A point cloud registration method and device based on the p-order maximum correlation entropy criterion

Abstract

This application discloses a point cloud registration method and device based on the p-order maximum correlation entropy criterion. The method involves inputting a shape point cloud and a model point cloud and initializing parameters; establishing a 3D point cloud rigid body registration optimization model based on the p-order maximum correlation entropy criterion; calculating the spatial correspondence between the shape point cloud and the model point cloud after rigid body transformation; and using the IWO algorithm to solve for the latest rigid body transformation R between the shape point cloud and the model point cloud after rigid body transformation. * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k The rotation matrix and translation vector are iteratively output to determine the optimal rigid body transformation between two point clouds. This application proposes an accurate and robust 3D point cloud rigid body registration method to effectively suppress the influence of noise and outliers on point cloud registration accuracy.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a point cloud registration method and device based on the p-order maximum correlation entropy criterion. Background Technology

[0002] Besl et al. proposed a rigid body registration method for 3D point clouds based on Iterative Closest Point (ICP). The solution process of this method is as follows:

[0003] Given two 3D point clouds, i.e., shape point clouds and model point cloud The goal of the Inter-square similarity algorithm (ICP) is to find the optimal spatial transformations R and t between shape point clouds and model point clouds, such that points in the two point clouds correspond in Euclidean distance space. The ICP algorithm uses the least square distance (LS) as the similarity metric, and its minimum optimization problem is:

[0004]

[0005] In the optimization model (1), two variables need to be optimized: the point cloud correspondence C and the rigid body transformation T (including the rotation matrix R and the translation vector t). The ICP algorithm uses an alternating iterative method to solve for these two variables simultaneously. After initializing the rotation matrix R0 and the translation vector t0, the ICP algorithm solves for the optimal rigid body transformation through alternating iteration. Each iteration includes the following steps:

[0006] Step 1: Based on the (k-1)th rigid body transformation (R k-1 ,t k-1 Establish the point correspondence between two point clouds to be registered.

[0007]

[0008] Step 2: Based on the point correspondence between the current point clouds, calculate the correspondence between the two point clouds. and New rigid body transformation R between * and t * :

[0009]

[0010] Then update the rigid body transformation R at step k. k and t k :

[0011]

[0012] Step 3: Calculate the value of rigid body transformation Rk and t k The mean square error (MSE) between two point clouds, given a mean square error threshold ε. min If ε k ≤ε min If the number of iterations k reaches a given threshold, the algorithm terminates; otherwise, it continues to iterate.

[0013] Since the ICP method uses the least square distance to establish a rigid body registration optimization model for three-dimensional point clouds (1), and the least square distance is very sensitive to noise and outliers, it is difficult to guarantee the registration accuracy of three-dimensional point clouds with noise and outliers. Therefore, the existing rigid body registration method for point clouds has poor registration accuracy and robustness for three-dimensional point clouds with noise and outliers. Summary of the Invention

[0014] This application provides a point cloud registration method and device based on the p-order maximum correlation entropy criterion, and proposes an accurate and robust three-dimensional point cloud rigid body registration method to effectively suppress the influence of noise and outliers on the point cloud registration accuracy.

[0015] This application provides a point cloud registration method based on the p-order maximum correlation entropy criterion, including:

[0016] Input shape point cloud and model point cloud N m ∈N), and initialize the parameters;

[0017] Establish a 3D point cloud rigid body registration optimization model based on the p-order maximum correlation entropy criterion, satisfying:

[0018]

[0019] Where σ represents the size of the kernel window used to control the p-th order maximum correlation entropy criterion, and the p-th order subfunction is used to control the steepness of the p-th order maximum correlation entropy criterion;

[0020] Calculate the shape point cloud after rigid body transformation and model point cloud Spatial correspondence;

[0021] Solving the shape point cloud after rigid body transformation using the IWO algorithm and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k ;

[0022] Iteratively output the rotation matrix and translation vector to determine the optimal rigid body transformation between the two point clouds.

[0023] Optional initialization parameters include: the minimum mean square error ε between two point clouds after a given rigid body transformation. min Maximum number of iterations k for rigid body registration max Set the initial value of the rotation matrix R0 and the initial value of the translation vector t0 for the rigid body transformation, and set the current iteration number k = 1.

[0024] Optionally, calculate the shape point cloud after rigid body transformation. and model point cloud Spatial correspondences include:

[0025] Using the Delaunay triangulation nearest-point search algorithm, and based on R in step k-1... k-1 and t k-1 The calculated values ​​are used to establish the correspondence between the two point clouds in the k-th step.

[0026] Optionally, the IWO algorithm can be used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k include:

[0027] The correspondence between the two point clouds updated after the k-th iteration Calculate two point clouds and Rigid body transformations between them satisfy:

[0028]

[0029] The rotation matrix R is described by three RPY attitude angles: roll angle φ, pitch angle θ, and yaw angle ψ, satisfying the following:

[0030]

[0031] Optionally, the IWO algorithm can be used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k Also includes:

[0032] Given the dimension D of the solution space, the initial population size m0, and the maximum population size mmax The minimum number of seeds s produced by a single weed min And the maximum number of seeds s max Maximum number of iterations J max The nonlinear modulation index p0 and the minimum standard deviation σ of each dimension of seed normal diffusion. min,i (i = 1, 2, ..., D) and the range of values ​​for feasible solutions to rigid body transformations [ρ min ,ρ max The objective function of rigid body transformation is used as the fitness function of weeds, and an initial population of m0 weeds is randomly initialized. Each weed Each consists of 3 RPY attitude angles Translation vector Composition, that is Let the current iteration number j of IWO be 1;

[0033] Calculate the population Ω j Each weed a (a = 1, 2, ..., m) j The number of seeds generated, w a j+1 :

[0034]

[0035] Calculate the standard deviations of each dimension of a seed normal random diffusion. Then, each seed is randomly and normally dispersed near its parent weed individual. Clamping operations are then performed on each seed after spatial dispersion, according to the range of feasible solutions.

[0036]

[0037] Calculate the total number of seeds generated in the (j+1)th iteration. If W j+1 +m j ≤m max Then it is determined by the population Ω j All weeds and their offspring constitute the population Ω after the (j+1)th iteration. j+1 And let m j+1 =W j+1 +m j Otherwise, the population Ω j Each weed and its offspring are sorted according to their fitness, and the first m with the lowest fitness are selected. max Individuals constitute a population Ω j+1 And let m j+1 =m max ;

[0038] If the iteration number j increases by 1, and j ≤ J max If true, then repeat the calculation of population Ω.j Number of seeds produced by each weed a Otherwise, according to the Jth step of the IWO algorithm max Population after the second iteration The location of weeds with minimum fitness As Obtain two point clouds and Rigid body transformation between and

[0039] Update the rigid body transformation R of point clouds Q and M at step k. k and t k :

[0040] R k =R * R k-1 , t k =R * t k-1 +t * .

[0041] Optionally, the IWO algorithm can be used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k Also includes:

[0042] Calculate rigid body transformation R k and R k Mean square error between the last two point clouds

[0043] If ε k ≤ε min or k≥k max If the condition is met, the algorithm iteration terminates and outputs the result.

[0044] Otherwise, the iteration count k is increased by 1, and the shape point cloud after the rigid body transformation is recalculated. and model point cloud Spatial correspondence.

[0045] This application also proposes a computer device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it implements the steps of the aforementioned point cloud registration method based on the p-order maximum correlation entropy criterion.

[0046] This application also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the aforementioned point cloud registration method based on the p-order maximum correlation entropy criterion.

[0047] This application proposes an accurate and robust rigid body registration method for three-dimensional point clouds to effectively suppress the influence of noise and outliers on the registration accuracy of point clouds.

[0048] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0050] Figure 1 This is a flowchart illustrating the point cloud registration method according to an embodiment of this application. Detailed Implementation

[0051] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0052] Given initial values ​​R0 (initial value of rotation matrix) and t0 (initial value of translation vector), the method of this application embodiment solves the rigid body transformation through an iterative method. Each iteration process includes three basic steps: (1) calculating the correspondence between two point clouds based on the rigid body transformation of the previous step; (2) calculating the new rigid body transformation of the point clouds using the Invasive Weed Optimization (IWO) algorithm based on the updated correspondence between the two point clouds; and (3) determining whether the iteration process has ended. Specifically, this application embodiment provides a point cloud registration method based on the p-order maximum correlation entropy criterion, including:

[0053] In step S101, input the shape point cloud. and model point cloud And initialize parameters. In some embodiments, initializing parameters includes: the minimum mean square error ε between two point clouds after a given rigid body transformation. min Maximum number of iterations k for rigid body registration max Set the initial value of the rotation matrix R0 and the initial value of the translation vector t0 for the rigid body transformation, and set the current iteration number k = 1.

[0054] In step S102, a three-dimensional point cloud rigid body registration optimization model based on the p-order maximum correlation entropy criterion is established, satisfying:

[0055]

[0056] stR T R = I3, det(R) = 1

[0057] The p-order maximum correlation entropy criterion has two parameters: the kernel function width σ and the p-order sub-function. σ controls the size of the kernel window of the p-order maximum correlation entropy criterion, and the p-order sub-function controls the steepness of the p-order maximum correlation entropy criterion.

[0058] In step S103, the shape point cloud after rigid body transformation is calculated. and model point cloud The spatial correspondence. In some embodiments, the Delaunay triangulated nearest-point search algorithm can be used, and based on the (k-1)th step R... k-1 and t k-1 The calculated values ​​are used to establish the correspondence between the two point clouds in the k-th step.

[0059] In step S104, the IWO algorithm is used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k .

[0060] In step S105, the rotation matrix and translation vector are iteratively output to determine the optimal rigid body transformation between the two point clouds.

[0061] This application proposes an accurate and robust rigid body registration method for three-dimensional point clouds, which can effectively suppress the influence of noise and outliers on the registration accuracy of point clouds.

[0062] In some embodiments, the IWO algorithm is used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t* And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k include:

[0063] The correspondence between the two point clouds updated after the k-th iteration Calculate two point clouds and Rigid body transformations between them satisfy:

[0064]

[0065] The rotation matrix R is described by three RPY attitude angles: roll angle φ, pitch angle θ, and yaw angle ψ, satisfying the following:

[0066]

[0067] Among them, the ranges of roll angle φ, pitch angle θ, and yaw angle ψ are respectively φ c ∈[-180°, 180°), θ c ∈[-90°, 90°] and ψ c ∈[-180°, 180°).

[0068] This application embodiment further employs the IWO algorithm to analyze two point clouds. and The rigid body transformations between the two are optimized and solved, where the initial values ​​of R and t are chosen as the identity matrix and the zero vector, respectively.

[0069] Based on this, the rigid body transformation R of the point cloud Q and M in the k-th step is updated using the following formula. k and t k :

[0070] R k =R * R k-1 , t k =R * t k-1 +t * (8)

[0071] Specifically, in some embodiments, the IWO algorithm is used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k Also includes:

[0072] Given the dimension D of the solution space, for example, D = 6, the initial population size m0, and the maximum population size mmax The minimum number of seeds s produced by a single weed min And the maximum number of seeds s max Maximum number of iterations J max The nonlinear modulation index p0 and the minimum standard deviation σ of each dimension of seed normal diffusion. min,i (i = 1, 2, ..., D) and the range of values ​​for feasible solutions to rigid body transformations [ρ min ,ρ max The objective function of rigid body transformation is used as the fitness function of weeds, and an initial population of m0 weeds is randomly initialized. Each weed Each consists of 3 RPY attitude angles Translation vector Composition, that is Let the current iteration number of IWO be j = 1.

[0073] Calculate the population Ω j Each weed a (a = 1, 2, ..., m) j Number of seeds generated

[0074]

[0075] Calculate the standard deviations of each dimension of a seed normal random diffusion. Then, each seed is randomly and normally dispersed near its parent weed individual. Clamping operations are then performed on each seed after spatial dispersion, according to the range of feasible solutions.

[0076]

[0077] Calculate the total number of seeds generated in the (j+1)th iteration. If W j+1 +m j ≤m max Then it is determined by the population Ω j All weeds and their offspring constitute the population Ω after the (j+1)th iteration. j+1 And let m j+1 =W j+1 +m j Otherwise, the population Ω j Each weed and its offspring are sorted according to their fitness, and the first m with the lowest fitness are selected. max Individuals constitute a population Ω j+1 And let m j+1 =m max .

[0078] If the iteration number j increases by 1, and j ≤ J max If true, then repeat the calculation of population Ω. jNumber of seeds produced by each weed a Otherwise, according to the Jth step of the IWO algorithm max Population after the second iteration The location of weeds with minimum fitness As Obtain two point clouds and Rigid body transformation between and

[0079] Update the rigid body transformation R of the point cloud Q and M in step k using formula (8). k and t k .

[0080] In some embodiments, the IWO algorithm is used to solve the shape point cloud after rigid body transformation. and model point cloud The latest rigid body transformation R between * and t * And update the rotation matrix R of the rigid body transformation in step k. k Translation vector t k Also includes:

[0081] Calculate rigid body transformation R k and R k Mean square error between the last two point clouds

[0082] If ε k ≤ε min or k≥k max If the condition is met, the algorithm iteration terminates and outputs the result.

[0083] Otherwise, the iteration count k is increased by 1, and the shape point cloud after the rigid body transformation is recalculated. and model point cloud Spatial correspondence.

[0084] For example, in the kth iteration ε k ≤ε min or k≥k max If true, then output the rotation matrix R obtained in the k-th iteration. k Translation vector t k Its value is the optimal rigid body transformation between the two point clouds.

[0085] To address the rigid body registration problem of 3D point clouds containing noise and outliers, this application proposes a point cloud registration method based on the p-order maximum correlation entropy criterion. To effectively suppress the impact of noise and outliers on registration accuracy, this application introduces a p-order maximum correlation entropy criterion robust to noise and outliers to establish an optimization model for 3D point cloud rigid body registration. This invention employs an alternating iterative approach to solve for the two unknown variables: the correspondence between point clouds and the rigid body transformation. The optimization model proposed in this application effectively suppresses noise and outliers in point clouds, thereby significantly improving the accuracy and robustness of the 3D point cloud registration method.

[0086] This application also proposes a computer device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it implements the steps of the aforementioned point cloud registration method based on the p-order maximum correlation entropy criterion.

[0087] This application also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the aforementioned point cloud registration method based on the p-order maximum correlation entropy criterion.

[0088] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0089] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0090] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0091] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method of registering point clouds based on p a maximum correlation entropy criterion, characterized in that Comprising: input shape point cloud ( , ) and model point cloud ( , ), and initialize parameters; A three-dimensional point cloud registration optimization model based on p The three-dimensional point cloud registration optimization model based on maximum correlation entropy criterion meets: wherein, denotes a control signal for controlling p the size of the kernel window of the order maximum correlation entropy criterion, p the order function for controlling p the steepness of the order maximum correlation entropy criterion; Computing spatial correspondence of a shape point cloud after rigid body transformation and a model point cloud ;​ Solving shape point cloud after rigid body transformation using IWO algorithm and model point cloud between the latest rigid body transformation R * and t * , and update the rotation matrix of the rigid body transformation in step k R k and the translation vector t k ;​ iteratively outputting a rotation matrix and a translation vector to determine an optimal rigid body transformation between two point clouds.

2. The method of claim 1, wherein the one or more parameters are selected based on a maximum correlation entropy criterion. p A point cloud registration method based on a maximum correlation entropy criterion, characterized in that, The initialization parameters include: minimum value of mean square error between two point clouds after a given rigid body transformation maximum number of iterations for rigid body registration k max initial value of rotation matrix of rigid body transformation R 0 and initial value of translation vector t 0, and let the current iteration number k =1.

3. The method of claim 1, wherein the one or more parameters are selected based on a maximum correlation entropy criterion. p The point cloud registration method based on a maximum correlation entropy criterion, wherein, Computing a spatial correspondence between a shape point cloud after rigid body transformation and a model point cloud includes: The nearest point search algorithm using Delaunay triangulation is used, and the correspondence between the two point clouds is established according to the first k -1 step R k-1 and t k-1 The calculation value of the first k step establishes the correspondence between the two point clouds .

4. The method based on claim 3 p A point cloud registration method based on the maximum correlation entropy criterion is characterized by, Solving shape point cloud after rigid body transformation using IWO algorithm and model point cloud between the latest rigid body transformation R * and t * and update the rotation matrix of the k-th step rigid body transformation R k and translation vector t k Comprising: For the first k Two point cloud correspondence relations after step iteration update , the rigid transformation between two point clouds and is calculated, which satisfies: where the rotation matrix R is described by the three RPY attitude angles , the pitch angle and the yaw angle satisfying: 。 5. The method of claim 4, wherein the plurality of points are based on p A point cloud registration method based on the maximum correlation entropy criterion, characterized in that, Solving the shape point cloud after rigid body transformation using the IWO algorithm and the model point cloud between the latest rigid body transformation R * and t * and update the rotation matrix of the k step rigid body transformation R k and the translation vector t k also comprises: Given solution space dimension D , initial population size , and maximum population size , minimum number of seeds for a single weed , and maximum number of seeds , maximum number of iterations , nonlinear modulation index p 0, minimum standard deviation of each dimension for normal seed diffusion , , and value range of feasible rigid body transformation , the objective function of rigid body transformation is taken as the fitness function of weeds, and the original population consisting of m 0 weeds is randomly initialized , where each weed is composed of 3 RPY attitude angles and a translation vector , that is ; let the current iteration number of IWO j =1; Computing population Each of the weeds , The number of seeds produced : ， standard deviation of each dimension of the normal random diffusion of the seed , , and the various seeds are normally randomly diffused around the parent weed individual, and the clamping operation is performed on each seed after the spatial diffusion according to the value range of the feasible solution: ， Calculate the () j The total number of seeds generated in +1) iterations ,like Then it is determined by the population All weeds and their offspring constitute the ( ) j Population after +1) iterations and order Otherwise, the population Each weed and its offspring are sorted according to their fitness, with the weeds having the lowest fitness ranked first. Individuals constitute a population And make ; Number of iterations j is increased by 1, and if is true, then the calculation of the population of weeds is repeated, and if is false, then the number of seeds produced by each weed in the population of weeds after the IWO algorithm has been iterated for the first time is determined, and if is the position of the weed with the smallest fitness in the population of weeds , then is obtained, and two point clouds and are obtained; a rigid body transformation between the two point clouds is obtained. updating the first k point cloud of the step Q and M rigid body transformation R k and t k : ， 。 6. The method of claim 5, wherein the at least one of the plurality of points is determined based on a maximum correlation entropy criterion. p The point cloud registration method based on a maximum correlation entropy criterion, characterized in that, Solving shape point cloud after rigid body transformation using IWO algorithm and model point cloud between the latest rigid body transformation R * and t * and update the rotation matrix of the rigid body transformation of the first k step R k and the translation vector t k also comprises: Computing rigid body transformations R k And R k Mean squared error between the last two point clouds , If or is true, the algorithm iteration terminates and the output is given. Otherwise iteration number Increase 1, repeat the calculation of the shape point cloud after the rigid transformation And the spatial correspondence relationship of the model point cloud. ​ 7. A computer device, characterized by A computer program product comprising a computer readable medium having stored thereon computer program means, the computer program means comprising computer program instructions executable by a processor to cause the processor to perform the method according to any one of claims 1 to 6. p Steps of a point cloud registration method based on the order maximum correlation entropy criterion.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method for registering point cloud based on the maximum correlation entropy criterion of order p The steps of the point cloud registration method based on the maximum correlation entropy criterion of order.

Images