On the X-axis, the maximum X-value of the scanned object is less than the minimum X-value of the machine tool hand; (3) On the Y axis, the minimum Y value of the machine tool hand is greater than the maximum Y value of the scanned object; (4) On the Y axis, the minimum Y value of the scanned object is greater than the maximum Y value of the machine tool hand; (5) On the Z axis, the minimum Z value of the machine tool hand is greater than the maximum Z value of the scanned object; (6) On the Z axis, the minimum Z value of the scanned object is greater than the maximum Z value of the machine tool hand; If the robot tool hand and the scanning target do not meet all the above conditions, the robot tool hand and the scanning target will collide and the corresponding configuration solution will be deleted.

4. The complex curved surface part reconstruction method based on a robot arm according to claim 1, characterized in that: Step S3 specifically includes: Convert the point cloud data scanned at different times into the same coordinate system; Project the point cloud data onto the same plane; The two-dimensional Delaunay growth method is used to connect the triangular meshes and establish the topological relationship of the two-dimensional space; According to the coordinate projection relationship of the point cloud data, the topology of the two-dimensional space is mapped to the three-dimensional space to form the topological relationship reconstruction of the potential space.

Images