A method and device for automatically tuning parameters of point cloud denoising algorithm

Abstract

The present application discloses a method and device for automatically tuning the parameters of a point cloud denoising algorithm, the method comprising: substituting each parameter in a pre-configured parameter group into an existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm, then denoising the denoised point cloud to obtain a denoising result, evaluating the smoothness and local density of the denoising result to obtain an evaluation result, using the evaluation result to adjust each parameter of the denoising algorithm to obtain an adjusted denoising algorithm, and returning to execute the step of denoising the denoised point cloud to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the parameters under the preset number of adjustments are combined into a target parameter group. It can be seen that the denoising effect is evaluated, the denoising evaluation result is obtained as the feedback result of the optimized denoising, the parameters of the point cloud denoising algorithm are updated, and finally the optimized parameters are obtained, which greatly saves the time of manual parameter adjustment.

Description

Technical Field

[0001] The present application relates to the field of three-dimensional point cloud denoising, and more specifically, to a method and device for automatically tuning point cloud denoising algorithm parameters. Background Art

[0002] With the continuous development of LiDAR detection technology, the depth information of the measured target can be quickly obtained and the point cloud data of the measured target can be generated. Noise is inevitable due to the influence of measurement equipment, measurement environment and surface characteristics of the measured object. The existing point cloud denoising algorithm based on model characteristics does not rely on massive training data compared to the general denoising method based on deep learning, and does not require a lot of training time to establish a training model.

[0003] However, if denoising is performed according to the default parameters of each point cloud denoising algorithm, the algorithm performance will drop significantly when the measurement environment and the object being measured change. At this time, manual adjustment of the parameters is required to obtain the ideal denoising effect, which is inefficient.

[0004] How to achieve automatic tuning of point cloud denoising algorithm parameters is an issue that needs attention. Summary of the invention

[0005] In view of the above problems, the present application is proposed to provide a method and device for automatically tuning the parameters of a point cloud denoising algorithm to save time for manual parameter adjustment.

[0006] In order to achieve the above objectives, the specific plan is proposed as follows:

[0007] A method for automatically tuning parameters of a point cloud denoising algorithm, comprising:

[0008] Substituting each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm;

[0009] De-noising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result;

[0010] Evaluating the smoothness and local density of the denoising result to obtain an evaluation result;

[0011] The evaluation result is used to adjust the various parameters of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

[0012] Optionally, the smoothness and local density of the denoising result are evaluated to obtain an evaluation result, including:

[0013] Performing local density denoising evaluation on the denoising result to obtain local density cost value information;

[0014] The denoising result is evaluated for smoothing denoising to obtain smoothness cost value information, and the local density cost value information and the smoothness cost value information are used as evaluation results.

[0015] Optionally, the denoising result includes a plurality of points;

[0016] Performing local density denoising evaluation on the denoising result to obtain local density cost value information includes:

[0017] For each point in the denoised result:

[0018] Determine a first preset number of neighboring points that are close to the point, and a reachable distance between the point and each of the neighboring points;

[0019] According to the reachable distance between the point and each neighboring point, local density cost value information of the denoising result is determined.

[0020] Optionally, each point in the denoising result has coordinate information;

[0021] The determining of a first preset number of neighboring points that are close to the point includes:

[0022] The Euclidean distance between the point and other points in the denoising result is calculated according to the coordinate information of the point and other points in the denoising result, and a first preset number of points in the calculated result having a shorter Euclidean distance to the point are taken as neighboring points.

[0023] Optionally, determining the reachable distance between the point and each neighboring point includes:

[0024] For each point in the denoising result and each neighboring point of the point:

[0025] Calculate the Euclidean distance between the neighboring point and the other points according to the coordinate information of the neighboring point and the other points, and determine the Euclidean distances whose ascending sorting numbers in the calculation results are the first preset number as the maximum neighboring distance of the neighboring points;

[0026] The Euclidean distance between the point and the neighboring point is compared with the distance between the point and the maximum neighboring distance of the neighboring point, and the longer distance is used as the reachable distance from the point to the neighboring point.

[0027] Optionally, determining local density cost value information of the denoising result according to the reachable distance between the point and each neighboring point includes:

[0028] Sum the reachable distances between the point and each of its neighboring points to obtain a total reachable distance, and divide the total number of points in the denoising result by the total reachable distance to obtain a local reachable density;

[0029] For each point in the denoising result, calculate the sum of the local reachable densities of the neighboring points of the point, and divide the sum by the total number of points in the denoising result to obtain an average local reachable density;

[0030] Determine a local density error between a local reachable density of each point in the denoising result and an average local reachable density;

[0031] The average value of the local density errors of each point in the denoising result is calculated as the local density cost value information of the denoising result.

[0032] Optionally, the denoising result includes a plurality of points, each point having coordinate information;

[0033] The denoising result is subjected to smoothing denoising evaluation to obtain smoothness cost value information, including:

[0034] For each point of the denoising result:

[0035] Determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point;

[0036] According to the coordinate matrix, obtaining the mean value of the coordinates of each neighboring point;

[0037] Calculating a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point;

[0038] Calculate the minimum eigenvalue of each neighboring point of the point in the covariance matrix, determine the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculate the information entropy of the point according to the projection length;

[0039] According to the information entropy of each point of the denoising result, smoothness cost value information of the denoising result is determined.

[0040] Optionally, adjusting various parameters of the intermediate point cloud denoising algorithm using the evaluation result to obtain an adjusted intermediate point cloud denoising algorithm includes:

[0041] The evaluation results are calculated using the following calculation method to obtain the adjusted parameters:

[0042] (a t+1 ,b t+1 ,c t+1 ,dt+1 ,……)=MOEA((a t ,b t ,c t ,d t ,……),(J1,J2))

[0043] Wherein, a, b, c, d represent different parameter values ​​of the parameter group, MOEA represents a multi-objective optimization algorithm, t represents the number of times adjusted, J1 represents the local density cost value information, and J2 represents the smoothness cost value information;

[0044] The adjusted parameters are used to replace the corresponding parameters in the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm.

[0045] Optionally, the configuration process of the parameter group includes:

[0046] Determine multiple types of parameters, and for each type of parameter, randomly generate a parameter value within a parameter value range of the parameter;

[0047] The parameter values ​​randomly generated for each type of parameter constitute a parameter group.

[0048] An automatic tuning device for point cloud denoising algorithm parameters, comprising:

[0049] A parameter substitution unit, used for substituting each parameter in a pre-configured parameter group into an existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm;

[0050] A point cloud denoising unit, used for denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result;

[0051] A denoising evaluation unit, used to evaluate the smoothness and local density of the denoising result to obtain an evaluation result;

[0052] A parameter adjustment unit is used to adjust the various parameters of the intermediate point cloud denoising algorithm using the evaluation result to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

[0053] Optionally, the denoising evaluation unit includes:

[0054] A local denoising evaluation unit, used to perform local density denoising evaluation on the denoising result to obtain local density cost value information;

[0055] The smoothing denoising evaluation unit is used to perform smoothing denoising evaluation on the denoising result to obtain smoothness cost value information, and use the local density cost value information and the smoothness cost value information as the evaluation result.

[0056] Optionally, the denoising result includes a plurality of points;

[0057] The local denoising evaluation unit comprises:

[0058] A neighboring point determination unit, configured to determine, for each point in the denoising result, a first preset number of neighboring points that are close to the point;

[0059] A reachable distance determination unit, used to determine, for each point in the denoising result, a reachable distance between the point and each neighboring point;

[0060] The local density cost value determining unit is used to determine the local density cost value information of the denoising result for each point in the denoising result according to the reachable distance between the point and each neighboring point.

[0061] Optionally, each point in the denoising result has coordinate information;

[0062] The neighboring point determination unit includes:

[0063] The first neighboring point determination subunit is used to calculate the Euclidean distance between the point and other points in the denoising result according to the coordinate information of the point and other points in the denoising result, and to use a first preset number of points in the calculation result with shorter Euclidean distances to the point as neighboring points.

[0064] Optionally, the reachable distance determining unit includes:

[0065] A first reachable distance determination subunit is configured to calculate, for each point in the denoising result and each neighboring point of the point, the Euclidean distance between the neighboring point and the other points according to the coordinate information of the neighboring point and the other points, and determine the Euclidean distances whose ascending sorting numbers in the calculation result are the first preset number as the maximum neighboring distances of the neighboring points;

[0066] The second reachable distance determination subunit is used to compare the Euclidean distance between the point and each neighboring point of the point with the length of the distance between the maximum neighboring distance of the neighboring points, and take the longer distance as the reachable distance from the point to the neighboring point.

[0067] Optionally, the local density cost value determining unit includes:

[0068] A first local density cost value determination subunit is used to sum the reachable distances between the point and each neighboring point of the point to obtain a total reachable distance, and divide the total number of points in the denoising result by the total reachable distance to obtain a local reachable density;

[0069] A second local density cost value determination subunit is used to calculate, for each point in the denoising result, the sum of the local reachable densities of the neighboring points of the point, and divide the sum by the total number of points in the denoising result to obtain an average local reachable density;

[0070] A third local density cost value determination subunit is used to determine a local density error between the local reachable density of each point in the denoising result and the average local reachable density;

[0071] The fourth local density cost value determining subunit is used to calculate the average value of the local density errors of each point in the denoising result as the local density cost value information of the denoising result.

[0072] Optionally, the denoising result includes a plurality of points, each point having coordinate information;

[0073] For each point of the denoising result, the smoothing denoising evaluation unit includes:

[0074] A first smoothing denoising evaluation subunit is used to determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point;

[0075] A second smoothing denoising evaluation subunit is used to obtain the mean value of the coordinates of each neighboring point according to the coordinate matrix;

[0076] A third smoothing denoising evaluation subunit is used to calculate a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point;

[0077] a fourth smoothing denoising evaluation subunit, configured to calculate the minimum eigenvalue of each neighboring point of the point in the covariance matrix, determine the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculate the information entropy of the point according to the projection length;

[0078] The fifth smoothing denoising evaluation subunit is used to determine the smoothness cost value information of the denoising result according to the information entropy of each point of the denoising result.

[0079] Optionally, the parameter adjustment unit includes:

[0080] The parameter updating unit is used to calculate the evaluation result using the following calculation method to obtain the adjusted parameters:

[0081] (a t+1 ,b t+1 ,c t+1 ,d t+1 ,……)=MOEA((a t ,b t ,c t ,d t ,……),(J1,J2))

[0082] Wherein, a, b, c, d represent different parameter values ​​of the parameter group, MOEA represents a multi-objective optimization algorithm, t represents the number of times adjusted, J1 represents the local density cost value information, and J2 represents the smoothness cost value information;

[0083] The parameter replacement unit is used to replace the corresponding parameters in the intermediate point cloud denoising algorithm with the adjusted parameters to obtain the adjusted intermediate point cloud denoising algorithm.

[0084] Optionally, the device further comprises:

[0085] A parameter generating unit, used to determine multiple types of parameters, and for each type of parameter, randomly generate a parameter value within a parameter value range of the parameter;

[0086] The parameter group forming unit is used to form a parameter group with parameter values ​​randomly generated from parameters of various types.

[0087] By means of the above technical solution, the present application substitutes each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm, denoises the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, evaluates the smoothness and local density of the denoising result to obtain an evaluation result, uses the evaluation result to adjust each parameter of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, and returns to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches the preset number of adjustments, and the parameters under the preset number of adjustments are combined into a target parameter group. It can be seen that the denoising effect is evaluated, the denoising evaluation result is obtained as the feedback result of the optimized denoising, the parameters of the point cloud denoising algorithm are updated, and finally the optimized parameters are obtained, which greatly saves the time of manual parameter adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0089] Figure 1 A schematic diagram of a process for automatically tuning the parameters of a point cloud denoising algorithm provided in an embodiment of the present application;

[0090] Figure 2 A schematic diagram of a device structure for realizing automatic tuning of point cloud denoising algorithm parameters provided in an embodiment of the present application;

[0091] Figure 3 A schematic diagram of the structure of a device for automatically tuning the point cloud denoising algorithm parameters provided in an embodiment of the present application. DETAILED DESCRIPTION

[0092] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0093] The present application solution can be implemented based on a terminal with data processing capabilities, which can be a computer, server, cloud, etc.

[0094] Next, combine Figure 1 The method for automatically tuning the point cloud denoising algorithm parameters of the present application may include the following steps:

[0095] Step S110: Substitute each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm.

[0096] Specifically, different point cloud denoising algorithms require different types of parameters to be adjusted, and different point cloud denoising algorithms require different numbers of parameters to be adjusted. The initial point cloud denoising algorithm can be a point cloud denoising algorithm to be substituted with parameters. By substituting various parameters into the initial point cloud denoising algorithm, an intermediate point cloud denoising algorithm can be obtained, which can be used to implement the denoising function of the point cloud.

[0097] Step S120: De-noising the point cloud to be de-noised based on the intermediate point cloud denoising algorithm to obtain a denoising result.

[0098] Specifically, the point cloud to be denoised may be composed of a number of points detected by a laser radar, and the data type of the point cloud to be denoised may be a three-dimensional image. The point cloud to be denoised may be input into an intermediate point cloud denoising algorithm, and a denoised point cloud may be output.

[0099] Step S130: Evaluate the smoothness and local density of the denoising result to obtain an evaluation result.

[0100] Specifically, the smoothness of the denoising result can represent the degree of regular distribution of points on the point cloud after denoising, and the local density of the denoising result can represent the degree of uniform distribution of points on the point cloud after denoising. The smoothness and local density of the denoising result can be analyzed simultaneously or in parallel, and the evaluation result obtained can represent the quality of the denoised point cloud.

[0101] Step S140: using the evaluation result to adjust various parameters of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm.

[0102] It is understandable that the denoising results of the denoised point cloud by the single point cloud denoising algorithm may be relatively rough. In order to improve the denoising effect of the point cloud denoising, it is necessary to readjust the parameters of the point cloud denoising algorithm according to the denoising evaluation results. By changing the parameters of the point cloud denoising algorithm, the denoising quality of the point cloud denoising algorithm for the denoised point cloud can be further optimized.

[0103] Step S150, determining whether the number of adjustment times reaches the preset number of adjustment times, if so, executing step S160, if not, returning to executing step S120.

[0104] Specifically, the preset number of adjustments may represent the number of adjustments required by the point cloud denoising algorithm to effectively denoise the denoised point cloud, and may be customized.

[0105] When the number of adjustments does not reach the preset number of adjustments, it may indicate that the point cloud denoising algorithm still fails to achieve effective denoising for the denoised point cloud, and the process returns to step S120 .

[0106] Step S160: assemble each parameter under the preset adjustment times into a target parameter group.

[0107] Specifically, when the number of adjustments reaches a preset number of adjustments, it can be said that the point cloud denoising algorithm has achieved effective denoising for the denoised point cloud, and the various parameters under the preset number of adjustments can be combined into a target parameter group, and the target parameter group can be used to substitute into the initial point cloud denoising algorithm to effectively denoise the denoised point cloud.

[0108] In the embodiment provided by the present application, each parameter in the pre-configured parameter group can be substituted into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm, and the denoised point cloud is denoised based on the intermediate point cloud denoising algorithm to obtain a denoising result, and the smoothness and local density of the denoising result are evaluated to obtain an evaluation result, and the evaluation result is used to adjust each parameter of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the parameters under the preset number of adjustments are combined into a target parameter group. It can be seen that the denoising effect is evaluated, the denoising evaluation result is obtained as the feedback result of the optimized denoising, the parameters of the point cloud denoising algorithm are updated, and finally the optimized parameters are obtained, which greatly saves the time of manual parameter adjustment.

[0109] In some embodiments of the present application, the process of configuring the parameter group mentioned in the above embodiment is introduced, and the specific configuration process may include:

[0110] S1. Determine multiple types of parameters, and for each type of parameter, randomly generate a parameter value within the parameter value range of the parameter.

[0111] Specifically, the parameter value range may represent a reasonable variation range of the parameter, and the parameter may be adjusted within the parameter value range during the parameter adjustment process.

[0112] It is understandable that the initial point cloud denoising algorithm requires parameters to be used for denoising the denoised point cloud, and the parameters can be adjusted multiple times, so each parameter that needs to be substituted into the initial point cloud denoising algorithm can be initialized.

[0113] In addition, each type of parameter can also use the default parameters of the point cloud denoising algorithm as initialization parameters.

[0114] S2. The randomly generated parameter values ​​of each type of parameter form a parameter group.

[0115] It is understandable that after multiple adjustments of several parameters in each parameter group, the parameters of the parameter group can be used as effective parameters for point cloud denoising algorithm to treat denoised point clouds. Due to the limited number of adjustments, each different randomly generated parameter may not be adjusted to the same effective parameter after a limited number of adjustments. Therefore, different parameter groups are configured and finally the parameter groups obtained after optimization and adjustment are different. Based on this, multiple parameter groups can be configured to obtain multiple parameter groups after optimization and adjustment, so as to select a more suitable parameter group according to different denoising directions.

[0116] In the embodiment provided in the present application, each type of parameter in the parameter group can be randomly generated in advance to be substituted into the initial point cloud denoising algorithm.

[0117] Furthermore, more parameter groups may be preconfigured so that after optimization and adjustment, a more appropriate parameter group can be selected from each parameter group according to different denoising directions.

[0118] In some embodiments of the present application, the process of evaluating the smoothness and local density of the denoising result in step S130 to obtain the evaluation result is introduced, and the process may include:

[0119] S1. Perform local density denoising evaluation on the denoising result to obtain local density cost value information.

[0120] Specifically, the denoising result can be evaluated and analyzed using a pre-established local density evaluation cost function to obtain local density cost information. The local density cost information can be a specific value, and its value can represent the uniformity of the point cloud after denoising. The smaller the value, the more uniform it is, and vice versa.

[0121] S2. Perform smoothing denoising evaluation on the denoising result to obtain smoothness cost value information, and use the local density cost value information and the smoothness cost value information as evaluation results.

[0122] Specifically, the denoising result can be evaluated and analyzed by using a pre-established local smoothness evaluation cost function to obtain smoothness cost information. The smoothness cost information can be a specific value, and its value can represent the smoothness of the denoised point cloud. The smaller the value, the smoother it is, and vice versa.

[0123] In the embodiments provided in the present application, the local density and smoothness of the denoising result can be analyzed respectively through the pre-established local density evaluation cost function and the local smoothness evaluation cost function to obtain the denoising evaluation result, so as to measure the denoising quality more effectively and accurately.

[0124] It can be understood that the denoising result may be a denoised image obtained after the image is denoised, so the denoising result may include several points. Based on this, in some embodiments of the present application, the above step S1, performing local density denoising evaluation on the denoising result to obtain the local density cost value information is introduced, and the process may include:

[0125] S1. For each point in the denoising result, determine a first preset number of neighboring points that are close to the point, and a reachable distance between the point and each neighboring point.

[0126] Specifically, the reachable distance may represent the farthest reachable distance between the point and each neighboring point within a proximity range.

[0127] It can be understood that each point has coordinate information in the denoising result, and the coordinate dimensions of points in different denoising results are different. For example, in the denoising result of a two-dimensional image, each point has two-dimensional coordinates, and in the denoising result of a three-dimensional image, each point has three-dimensional coordinates.

[0128] The first preset number may represent the number of points close to each point in the denoising result, and the specific number may be customized. The process of determining the first preset number of neighboring points close to the point may be specifically as follows:

[0129] S11, calculating the Euclidean distance between the point and other points in the denoising result according to the coordinate information of the point and other points, and taking a first preset number of points in the calculation result having a shorter Euclidean distance to the point as neighboring points.

[0130] Specifically, if the denoising result is a three-dimensional image, the following formula can be used to calculate the relationship between the point p(x, y, z) and other points p j (x j ,y j ,z j ) of the Euclidean distance d(p,p j ):

[0131]

[0132] Wherein, (x, y, z) represents the three-dimensional coordinates of a point, and N represents the number of points other than point p in the denoising result.

[0133] The process of determining the reachable distance between the point and each neighboring point may specifically include:

[0134] S12. For each point in the denoising result and each neighboring point of the point, calculate the Euclidean distance between the neighboring point and the other points according to the coordinate information of the neighboring point and the other points, and determine the Euclidean distance whose ascending sorting sequence number in the calculation result is the first preset number as the maximum neighboring distance of the neighboring points.

[0135] For example, for point A in the denoising result, the Euclidean distance to its neighboring point B can be the closest Euclidean distance between point A and all other points, so the ascending sorting number of this Euclidean distance is 1, the Euclidean distance to its neighboring point C can be the second closest Euclidean distance, so the ascending sorting number of this Euclidean distance is 2, and the Euclidean distance to its neighboring point D can be the third closest Euclidean distance, so the ascending sorting number of this Euclidean distance is 3. Therefore, if the preset number is 3, it can be determined that the maximum neighboring distance for the neighboring points of point A is the Euclidean distance between point A and its neighboring point D.

[0136] S13. Compare the Euclidean distance between the point and the neighboring point and the distance between the point and the maximum neighboring distance of the neighboring point, and take the longer distance as the reachable distance from the point to the neighboring point.

[0137] To further illustrate based on the example given in S12 above, if the preset number is 3, the reachable distance between point A and point B can be the maximum value between the Euclidean distance between A and B and the Euclidean distance to the third closest point to point B. The reachable distance between point A and point C can be the maximum value between the Euclidean distance between A and C and the Euclidean distance to the third closest point to point C.

[0138] S2. For each point in the denoising result, determine local density cost value information of the denoising result according to the reachable distance between the point and each neighboring point.

[0139] Specifically, the local reachable density of the point can be calculated by the reachable distance between each point and each neighboring point of the point. Further, the average local reachable density of each point is analyzed to obtain the local density error value of each point to represent the local density cost value information of the denoising result. The specific process may include:

[0140] S21. Sum the reachable distances between the point and each of its neighboring points to obtain a total reachable distance, and divide the total number of points in the denoising result by the total reachable distance to obtain a local reachable density.

[0141] Specifically, the local reachable density can represent the sparseness of the points in the central point and its surrounding area. The local reachable density of point p in the denoising result can be calculated using the following formula:

[0142]

[0143] Among them, N can represent the number of points other than point p in the denoising result, k can represent the number of neighboring points of point p, o can represent the neighboring points of point p, and reach-distance k (p,o j ) can represent point p to point o jThe reachable distance.

[0144] S22. For each point in the denoising result, calculate the sum of the local reachable densities of the neighboring points of the point, and divide the sum by the total number of points in the denoising result to obtain an average local reachable density.

[0145] Specifically, the average local reachable density can represent the density of points in the surrounding area near the center point. The average local reachable density of point p in the denoising result can be calculated using the following formula:

[0146]

[0147] Among them, lrd k (o j ) can represent the point o in the denoising result j The local reachable density of .

[0148] S23, determining a local density error between a local reachable density of each point in the denoising result and an average local reachable density.

[0149] It can be understood that, for each point in the denoising result, when the difference between its local reachable density and the average local reachable density is small, the noise in the neighboring area of ​​the point is small, that is, the denoising effect is better.

[0150] Specifically, the local density error between the local reachable density of the points in the denoising result and the average local reachable density can be expressed as

[0151] S24, calculating the average value of the local density errors of each point in the denoising result as the local density cost value information of the denoising result.

[0152] It can be understood that the local density error of each point is summarized and analyzed, and the result obtained can represent the denoising effect of the denoising result, and the denoising effect can be represented by the local density cost information.

[0153] Specifically, the local density cost value of the denoising result can be calculated using the following formula:

[0154]

[0155] Among them, M can represent the total number of points in the denoising result.

[0156] In the embodiment provided by the present application, by determining the neighboring points of each point in the denoising result and the reachable distance between each point and its neighboring points, and then calculating the local reachable density and the average local reachable density of each point based on the reachable distance, the error between the local reachable density and the average local reachable density is statistically obtained, and then the local density cost value information of the denoising result is obtained, which can more accurately quantify the denoising effect.

[0157] It can be understood that the denoising result may be a denoised image obtained after the image is denoised, so the denoising result may include a number of points, and these points may include specific coordinate value information. Based on this, in some embodiments of the present application, the above step S2, the process of performing smoothing denoising evaluation on the denoising result to obtain smoothness cost value information is introduced, and the process may include:

[0158] In this embodiment, the object analyzed below may be each point in the denoising result.

[0159] S21. Determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point.

[0160] Among them, the process of determining a second preset number of neighboring points that are close to the point can calculate the Euclidean distance between the point and other points in the denoising result based on the coordinate information of the point and other points in the denoising result, and take the second preset number of points in the calculation result that have a shorter Euclidean distance to the point as neighboring points.

[0161] Taking into account data operation pressure, the second preset number may be smaller than the first preset number.

[0162] Specifically, the coordinate matrix may represent coordinate distribution information of neighboring points of the point.

[0163] For example, the denoising result can be a three-dimensional image, then the points of the denoising result can have three-dimensional coordinates, the horizontal coordinates of each neighboring point can be used as the first row of the matrix, the vertical coordinates of each neighboring point can be used as the second row of the matrix, and the vertical coordinates of each neighboring point can be used as the third row of the matrix, for example:

[0164]

[0165] Among them, (x, y, z) can represent three-dimensional coordinates, p can represent the analysis object point, and m can represent the number of some neighboring points among all k neighboring points.

[0166] S22. Obtain the mean of the coordinates of the adjacent points according to the coordinate matrix.

[0167] Specifically, the mean of the coordinates of each neighboring point can represent the average position of the neighboring point in each dimensional direction.

[0168] Taking the example of S21 in this embodiment as an example, the mean value of the three-dimensional coordinates can be expressed as:

[0169]

[0170] S23. Calculate a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point.

[0171] Specifically, the covariance matrix can represent the degree of dispersion of neighboring points centered on the mean. The covariance matrix can be calculated using the following formula:

[0172]

[0173] S24, calculating the minimum eigenvalue of each neighboring point of the point in the covariance matrix, and determining the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculating the information entropy of the point based on the projection length.

[0174] Specifically, information entropy can represent the degree of disorder of the distribution of points around the point.

[0175] The eigenvector matrix E(p) can be obtained by decomposing the covariance matrix Cov(p), and the eigenvalue diagonal matrix can be calculated using the following formula:

[0176] Λ=E T (p)Cov(p)E(p)

[0177] Among them, the eigenvector corresponding to the minimum eigenvalue of the eigenvalue diagonal matrix Λ is the normal vector n(p) of point p. Based on this, the projection length can be calculated using the following formula:

[0178]

[0179] Among them, (x, y, z) can represent the three-dimensional coordinates of point p, (x j ,y j , z j ) can represent a point o j The three-dimensional coordinates of .

[0180] The information entropy of the point is calculated using the following equation:

[0181]

[0182] Wherein, H(p) may represent the information entropy of the point, and k may represent the total number of neighboring points of point p.

[0183] S25. Determine smoothness cost value information of the denoising result according to the information entropy of each point of the denoising result.

[0184] It can be understood that the information entropy of each point is summarized and analyzed, and the result obtained can represent the denoising effect of the denoising result, and the denoising effect can be represented by the smoothness representative value information.

[0185] The smoothness cost value of the denoising result can be calculated using the following formula:

[0186]

[0187] Wherein, M can represent the total number of points in the denoising result.

[0188] In the embodiment provided by the present application, a coordinate matrix is ​​generated by using the coordinates of several neighboring points of each point in the denoising result, and the means of several neighboring points in each dimension are further calculated to obtain a covariance matrix. On this basis, the information entropy of the point is calculated, and the degree of chaos of the points in the area around the point can be accurately analyzed. Finally, the information entropy of each point is analyzed to obtain the smoothness cost value information of the denoising result, which can accurately quantify the smoothness of the denoised image.

[0189] In some embodiments of the present application, the process of adjusting various parameters of the intermediate point cloud denoising algorithm using the evaluation result in the above step 140 to obtain an adjusted intermediate point cloud denoising algorithm is introduced, and the process may include:

[0190] S1. Calculate the evaluation results using the following calculation method to obtain the adjusted parameters:

[0191] (a t+1 ,b t+1 ,c t+1 ,d t+1 ,……)=MOEA((a t ,b t ,c t ,d t ,……),(J1,J2))

[0192] Among them, a, b, c, d represent different parameter values ​​of the parameter group, MOEA represents a multi-objective optimization algorithm, t represents the number of adjustments, J1 represents the local density cost value information, and J2 represents the smoothness cost value information.

[0193] S2. Use the adjusted parameters to replace the corresponding parameters in the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm.

[0194] It is understandable that by a t ,b t,c t ,d t ...as the parameters of the intermediate point cloud denoising algorithm, the denoising effect is obtained and J1 and J2 are obtained, and as feedback information, a is updated t+1 ,b t+1 ,c t+1 ,d t+1 ..., so a t+1 ,b t+1 ,c t+1 ,d t+1 ...is the optimized parameter, and the optimized parameter replaces the parameter a in the current intermediate point cloud denoising algorithm t ,b t ,c t ,d t , the new J1 and J2 are analyzed to be smaller in the denoising results, achieving the effect of progressive denoising.

[0195] In the embodiment provided in the present application, various parameters of the point cloud denoising algorithm are updated through a multi-objective optimization algorithm and the local density cost value information and smoothness cost value information of the denoising evaluation results, so that after the updated parameters are substituted into the point cloud denoising algorithm, the local density cost value information and smoothness cost value information analyzed in the obtained denoising results will be smaller, thereby achieving the effect of automatic optimization denoising.

[0196] See also Figure 2 , Figure 2 A schematic diagram of the structure of a device for automatically tuning the parameters of a point cloud denoising algorithm disclosed in an embodiment of the present application.

[0197] like Figure 2 As shown, the device may include:

[0198] An automatic tuning device for point cloud denoising algorithm parameters, comprising:

[0199] A parameter substitution unit 11 is used to substitute each parameter in a pre-configured parameter group into an existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm;

[0200] A point cloud denoising unit 12, configured to denoise the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result;

[0201] A denoising evaluation unit 13, used to evaluate the smoothness and local density of the denoising result to obtain an evaluation result;

[0202] The parameter adjustment unit 14 is used to adjust the various parameters of the intermediate point cloud denoising algorithm using the evaluation result to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

[0203] Optionally, the denoising evaluation unit 13 includes:

[0204] A local denoising evaluation unit, used to perform local density denoising evaluation on the denoising result to obtain local density cost value information;

[0205] The smoothing denoising evaluation unit is used to perform smoothing denoising evaluation on the denoising result to obtain smoothness cost value information, and use the local density cost value information and the smoothness cost value information as the evaluation result.

[0206] Optionally, the denoising result includes a plurality of points;

[0207] The local denoising evaluation unit comprises:

[0208] A neighboring point determination unit, configured to determine, for each point in the denoising result, a first preset number of neighboring points that are close to the point;

[0209] A reachable distance determination unit, used to determine, for each point in the denoising result, a reachable distance between the point and each neighboring point;

[0210] The local density cost value determining unit is used to determine the local density cost value information of the denoising result for each point in the denoising result according to the reachable distance between the point and each neighboring point.

[0211] Optionally, each point in the denoising result has coordinate information;

[0212] The neighboring point determination unit includes:

[0213] The first neighboring point determination subunit is used to calculate the Euclidean distance between the point and other points in the denoising result according to the coordinate information of the point and other points in the denoising result, and to use a first preset number of points in the calculation result with a shorter Euclidean distance to the point as neighboring points.

[0214] Optionally, the reachable distance determining unit includes:

[0215] A first reachable distance determination subunit is configured to calculate, for each point in the denoising result and each neighboring point of the point, the Euclidean distance between the neighboring point and the other points according to the coordinate information of the neighboring point and the other points, and determine the Euclidean distances whose ascending sorting numbers in the calculation result are the first preset number as the maximum neighboring distances of the neighboring points;

[0216] The second reachable distance determination subunit is used to compare the Euclidean distance between the point and each neighboring point of the point with the length of the distance between the maximum neighboring distance of the neighboring points, and take the longer distance as the reachable distance from the point to the neighboring point.

[0217] Optionally, the local density cost value determining unit includes:

[0218] A first local density cost value determination subunit is used to sum the reachable distances between the point and each neighboring point of the point to obtain a total reachable distance, and divide the total number of points in the denoising result by the total reachable distance to obtain a local reachable density;

[0219] A second local density cost value determination subunit is used to calculate, for each point in the denoising result, the sum of the local reachable densities of the neighboring points of the point, and divide the sum by the total number of points in the denoising result to obtain an average local reachable density;

[0220] A third local density cost value determination subunit is used to determine a local density error between the local reachable density of each point in the denoising result and the average local reachable density;

[0221] The fourth local density cost value determining subunit is used to calculate the average value of the local density errors of each point in the denoising result as the local density cost value information of the denoising result.

[0222] Optionally, the denoising result includes a plurality of points, each point having coordinate information;

[0223] For each point of the denoising result, the smoothing denoising evaluation unit includes:

[0224] A first smoothing denoising evaluation subunit is used to determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point;

[0225] A second smoothing denoising evaluation subunit is used to obtain the mean value of the coordinates of each neighboring point according to the coordinate matrix;

[0226] A third smoothing denoising evaluation subunit is used to calculate a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point;

[0227] a fourth smoothing denoising evaluation subunit, configured to calculate the minimum eigenvalue of each neighboring point of the point in the covariance matrix, determine the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculate the information entropy of the point according to the projection length;

[0228] The fifth smoothing denoising evaluation subunit is used to determine the smoothness cost value information of the denoising result according to the information entropy of each point of the denoising result.

[0229] Optionally, the parameter adjustment unit 14 includes:

[0230] The parameter updating unit is used to calculate the evaluation result using the following calculation method to obtain the adjusted parameters:

[0231] (a t+1 ,b t+1 ,c t+1 ,d t+1 ,……)=MOEA((a t ,b t ,c t ,d t ,……),(J1,J2))

[0232] Wherein, a, b, c, d represent different parameter values ​​of the parameter group, MOEA represents a multi-objective optimization algorithm, t represents the number of times adjusted, J1 represents the local density cost value information, and J2 represents the smoothness cost value information;

[0233] The parameter replacement unit is used to replace the corresponding parameters in the intermediate point cloud denoising algorithm with the adjusted parameters to obtain the adjusted intermediate point cloud denoising algorithm.

[0234] Optionally, the device further comprises:

[0235] A parameter generating unit, used to determine multiple types of parameters, and for each type of parameter, randomly generate a parameter value within a parameter value range of the parameter;

[0236] The parameter group forming unit is used to form a parameter group with parameter values ​​randomly generated from parameters of various types.

[0237] The device for automatically tuning the parameters of the point cloud denoising algorithm provided in the embodiment of the present application can be applied to a device for automatically tuning the parameters of the point cloud denoising algorithm, such as a terminal: a cloud, a computer, etc. Optionally, Figure 3 The hardware structure diagram of the automatic tuning of the point cloud denoising algorithm parameters is shown in FIG. Figure 3 ,The hardware structure of the automatic tuning device for point cloud denoising algorithm parameters may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0238] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0239] The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention, etc.;

[0240] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory;

[0241] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0242] Substituting each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm;

[0243] De-noising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result;

[0244] Evaluating the smoothness and local density of the denoising result to obtain an evaluation result;

[0245] The evaluation result is used to adjust the various parameters of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

[0246] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0247] The embodiment of the present application further provides a storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0248] Substituting each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm;

[0249] De-noising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result;

[0250] Evaluating the smoothness and local density of the denoising result to obtain an evaluation result;

[0251] The evaluation result is used to adjust the various parameters of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

[0252] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0253] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0254] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0255] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatically tuning the parameters of a point cloud denoising algorithm, characterized in that: include: Substituting each parameter in the pre-configured parameter group into the existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm; De-noising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result; Evaluating the smoothness and local density of the denoising result to obtain an evaluation result; Using the evaluation result, various parameters of the intermediate point cloud denoising algorithm are adjusted to obtain an adjusted intermediate point cloud denoising algorithm, and returning to the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments form a target parameter group; The smoothness and local density of the denoising result are evaluated to obtain an evaluation result, including: Performing local density denoising evaluation on the denoising result to obtain local density cost value information; Performing smoothing denoising evaluation on the denoising result to obtain smoothness cost value information, and using the local density cost value information and the smoothness cost value information as evaluation results; The denoising result includes a number of points; The performing local density denoising evaluation on the denoising result to obtain local density cost value information includes: For each point in the denoised result: Determine a first preset number of neighboring points that are close to the point, and a reachable distance between the point and each of the neighboring points; Determining local density cost value information of the denoising result according to the reachable distance between the point and each neighboring point; The denoising result includes a plurality of points, each point having coordinate information; The step of performing smoothing denoising evaluation on the denoising result to obtain smoothness cost value information includes: For each point of the denoising result: Determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point; According to the coordinate matrix, obtaining the mean value of the coordinates of each neighboring point; Calculating a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point; Calculate the minimum eigenvalue of each neighboring point of the point in the covariance matrix, determine the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculate the information entropy of the point according to the projection length; According to the information entropy of each point of the denoising result, smoothness cost value information of the denoising result is determined.

2. The method according to claim 1, characterized in that Each point in the denoising result has coordinate information; The determining of a first preset number of neighboring points that are close to the point includes: The Euclidean distance between the point and other points in the denoising result is calculated according to the coordinate information of the point and other points in the denoising result, and a first preset number of points in the calculated result having a shorter Euclidean distance to the point are taken as neighboring points.

3. The method according to claim 1, characterized in that Determine the reachable distance between the point and each neighboring point, including: For each point in the denoising result and each neighboring point of the point: Calculate the Euclidean distance between the neighboring point and the other point according to the coordinate information of the neighboring point and the other point, and determine the Euclidean distances whose ascending sorting numbers in the calculation results are the first preset number as the maximum neighboring distance of the neighboring point; The Euclidean distance between the point and the neighboring point is compared with the distance between the point and the maximum neighboring distance of the neighboring point, and the longer distance is used as the reachable distance from the point to the neighboring point.

4. The method according to claim 1, characterized in that Determining local density cost value information of the denoising result according to the reachable distance between the point and each neighboring point includes: Sum the reachable distances between the point and each of its neighboring points to obtain a total reachable distance, and divide the total number of points in the denoising result by the total reachable distance to obtain a local reachable density; For each point in the denoising result, calculate the sum of the local reachable densities of the neighboring points of the point, and divide the sum by the total number of points in the denoising result to obtain an average local reachable density; Determine a local density error between a local reachable density of each point in the denoising result and an average local reachable density; The average value of the local density errors of each point in the denoising result is calculated as the local density cost value information of the denoising result.

5. The method according to claim 1, characterized in that The denoising result includes a plurality of points, each point having coordinate information; The denoising result is subjected to smoothing denoising evaluation to obtain smoothness cost value information, including: For each point of the denoising result: Determine a second preset number of neighboring points that are close to the point, and use the coordinates of each dimension of each neighboring point as each row matrix of the point to generate a coordinate matrix of the point; According to the coordinate matrix, obtaining the mean value of the coordinates of each neighboring point; Calculating a covariance matrix according to the coordinate matrix and the mean of the coordinates of each neighboring point; Calculate the minimum eigenvalue of each neighboring point of the point in the covariance matrix, determine the projection length in the direction of the eigenvector corresponding to the minimum eigenvalue, and calculate the information entropy of the point according to the projection length; According to the information entropy of each point of the denoising result, smoothness cost value information of the denoising result is determined.

6. The method according to claim 1, characterized in that The evaluation result is used to adjust various parameters of the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm, including: The evaluation results are calculated using the following calculation method to obtain the adjusted parameters: (a t+1 ,b t+1 ,c t+1 ,d t+1 ,……)=MOEA((a t ,b t ,c t ,d t ,……),(J1,J2)) Wherein, a, b, c, d represent different parameter values ​​of the parameter group, MOEA represents a multi-objective optimization algorithm, t represents the number of times adjusted, J1 represents the local density cost value information, and J2 represents the smoothness cost value information; The adjusted parameters are used to replace the corresponding parameters in the intermediate point cloud denoising algorithm to obtain an adjusted intermediate point cloud denoising algorithm.

7. The method according to claim 1, characterized in that The configuration process of the parameter group includes: Determine multiple types of parameters, and for each type of parameter, randomly generate a parameter value within a parameter value range of the parameter; The parameter values ​​randomly generated for each type of parameter constitute a parameter group.

8. An automatic tuning device for point cloud denoising algorithm parameters, characterized in that: Applied to the method according to claim 1, the device comprises: A parameter substitution unit, used for substituting each parameter in a pre-configured parameter group into an existing initial point cloud denoising algorithm to obtain an intermediate point cloud denoising algorithm; A point cloud denoising unit, used for denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result; A denoising evaluation unit, used to evaluate the smoothness and local density of the denoising result to obtain an evaluation result; A parameter adjustment unit is used to adjust the various parameters of the intermediate point cloud denoising algorithm using the evaluation result to obtain an adjusted intermediate point cloud denoising algorithm, and return to execute the step of denoising the denoised point cloud based on the intermediate point cloud denoising algorithm to obtain a denoising result, until the number of adjustments reaches a preset number of adjustments, and the various parameters under the preset number of adjustments are combined into a target parameter group.

Images