An adaptive grid search method for spherical target fitting
Through the adaptive grid search method combined with the geometric features of the spherical target, the fitting of the spherical target center and radius is optimized, and the problem of noise interference and low computational efficiency of the least squares method in the fitting of spherical target point cloud is solved, and the problem of low computational efficiency of the least squares method is achieved. High-precision and efficient solution.
Patent Information
- Application Number
- CN202111313264.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-08
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-11-08
AI Technical Summary
The existing least squares method is susceptible to noise interference in spherical target point cloud fitting, resulting in inaccurate solution results and low computational efficiency, especially under a large amount of measurement point data.
Adaptive grid search method is used, combined with the geometric characteristics of the spherical target, and through initial parameter setting, sample space construction and error determination criteria, the center and radius of the spherical target that meet the predetermined accuracy are found through the cyclic optimization process.
It improves the accuracy and efficiency of spherical target center solution, avoids accuracy loss in linearization, and is suitable for ground and vehicle-mounted three-dimensional laser scanning systems.
Smart Images

Figure CN114092545B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of engineering surveying and three-dimensional laser scanning, and in particular relates to an adaptive grid search method for spherical target fitting suitable for ground laser scanning. Background Art
[0002] Terrestrial 3D laser scanning technology is a new type of measurement technology that has emerged in recent years. Through high-speed laser scanning, high-resolution point cloud data of the target surface can be quickly obtained, bringing traditional single-point measurement into the era of surface measurement. It has the characteristics of active, non-contact, high-resolution, high-precision, fast and flexible data acquisition, and has been widely used in engineering measurement, cultural relics protection, disaster monitoring, reverse 3D reconstruction and other work. Spherical targets have typical spatial rotational symmetry and standard parameterized forms, and play an important role in various types of terrestrial 3D laser scanning research, such as terrestrial laser scanner calibration and verification, scanning accuracy evaluation, point cloud registration and coordinate transformation, etc. The center of the spherical target is located inside the sphere and cannot be obtained by direct measurement means, so the fitting method is usually used to obtain its center.
[0003] Fitting the center coordinates of the point cloud of a spherical target acquired by three-dimensional laser scanning is a nonlinear problem in itself. At present, it is usually solved by various least squares methods, such as classical least squares, global least squares, and weighted global least squares. From the theoretical perspective of least squares, the least squares estimation assumes that the mean of the data noise is zero, thus producing an unbiased parameter estimate. If the variance of the noise is known, the minimum variance parameter estimate can be obtained by selecting appropriate weights on the data. In addition, the least squares estimation implicitly assumes that the entire data set can only be explained by one parameter vector of a given model. A large number of studies have clearly shown that the least squares estimation can easily violate these assumptions. Sometimes, even if the data contains only one bad data, the least squares estimation may be seriously disturbed, resulting in poor solution results. Therefore, this type of method is mostly based on the solution of pure measurement point data, which is prone to problems such as accuracy being greatly affected by initial values and being easily trapped in local optimality. At the same time, the number of measurement points of the point cloud of a spherical target is usually more than thousands, which will result in a huge calculation matrix and low operating efficiency.
[0004] In the scanning work, the spherical target we use has a certain geometric size, that is, for any spherical target in space, the distribution of its point cloud in space has a certain range, which is called a bounding box. This bounding box contains all the measurement point data and noise of the spherical target. At the same time, from the geometric characteristics of the spherical target, its center and radius must be in this bounding box. In this way, a search strategy can be used to find the optimal center and radius in the bounding box that meet specific error judgment rules. In this context, we combine the point cloud and geometric characteristics of the spherical target and propose an adaptive grid search method suitable for fitting various types of spherical targets. Summary of the invention
[0005] The purpose of the present invention is to design an adaptive grid search method suitable for spherical target fitting by combining the measurement point information and geometric feature information of the spherical target. This method not only avoids the accuracy loss of the linearization problem, improves the solution accuracy of the spherical target center, but also improves the solution efficiency by constraining the geometric features of the spherical target.
[0006] To achieve the above object, the technical solution adopted by the present invention is: an adaptive grid search method suitable for spherical target fitting, which comprises the following steps:
[0007] Step 1: Use a 3D laser scanning system to obtain a spherical target point cloud;
[0008] Step 2: Set initial parameters;
[0009] Step 3: Set the fitting error judgment criteria to obtain the optimal spherical target center and radius;
[0010] Step 4: Determine the initial scale The initial scale is composed of the scales of four feature quantities: X, Y, Z, and R, mainly the geometric center of gravity of the point cloud. With the preset geometric radius R set jointly determined;
[0011] Step 5: Construct the initial sample space U (1) ; Initial sample space Each sample in is composed of four features, is the possible geometric center of the spherical target, is the possible geometric radius of the spherical target;
[0012] Step 6: Determine the initial optimal sample;
[0013] Step 7: Update the scale
[0014] Step 8: Update the sample space U (i) ;
[0015] Step 9: Find the optimal sample;
[0016] Step 10: Loop through steps 7, 8, and 9 until a sample that meets the predetermined accuracy is found or the predetermined number of loops N is reached. opt .
[0017] Furthermore, in step 1, a ground-based 3D laser scanning system or a vehicle-mounted 3D laser scanning system is used to collect a point cloud P of a spherical target = {(x i ,y i ,z i )|i=1,2,…,n}, the spherical target point cloud is composed of the measurement point data on the surface of the spherical target, and each measurement point is composed of three coordinate components: X, Y, and Z.
[0018] Furthermore, in step 2, the initial parameters are mainly determined by the preset geometric radius R of the spherical target. set , grid sampling number N sub , iterative optimization times N opt , fitting error threshold E min The spherical target has a preset geometric radius R set is the estimated radius of the spherical target to be fitted; the number of grid samples N sub is the number of samples in each dimension (X, Y, Z, R) when constructing the spatial grid; the number of iterative optimizations N opt The maximum number of executions of the predetermined iterative optimization is used to limit the number of iterations of the fitting process to prevent infinite loops; the fitting error threshold E min The overall error of the spherical target point cloud for determining the end of the search condition.
[0019] Furthermore, in step 3, in order to obtain the optimal center and radius of the spherical target, the optimal center and radius determination criteria are set; here, the arithmetic error measure is used to fit the error E fit Evaluation: Assuming that the fitting center of the spherical target is (a, b, c) and the fitting radius is R, then the fitting error E fit can be determined by formula (1):
[0020]
[0021] In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of the spherical target point cloud, and n is the number of measuring points contained in the spherical target point cloud.
[0022] Furthermore, in step 4, the geometric center of gravity According to formula (2), the initial scale According to formula (3), we can get:
[0023]
[0024] In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of point cloud P, n is the number of measuring points of point cloud P;
[0025]
[0026] Furthermore, in step 5, the initial sample space U (1) The samples in the initial scale The preset number of grid samples N sub Determine together that X (1) , Y (1) , Z (1) , R (1) All possible combinations of the four feature quantities are determined by formula (4):
[0027]
[0028] In the formula, The value is 0 because the geometric radius of the spherical target is always greater than 0.
[0029] Furthermore, in step 6, the initial sample space U is determined by using the judgment criteria of step 3. (1) The fitting error is calculated for each sample in the equation, and the sample with the smallest fitting error is selected as the initial optimal solution. In the process of calculating the fitting error, it is necessary to detect whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
[0030] Furthermore, in step 7, during the grid search process, the probability of obtaining the optimal value is related to the accuracy of the grid space. The finer the grid space, the higher the probability of finding the optimal value. If no sample points that meet the predetermined accuracy are found, the search scale needs to be further optimized to improve the fineness of the sample; the scale update is calculated using formula (5):
[0031]
[0032] Furthermore, in step 8, based on the optimal center (X (i-1) ,Y (i-1) ,Z (i-1) ) and radius R (i-1), and the updated scale Generate a new sample space U (i) ; Sample space For X (i) , Y (i) , Z (i) , R (i) All possible combinations of the four feature values can be expressed as follows:
[0033]
[0034] Furthermore, in step 9, the updated sample space U is updated using the judgment criteria of step 3. (i) Each sample in the equation is tested for error one by one, and the sample with the smallest fitting error is selected as the optimal solution. In the process of calculating the fitting error, it is tested whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
[0035] In summary, the spherical target fitting of this method is to extract the center coordinates and radius of the spherical target from the point cloud with unknown distribution and containing outliers, which is actually a problem of optimal parameter estimation. In this optimal parameter estimation problem, the center coordinates (X, Y, Z) and radius R of the spherical target are regarded as parameters to be solved, and the target point cloud is used as the observation value. The process of fitting the center coordinates and radius of the spherical target point cloud can be regarded as a process of finding the optimal parameters that meet specific judgment rules. With the geometric center of gravity of the spherical target point cloud as the center and more than 2 times the radius length as a constraint, an initial bounding box is constructed, and according to the geometric characteristics of the spherical target, its center and radius must be within this bounding box. Based on this feature, the spherical target fitting problem can be solved from a probabilistic perspective.
[0036] The beneficial effects of the present invention are as follows: the present invention combines the point cloud and geometric feature information of the spherical target, and uses an adaptive grid search method to achieve accurate fitting of the center and radius of the spherical target, thereby avoiding the problem of precision loss in the linearization process. The present invention can not only achieve high-precision solution of the center and radius of the spherical target point cloud, but also has high execution efficiency.
[0037] At the same time, the invention is not only applicable to the center solution of ground or vehicle-mounted three-dimensional laser scanning spherical targets, but can also be applied to the extraction of center coordinates of other spherical objects. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 is a flow chart of the present invention;
[0039] Figure 2 It is a simulated point cloud image of a spherical target;
[0040] Figure 3 is the constructed initial sample space diagram;
[0041] Figure 4 is the updated sample space diagram. DETAILED DESCRIPTION
[0042] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] like Figure 1 The present invention provides an adaptive grid search algorithm suitable for fitting various spherical targets. The specific implementation steps are as follows:
[0044] Step 1: Use a 3D laser scanning system to obtain a point cloud of a spherical target. Use a ground-based 3D laser scanning system or a vehicle-mounted 3D laser scanning system to collect a point cloud of a spherical target P = {(x i ,y i ,z i )|i=1,2,…,n}, the spherical target point cloud is composed of the measurement point data on the surface of the spherical target, and each measurement point is composed of three coordinate components: X, Y, and Z.
[0045] Step 2: Set the initial parameters. The initial parameters are mainly determined by the preset geometric radius R of the spherical target. set , grid sampling number N sub , iterative optimization times N opt , fitting error threshold E min The preset geometric radius R of the spherical target set Refers to the estimated value of the radius of the spherical target to be fitted, which is usually easy to determine. In 3D laser scanning operations, the geometric radius of the spherical target used is often clearly stated in the technical specifications. If the geometric radius of the spherical target to be fitted cannot be determined, a relatively larger empirical estimate can be set. Grid sampling number N sub It refers to the number of samples in each dimension (X, Y, Z, R) when constructing the spatial grid. Iteration optimization times N opt It refers to the maximum number of executions of the predetermined iterative optimization. This value is used to limit the number of iterations of the fitting process to prevent infinite loops. Fitting error threshold E min It refers to the overall error of the spherical target point cloud for determining the end of the search condition. This value can be calculated by the error judgment criterion in step 3. By setting the threshold here, the subsequent search can be directly terminated after the predetermined fitting error is reached, so as to improve the efficiency of the algorithm.
[0046] Step 3: Set the criterion for fitting error. In order to obtain the optimal center and radius of the spherical target, the criterion for the optimal center and radius must be set. Here, an arithmetic error measure is used to calculate the fitting error E.fit Evaluation, which is obtained by the sum of the errors of all measurement points. Assuming that the fitting center of the spherical target is (a, b, c) and the fitting radius is R, then the fitting error E fit can be determined by formula (1).
[0047]
[0048] In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of the spherical target point cloud, and n is the number of measuring points contained in the spherical target point cloud.
[0049] Step 4: Determine the initial scale The initial scale is composed of the scales of four feature quantities: X, Y, Z, and R, mainly the geometric center of gravity of the point cloud. With the preset geometric radius R set Jointly determine. Geometric center of gravity According to formula (2), the initial scale It can be calculated by formula (3).
[0050]
[0051] In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of point cloud P, and n is the number of measuring points of point cloud P.
[0052]
[0053] Step 5: Construct the initial sample space U (1) The initial sample space Each sample in is composed of four features, is the possible geometric center of the spherical target, is the possible geometric radius of the spherical target. Initial sample space U (1) The samples in the main are composed of the initial scale The preset number of grid samples N sub Determine together that X (1) , Y (1) , Z (1) , R (1) All possible combinations of the four feature values are determined by formula (4).
[0054]
[0055] In the formula, The value is 0 because the geometric radius of the spherical target is always greater than 0.
[0056] Step 6: Determine the initial optimal sample. Using the judgment criteria in step 3, the initial sample space U (1) The fitting error is calculated for each sample in the equation, and the sample with the smallest fitting error is selected as the initial optimal solution. In the process of calculating the fitting error, it is necessary to check whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
[0057] Step 7: Update the scale According to probability statistics theory, in the grid search process, the probability of obtaining the optimal value is related to the accuracy of the grid space. The finer the grid space, the higher the probability of finding the optimal value. If no sample points that meet the predetermined accuracy are found, the search scale needs to be further optimized to improve the fineness of the sample. The scale update is calculated using formula (5).
[0058]
[0059] Step 8: Update the sample space U (i) Based on the optimal center (X (i-1) ,Y (i-1) ,Z (i -1) ) and radius R (i-1) , and the updated scale Generate a new sample space U (i) Sample space For X (i) , Y (i) , Z (i) , R (i) All possible combinations of the four feature values can be expressed by equation (6).
[0060]
[0061] Step 9: Find the optimal sample. Using the judgment criteria in step 3, update the sample space U (i) Each sample in the equation is tested for error one by one, and the sample with the smallest fitting error is selected as the optimal solution. In the process of calculating the fitting error, it is tested whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
[0062] Step 10: Loop through steps 7, 8, and 9 until a sample that meets the predetermined accuracy is found or the predetermined number of loops N is reached.opt At this point, the optimal sample is taken as the final fitting result and the fitting is ended. Generally, after more than a dozen iterations of optimization, the optimal solution can be basically determined.
[0063] The solution of the above embodiment is further described below in combination with specific data:
[0064] Considering that the geometric center of the spherical target is located inside the sphere, it cannot be obtained by direct measurement means, making it difficult to evaluate the fitting accuracy. Here, simulation data with known geometric center and radius are used to illustrate the implementation method and fitting accuracy of the present invention.
[0065] (1) Using the true center coordinates (1000, 1000, 100) and a radius of 0.0725 m, simulate the 3260 measurement point data of a spherical target, such as Figure 2 As shown, the black points are simulated point cloud data.
[0066] (2) Follow the method in step 2 to set the initial parameters. Preset the geometric radius R set is 0.0725m, grid sampling number N sub is 10, the number of iterations is N opt is 30, the fitting error threshold E min is 0.001.
[0067] (3) Determine the initial scale according to step 4 The calculation result is (0.2900, 0.2900, 0.2900, 0.1450).
[0068] (5) Construct the initial sample space U according to step 5 (1) ,like Figure 3 As shown in the figure, “*” indicates the location of the sample point. At this time, the sample space contains 10,000 samples in total, which are evenly distributed in the space.
[0069] (4) Using the judgment criteria of step 3, determine the initial optimal sample according to step 6. At this time, the optimal sample obtained is (1000.0159, 1000.0161, 100.0342, 0.0805), and its fitting error 5.0608 is greater than the predetermined fitting error 0.001, which does not meet the predetermined accuracy, so it is necessary to update the scale and sample space.
[0070] (5) Update the scale according to step 7. Updated scale is (0.0725,0.0725,0.0725,0.0363).
[0071] (6) Update the sample space U according to step 8 (2) ,like Figure 4As shown in the figure, “*” indicates the location of the sample point. At this time, the sample space contains 10,000 samples in total, which are evenly distributed in the space.
[0072] (7) Find the optimal sample after updating the scale and sample space according to step 9. At this time, the optimal sample obtained is (1000.0078, 1000.0080, 99.9954, 0.0765), and its fitting error 2.0472 is greater than the predetermined fitting error threshold 0.001, which does not meet the predetermined accuracy. It is necessary to continue to update the scale and sample space.
[0073] (8) According to step 10, step 7, step 8, and step 9 are executed repeatedly. After 17 iterations of optimization, the optimal sample obtained is (1000.0000, 1000.0000, 100.0000, 0.0725). Its fitting error 0.0002 is less than the predetermined fitting error threshold 0.001, reaching the predetermined accuracy, and the iteration process ends. At the same time, the execution time is calculated to be about 1.5396 seconds.
[0074] From the experimental results, it can be seen that the method designed by the present invention can accurately fit the coordinates and radius of the center point of the spherical target, and the execution efficiency is relatively high.
[0075] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the protection scope of the present invention in any form, and all technical solutions obtained by equivalent replacement and the like fall within the protection scope of the present invention.
[0076] The parts not involved in the present invention are the same as the prior art or can be implemented by using the prior art.
Claims
1. An adaptive grid search method for spherical target fitting, Features The following steps are involved: Step 1: Use a 3D laser scanning system to obtain a spherical target point cloud; Step 2: Set the initial parameters; the initial parameters are determined by the preset geometric radius R of the spherical target. set , grid sampling number N sub , iterative optimization times N opt , fitting error threshold E min The spherical target has a preset geometric radius R set is the estimated radius of the spherical target to be fitted; the number of grid samples N sub is the number of samples in each dimension (X, Y, Z, R) when constructing the spatial grid; the number of iterative optimizations N opt The maximum number of executions of the predetermined iterative optimization is used to limit the number of iterations of the fitting process to prevent infinite loops; the fitting error threshold E min The overall error of the spherical target point cloud for determining the end of the search condition; Step 3: Set the fitting error judgment criteria to obtain the optimal spherical target center and radius; Step 4: Determine the initial scale The initial scale is composed of the scales of four feature quantities: X, Y, Z, and R, which are determined by the geometric center of gravity of the point cloud. With the preset geometric radius R set jointly determined; Step 5: Construct the initial sample space U (1) ; Initial sample space Each sample in is composed of four features, is the possible geometric center of the spherical target, is the possible geometric radius of the spherical target; Step 6: Determine the initial optimal sample; Step 7: Update the scale In the grid search process, the probability of obtaining the optimal value is related to the accuracy of the grid space. The finer the grid space, the higher the probability of finding the optimal value. If no sample points that meet the predetermined accuracy are found, the search scale needs to be further optimized to improve the fineness of the sample; the scale update is calculated using formula (5): Step 8: Update the sample space U (m) ; Based on the optimal center (X (m-1) ,Y (m-1) ,Z (m-1) ) and radius R (m-1) , and the updated scale Generate a new sample space U (m) ; Sample space For X (m) , Y (m) , Z (m) , R (m) All possible combinations of the four feature values can be obtained from formula (6): Step 9: Find the optimal sample; Step 10: Loop through steps 7, 8, and 9 until a sample that meets the predetermined accuracy is found or the predetermined number of iterations N is reached. opt .
2. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 1, a ground-based 3D laser scanning system or a vehicle-mounted 3D laser scanning system is used to collect a point cloud P of a spherical target. i ,y i ,z i )|i=1,2,…,n}, the spherical target point cloud is composed of the measurement point data on the surface of the spherical target, and each measurement point is composed of three coordinate components: X, Y, and Z.
3. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 3, in order to obtain the optimal center and radius of the spherical target, the optimal center and radius judgment criteria are set; here, the arithmetic error measure is used to fit the error E fit Evaluation: Assuming that the fitting center of the spherical target is (a, b, c) and the fitting radius is R, then the fitting error E fit can be determined by formula (1): In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of the spherical target point cloud, and n is the number of measuring points contained in the spherical target point cloud.
4. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 4, the geometric center of gravity According to formula (2), the initial scale According to formula (3), we can get: In the formula, (x i ,y i ,z i ) is the coordinate of the measuring point of point cloud P, n is the number of measuring points of point cloud P; 5. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 5, the initial sample space U (1) The samples in the initial scale The preset number of grid samples N sub Determine together that X (1) , Y (1) , Z (1) , R (1) All possible combinations of the four feature quantities are determined by formula (4): In the formula, The value is 0 because the geometric radius of the spherical target is always greater than 0.
6. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 6, the initial sample space U is determined by using the judgment criteria of step 3. (1) The fitting error is calculated for each sample in the equation, and the sample with the smallest fitting error is selected as the initial optimal solution. In the process of calculating the fitting error, it is necessary to detect whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
7. The adaptive grid search method for spherical target fitting according to claim 1, It is characterized in that In step 9, the updated sample space U is updated using the judgment criteria of step 3. (m) Each sample in the equation is tested for error one by one, and the sample with the smallest fitting error is selected as the optimal solution. In the process of calculating the fitting error, it is tested whether the fitting error of each sample reaches the preset fitting error threshold E. min , if it is reached, the current sample can be used as the final fitting result, otherwise, continue to detect.
