Low overlap ratio warhead point cloud registration method based on line feature detection

Abstract

The present application relates to the field of point cloud registration, and more particularly to a low overlap ratio warhead point cloud registration method based on line feature detection, comprising S1: voxel down-sampling source point cloud and target point cloud; S2: fitting the central axis of the voxel down-sampled source point cloud and target point cloud respectively, and registering the two fitted central axes to realize the coarse registration of the source point cloud and the target point cloud; S3: using Hough transform method to extract linear bullet mark features from the coarse registered source point cloud and target point cloud; S4: finding the best rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud; S5: aligning the source point cloud with the target point cloud by rotating the best rotation angle, and completing the fine registration of the source point cloud and the target point cloud. The present application can complete the accurate registration of bullet mark point cloud data under the condition that the source point cloud and the target point cloud have low overlap ratio.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of point cloud registration, and particularly relates to a low-overlap-ratio warhead point cloud registration method based on line feature detection. BACKGROUND

[0002] Traditional bullet mark detection is to observe and compare bullet mark features by a technician operating a microscope device combined with image processing technology, but this comparison method lacks objective quantitative analysis, the technician has strong subjective judgment, and the bullet mark detection result is not objective and comprehensive. With the development and popularization of various three-dimensional scanning technologies, current bullet mark detection is gradually changing from 2D to 3D measurement, comparison and identification, and the surface topography of a warhead is usually scanned at different angles, and the obtained part point cloud data needs to be registered by using registration technology, so as to register the part point cloud data to the same coordinate system, that is, to register into a complete warhead three-dimensional point cloud model.

[0003] For low-overlap-ratio (the overlap ratio of source point cloud and target point cloud is less than 60%) warhead point cloud data, since the amount of point cloud data in the overlapping part is small, the features that can be extracted are limited, and direct use of source point cloud registration will produce many outlier subsets, the registration process is time-consuming and prone to matching errors. The point cloud registration technology affects the quality of the three-dimensional point cloud model, therefore, the low-overlap-ratio point cloud registration problem is a difficulty in bullet mark detection registration. SUMMARY

[0004] Therefore, the application creates a low-overlap-ratio warhead point cloud registration method based on line feature detection to solve the problem of poor registration accuracy of low-overlap-ratio warhead point cloud data.

[0005] To achieve the above-mentioned purpose, the technical solution of the application creation is as follows:

[0006] A low-overlap-ratio warhead point cloud registration method based on line feature detection, comprising the following steps:

[0007] S1: voxel down-sampling the source point cloud and the target point cloud;

[0008] S2: fitting the middle axis line of the voxel down-sampled source point cloud and target point cloud respectively, and registering the two fitted middle axis lines to realize the coarse registration of the source point cloud and the target point cloud;

[0009] S3: extracting linear bullet mark features from the coarse-registered source point cloud and target point cloud by using the Hough transform method;

[0010] S4: finding the best rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud;

[0011] S5: Rotating the source point cloud by the optimal rotation angle to align with the target point cloud, and completing the fine registration of the source point cloud and the target point cloud.

[0012] Further, in step S1, the specific process of voxel down-sampling the source point cloud and the target point cloud includes:

[0013] A three-dimensional voxel grid is constructed using a loop voxel filtering method, the source point cloud and the target point cloud are gridded, each cube after gridding is a voxel, and all points in each voxel are replaced by a barycenter point, thereby completing the voxel down-sampling.

[0014] Further, in step S2, the specific process of fitting the central axis of the source point cloud and the target point cloud after voxel down-sampling includes:

[0015] S21: Layering the source point cloud and the target point cloud along the Y-axis in the vertical direction, and then selecting four points not in the same plane for each layer 、 、 、 calculating the center coordinates of the circumscribed sphere of the points 、 、 、 The calculation equation is as follows:

[0016] ;

[0017] Wherein, r represents the radius of the circumscribed sphere;

[0018] S22: Fitting the center coordinates calculated for each layer into a straight line, and centering all point coordinates of the source point cloud and the target point cloud based on the fitted straight line to form a new point set , so that each point coordinate in the point set , represents the coordinate of the i-th point, represents the center point coordinate of the point cloud;

[0019] S23: By solving the eigenvalues and eigenvectors of the covariance matrix , find the direction vector , so that the projection variance of the point set in the direction is maximized; wherein, , n represents the number of point coordinates in the point set , represents the transpose of the point set ; after finding the first principal component direction , along the direction , the central axis of the source point cloud and the target point cloud is fitted.​ The direction selects a series of point coordinates, and calculates the projection of each centered point coordinate in the direction , The coordinates of the i-th point in the point set A series of values are uniformly selected between the maximum and minimum , and the point coordinates corresponding to each value on the fitted straight line are obtained .

[0020] Further, in step S2, the two fitted center axes are registered to realize the specific process of coarse registration of the source point cloud and the target point cloud, which includes:

[0021] S24: Determine the direction vector and center point of the two center axes;

[0022] S25: Calculate the rotation matrix using the direction vectors of the two center axes;

[0023] S26: Rotate the center axis of the source point cloud, and obtain the translation vector according to the difference between the center point of the rotated center axis of the source point cloud and the center point of the center axis of the target point cloud;

[0024] S27: Combine the rotation matrix and the translation vector into a transformation matrix to realize the coarse registration of the source point cloud and the target point cloud.

[0025] Further, in step S3, the specific process of linear bullet hole feature extraction of the source point cloud and the target point cloud after coarse registration by using the Hough transform method includes:

[0026] S31: The value of each point in the source point cloud and the target point cloud after voxel downsampling is defined as , which is calculated by the following formula:

[0027]

[0028] , wherein , , represents the K-field covariance matrix eigenvalue of point ;

[0029] If the calculated is greater than or equal to the set threshold τ, the point is identified as a sharp feature point, and is attributed to the target point cloud or the source point cloud ;

[0030] S32: Discretize the parameter space of the Hough transform into a regular icosahedron; ​​​

[0031] S33: mapping the voting results of each sharp feature point in the target point cloud or the source point cloud to each face of the regular icosahedron to form a voting matrix of the parameter space;

[0032] S34: performing linear bullet hole feature detection using the improved Hough transform algorithm.

[0033] Further, the specific process of performing linear bullet hole feature detection using the improved Hough transform algorithm in step S34 includes:

[0034] S341: in the voting matrix, a straight line is represented as parameters (p, q, f), where p is the shortest distance from a point to the straight line, q and f are direction angles;

[0035] S342: an accumulator is established, each point in the source point cloud and the target point cloud votes in the accumulator according to its normal vector or local feature, and a combination of parameters (p, q, f) higher than a set number of votes is found through peak detection to obtain bullet hole straight line features or bullet hole curve features.

[0036] Further, in the process of performing linear bullet hole feature detection using the improved Hough transform algorithm, an iterative method is used to optimize the estimation of linear parameters, and in each iteration, least squares is used as an error function for linear fitting to correct the current linear parameters.

[0037] Further, in step S4, the specific process of finding the best rotation angle based on the linear bullet hole features extracted from the source point cloud and the target point cloud includes:

[0038] S41: using a three-dimensional rotation matrix to rotate the source point cloud to obtain a rotated source point cloud ; wherein represents the rotation angle;

[0039] S42: according to the target point cloud and the rotated source point cloud , calculate the fitness function , the calculation formula is:

[0040]

[0041] wherein represents the dot product, represents the absolute value of the dot product of vectors and , and vectors and belong to the target point cloud and the source point cloud , and The parameters represent the Gaussian function. and Indicates the target point cloud Heyuan Point Cloud Feature descriptor for the coordinates of the corresponding point in the middle;

[0042] S43: Repeat steps S41 and S42 to find the maximum fitness function within the number of iterations. The corresponding rotation angle , as the optimal rotation angle.

[0043] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0044] This invention first filters the source and target point clouds, then achieves coarse registration of the source and target point clouds by fitting the central axis. Next, it uses the Hough transform method to detect linear bullet hole features. Finally, it finds the optimal rotation angle based on the detected linear bullet hole features and achieves fine registration of the source and target point clouds based on the optimal rotation angle. Based on the above operations, this invention can complete the accurate registration of bullet hole point cloud data when the overlap rate between the source and target point clouds is 20% to 60%. Attached Figure Description

[0045] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0046] Figure 1 This is a flowchart illustrating the low overlap warhead point cloud registration method based on line feature detection according to an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram illustrating the principle of the low overlap rate warhead point cloud registration method based on line feature detection according to an embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating the principle of finding the center of the circumscribed ball according to an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the process for finding the optimal rotation angle according to an embodiment of the present invention. Detailed Implementation

[0050] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not constitute a limitation on the present application.

[0051] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0052] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second" and the like are only used for description purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined with "first", "second" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0053] In the description of the present application, it should be noted that unless otherwise specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood through specific circumstances.

[0054] The present application will be described in detail below with reference to the drawings Figures 1-4 and embodiments.

[0055] As shown in Figure 1 and Figure 2 , the low overlap ratio warhead point cloud registration method based on line feature detection provided by the embodiments of the present application comprises the following steps:

[0056] S1: voxel down-sampling is performed on the source point cloud and the target point cloud.

[0057] Since the collected point cloud data of the bullet surface is large, the point cloud data with noise will reduce the data quality, and the large point cloud data will greatly increase the running time, in order to reduce the point cloud data and filter out the noise, the application uses a loop voxel filtering method to construct a three-dimensional voxel grid, grids the source point cloud and the target point cloud, each cube after gridding is a voxel, calculates the maximum and minimum values of the X, Y and Z axes (a conventional three-dimensional coordinate system in the point cloud, the Y axis is the coordinate axis of the vertical direction, and the X and Z axes are the coordinate axes of the horizontal direction) of the source point cloud and the target point cloud, establishes a three-dimensional bounding box according to the values, and divides the bounding box into cubes with a specified voxel size, replaces all points in each cube with the center of gravity of the cube, so as to reduce the size of the point cloud and complete the voxel down-sampling of the source point cloud and the target point cloud.

[0058] A large number of isolated point noises will occupy the operation and storage resources of the data processor, first, the bullet point cloud data is filtered for noise points, which can greatly reduce the calculation amount of data processing, prevent data blocking, and improve the efficiency of subsequent point cloud processing.

[0059] S2: The central axis fitting is respectively performed on the source point cloud and the target point cloud after the voxel down-sampling, and the two fitted central axes are registered to realize the coarse registration of the source point cloud and the target point cloud.

[0060] As shown in Figure 3 , the specific process of fitting the central axis of the source point cloud and the target point cloud after the voxel down-sampling includes:

[0061] S21: The source point cloud and the target point cloud are layered along the Y axis, and then four points not in the same plane are selected from each layer 、 、 、 The center coordinates of the circumscribed sphere of the points 、 、 、 are calculated , and the equation is as follows:

[0062] ;

[0063] Wherein, r represents the radius of the circumscribed sphere.

[0064] By constructing a linear equation, the SVD is applied to find the best solution of the sphere center coordinates in the least square sense .

[0065] S22: The center coordinates calculated from each layer are fitted into a straight line, and the center coordinates of all points of the source point cloud and the target point cloud are centralized based on the fitted straight line to form a new point set , such that the point set has the maximum variance in the direction of , represents the coordinate of the i-th point in the point set , and

[0066] The center of the source point cloud and the target point cloud is taken as the origin (0, 0, 0), and the coordinates of all points in the source point cloud and the target point cloud are centralized, that is, the distribution of the coordinates of all points in the source point cloud and the target point cloud is moved to a coordinate system with the origin (0, 0, 0) as the center, forming a new point set .

[0067] S23: Find the direction vector by solving the eigenvalues and eigenvectors of the covariance matrix , such that the point set has the maximum variance in the direction of ; wherein, , n represents the number of point coordinates in the point set , that is, the point set includes n point coordinates, represents the transpose of the point set ; after finding the first principal component direction , a series of point coordinates are selected along the direction , and the projection of each centralized point coordinate in the direction is calculated , represents the coordinate of the i-th point in the point set , and a series of values are uniformly selected between the maximum and minimum , to obtain the point coordinates on the fitted straight line corresponding to each value.

[0068] Finally, the fitted center axis of the source point cloud and the fitted center axis of the target point cloud are registered to obtain a better coarse registration position of the source point cloud and the target point cloud with low overlap on the same axis, and the specific process is as follows:

[0069] S24: Determine the direction vector and center point of the two center axes.

[0070] S25: Calculate the rotation matrix using the direction vectors of the two center axes.

[0071] S26: Obtain the translation vector by rotating the center axis of the source point cloud according to the difference between the center point of the rotated center axis of the source point cloud and the center point of the center axis of the target point cloud.

[0072] ​​S27: Combine the rotation matrix and the translation vector into a transformation matrix to achieve the coarse registration of the source point cloud and the target point cloud.

[0073] S3: Extract the linear bullet hole feature of the source point cloud and the target point cloud after coarse registration by using the Hough transform method.

[0074] The specific process of step S3 includes:

[0075] S31: For each point in the source point cloud and the target point cloud after voxel downsampling , the value is defined as , which is calculated by the following formula:

[0076]

[0077] , wherein , , represents the K-field covariance matrix eigenvalue of point .

[0078] If the calculated is greater than or equal to the set threshold τ, the point is identified as a sharp feature point, and is attributed to the target point cloud or the source point cloud . If the calculated is less than the set threshold τ, the point is identified as a smooth feature point, and is not attributed to the target point cloud or the source point cloud . In this way, the purpose of enhancing the sharp feature of the bullet hole point cloud is achieved.

[0079] S32: Discretize the parameter space of the Hough transform into a regular icosahedron.

[0080] This step can effectively find the linear bullet hole feature.

[0081] S33: Map the voting results of each sharp feature point in the target point cloud or the source point cloud to each face of the regular icosahedron to form a voting matrix of the parameter space.

[0082] S34: Perform linear bullet hole feature detection using the improved Hough transform algorithm.

[0083] In step S34, the specific process of performing linear bullet hole feature detection using the improved Hough transform algorithm includes:

[0084] S341: In the voting matrix, represent a straight line as parameters (ρ, θ, φ), where ρ is the shortest distance from the point to the straight line, and θ and φ are the direction angles.

[0085] S342: Establish an accumulator, and vote in the accumulator according to the normal vector or local feature of each point in the source point cloud and the target point cloud, find the combination of parameters (p, q, f) higher than the set number of votes through peak detection, obtain the bullet hole straight line feature or bullet hole curve feature, map the bullet hole straight line feature or bullet hole curve feature back to the three-dimensional space, and complete the detection of the linear bullet hole feature.

[0086] In the process of detecting the linear bullet hole feature using the improved Hough transform algorithm, the iterative method is used to optimize the estimation of the linear parameters, and in each iteration, the least square is used as the error function for linear fitting to correct the current linear parameters, so as to improve the accuracy of detection.

[0087] The Hough transform method can more accurately identify the fine bullet hole features on the surface of the bullet.

[0088] S4: Find the best rotation angle based on the linear bullet hole features extracted from the source point cloud and the target point cloud.

[0089] The specific process of step S4 includes:

[0090] S41: Use a three-dimensional rotation matrix to rotate the source point cloud to obtain the rotated source point cloud ; wherein, represents the rotation angle.

[0091] S42: According to the target point cloud and the rotated source point cloud , calculate the fitness function , and the calculation formula is:

[0092]

[0093] wherein, represents the dot product; represents the absolute value of the dot product of the vector and the vector ; the vector and belong to the target point cloud and the source point cloud , respectively, and have been normalized; and represent the feature descriptors of the corresponding point coordinates in the target point cloud and the source point cloud ; and The parameters of the Gaussian function, adjusting the form of the function, the exponential part inside the function is the square of the difference of the vector dot product, for the Gaussian term, give a certain "reward" to the matching pair of vectors, while almost no effect on the non-matching pair of vectors.

[0094] S43: repeat steps S41 and S42, find the maximum fitness function in the number of iterations The corresponding rotation angle , as the best rotation angle.

[0095] According to the flowchart shown in Figure 4 , find the maximum fitness function in the number of iterations The corresponding Angle, that is, the best rotation angle.

[0096] S5: Rotate the source point cloud by the best rotation angle and align it with the target point cloud, complete the fine registration of the source point cloud and the target point cloud.

[0097] It should be understood that various forms of the flowchart shown above can be used to reorder, add or delete steps. For example, each step described in the present disclosure can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present disclosure can be achieved, which is not limited herein.

[0098] The above specific embodiments do not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for registering low-overlap warhead point clouds based on line feature detection, characterized in that, Includes the following steps: S1: Perform voxel downsampling on the source point cloud and the target point cloud; S2: Perform centerline fitting on the source point cloud and target point cloud after voxel downsampling, and register the two fitted centerlines to achieve coarse registration between the source point cloud and the target point cloud. S3: Use the Hough transform method to extract linear bullet hole features from the coarsely registered source and target point clouds; In step S3, the specific process of extracting linear bullet hole features from the coarsely registered source and target point clouds using the Hough transform method includes: S31: For each point in the source and target point clouds after voxel downsampling The value is defined as Calculated using the following formula: in, , , Point eigenvalues ​​of the K-domain covariance matrix; If calculated If the value is greater than or equal to the set threshold τ, it is considered a valid point. Sharp feature points are assigned to the target point cloud. Or source cloud ; S32: Discretize the parameter space of the Hough transform into a regular icosahedron; S33: Transfer the target point cloud Or source cloud The voting result of each sharp feature point in the matrix is ​​mapped to each face of the icosahedron, forming a voting matrix in the parameter space; S34: Linear bullet hole feature detection using an improved Hough transform algorithm; S4: Find the optimal rotation angle based on the linear bullet hole features extracted from the source point cloud and the target point cloud; In step S4, the specific process of finding the optimal rotation angle based on the linear bullet hole features extracted from the source point cloud and the target point cloud includes: S41: Using a 3D rotation matrix Source point cloud Rotate the source point cloud to obtain the rotated source point cloud. ;in, Indicates the rotation angle; S42: Based on the target point cloud With the rotated source cloud Calculate the fitness function The calculation formula is: in, Dot product, Representing vectors sum vector The absolute value of the dot product of vectors and They belong to the target point cloud. Heyuan Point Cloud , and The parameters represent the Gaussian function. and Indicates the target point cloud Heyuan Point Cloud Feature descriptor for the coordinates of the corresponding point in the middle; S43: Repeat steps S41 and S42 to find the maximum fitness function within the number of iterations. The corresponding rotation angle , as the optimal rotation angle; S5: Rotate the source point cloud to the optimal rotation angle and align it with the target point cloud to complete the fine registration of the source and target point clouds.

2. The method for low-overlap warhead point cloud registration based on line feature detection according to claim 1, characterized in that, In step S1, the specific process of voxel downsampling of the source point cloud and the target point cloud includes: A 3D voxel mesh is constructed using the cyclic voxel filtering method. The source point cloud and the target point cloud are meshed. Each cube after meshing is a voxel, and all points in each voxel are replaced by the centroid point to complete the voxel downsampling.

3. The method for low-overlap warhead point cloud registration based on line feature detection according to claim 1, characterized in that, In step S2, the specific process of fitting the centerline of the source point cloud and the target point cloud after voxel downsampling includes: S21: Divide the source point cloud and the target point cloud into layers along the vertical Y-axis, and then arbitrarily select four non-coplanar points for each layer. , , , Calculation points , , , Coordinates of the center of the circumsphere The calculation equation is as follows: ； in, r Indicates the radius of the circumscribed sphere; S22: Fit the center coordinates of the spheres calculated at each layer to a straight line, and center the coordinates of all points in the source and target point clouds based on the fitted straight line to form a new point set. This makes the point set Coordinates of each point , This represents the coordinates of the nth point. Indicates the coordinates of the center point of the point cloud; S23: Solving the covariance matrix Using eigenvalues ​​and eigenvectors, find the direction vector. This makes the point set exist Maximize the projection variance in the direction; where, , 𝑛 represents the point set The number of midpoint coordinates Represents a point set The transpose; in finding the direction of the first principal component Afterwards, along Select a series of point coordinates and calculate the coordinates of each centered point. Projection in direction , Represents a point set The coordinates of the i-th point in the range of maximum and minimum values ​​are... A series of uniformly selected The value is obtained by fitting the line corresponding to each... The point coordinates of the value .

4. The method for low-overlap warhead point cloud registration based on line feature detection according to claim 1 or 3, characterized in that, In step S2, the two fitted central axes are registered to achieve coarse registration between the source point cloud and the target point cloud. The specific process includes: S24: Determine the direction vectors and center points of the two central axes; S25: Calculate the rotation matrix using the direction vectors of the two central axes; S26: By rotating the central axis of the source point cloud, the translation vector is obtained based on the difference between the center point of the central axis of the rotated source point cloud and the center point of the central line of the target point cloud. S27: Combine the rotation matrix and translation vector into a single transformation matrix to achieve coarse registration between the source point cloud and the target point cloud.

5. The method for low-overlap warhead point cloud registration based on line feature detection according to claim 1, characterized in that, In step S34, the specific process of using the improved Hough transform algorithm for linear bullet hole feature detection includes: S341: In the voting matrix, the line is represented as parameters (ρ, θ, φ), where ρ is the shortest distance from the point to the line, and θ and φ are the direction angles; S342: Establish an accumulator. For each point in the source point cloud and the target point cloud, vote in the accumulator based on its own normal vector or local features. Find the combination of parameters (ρ, θ, φ) that are higher than the set number of votes through peak detection to obtain the bullet mark straight line feature or bullet mark curve feature.

6. The method for low-overlap warhead point cloud registration based on line feature detection according to claim 1, characterized in that, In the process of linear bullet hole feature detection using the improved Hough transform algorithm, the estimation of linear parameters is optimized by using an iterative method. In each iteration, least squares is used as the error function for linear fitting to correct the current linear parameters.

Images