Point cloud registration method and device

Abstract

The invention discloses a point cloud registration method and device, and relates to the technical field of computer vision and three-dimensional reconstruction, and the method comprises the steps: obtaining an initial registration point pair of first point cloud data and second point cloud data; calculating an Euclidean distance value between the first point and the second point in each initial registration point pair; analyzing the Euclidean distance values of the plurality of initial registration point pairs, and constructing a point pair weight function; optimizing parameters of a preset point cloud registration model according to the point pair weighting function and the initial registration point pair; and performing point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result. According to the method, high-precision, high-robustness and low-calculation-cost point cloud registration can be realized in a low-overlapping-rate and high-noise environment, and the application requirements of automation, high efficiency and high robustness are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer vision and three-dimensional reconstruction, and in particular to a point cloud registration method and device. BACKGROUND

[0002] At present, the commonly used point cloud registration method is the Iterative Closest Point (ICP) algorithm, and the process thereof includes: finding the closest points to establish point pairs, and then estimating the rigid transformation to reduce the registration error. However, the ICP algorithm is highly sensitive to the initial registration accuracy, noise, outliers, and the overlap rate between point clouds. In the case of low overlap rate, the wrong point pair matching will dominate the optimal rigid transformation estimation, resulting in the ICP algorithm converging to the wrong local extremum; and in the case of high noise or a large number of outliers, the traditional ICP algorithm will include the wrong pairing into the optimization target, which seriously affects the final registration accuracy.

[0003] In order to solve the above problems, the existing improved methods include: using fixed or adaptive distance threshold to eliminate wrong point pairs (such as KISS-ICP); designing a robust weighting function to reduce the influence of abnormal point pairs (such as Robust ICP); using sparse or statistical models (such as GMM, Sparse ICP) to model and eliminate outliers, etc. However, most of these methods rely on manual parameter tuning or external loop search for hyperparameters, which increases the computational overhead and system complexity, and is difficult to adapt to the characteristics of high variability of point cloud density, overlap rate and noise in actual applications.

[0004] Therefore, there is still a need for a point cloud registration method with high accuracy, high robustness, and low computational cost in a low overlap rate and high noise environment to meet the application requirements of automation, high efficiency, and high robustness. SUMMARY

[0005] The embodiments of the present application provide a point cloud registration method to achieve high accuracy, high robustness, and low computational cost of point cloud registration in a low overlap rate and high noise environment, meet the application requirements of automation, high efficiency, and high robustness, and the method comprises: obtaining initial registration point pairs of first point cloud data and second point cloud data, each of the initial registration point pairs comprising a first point in the first point cloud data and a second point in the second point cloud data most adjacent to the first point; calculating the Euclidean distance value between the first point and the second point in each of the initial registration point pairs; analyzing the Euclidean distance values of a plurality of the initial registration point pairs to construct a point pair weight function; optimizing the parameters of a preset point cloud registration model according to the point pair weight function and the initial registration point pairs; According to the optimized point cloud registration model, the first point cloud data and the second point cloud data are subjected to point cloud registration to obtain a point cloud registration result.

[0006] Further, the analysis of the Euclidean distance values of the plurality of initial registration point pairs comprises: analyzing the Euclidean distance values of the plurality of initial registration point pairs to identify a mutation point pair in which the Euclidean distance value is mutated; and taking the Euclidean distance value corresponding to the mutation point pair as a screening threshold value; determining a confidence degree of the screening threshold value according to the screening threshold value, a total number of the initial registration point pairs, a maximum value of the Euclidean distance values of the plurality of initial registration point pairs, and a number of initial registration point pairs satisfying a preset distance condition; the preset distance condition is that the Euclidean distance value of the initial registration point pair is less than or equal to the screening threshold value; constructing a point pair weight function according to the screening threshold value and the confidence degree of the screening threshold value.

[0007] Further, the analysis of the Euclidean distance values of the plurality of initial registration point pairs to identify a mutation point pair in which the Euclidean distance value is mutated comprises: arranging the Euclidean distance values of the plurality of initial registration point pairs in ascending order to generate a distance ascending curve; identifying a mutation position in the distance ascending curve by using a preset heuristic hard threshold algorithm; the initial registration point pair corresponding to the mutation position is the mutation point pair in which the Euclidean distance value is mutated.

[0008] Further, the heuristic hard threshold algorithm is implemented by the following formula: wherein, is a point pair serial number of the mutation point pair; is an initial registration point pair in the distance ascending curve; is an initial registration point pair in the distance ascending curve; is an initial registration point pair in the distance ascending curve; is an Euclidean distance value of the initial registration point pair in the distance ascending curve; is an Euclidean distance value of the first initial registration point pair in the distance ascending curve; is an Euclidean distance value of the initial registration point pair in the distance ascending curve; is a set of point pair serial numbers in the distance ascending curve whose Euclidean distance values are less than the Euclidean distance value of the point pair serial number . .

[0009] ​Further, the method further comprises: The confidence of the screening threshold is determined according to the following formula: Wherein, M is the total number of initial registration point pairs; The number of initial registration point pairs satisfying the preset distance condition The number of initial registration point pairs satisfying the preset distance condition The screening threshold; The Euclidean distance value of the i th initial registration point pair; The maximum value of the Euclidean distance values of the plurality of initial registration point pairs; The confidence of the screening threshold.

[0010] Further, the method further comprises: The point pair weight function is constructed according to the following formula: Wherein, The Euclidean distance value of the i th initial registration point pair; The screening threshold; The confidence of the screening threshold; The weight of the i th initial registration point pair.

[0011] Further, the method further comprises: Setting the maximum number of iterations and the convergence threshold; After the point cloud registration result is obtained by performing the point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model, the method further comprises: Judging whether the number of iterations of the current iteration reaches the maximum number of iterations, or calculating the parameter error between the point cloud registration model optimized in the current iteration and the point cloud registration model optimized in the last iteration, and in the case that the parameter error is less than or equal to the convergence threshold, stopping the iteration and taking the point cloud registration result obtained in the current iteration as the final point cloud registration result.

[0012] The embodiment of the application also provides a point cloud registration device for realizing high-precision, high-robustness and low-computing-cost point cloud registration in a low-overlap-rate and high-noise environment, and meeting the application requirements of automation, high efficiency and high robustness. The device comprises: The point pair acquisition module is configured to acquire initial registration point pairs of the first point cloud data and the second point cloud data, each of the initial registration point pairs including a first point in the first point cloud data and a second point in the second point cloud data that is most adjacent to the first point; The calculation module is configured to calculate an Euclidean distance value between the first point and the second point in each of the initial registration point pairs; The weight construction module is configured to analyze the Euclidean distance values of the plurality of initial registration point pairs to construct a point pair weight function; The model optimization module is configured to optimize parameters of a preset point cloud registration model according to the point pair weight function and the initial registration point pairs; The registration module is configured to perform point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result.

[0013] The embodiment of the present application also provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the point cloud registration method when executing the computer program.

[0014] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the point cloud registration method.

[0015] The embodiment of the present application also provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the point cloud registration method.

[0016] In the embodiment of the present application, the initial registration point pairs of the first point cloud data and the second point cloud data are acquired, each of the initial registration point pairs including a first point in the first point cloud data and a second point in the second point cloud data that is most adjacent to the first point; the Euclidean distance value between the first point and the second point in each of the initial registration point pairs is calculated; the Euclidean distance values of the plurality of initial registration point pairs are analyzed to construct a point pair weight function; the parameters of a preset point cloud registration model are optimized according to the point pair weight function and the initial registration point pairs; and the first point cloud data and the second point cloud data are subjected to point cloud registration according to the optimized point cloud registration model to obtain a point cloud registration result. Compared with the existing point cloud registration method, the point pair weight function is constructed by analyzing the Euclidean distance values of the plurality of initial registration point pairs, so that manual parameter adjustment or external loop search of hyperparameters is not required, the weight of each of the initial registration point pairs can be adaptively adjusted based on data, and high-precision, high-robustness and low-computing-cost point cloud registration can be realized in a low-overlap and high-noise environment, thereby meeting the application requirements of automation, high efficiency and high robustness. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0018] In the drawings: Figure 1 A flow chart of a point cloud registration method provided in an embodiment of the present application; Figure 2 A flow chart of a method for analyzing the Euclidean distance values of a plurality of initial registration point pairs and constructing a point pair weight function provided in an embodiment of the present application; Figure 3 A schematic diagram of a distance ascending order curve provided in an embodiment of the present application; Figure 4 A schematic diagram of a mutation point pair identified by a three-part method-based mutation point geometric searching method provided in an embodiment of the present application; Figure 5 A schematic diagram of a distance ascending order curve and a point pair weight function obtained through multiple iterations provided in an embodiment of the present application; Figure 6 A comparison diagram of a point cloud registration result obtained by using the point cloud registration method provided in an embodiment of the present application and a point cloud registration result obtained by using a traditional ICP method provided in an embodiment of the present application; Figure 7 A schematic diagram of a point cloud registration device provided in an embodiment of the present application; Figure 8 A schematic diagram of a weight construction module provided in an embodiment of the present application; Figure 9 A schematic diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will further describe the embodiments of the present application in combination with the drawings. Herein, the schematic embodiments of the present application and the descriptions thereof are used to explain the present application, but not as a limitation of the present application.

[0020] In the description of the present specification, "include", "includes", "have", "has", and the like are open terms, that is, mean to include but not limited to. The description referring to the terms "one embodiment", "one specific embodiment", "some embodiments", "for example", and the like means that the specific features, structures, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. The order of the steps involved in each embodiment is used to illustrate the implementation of the present application, and the order of the steps is not limited, and can be appropriately adjusted as needed.

[0021] In order to solve the problem that the existing point cloud registration method cannot realize high-precision, high-robustness, and low-computing-cost point cloud registration in a low-overlap rate and high-noise environment, an embodiment of the present application provides a point cloud registration method, which can realize high-precision, high-robustness, and low-computing-cost three-dimensional point cloud registration in an extreme environment such as a low-overlap rate and high-noise interference.

[0022] As shown in Figure 1 , a flowchart of a point cloud registration method provided by an embodiment of the present application, the method can include: Step 101, obtaining initial registration point pairs of first point cloud data and second point cloud data, each initial registration point pair including a first point in the first point cloud data and a second point in the second point cloud data closest to the first point; Step 102, calculating the Euclidean distance value between the first point and the second point in each initial registration point pair; Step 103, analyzing the Euclidean distance values of the plurality of initial registration point pairs to construct a point pair weight function; Step 104, optimizing the parameters of a preset point cloud registration model according to the point pair weight function and the initial registration point pairs; Step 105, performing point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result.

[0023] In the embodiment of the present application, by acquiring initial registration point pairs of the first point cloud data and the second point cloud data, each initial registration point pair includes a first point in the first point cloud data and a second point in the second point cloud data which is most adjacent to the first point; the Euclidean distance value between the first point and the second point in each initial registration point pair is calculated; the Euclidean distance values of the plurality of initial registration point pairs are analyzed to construct a point pair weight function; the parameters of a preset point cloud registration model are optimized according to the point pair weight function and the initial registration point pairs; and the point cloud registration is performed on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result. Compared with the existing point cloud registration method, the embodiment of the present application analyzes the Euclidean distance values of the plurality of initial registration point pairs to construct a point pair weight function, so that manual parameter adjustment or external loop search of hyperparameters is not required, the weight of each initial registration point pair can be adaptively adjusted based on data, and high-precision, high-robustness and low-computing-cost point cloud registration can be realized in a low-overlap and high-noise environment, thereby meeting the application requirements of automation, high efficiency and high robustness.

[0024] The point cloud registration method shown in FIG. 1 will be described in detail below. Figure 1 The point cloud registration method shown in FIG. 1 will be described in detail below.

[0025] In step 101, the initial registration point pairs of the first point cloud data P and the second point cloud data Q are acquired.

[0026] In specific implementation, if there is an initial transformation such as known rigid transformation parameters, i.e., an initial rotation matrix R0 and a translation vector t0, between the first point cloud data P and the second point cloud data Q to be registered, for each first point , the most adjacent second point to the first point in the second point cloud data Q can be found according to the following formula (1) to obtain an initial registration point pair { }: Formula (1) Wherein, is a set of ; i=1, 2, …, M, and M is the total number of initial registration point pairs.

[0027] If there is no initial transformation between the first point cloud data P and the second point cloud data Q to be registered, the initial registration point pairs can be obtained by performing coarse registration on the first point cloud data P and the second point cloud data Q; wherein the coarse registration can be realized by using existing algorithms such as a random sample consensus (RANSAC) algorithm, a TEASER++ method, etc.

[0028] In step 102 above, based on the multiple initial registration point pairs obtained in step 101, each initial registration point pair can be calculated using the following formula (2): The first point in} Second point Euclidean distance between : Formula (2) in, yes A set that consists of.

[0029] In step 103 above, the Euclidean distance values ​​of multiple initial registration point pairs are analyzed to determine the screening threshold and the confidence level of the screening threshold. The screening threshold can refer to the threshold for distinguishing between correct and incorrect initial registration point pairs.

[0030] In one embodiment, such as Figure 2 As shown, step 103 above may specifically include: Step 201: Analyze the Euclidean distance values ​​of multiple initial registration point pairs, identify abrupt change points where the Euclidean distance values ​​change abruptly, and use the Euclidean distance values ​​corresponding to the abrupt change points as the screening threshold. Step 202: Determine the confidence level of the screening threshold based on the screening threshold, the total number of initial registration point pairs, the maximum value of the Euclidean distance of multiple initial registration point pairs, and the number of initial registration point pairs that meet the preset distance condition; the preset distance condition is that the Euclidean distance of the initial registration point pairs is less than or equal to the screening threshold. Step 203: Construct a point-to-point weight function based on the screening threshold and the confidence level of the screening threshold.

[0031] In specific implementation, in step 201, a mutation point pair refers to the boundary position between the correct initial registration point pair and the incorrect initial registration point pair among multiple initial registration point pairs. This can be achieved by sequentially arranging the Euclidean distance values ​​of multiple initial registration point pairs, analyzing the mutation point pairs based on the sequentially arranged Euclidean distance values, and using the Euclidean distance value corresponding to the mutation point pair as a screening threshold.

[0032] In one embodiment, step 201 may specifically include: first arranging the Euclidean distance values ​​of multiple initial registration point pairs in ascending order, i.e. , This represents the point pair index obtained by sorting the Euclidean distance values ​​in ascending order; a Cumulative Distance Ascending Curve (CDAC) is formed based on the ascending Euclidean distance values, as shown in the figure. Figure 3As shown, the horizontal axis represents the index of the point pairs in ascending order, and the vertical axis represents the Euclidean distance between the point pairs. Then, the ascending distance curve can be analyzed to identify abrupt change points. Specifically, a pre-defined heuristic hard thresholding algorithm (Robust Least Squares via Heuristic Hard Thresholding, RLHH) can be used to identify abrupt change points in the ascending distance curve. The initial registration point pairs corresponding to these abrupt change points are the abrupt change point pairs where the Euclidean distance value changes abruptly.

[0033] Furthermore, the heuristic hard thresholding algorithm can be implemented using the following formula (3): in, The point pair number is the mutation point pair number; The first in the distance ascending curve One initial registration point pair; The first in the distance ascending curve The Euclidean distance between the initial registration point pairs; This is the Euclidean distance value of the first initial registration point pair in the distance ascending curve, i.e., the minimum Euclidean distance value; The first in the distance ascending curve The Euclidean distance value of the initial registration point pair, i.e. the maximum Euclidean distance value; For points on the ascending distance curve, the Euclidean distance value is less than the point-to-index value. The set of point pairs with Euclidean distance values. .

[0034] in, express The floor function, for example, when M is 5. If it is 2.5, then The value is 3.

[0035] Identify mutation point pairs using the above formula (3) , to pair the mutation points Corresponding Euclidean distance value As a filtering threshold .

[0036] In this way, by arranging the Euclidean distance values of the plurality of initial registration point pairs in ascending order, a distance ascending curve is generated, which can reflect the distribution law of the matching error of the point pairs in the initial registration state, and is conducive to further identifying the mutant point pairs; then, the mutant point pairs are identified based on the RLHH algorithm. Since the RLHH algorithm does not require hyperparameters, has stable local minimum error judgment ability, and is suitable for different scales of data noise and coverage rate changes, etc., the accuracy of the identification of the mutant point pairs can be improved.

[0037] In one embodiment, after the distance ascending curve is generated in step 201, the mutant point pairs can also be identified based on the three-point geometric searching method. For example, as shown in FIG. 2B, first, two end points in the distance ascending curve CDAC are connected to form a straight line, then the distances of other points of CDAC to the straight line are calculated, and the point with the maximum distance is identified as the mutant point pair. Figure 4

[0038] In step 202, the confidence of the screening threshold is determined according to the screening threshold, and the total number of initial registration point pairs, the maximum value of the Euclidean distance values of the plurality of initial registration point pairs, and the number of initial registration point pairs satisfying the preset distance condition. The preset distance condition is that the Euclidean distance value of the initial registration point pair is less than or equal to the screening threshold.

[0039] In specific implementation, the screening threshold is determined according to the number of Euclidean distance values in or the M initial registration point pairs known in step 101. The total number of initial registration point pairs can be determined as M according to . The maximum value of the Euclidean distance values of the initial registration point pairs = can be determined according to . Assuming that the distance condition is , the number of initial registration point pairs satisfying , is the Euclidean distance value of the i-th initial registration point pair in the plurality of initial registration point pairs.

[0040] In one embodiment, after the above parameter values are determined, the confidence of the screening threshold can be determined according to the following formula (4): The formula (4) In this way, according to the formula (4), the confidence of the screening threshold can be calculated. The confidence is used to measure the credibility of the screening threshold , and the value is closer to 1, indicating that the screening threshold is more reliable. ​​​The correct initial registration point pair and the false initial registration point pair can be separated more accurately.

[0041] In step 203, the point pair weight function can be constructed according to the screening threshold , the confidence of the screening threshold .

[0042] In an embodiment, the point pair weight function can be constructed according to formula (5) as follows: Formula (5) Wherein, is the Euclidean distance value of the i th initial registration point pair; is the screening threshold; is the confidence of the screening threshold; is the weight of the i th initial registration point pair.

[0043] In this way, the point pair weight function can assign different weights to each initial registration point pair, the point pair weight function has a gradual nature, and a continuous transition from "soft weighting" to "hard rejection" can be realized, and then the adaptive regulation of the weight of the initial registration point pair can be realized without manually setting the hyperparameter, and the robustness of registration can be significantly improved.

[0044] In step 104, the parameters of the preset point cloud registration model can be optimized by using the initial registration point pair and the constructed point pair weight function, that is, the rigid transformation parameters of the point cloud registration model are re-estimated on the plurality of initial registration point pairs, and the rigid transformation parameters include the rotation matrix R and the translation vector t.

[0045] In implementation, the preset point cloud registration model can be a point-to-point registration model, a point-to-plane registration model, or a symmetric metric registration model.

[0046] Taking the point-to-point registration model as an example, the rigid transformation parameters R and t can be obtained by solving the optimization formula (6) : Formula (6) Wherein, is the point pair weight function.

[0047] Taking the point-to-plane registration model as an example, the rigid transformation parameters R and t can be obtained by solving the optimization formula (7) : Formula (7) Wherein, is the point pair weight function; is The normal vector.

[0048] Taking the symmetric registration model as an example, the rigid transformation parameters can be obtained by solving the optimization formula (8). : (Formula 8) in, For point-to-point weighting functions; yes The normal vector; yes The normal vector.

[0049] In step 105 above, point cloud registration is performed on the first point cloud data and the second point cloud data according to the optimized point cloud registration model.

[0050] In practice, the optimized point cloud registration model can be used to register the first point cloud data and the second point cloud data to obtain the point cloud registration result.

[0051] In one embodiment of the present invention, in order to further improve the accuracy of point cloud registration, a maximum number of iterations is also set. and convergence threshold In each iteration, repeat steps 101-105 above. Each iteration, following step 105 above, may also include: After obtaining the point cloud registration result in each iteration, it is determined whether the number of iterations in this iteration has reached the maximum number of iterations, or the parameter error between the point cloud registration model optimized in this iteration and the point cloud registration model optimized in the previous iteration is calculated. If the parameter error is less than or equal to the convergence threshold, the iteration is stopped, and the point cloud registration result obtained in this iteration is taken as the final point cloud registration result.

[0052] In practice, at each iteration, the iteration can be terminated by checking whether the maximum number of iterations has been reached. Alternatively, the parameter error between the optimized point cloud registration model in the current iteration and the optimized point cloud registration model in the previous iteration can be calculated. If the parameter error is less than or equal to the convergence threshold, the iteration is terminated, and the point cloud registration result obtained in this iteration is taken as the final point cloud registration result. If the iteration cannot be terminated, the point cloud registration result in this iteration can be taken as the initial registration point pair for the next iteration.

[0053] Taking the point cloud registration model as a rigid body transformation as an example, the parameter error between the point cloud registration model optimized in this iteration and the point cloud registration model optimized in the previous iteration can be calculated by the following formula (9): Equation (9) wherein the convergence threshold in the embodiment of the present application is can be set as ; using T to represent the rigid transformation matrix, that is, is the rigid transformation matrix of the optimized point cloud registration model at the k+1th iteration; is the rigid transformation matrix of the optimized point cloud registration model at the kth iteration; represents the element of the i-th row and the j-th column of the rigid transformation matrix at the kth iteration; represents the element of the i-th row and the j-th column of the rigid transformation matrix at the k+1th iteration.

[0054] For example, Figure 5 is a schematic diagram of the distance ascending curve and the point pair weight function obtained through multiple iterations in the embodiment of the present application, Figure 6 is a comparison diagram of the point cloud registration result obtained by the point cloud registration method provided by the embodiment of the present application and the point cloud registration result obtained by the traditional ICP method, Figure 6 The initial registration in the above formula is the initial registration schematic diagram corresponding to the initial registration point pair in step 101. Compared with the traditional ICP method, the point cloud registration method provided by the embodiment of the present application can still accurately and quickly register under low overlap and strong noise conditions (such as under the extreme test condition of only 10% overlap rate or 100% outlier interference), has better convergence speed and registration accuracy than the existing mainstream robust ICP method, and is suitable for scenes such as SLAM mapping, multi-view model splicing, laser radar fusion, etc. which require high accuracy and efficiency.

[0055] In summary, the point cloud registration method provided by the embodiment of the present application has the following beneficial effects: 1. No super parameter design: The construction of the weight function in the embodiment of the present application is completely based on data adaptive estimation, without manual parameter tuning or external loop search of super parameters, which can improve the simplicity and robustness of actual deployment.

[0056] 2. Strong robustness and wide adaptability: The point cloud registration method provided by the embodiment of the present application can identify mutant point pairs through the CDAC curve and adaptively regulate the weight based on the confidence, can stably converge under low overlap and high noise conditions, can automatically identify and suppress the interference of error point pairs in the iteration process, and can realize high-precision and high-robustness registration of three-dimensional point clouds.

[0057] 3. High efficiency and high precision: The point cloud registration method provided by the embodiment of the present application can complete all steps in the traditional ICP standard inner loop structure, avoiding external loop tuning and complex statistical modeling. 4. Good engineering compatibility: The point cloud registration method provided in this embodiment of the invention can be seamlessly integrated with mainstream point cloud processing modules, feature extractors and pre-registration schemes, and is suitable for industrial-grade scenarios such as SLAM mapping, point cloud stitching and LiDAR calibration.

[0058] This invention also provides a point cloud registration device, as described in the following embodiments. Since the principle by which this device solves the problem is similar to the point cloud registration method described above, the implementation of this method can be found in the implementation of the point cloud registration method, and repeated details will not be elaborated further.

[0059] like Figure 7 The diagram shown is a schematic of a point cloud registration device provided in an embodiment of the present invention. The device may include: The point-to-point acquisition module 701 is used to acquire the initial registration point pair of the first point cloud data and the second point cloud data. Each initial registration point pair includes a first point in the first point cloud data and a second point in the second point cloud data that is closest to the first point. Calculation module 702 is used to calculate the Euclidean distance between the first point and the second point in each of the initial registration point pairs; The weight construction module 703 is used to analyze the Euclidean distance values ​​of multiple initial registration point pairs and construct a point pair weight function; The model optimization module 704 is used to optimize the parameters of the preset point cloud registration model based on the point pair weight function and the initial registration point pair. The registration module 705 is used to perform point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model, so as to obtain the point cloud registration result.

[0060] In one embodiment, such as Figure 8 As shown, the weight construction module 703 may specifically include: The threshold identification unit 7031 is used to analyze the Euclidean distance values ​​of multiple initial registration point pairs, identify abrupt change point pairs where the Euclidean distance values ​​change abruptly, and use the Euclidean distance value corresponding to the abrupt change point pair as a screening threshold. The confidence determination unit 7032 is used to determine the confidence of the filtering threshold based on the filtering threshold, the total number of initial registration point pairs, the maximum value of the Euclidean distance of multiple initial registration point pairs, and the number of initial registration point pairs that meet the preset distance condition; the preset distance condition is that the Euclidean distance of the initial registration point pairs is less than or equal to the filtering threshold. The weight construction unit 7033 is used to construct a point-to-point weight function based on the filtering threshold and the confidence level of the filtering threshold.

[0061] In one embodiment, the threshold recognition unit 7031 can be specifically used for: Arrange the Euclidean distance values ​​of multiple initial registration point pairs in ascending order to generate a distance ascending curve; Using a preset heuristic hard threshold algorithm, the abrupt change locations in the distance ascending curve are identified, and the initial registration point pairs corresponding to the abrupt change locations are the abrupt change point pairs where the Euclidean distance value changes abruptly.

[0062] In one embodiment, the heuristic hard thresholding algorithm is implemented using the following formula: in, The point pair number is the mutation point pair number; The first in the distance ascending curve One initial registration point pair; The first in the distance ascending curve The Euclidean distance between the initial registration point pairs; This is the Euclidean distance value for the first initial registration point pair in the distance ascending curve; The first in the distance ascending curve The Euclidean distance between the initial registration point pairs; For points on the ascending distance curve, the Euclidean distance value is less than the point-to-index value. The set of point pairs whose Euclidean distance values ​​are . .

[0063] In one embodiment, the confidence determination unit 7032 can be specifically used for: The confidence level of the screening threshold is determined according to the following formula: Where M is the total number of initial registration point pairs; To meet the preset distance conditions The number of initial registration point pairs; The filtering threshold; Let be the Euclidean distance value of the i-th initial registration point pair; The maximum value of the Euclidean distance between multiple initial registration point pairs; The confidence level for the screening threshold.

[0064] In one embodiment, the weight construction unit 7033 can be specifically used for: The point-pair weight function is constructed according to the following formula: in, Let be the Euclidean distance value of the i-th initial registration point pair; a screening threshold value; a confidence level of the screening threshold value; a weight of the ith initial registration point pair.

[0065] In an embodiment, the method further comprises an iteration parameter setting module configured to set a maximum iteration number and a convergence threshold value. The judging module is configured to, after the registration module 705 performs point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result: determine whether the iteration number of the current iteration reaches the maximum iteration number, or calculate a parameter error between the point cloud registration model optimized in the current iteration and the point cloud registration model optimized in the last iteration, and in a case where the parameter error is less than or equal to the convergence threshold value, stop the iteration and take the point cloud registration result obtained in the current iteration as a final point cloud registration result.

[0066] The embodiment of the present application further provides a computer device, Figure 9 FIG. 9 is a schematic diagram of a computer device according to the embodiment of the present application, and the computer device 900 comprises a memory 910, a processor 920, and a computer program 930 stored in the memory 910 and capable of running on the processor 920, wherein the processor 920 implements the point cloud registration method described above when executing the computer program 930.

[0067] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the point cloud registration method described above.

[0068] The embodiment of the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the point cloud registration method described above.

[0069] In the embodiment of the present application, by acquiring initial registration point pairs of the first point cloud data and the second point cloud data, each initial registration point pair includes a first point in the first point cloud data and a second point in the second point cloud data which is most adjacent to the first point; the Euclidean distance value between the first point and the second point in each initial registration point pair is calculated; the Euclidean distance values of the plurality of initial registration point pairs are analyzed to construct a point pair weight function; the parameters of a preset point cloud registration model are optimized according to the point pair weight function and the initial registration point pairs; and the point cloud registration is performed on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result. Compared with the existing point cloud registration method, the embodiment of the present application analyzes the Euclidean distance values of the plurality of initial registration point pairs to construct the point pair weight function, so that the manual parameter adjustment or the external loop search of the hyperparameters is not required, the weight of each initial registration point pair can be adaptively adjusted based on the data, the point cloud registration with high precision, high robustness and low calculation cost can be realized in a low overlap rate and high noise environment, and the application requirements of automation, high efficiency and high robustness are met.

[0070] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0071] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device implemented in the flowcharts and / or block diagrams. Figure 1 The function specified in one or more flows and / or blocks. Figure 1 The means for performing the function specified in one or more flows and / or blocks.

[0072] These computer program instructions can also be stored in a computer-readable memory capable of guiding the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means, which implements the flowcharts and / or block diagrams. Figure 1 The function specified in one or more flows and / or blocks. Figure 1 The means for performing the function specified in one or more flows and / or blocks.

[0073] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable data processing devices provide processes for implementing the flows Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0074] The above-described specific embodiments, the purpose, technical solutions and beneficial effects of the present application are further described in detail, it should be understood that the above-described is only a specific embodiment of the present application, and is not used to limit the protection scope of the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A point cloud registration method, characterized in that, The method comprises the following steps: obtaining initial registration point pairs of first point cloud data and second point cloud data, each of the initial registration point pairs comprising a first point in the first point cloud data and a second point in the second point cloud data closest to the first point; calculating an Euclidean distance value between the first point and the second point in each of the initial registration point pairs; analyzing the Euclidean distance values of the initial registration point pairs to construct a point pair weight function; optimizing parameters of a preset point cloud registration model according to the point pair weight function and the initial registration point pairs; performing point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result.

2. The method of claim 1, wherein, The method further comprises the following steps: analyzing the Euclidean distance values of the initial registration point pairs to identify mutation point pairs in which the Euclidean distance values mutate; taking the Euclidean distance value corresponding to the mutation point pairs as a screening threshold value; determining a confidence degree of the screening threshold value according to the screening threshold value, the total number of the initial registration point pairs, the maximum value of the Euclidean distance values of the initial registration point pairs, and the number of the initial registration point pairs satisfying a preset distance condition, wherein the preset distance condition is that the Euclidean distance value of the initial registration point pairs is less than or equal to the screening threshold value; 3. The method of claim 2, wherein, constructing the point pair weight function according to the screening threshold value and the confidence degree of the screening threshold value. The method further comprises the following steps: arranging the Euclidean distance values of the initial registration point pairs in ascending order to generate a distance ascending curve; 4. The method of claim 3, wherein, identifying a mutation position in the distance ascending curve by using a preset heuristic hard threshold algorithm, wherein the initial registration point pairs corresponding to the mutation position are the mutation point pairs in which the Euclidean distance values mutate. wherein, is a point pair index of the mutation point pair; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a point pair index of the 1st initial registration point pair in the distance ascending curve; is a set of point pair indexes whose Euclidean distance values are less than the Euclidean distance value of the point pair index .​ 5. The method of claim 2, wherein, The heuristic hard threshold algorithm is implemented by the following formula: The method further comprises the following steps: wherein M is the total number of initial registration point pairs; to meet the preset distance condition the number of initial registration point pairs; is a screening threshold value; is the Euclidean distance value of the i-th initial registration point pair; is the maximum value of the Euclidean distance values of the plurality of initial registration point pairs; is the confidence of the screening threshold value.

6. The method of claim 2, wherein, determining the confidence degree of the screening threshold value according to the following formula: The method further comprises the following steps: wherein, is a Euclidean distance value for the i-th initial registration point pair; is a screening threshold value; is a confidence of the screening threshold value; is a weight of the i-th initial registration point pair.

7. The method of claim 1, wherein, constructing the point pair weight function according to the following formula: The method further comprises the following steps: setting a maximum iteration number and a convergence threshold value; The method further comprises the following steps:

8. A point cloud registration apparatus, characterized by, after performing the point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain the point cloud registration result, determining whether the iteration number of the current iteration reaches the maximum iteration number, or calculating a parameter error between the point cloud registration model optimized in the current iteration and the point cloud registration model optimized in the last iteration, and in the case that the parameter error is less than or equal to the convergence threshold value, stopping the iteration and taking the point cloud registration result obtained in the current iteration as the final point cloud registration result. The method comprises the following steps: a point pair acquisition module, configured to acquire initial registration point pairs of the first point cloud data and the second point cloud data, each of the initial registration point pairs comprising a first point in the first point cloud data and a second point in the second point cloud data most adjacent to the first point; a calculation module, configured to calculate an Euclidean distance value between the first point and the second point in each of the initial registration point pairs; a weight construction module, configured to analyze the Euclidean distance values of a plurality of the initial registration point pairs to construct a point pair weight function; a model optimization module, configured to optimize parameters of a preset point cloud registration model according to the point pair weight function and the initial registration point pairs; a registration module, configured to perform point cloud registration on the first point cloud data and the second point cloud data according to the optimized point cloud registration model to obtain a point cloud registration result.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the point cloud registration method in any of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores the computer program, and the computer program is executed by the processor to implement the point cloud registration method in any of claims 1 to 7.

Images