Adaptive representation constraint-based point cloud registration method and system

Abstract

The invention provides a point cloud registration method and system based on adaptive representation constraint, and the method comprises the steps: enabling to-be-registered point cloud data to pass through a registration constraint model based on adaptive representation, and analyzing the spatial position relation between a to-be-registered source point set and a to-be-registered target point set, so as to determine a registration region and a registration point location, the method comprises the following steps of: performing robust estimation on a point set with abnormal points at a registration point position, performing fitness evaluation, optimizing registration parameters, and performing fitness evaluation and optimization updating on the registration parameters through iteration to obtain optimal registration parameters, thereby realizing registration and alignment of point cloud data.

Description

Technical Field

[0001] The present invention belongs to the technical fields such as point cloud registration, and particularly relates to a point cloud registration method and system based on adaptive representation constraint. Background Art

[0002] In recent years, with the rapid development of measurement technologies, point cloud registration technology has been widely applied in fields such as medical image processing, robot navigation, and autonomous driving, and has become an important means for acquiring and analyzing spatial information. In order to achieve high-precision registration, researchers have proposed various registration methods for different application scenarios and achieved good registration effects to a certain extent. However, in practical applications, due to the influence of factors such as measurement errors and environmental noises, point cloud data often contains a large amount of noises and gross errors, resulting in the degradation of the performance of existing registration methods and even the inability to achieve robust registration. Especially in scenarios with large-scale, high-noise or missing data, how to improve the registration accuracy and enhance the robustness to noises and outliers is still a key issue in current research. Summary of the Invention

[0003] Aiming at the defects and deficiencies existing in the prior art, the present invention proposes a point cloud registration method based on adaptive representation constraint, which belongs to a brand-new robust point cloud registration method. By simply inputting the measurement data into the method, high-precision point cloud registration can be achieved.

[0004] Based on the registration constraint model of adaptive representation of the present invention, the fitness of the registration parameters is evaluated, and the improved snake optimization algorithm is used as the optimization method for the registration parameters. The registration parameters are optimized and updated iteratively to obtain the optimal registration parameters.

[0005] The technical solution specifically adopted by the present invention to solve its technical problems is as follows:

[0006] A point cloud registration method based on adaptive representation constraint: The point cloud data to be registered is analyzed through the registration constraint model of adaptive representation to determine the spatial position relationship between the source point set and the target point set to be registered, so as to determine the registration region and registration points. The point set with outliers is robustly estimated at the registration points, and then the fitness evaluation is carried out, and the registration parameters are optimized. The fitness evaluation and optimization update of the registration parameters are carried out iteratively to obtain the optimal registration parameters, so as to achieve the registration alignment of the point cloud data.

[0007] Furthermore, the construction process of the registration constraint model of adaptive representation is specifically as follows:

[0008] For the source point set P and the target point set Q to be registered, assuming there are outliers in the source point set P, after the source point set P undergoes rotation and translation and reaches a new position, denoted as the source point set P', then according to the spatial positions of the source point set P' and the target point set Q, the spatial intersection region of the two point sets is determined; within the said spatial intersection region, several points are randomly selected from the target point set Q, denoted as the matching points of the target point set; the number of matching points in the target point set within the intersection region is defined as:

[0009]

[0010] where N is the number of selected matching points, l is the range of the intersection region, c is a proportionality parameter, and md is the average projection distance of all points in the target point set Q within the intersection region;

[0011] Taking the positions of the matching points of the target point set as the positions of the points to be estimated in the source point set P', the source point set is divided into corresponding numbers of influence domains, and within each influence domain, a robust estimation method is used to obtain the estimated values, and the fitness is calculated using the distances d = {d1, d2, d3, d4,...} between the matching points in the target point set and the estimated values of the points to be estimated in the source point set.

[0012] Furthermore, the fitness for the fitness calculation is the arithmetic mean of the distances.

[0013] Furthermore, the specific robust estimation method is as follows;

[0014] Filter out the discrete points of the source point set within the influence domain;

[0015] Generate a reference curve through the student t-distribution parameter regression method based on ECM estimation;

[0016] According to the set iteration stop condition, perform density ratio clustering on the absolute values of the residuals between the discrete points within the influence domain and the reference curve, record the absolute values of the residuals and their corresponding point sets for each class after each clustering, and remove one class with the largest mean absolute value of the residuals each time;

[0017] Using the weighted Legendre basis function method, introducing a compactly supported weight function, and solving using the least squares criterion, complete the robust estimation of the points to be estimated, obtain the estimated values, and as the influence domain moves step by step, realize the robust estimation of all points to be estimated in the source point set within the intersection region.

[0018] Furthermore, the robust estimation method uses the second-order Legendre basis function; the expression of the compactly supported weight function used is:

[0019]

[0020] where d is the projection distance between each discrete point within the influence domain and the point to be estimated.

[0021] Further, the specific process of the density ratio clustering is as follows:

[0022] Mark all data points as "unvisited", set the small neighborhood radius ∈, the large neighborhood radius η, the density ratio threshold τ, and let the cluster number C = 0;

[0023] Select an unvisited point p from the dataset, mark it as visited, calculate the number of points in the ∈-neighborhood and the η-neighborhood respectively, so as to obtain the density ratio R(p); if R(p) ≥ τ, create a new cluster C and add the point p to it;

[0024] Expand the cluster, include all points in the ∈-neighborhood of point p into a list to be processed, take out point p' one by one and calculate the density ratio. If R(p') ≥ τ, then incorporate the ∈-neighborhood of point p into the list to be processed; meanwhile, if point p' does not belong to any cluster, incorporate it into the current cluster C; if the density ratio R(p) of a certain point p < τ, mark it as a noise point;

[0025] Repeat the above process continuously until all points are visited, so as to obtain clusters that can adapt to different densities and some noise points.

[0026] Further, an improved snake optimization algorithm is used as the iterative optimization method for registration parameters. The improved snake optimization algorithm uses the Bernoulli chaotic map to generate the initial population; divides the population into male and female groups, and controls the behavior pattern switching through the food amount and temperature as dynamic thresholds; introduces the lens imaging principle to generate reverse solutions, and dynamically adjusts the reverse point positions through the scaling factor; and when the change amplitude of the fitness is less than the threshold, applies a random perturbation near the optimal solution to generate temporary individuals for exploration.

[0027] Further, the improved snake optimization algorithm specifically includes the following steps:

[0028] Step 1: Read data and initialize parameters; the initial value of the population is randomly generated by the rand function and then updated by the Bernoulli chaotic map. Initialize the initial position of the snake individual as:

[0029] X i = X min + B i × (X max - X min )

[0030] where, X i represents the position of the i-th snake, i = 1, 2,..., N; X min and X max represent the lower and upper bounds of the solution space; B i represents the i-th vector of the chaotic random number B(i, j); obtain the initial population:

[0031] X = [X1, X2, …, X N

[0032] The population is then divided into two groups: male and female;

[0033] Step 2: Evaluate the initial population; calculate the objective function value for each snake's position and find the current optimal solution and its corresponding fitness;

[0034] Step 3: Simulate the "amount of food" Q and "temperature" Temp based on the number of iterations and random parameters, and decide whether the snake enters the foraging mode, combat mode, or mating mode;

[0035] Step 4: If Q < Threshold, it means there is insufficient food, and the snake enters the foraging mode; if Q ≥ Threshold and Temp > Threshold2, it means the snake will only move to the food source; if Q ≥ Threshold and Temp < Threshold2, it means there is sufficient food and suitable temperature, and the snake enters the combat mode or mating mode;

[0036] Step 5: Update the population using the lens imaging reverse learning strategy. The specific process is as follows: Map the snake individuals to the x-axis to obtain the position X new , and the reverse snake individuals obtained through lens imaging are mapped to the x-axis to obtain X new *, and the calculation formula is:

[0037]

[0038] Let h / h* = k, where k represents the scaling factor and can be expressed as:

[0039]

[0040] The formula for the position of the reverse point in space is:

[0041]

[0042] Then map the position of the reverse point X new * back to the snake individuals;

[0043] Step 6: Calculate the objective function value for the updated population. If the new solution is better than the current solution, replace it and update the global optimal solution;

[0044] Step 7: Detect the difference between the latest fitness and the average result within the previous K iterations. If the change amplitude is less than the predetermined threshold, initiate a micro-perturbation, and the perturbation calculation formula is:

[0045] X p = (rand - 0.5)·2·S·(X max - X min ​)

[0046] Among them, X p represents the perturbation increment, S represents the perturbation intensity, X min and X max represent the lower and upper bounds of the solution space; K is set to the maximum number of iterations * proportionality coefficient; generate a temporary individual based on the current optimal solution and add perturbations, while ensuring that it is still within the boundaries after perturbation:

[0047] X temp = X best + X p

[0048] X temp = max(min(X temp , X max ), X min )

[0049] Calculate the objective function value. If the fitness is better than the original optimal solution, update the optimal solution; otherwise, discard the solution; the number of perturbation times is a set value;

[0050] Repeat steps three to seven until the maximum number of iterations T max or the convergence condition is met;

[0051] The objective function value is obtained through the registration constraint model based on adaptive representation.

[0052] And, a point cloud registration system based on adaptive representation constraint, including: a registration constraint model based on adaptive representation, passing the point cloud data to be registered through the registration constraint model based on adaptive representation, analyzing the spatial position relationship between the source point set and the target point set to be registered to determine the registration area and registration points, performing robust estimation on the point set with abnormal points at the registration points, and then performing fitness evaluation; and a parameter optimization module for optimizing the registration parameters, performing fitness evaluation and optimization update of the registration parameters through iteration to obtain the optimal registration parameters, so as to achieve the registration alignment of the point cloud data.

[0053] An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and the steps of the above method are implemented when the processor executes the program.

[0054] A non-transitory computer-readable storage medium, on which a computer program is stored, and the steps of the above method are implemented when the computer program is executed by the processor.

[0055] Compared with the prior art, in the registration problem, the present invention and its preferred solutions can automatically determine the registration area and registration points for contaminated registration point cloud data through a registration constraint model with adaptive representation, robustly estimate the point set with abnormal points at the registration points, then conduct fitness evaluation, and use an improved snake optimization algorithm as the optimization method for registration parameters. Through iterative fitness evaluation and optimization update of the registration parameters, the optimal registration parameters can be obtained robustly and efficiently, so as to achieve automatic registration alignment only by inputting the point cloud measurement data of complex curves or surfaces. Its outstanding features include high precision, excellent robustness, etc. This method successfully suppresses highly contaminated abnormal points and provides registration parameters that best conform to the true shape of the registration point cloud data. The achievements of the present invention can be widely applied to fields such as 3D modeling, reverse engineering, and computer graphics. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0057] Figure 1 Schematic diagram of the registration constraint model based on adaptive representation for the embodiment of the present invention;

[0058] Figure 2 Schematic diagram of the robust estimation method for the embodiment of the present invention;

[0059] Figure 3 Flowchart of the improved snake optimization algorithm for the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] To make the features and advantages of this patent more obvious and understandable, specific embodiments are given below and described in detail as follows:

[0061] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations for this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0062] It should be noted that the terms used here are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to this application. As used here, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "include" and / or "comprise" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0063] The registration constraint model is the key to evaluating fitness. The registration constraint model based on adaptive representation proposed in the embodiments of the present invention automatically determines the registration area and registration points by analyzing the spatial position relationship between the source point set and the target point set to be registered, robustly estimates the point set with outliers at the registration points, and then conducts fitness evaluation. The following is a detailed introduction to the registration model method with adaptive representation constraints, and its construction principle is as Figure 1 shown.

[0064] First, as Figure 1 (a) shows, there are two sets of point sets to be registered, which are respectively called the source point set P and the target point set Q, and there are outliers in the source point set P.

[0065] Second, after the source point set P is rotated and translated to reach a new position, denoted as the source point set P', then according to the spatial positions of the source point set P' and the target point set Q, the spatial intersection area of the two sets of point sets is determined. In this area, several points are randomly selected from the target point set Q, denoted as the matching points of the target point set. As Figure 1 (b) shows, the number of matching points in the figure is only for illustration. The number of matching points in the target point set in the intersection area is defined as:

[0066]

[0067] where N is the number of selected matching points, l is the range of the intersection area, c is a proportionality parameter, and md is the average projection distance of all points in the target point set Q in the intersection area. The calculation formula is:

[0068]

[0069] Then, in the source point set P', taking the position of the matching points of the target point set as the position of the points to be estimated (in the registration of curve point sets, the x coordinate position of the matching point is the coordinate position of the point to be estimated), the source point set is divided into corresponding numbers of influence domains, and robust estimation methods are used to obtain estimated values in each influence domain. As Figure 1 (c) shows.

[0070] Finally, as Figure 1 (d) shows, the fitness is calculated using the distances d = {d1, d2, d3, d4,...} between the matching points in the target point set and the estimated values of the points to be estimated in the source point set, and the fitness is set as the arithmetic mean of the distances.

[0071] To perform a robust estimation of the points to be estimated in the source point set P', the present invention proposes a preferred robust estimation method. (It should be noted that within the framework of the registration constraint model based on adaptive representation of the present invention, other existing robust estimation schemes can also be adopted. The scheme given in the embodiment is only a preferred scheme rather than a limitation on the only implementation manner of the overall scheme of the present invention.) The following is a detailed introduction to this robust estimation method, and its principle is as Figure 2 shown.

[0072] First, as Figure 2 (a) shows, the discrete points of the source point set within the influence domain are screened out.

[0073] Secondly, a reference curve is generated by the student t-distribution parameter regression method based on ECM estimation, as Figure 2 (b) shows. The ECM estimation method maximizes the location vector μ, covariance Σ, and degrees of freedom ν of the multivariate t-distribution in three steps: In the E step, the conditional expectation is calculated:

[0074]

[0075] CM step 1: Fix the degrees of freedom ν and update μ and Σ:

[0076]

[0077] CM step 2: Fix the new μ and Σ and then solve for the degrees of freedom ν:

[0078]

[0079] Then, according to the set iteration stop condition, density ratio clustering is performed on the absolute value of the residual between the discrete points within the influence domain and the reference curve, and the absolute value of the residual of each class and its corresponding point set after each clustering are recorded. Each time, the class with the largest mean absolute value of the residual is removed, as Figure 2 (c) shows. The clustering process mainly includes the following steps:

[0080] Step 1: Mark all data points as "unvisited", set a small neighborhood radius ∈, a large neighborhood radius η, a density ratio threshold τ, and let the clustering number C = 0;

[0081] Step 2: Select an unvisited point p from the data set, mark it as visited, and calculate the number of points in the ∈ neighborhood and the η neighborhood respectively, so as to obtain the density ratio R(p). If R(p) ≥ τ, it means that p is in a high-density area, and a new cluster C can be created and p can be added to it;

[0082] Step 3: Expand clustering. Incorporate all points in the ∈-neighborhood of p into a list to be processed. Take out point p' one by one and calculate its density ratio. If R(p') ≥ τ, then incorporate the ∈-neighborhood of p into the list to be processed as well. Meanwhile, if p' does not belong to any cluster, incorporate it into the current cluster C. If the density ratio R(p) of a certain point p < τ, then mark it as a noise point;

[0083] Step 4: Continuously repeat the above process until all points are visited, thereby obtaining clusters that can adapt to different densities and some noise points.

[0084] Finally, use the weighted Legendre basis function method, introduce a compactly supported weight function, and solve using the least squares criterion to complete the robust estimation of the points to be estimated, as shown in Figure 2 (d). The expression of the Legendre basis function is:

[0085]

[0086] In this embodiment, the second-order Legendre basis function is preferably adopted. The expression of the compactly supported weight function is:

[0087]

[0088] where d is the projection distance between each discrete point in the influence domain and the point to be estimated.

[0089] As the influence domain moves step by step, the robust estimation of all points to be estimated in the source points in the intersection area is thus realized.

[0090] Based on the constructed registration constraint model, the present invention further proposes an improved snake optimization algorithm to iteratively optimize the registration parameters in the constraint model (it should be noted that within the framework of the above-introduced solution of the present invention, other existing similar meta-heuristic optimization algorithms or swarm intelligence optimization algorithms can also be used. The solution given in the embodiment is only taken as a preferred solution rather than a limitation on the only implementation method of the overall solution of the present invention). The following is a detailed introduction to this algorithm, and its construction principle is as shown in Figure 3 shown.

[0091] Step 1: Read data and initialize parameters. The initial value of the population is randomly generated by the rand function and then updated by the Bernoulli chaotic mapping. The mathematical expression of the Bernoulli mapping is:

[0092]

[0093] where B(i, j) represents a chaotic random number, the range of i is [1, N], the range of j is [1, dim], and the range of B(i, j) is [0, 1]. Therefore, the initial position of the snake individual is initialized as:

[0094] X i = X min + B i × (X max - X min )(10)

[0095] wherein, X i represents the position of the i-th snake (i = 1, 2,..., N); X min and X max represent the lower and upper bounds of the solution space; B i represents the i-th vector of B. The initial population is obtained as follows:

[0096] X = [X1, X2,..., X N (11)

[0097] The population is then divided into two groups: male and female.

[0098] Step 2: Evaluate the initial population. Calculate the objective function value for each snake's position and find the current optimal solution and the corresponding fitness.

[0099] Step 3: Simulate the "amount of food" Q and the "temperature" Temp according to the number of iterations and random parameters, and determine whether the snake enters the foraging mode, combat mode, or mating mode.

[0100] Step 4: If Q < Threshold, it means the food is insufficient, and the snake enters the foraging mode. If Q ≥ Threshold and Temp > Threshold2, it means the snake only moves to the food. If Q ≥ Threshold and Temp < Threshold2, it means the food is sufficient and the temperature is suitable, and the snake enters the combat mode or mating mode. Preferably, Threshold is set to 0.25 and Threshold2 is set to 0.6.

[0101] Step 5: Update the population using the lens imaging reverse learning strategy. The specific process is as follows: Map the snake individual to the x-axis to obtain the position X new , and after lens imaging, the reverse snake individual is obtained and mapped to the x-axis to obtain X new *, and the calculation formula is:

[0102]

[0103] Let h / h* = k, where k represents the scaling factor and can be expressed in the present invention as:

[0104]

[0105] As the number of iterations increases continuously, the value of k will continuously increase. According to the mapping formula transformation, the formula for the position of the reverse point in space is:

[0106]

[0107] Then, the position X of the reverse point new * is mapped back to the snake individual.

[0108] Step 6: Calculate the objective function value for the updated population. If the new solution is better than the current solution, replace it and update the global optimal solution.

[0109] Step 7: Detect the difference between the latest fitness and the average result within the previous K iterations. If the change amplitude is less than the predetermined threshold, initiate micro-perturbation. The perturbation calculation formula is:

[0110] X p = (rand - 0.5)·2·S·(X max - X min )(15)

[0111] where, X p represents the perturbation increment, S represents the perturbation intensity, X min and X max represent the lower and upper bounds of the solution space. K is set to the maximum number of iterations * proportionality coefficient, and this proportionality coefficient can be set from 0 to 1. Generate a temporary individual based on the current optimal solution and add perturbation, while ensuring that it is still within the boundary after perturbation:

[0112]

[0113] Calculate the objective function value. If this fitness is better than the original optimal solution, update the optimal solution; otherwise, discard this solution. Perturbation can be repeated multiple times.

[0114] Repeat Steps 3 to 7 until the maximum number of iterations T max or the convergence condition is met.

[0115] When calculating the objective function value in the above steps, this objective function value is obtained by using the registration constraint model method based on adaptive representation proposed by the present invention.

[0116] Combining according to the above steps can obtain the point cloud registration scheme based on adaptive representation constraint finally and preferably provided by the embodiments of the present invention.

[0117] Based on the same inventive concept, the present invention further provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is used to implement one or more instructions, specifically for loading and executing one or more instructions in the computer storage medium to implement the above method.

[0118] It should be further noted that, based on the same inventive concept, the present invention further provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or combined with an instruction execution system, apparatus, or device.

[0119] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0120] As described above, it is only the preferred embodiment of the present invention, and it is not a limitation of the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

[0121] This patent is not limited to the above best implementation mode. Anyone inspired by this patent can obtain various other forms of a point cloud registration method and system based on adaptive representation constraints. All equal changes and modifications made according to the scope of the patent application of the present invention shall fall within the coverage scope of this patent.

Claims

1. A point cloud registration method based on adaptive representation constraint, characterized in that: The point cloud data to be registered is analyzed through a registration constraint model based on adaptive representation to determine the spatial position relationship between the source point set and the target point set to be registered, so as to determine the registration area and registration points. For the point set with abnormal points, a robust estimation is performed at the registration points, and then a fitness evaluation is carried out, and the registration parameters are optimized. Through iterative fitness evaluation and optimization update of the registration parameters, the optimal registration parameters are obtained, so as to achieve the registration alignment of the point cloud data.

2. The point cloud registration method based on adaptive representation constraint according to claim 1, characterized in that: The construction process of the registration constraint model based on adaptive representation is specifically as follows: For the source point set P and the target point set Q to be registered, assuming that there are abnormal points in the source point set P, after the source point set P is rotated and translated to reach a new position, denoted as the source point set P', then according to the spatial positions of the source point set P' and the target point set Q, the spatial intersection area of the two point sets is determined; within the spatial intersection area, several points are randomly selected from the target point set Q, denoted as the matching points of the target point set; the number of matching points in the target point set of the intersection area is defined as: where N is the number of selected matching points, l is the range of the intersection area, c is the proportionality parameter, and md is the average projection distance of all points in the target point set Q in the intersection area; Taking the positions of the matching points of the target point set as the positions of the points to be estimated in the source point set P', the source point set is divided into corresponding numbers of influence domains, and robust estimation methods are used to obtain estimated values within each influence domain. The fitness is calculated using the distances d = {d1, d2, d3, d4,...} between the matching points in the target point set and the estimated values of the points to be estimated in the source point set.

3. The point cloud registration method based on adaptive representation constraint according to claim 2, characterized in that: The fitness of the fitness calculation is the arithmetic mean of the distances.

4. The point cloud registration method based on adaptive representation constraint according to claim 2, characterized in that: The robust estimation method is specifically as follows; Filter out the discrete points of the source point set within the influence domain; Generate a reference curve through the student t-distribution parameter regression method based on ECM estimation; According to the set iteration stop condition, perform density ratio clustering on the absolute values of the residuals between the discrete points within the influence domain and the reference curve, record the absolute values of the residuals and their corresponding point sets for each class after each clustering, and remove the class with the largest mean absolute value of the residuals each time; Use the weighted Legendre basis function method, introduce a compactly supported weight function, and solve using the least squares criterion to complete the robust estimation of the points to be estimated and obtain the estimated values. As the influence domain moves gradually, the robust estimation of all points to be estimated in the source point set within the intersection area is realized.

5. The point cloud registration method based on adaptive representation constraint according to claim 4, characterized in that: The robust estimation method uses a second-order Legendre basis function; the expression of the compactly supported weight function used is: where d is the projection distance between each discrete point within the influence domain and the point to be estimated.

6. The point cloud registration method based on adaptive representation constraint according to claim 4, characterized in that: The specific process of the density ratio clustering is: Mark all data points as "unvisited", set the small neighborhood radius ∈, the large neighborhood radius η, the density ratio threshold τ, and let the cluster number C = 0; Select an unvisited point p from the dataset, mark it as visited, calculate the number of points in the ∈-neighborhood and η-neighborhood respectively, so as to obtain the density ratio R(p); if R(p) ≥ τ, create a new cluster C and add point p to it; Expand the cluster, include all points in the ∈-neighborhood of point p into a list to be processed, take out point p' one by one and calculate the density ratio. If R(p') ≥ τ, then incorporate the ∈-neighborhood of point p into the list to be processed; at the same time, if point p' does not belong to any cluster, incorporate it into the current cluster C; if the density ratio R(p) of a certain point p < τ, mark it as a noise point; Keep repeating the above process until all points are visited, so as to obtain clusters that can adapt to different densities and some noise points.

7. A point cloud registration method based on adaptive representation constraint according to claim 1, characterized in that: Adopt an improved snake optimization algorithm as the iterative optimization method for registration parameters. The improved snake optimization algorithm uses Bernoulli chaotic mapping to generate the initial population; Divide the population into male and female groups, and control the behavior pattern switching through the food quantity and temperature as dynamic thresholds; introduce the lens imaging principle to generate reverse solutions, and dynamically adjust the reverse point positions through the scaling factor; And when the change amplitude of the fitness is less than the threshold, apply random perturbations near the optimal solution to generate temporary individuals for exploration.

8. A point cloud registration method based on adaptive feature constraint according to claim 4, characterized in that: The improved snake optimization algorithm specifically includes the following steps: Step 1: Read data and initialize parameters; the initial value of the population is randomly generated by the rand function and then updated by Bernoulli chaotic mapping. Initialize the initial position of the snake individual as: X i = X min + B i × (X max - X min ) Among them, X i represents the position of the i-th snake, where i = 1, 2, …, N; X min and X max represent the lower and upper bounds of the solution space; B i represents the i-th vector of the chaotic random number B(i, j); obtaining the initial population: X = [X1, X2, …, X N ​ The population is further divided into two groups, male and female; Step 2: Evaluate the initial population; calculate the objective function value for the position of each snake, and find the current optimal solution and the corresponding fitness; Step 3: Simulate the "food quantity" Q and "temperature" Temp according to the number of iterations and random parameters, and decide whether the snake enters the foraging mode, fighting mode or mating mode; Step 4: If Q < Threshold, it means that the food is insufficient and enter the foraging mode; if Q ≥ Threshold and Temp > Threshold2, it means that the snake will only move to the food; if Q ≥ Threshold and Temp < Threshold2, it means that the food is sufficient and the temperature is suitable, and enter the fighting mode or mating mode; Step 5: Update the population using the lens imaging reverse learning strategy. The specific process is as follows: Map the snake individuals to the x-axis to obtain the position X new , and the reverse snake individuals obtained through lens imaging are mapped to the x-axis to obtain X new *, and the calculation formula is: Let h / h* = k, where k represents the scaling factor, which is expressed as: The formula for the reverse point position in space is: Then, the reverse point position X new * is mapped back to the snake individual; Step 6: Calculate the objective function value for the updated population. If the new solution is better than the current solution, replace it and update the global optimal solution; Step 7: Detect the difference between the latest fitness and the average result within the previous K iterations. If the change amplitude is less than the predetermined threshold, start the micro-perturbation. The perturbation calculation formula is: X p = (rand - 0.5)·2·S·(X max - X min ) Among them, X p represents the disturbance increment, S represents the disturbance intensity, X min and X max represent the lower and upper bounds of the solution space; K is set to the maximum number of iterations * proportionality coefficient; generate a temporary individual based on the current optimal solution and add disturbances, while ensuring that it is still within the boundaries after the disturbance: X temp = X best + X p X temp = max(min(X temp , X max ), X min ) Calculate the objective function value. If this fitness is better than the original optimal solution, update the optimal solution; otherwise, discard this solution; the number of perturbation times is the set value; Repeat steps three to seven until the maximum number of iterations T is reached max or the convergence condition is met; The objective function value is obtained through the registration constraint model based on adaptive representation.

9. A point cloud registration system based on adaptive representation constraints, characterized in that: It includes: The registration constraint model based on adaptive representation, which analyzes the spatial position relationship between the source point set and the target point set to be registered by passing the point cloud data to be registered through the registration constraint model based on adaptive representation, determines the registration area and registration points, robustly estimates the point set with abnormal points at the registration points, and then conducts fitness evaluation; And a parameter optimization module, which is used to optimize the registration parameters, and through iterative fitness evaluation and optimization update of the registration parameters, obtains the optimal registration parameters, so as to realize the registration alignment of the point cloud data.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1-8.

Images