阅读说明：本技术 一种复杂自由曲面迭代双向法线投影配准方法 (Iterative bidirectional normal projection registration method for complex free-form surface ) 是由 张海涛 毛晴 张璇 李杏华 于 2021-09-30 设计创作，主要内容包括：本发明属于复杂自由曲面点云数据配准技术领域,在三维高精度自动配准领域中,已有的改进算法在一定程度、一定条件下拓宽原始ICP算法的性能,但初始位置对算法收敛的影响、对应关系的搜索及计算仍是该类算法的瓶颈问题；本发明提供一种复杂自由曲面迭代双向法线投影配准方法,在粗配准环节基于数据自身的固有几何特征给出不同的粗配准实现算法。精确配准方法建立在经典的ICP算法框架之上,针对对应关系的建立、伪对应点对的剔除,对ICP算法进行改进,本发明能够更大程度利用隐含的特征信息,配准精度高,且所提供的粗配准方法能够满足不同情况下对初值的需求。(The invention belongs to the technical field of point cloud data registration of complex free-form surfaces, and in the field of three-dimensional high-precision automatic registration, the existing improved algorithm widens the performance of the original ICP algorithm to a certain extent and under certain conditions, but the influence of the initial position on algorithm convergence and the search and calculation of corresponding relations are still the bottleneck problems of the algorithm; the invention provides a complex free-form surface iterative bidirectional normal projection registration method, which provides different rough registration realization algorithms based on inherent geometric characteristics of data in a rough registration link. The accurate registration method is established on a classic ICP algorithm framework, the ICP algorithm is improved aiming at the establishment of the corresponding relation and the elimination of the false corresponding point pair, the method can utilize the implicit characteristic information to a greater extent, the registration accuracy is high, and the provided coarse registration method can meet the requirements of initial values under different conditions.)

1. A registration method for iterative bidirectional normal projection of a complex free-form surface is characterized by comprising the following steps:

step 1: coarse registration of point set P1 and point set P2: curvature information Cur (Q) is calculated for each data point in the point set P1 and P2_i) And angle of normal information ang (Q)_i) (ii) a Generating feature descriptors fea (Q) containing curvature information, normal angle information for each data point in the point set P1 and P2_i) (ii) a According to the feature descriptor fea (Q)_i) Screening a data point Q 'and a data point M' in registration in a point set P1 and a point set P2, determining corresponding point pairs on two planes according to curvature and a normal included angle, and calculating a transformation matrix according to a quaternion method;

step 2: fine registration of point set P1 and point set P2: establishing a preselected corresponding point according to a bidirectional normal vector projection method, wherein the result of forward projection isThe result of the back projection isTo pairPerforming pseudo correspondence elimination and reserving forward correspondence meeting the conditionsTo pairPerforming false corresponding elimination; if the data is processed in blocks, carrying out pseudo-corresponding elimination on each block of data in sequence; obtain a corresponding point set Q^kAnd M^kThe number of the contained corresponding point pairs is N^k(ii) a Computing a rotation matrix R according to a singular value decomposition method^kTranslation vector t^k(ii) a And (3) carrying out data transformation: q^k+1＝R^kQ^k+t^k(ii) a Calculating an error amount:if d is^k+1D, completing registration when the difference is less than or equal to delta; if d is^k+1＞δ，d^k-d^k+1If the difference is greater than epsilon and the maximum iteration number kmax is not reached, the pre-selected corresponding points are re-established for calculation, delta is a given threshold value of the average error of the corresponding points, and epsilon is the variation of the error amount in two adjacent iterations; otherwise, the iteration is terminated.

2. The iterative bi-directional normal projection registration method for complex free-form surfaces according to claim 1, wherein in step 1, any point Q is selected_iImplicit feature descriptors ofWherein the content of the first and second substances,the coefficient corresponding to the curvature; by passingCalculating Q_iNormal vector angle to all neighborhood points, where Q_jIs Q_iA neighborhood point of, Q_jAnd Q_iThe angle between the normal vectors is recorded asQ_i、Q_jNormal vector of (a) is n_i、n_jTo obtain Q_iInformation of normal included angleQ_j∈nbhd(Q_i)，nbhd(Q_i) Is Q_iK adjacent points.

3. The iterative bi-directional normal projection registration method for complex free-form surfaces according to claim 1, wherein the iterative bi-directional normal projection registration method is characterized in thatIn step 1, for any point Q in the point set P1_iIf fea (Q)_i) Kappa, then Q_iSelecting a point in the registered data point set Q 'in the point set P1, and similarly selecting a registered data point set M' and kappa in the point set P2 as a set threshold; if Q'_iAnd M'_jSatisfy the requirement ofThen Q'_iAnd M'_jCorresponding point pairs, κ, of point set P1 and point set P2_curvFor a set curvature threshold, κ_angIs a set normal angle threshold value.

4. The iterative bidirectional normal projection registration method for a complex free-form surface according to claim 1, wherein in the step 2, the method for establishing the preselected corresponding points according to the bidirectional normal vector projection method comprises: fitting NURBS through the point set P1 and the point set P2 to the surface slice to obtain S₁(u,v)、S₂(u, v) is a set of corresponding NURBS patches on two NURBS fitting surfaces, and the patches of the two point sets are respectively solved at data points (u, v)_i,v_i) The unit normal vector of (A) is:S_u(u_i,v_i) And S_v(u_i,v_i) Tangent lines of the NURBS curved surface in the u direction and the v direction are respectively formed; calculating the surface S₁(u, v) at data point { Q }₁,Q₂,...,Q_hThe normal vector n at_AAnd a curved surface S₂(u, v) and the point B is the point A on the curved surface S₂(u, v) a preselected corresponding point; at S₂(u, v) selecting a region adjacent to the point B, randomly extracting the point B', and calculating the curved surface S by the method₂(u, v) normal vector at B' and surface S₁Intersection A' of (u, v); completing from point B' to surface S₁(u, v) backprojecting, points B ', a' being a pair of preselected correspondences determined for the backprojection;

for S₁H points on (u, v) { Q₁,Q₂,...,Q_hAnd sequentially carrying out the bidirectional normal projection, wherein the pre-selected corresponding point pair established by the forward projection is { (Q)_i,M_i) 1, 2.., h }, the preselected correspondence established by the reverse projection is { (M'_i,Q′_i),i＝1,2,...,h}。

5. The iterative bidirectional normal projection registration method for a complex free-form surface according to claim 1, wherein in the step 2, the method for deleting the pseudo corresponding point pairs is as follows: for any set of two-way projection results (Q)_i,M_i) And (M'_i,Q′_i): at the curved surface S₁(u, v) above, point Q_i、Q′_iDistance between is dist (Q)_i-Q′_i) (ii) a At the curved surface S₂(u, v) above, point M_i、M′_iDistance between is dist (M)_i-M′_i) If any set of two-way projection results (Q)_i,M_i) And (M'_i,Q′_i) Satisfies the constraint of_i-Q′_i)-dist(M_i-M′_i) Eta is less than or equal to the set threshold value, then (Q)_i,M_i) And (M'_i,Q′_i) The contained characteristic information is similar, contributes the same to the subsequent registration, and keeps positive correspondence (Q)_i,M_i) Performing subsequent calculation, and removing the pseudo corresponding point pairs (M'_i,Q′_i)。

Technical Field

The invention belongs to the technical field of point cloud data registration of complex free-form surfaces, and particularly relates to a point cloud data registration method of a complex free-form surface by iterative bidirectional normal projection.

Background

For many years, the registration technology is a research hotspot, and the registration is a basic and key link in the fields of computer vision, digital image processing, pattern recognition, reverse engineering and the like. In the measurement and detection of the complex curved surface, the measurement difficulty of the complex free-form curved surface is high, and in the data acquisition process, it is difficult to acquire all required data at the same pose of a measured object by using one sensor. The measured data under different poses and visual angles need to be registered, and the data obtained by different sensors also needs to be considered for data fusion. The purpose of registration is to achieve uniformity between different coordinate systems. In the measurement and detection of the complex curved surface, the unification between each measurement coordinate system and the design coordinate system is the basis and the premise for ensuring the effectiveness of error evaluation. It is a high accuracy auto-registration problem. Since the ICP algorithm was proposed, the ICP algorithm gradually becomes the method with the widest application range in the field of high-precision automatic registration due to its excellent performance. However, theoretically, the ICP algorithm converges on a local minimum value in the sense of a square distance scale, the algorithm has a high requirement for an initial position, and only by providing an initial value close to a global minimum value, the convergence direction can be ensured, and a global optimal solution for curved surface registration is obtained. The original ICP algorithm has strong constraint requirements on the inclusion relationship of two groups of point clouds, namely the algorithm requires that all or a large proportion part of one point cloud is a subset of the other point cloud in the two point clouds, otherwise, the final convergence result is influenced, and even error matching is generated. In addition, the ICP algorithm is a computationally intensive method, the computational overhead needs to be considered in practical application, and the objective function of the method is established on the square term of the error of all data points, which enlarges the proportion of local errors in the registration process. The ICP algorithm has prominent advantages and clear defects, and the ICP algorithm and related variant methods thereof occupy the mainstream position in the field of three-dimensional high-precision automatic registration at present. However, the performance of the original ICP algorithm can be broadened to a certain extent and under certain conditions by the existing improved algorithm, and the influence of the initial position on the convergence condition of the algorithm and the search and calculation of the corresponding relationship are still the bottleneck problems of the algorithm.

Disclosure of Invention

The invention overcomes the defects of the prior art, and solves the technical problems that: the invention provides a complex free-form surface iterative bidirectional normal projection registration method, aiming at the establishment of a corresponding relation and the elimination of a pseudo corresponding point pair, an ICP (inductively coupled plasma) algorithm is improved, and the coarse registration method provided by the invention can meet the requirements on initial values under different conditions.

In order to achieve the purpose, the invention provides the following technical scheme:

a registration method for iterative bidirectional normal projection of a complex free-form surface comprises the following steps:

step 1: coarse registration of point set P1 and point set P2: curvature information Cur (Q) is calculated for each data point in the point set P1 and P2_i) And angle of normal information ang (Q)_i) (ii) a Generating feature descriptors fea (Q) containing curvature information, normal angle information for each data point in the point set P1 and P2_i) (ii) a According to the feature descriptor fea (Q)_i) Screening a data point Q 'and a data point M' in registration in a point set P1 and a point set P2, determining corresponding point pairs on two planes according to curvature and a normal included angle, and calculating a transformation matrix according to a quaternion method;

step 2: fine registration of point set P1 and point set P2: establishing a preselected corresponding point according to a bidirectional normal vector projection method, wherein the result of forward projection isThe result of back-projection is { (M'_i ^k，Q′_i ^k)，i＝0，1，...，h^k}; to pairPerforming pseudo correspondence elimination and reserving forward correspondence meeting the conditionsTo pairPerforming false corresponding elimination; if the data is processed in blocks, carrying out pseudo-corresponding elimination on each block of data in sequence; obtain a corresponding point set Q^kAnd M^kThe number of the contained corresponding point pairs is N^k(ii) a Computing a rotation matrix R according to a singular value decomposition method^kTranslation vector t^k(ii) a And (3) carrying out data transformation: q^k+1＝R^kQ^k+t^k(ii) a Calculating an error amount:if d is^k+1D, completing registration when the difference is less than or equal to delta; if d is^k+1＞δ，d^k-d^k+1If the difference is greater than epsilon and the maximum iteration number kmax is not reached, the pre-selected corresponding points are re-established for calculation, delta is a given threshold value of the average error of the corresponding points, and epsilon is the variation of the error amount in two adjacent iterations; otherwise, the iteration is terminated.

Further, in step 1, any point Q_iImplicit feature descriptors ofWherein the content of the first and second substances,the coefficient corresponding to the curvature; by passingCalculating Q_iNormal vector angle to all neighborhood points, where Q_jIs Q_iA neighborhood point of, Q_jAnd Q_iThe angle between the normal vectors is recorded asQ_i、Q_jNormal vector of (a) is n_i、n_jTo obtain Q_iInformation of normal included angleQ_j∈nbhd(Q_i)，nbhd(Q_i) Is Q_iK adjacent points.

Further, in step 1, Q is assigned to any point in the point set P1_iIf fea (Q)_i) Kappa, then Q_iSelecting a point in the registered data point set Q 'in the point set P1, and similarly selecting a registered data point set M' and kappa in the point set P2 as a set threshold; if Q'_iAnd M'_jSatisfy the requirement ofThen Q'_iAnd M'_jCorresponding point pairs, κ, of point set P1 and point set P2_curvFor a set curvature threshold, κ_angIs a set normal angle threshold value.

Further, in step 2, the method for establishing the preselected corresponding point according to the bidirectional normal vector projection method comprises the following steps: fitting NURBS through the point set P1 and the point set P2 to the surface slice to obtain S₁(u，v)、S₂(u, v) is a set of corresponding NURBS patches on two NURBS fitting surfaces, and the patches of the two point sets are respectively solved at data points (u, v)_i，v_i) The unit normal vector of (A) is:S_u(u_i，v_i) And S_v(u_i，v_i) Tangent lines of the NURBS curved surface in the u direction and the v direction are respectively formed; calculating the surface S₁(u, v) at data point { Q }₁，Q₂，...，Q_hThe normal vector n at_AAnd a curved surface S₂(u, v) and the point B is the point A on the curved surface S₂(u, v) a preselected corresponding point; at S₂(u, v) selecting a region adjacent to the point B, randomly extracting the point B', and calculating the curved surface S by the method₂(u, v) normal vector at B' and surface S₁Intersection A' of (u, v); completing from point B' to surface S₁(u, v) backprojecting, points B ', a' being a pair of preselected correspondences determined for the backprojection; for S₁H points on (u, v) { Q₁，Q₂，...，Q_hAnd sequentially carrying out the bidirectional normal projection, wherein the pre-selected corresponding point pair established by the forward projection is { (Q)_i，M_i) 1, 2.., h }, the preselected correspondence established by the reverse projection is { (M'_i，Q′_i)，i＝1，2，...，h}。

Further, in step 2, the method for deleting the pseudo corresponding point pair includes: for any set of two-way projection results (Q)_i，M_i) And (M'_i，Q′_i): at the curved surface S₁(u, v) above, point Q_i、Q′_iDistance between is dist (Q)_i-Q′_i) (ii) a At the curved surface S₂(u, v) above, point M_i、M′_iDistance between is dist (M)_i-M′_i) If any set of two-way projection results (Q)_i，M_i) And (M'_i，Q′_i) Satisfies the constraint of_i-Q′_i)-dist(M_i-M′_i) Eta is less than or equal to the set threshold value, then (Q)_i，M_i) And (M'_i，Q′_i) The contained characteristic information is similar, contributes the same to the subsequent registration, and keeps positive correspondence (Q)_i，M_i) Performing subsequent calculation, and removing the pseudo corresponding point pairs (M'_i，Q′_i)。

Compared with the prior art, the invention has the following beneficial effects:

1. the coarse registration method based on the inherent geometric characteristics has no requirement on the initial positions of two groups of data, is not strict in the constraint on the overlapping area, and has good universality. In addition, the initial screening based on the descriptors including curvature information and angle information, which is provided by the invention, can effectively delete data, and well reserve points with obvious geometric characteristics, thereby reducing the calculation amount for establishing the corresponding relation.

2. The method comprises the steps of respectively performing surface fitting on measured point cloud data and designed point cloud data, determining preselected corresponding point pairs of two curved surfaces by adopting a bidirectional normal projection method, removing pseudo corresponding relations by utilizing curved surface continuity constraint, distance constraint and curvature constraint, and calculating a coordinate transformation relation according to the obtained corresponding point pairs. Compared with the conventional registration algorithm, the method has the advantage that the accurate automatic registration of the measured data and the design data can be realized under the condition that the original data does not contain the corresponding relation.

Drawings

Fig. 1 shows the main flow of the ICP improvement algorithm based on bi-directional normal projection.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, the invention discloses a registration method for iterative bidirectional normal projection of a complex free-form surface, which comprises the following steps:

step 1: coarse registration of point set P1 and point set P2: curvature information Cur (Q) is calculated for each data point in the point set P1 and P2_i) And angle of normal information ang (Q)_i) (ii) a Generating feature descriptors fea (Q) containing curvature information, normal angle information for each data point in the point set P1 and P2_i) (ii) a According to the feature descriptor fea (Q)_i) Screening a data point Q 'and a data point M' in registration in a point set P1 and a point set P2, determining corresponding point pairs on two planes according to curvature and a normal included angle, and calculating a transformation matrix according to a quaternion method; any point Q_iImplicit feature descriptors ofWherein the content of the first and second substances,the coefficient corresponding to the curvature; let Q_jIs Q_iA normal vector angle between the neighboring points is recorded asQ_i、Q_jNormal vector of (a) is n_i、n_jThen, the first step is executed,calculating Q_iThe normal vector included angles with all the neighborhood points are obtained to obtain Q_iInformation of normal included angleQ_j∈nbhd(Q_i)，nbhd(Q_i) Is Q_iK adjacent points.

For any point Q in point set P1_iIf fea (Q)_i) Kappa, then Q_iSelecting a point in the registered data point set Q 'in the point set P1, and similarly selecting a registered data point set M' and kappa in the point set P2 as a set threshold; if Q'_iAnd M'_jSatisfy the requirement ofThen Q'_iAnd M'_jCorresponding point pairs, κ, of point set P1 and point set P2_curvFor a set curvature threshold, κ_angIs a set normal angle threshold value.

Step 2: fine registration of point set P1 and point set P2: the method for establishing the preselected corresponding points according to the bidirectional normal vector projection method comprises the following steps: fitting NURBS through the point set P1 and the point set P2 to the surface slice to obtain S₁(u，v)、S₂(u, v) is a set of corresponding NURBS patches on two NURBS fitting surfaces, and the patches of the two point sets are respectively solved at data points (u, v)_i，v_i) The unit normal vector of (A) is:S_u(u_i，v_i) And S_v(u_i，v_i) Tangent lines of the NURBS curved surface in the u direction and the v direction are respectively formed; calculating the surface S₁(u, v) at data point { Q }₁，Q₂，...，Q_hThe normal vector n at_AAnd a curved surface S₂(u, v) and the point B is the point A on the curved surface S₂(u, v) a preselected corresponding point; at S₂(u, v) selecting a region adjacent to the point B, randomly extracting the point B', and calculating the curved surface S by the method₂(u, v) normal vector at B' and surface S₁Intersection A' of (u, v); completing from point B' to surface S₁(u, v) and points B ', A' are a pair of preselected corresponding points determined for the backprojection(ii) a For S₁H points on (u, v) { Q₁，Q₂，...，Q_hAnd sequentially carrying out the bidirectional normal projection, wherein the pre-selected corresponding point pair established by the forward projection is { (Q)_i，M_i) 1, 2.., h }, the preselected correspondence established by the reverse projection is { (M'_i，Q′_i) I 1,2,.., h }, the result of forward projection isThe result of back-projection is { (M'_i ^k，Q′_i ^k)，i＝0，1，...，h^k}; to pairPerforming pseudo correspondence elimination and reserving forward correspondence meeting the conditionsTo pairThe method for removing the pseudo corresponding points comprises the following steps: for any set of two-way projection results (Q)_i，M_i) And (M'_i，Q′_i): at the curved surface S₁(u, v) above, point Q_i、Q′_iDistance between is dist (Q)_i-Q′_i) (ii) a At the curved surface S₂(u, v) above, point M_i、M′_iDistance between is dist (M)_i-M′_i) If any set of two-way projection results (Q)_i，M_i) And (M'_i，Q′_i) Satisfies the constraint of_i-Q′_i)-dist(M_i-M′_i) Eta is less than or equal to the set threshold value, then (Q)_i，M_i) And (M'_i，Q′_i) The contained characteristic information is similar, contributes the same to the subsequent registration, and keeps positive correspondence (Q)_i，M_i) Performing subsequent calculation, and removing the pseudo corresponding point pairs (M'_i，Q′_i) (ii) a If the data is processed in blocks, for each block in turnPerforming pseudo-corresponding elimination on the block data; obtain a corresponding point set Q^kAnd M^kThe number of the contained corresponding point pairs is N^k(ii) a Computing a rotation matrix R according to a singular value decomposition method^kTranslation vector t^k(ii) a And (3) carrying out data transformation: q^k+1＝R^kQ^k+t^k(ii) a Calculating an error amount:if d is^k+1D, completing registration when the difference is less than or equal to delta; if d is^k+1δ is a given threshold for the mean error of the corresponding point, d^k-d^k+1If the difference is more than epsilon, the epsilon is the change quantity of the error quantity in the two adjacent iterations, and the maximum iteration times kmax is not reached, and the establishment of the preselected corresponding points for calculation is restarted; otherwise, the iteration is terminated.

The invention provides a high-precision automatic registration method for ensuring the precision, the automation degree and the universality of the detection of a complex curved surface. The registration process inherits the classic idea from coarse to fine, and different coarse registration realization algorithms are provided in the coarse registration link based on the inherent geometric characteristics of data. The accurate registration method is established on a classic ICP algorithm framework, and the ICP algorithm is improved aiming at establishment of the corresponding relation and elimination of the pseudo corresponding point pair. In the registration process, the registration between data points is converted into the registration between fitting patches by adopting a curved surface fitting technology, the implicit characteristic information is utilized to a greater extent, the registration precision is high, and the provided coarse registration method can meet the requirements on initial values under different conditions.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.