Low-overlapping-rate bullet point cloud registration method based on line feature detection

Abstract

The invention relates to the field of point cloud registration, in particular to a low-overlapping-rate bullet point cloud registration method based on line feature detection, which comprises the following steps: S1, performing voxel downsampling on a source point cloud and a target point cloud; s2, respectively performing central axis fitting on the source point cloud and the target point cloud after voxel downsampling, and performing registration on the two fitted central axes to realize coarse registration of the source point cloud and the target point cloud; s3, performing linear bullet mark feature extraction on the source point cloud and the target point cloud after coarse registration by using a Hough transform method; s4, searching an optimal rotation angle based on linear bullet mark features extracted from the source point cloud and the target point cloud; and S5, rotating the source point cloud by the optimal rotation angle to align the source point cloud with the target point cloud, and finishing fine registration of the source point cloud and the target point cloud. According to the invention, under the condition that the overlapping rate of the source point cloud and the target point cloud is low, accurate registration of the bullet mark point cloud data can be completed.

Description

Technical Field

[0001] The invention belongs to the technical field of point cloud registration, and in particular relates to a low-overlap rate bullet point cloud registration method based on line feature detection. Background Art

[0002] my country has strict control over guns, but the number of gun-related cases has increased year by year. Since the marks on bullets and shells are highly reflective of guns, the bullet marks in gun-related cases are of great significance to the detection of gun-related cases. Bullet marks are very subtle, usually at the micron level, so traditional bullet mark detection is done by technicians using microscope equipment combined with image processing technology to observe and compare bullet mark features. However, this comparison method lacks objective quantitative analysis, and technicians are highly subjective, so the bullet mark detection results are not objective and comprehensive enough. With the development and popularization of various three-dimensional scanning technologies, bullet mark detection is gradually shifting from 2D to 3D measurement, comparison and identification. To obtain the surface morphology of the bullet, the bullet is usually scanned at different angles, and the point cloud data of each part needs to be registered using registration technology, so that the point cloud data of each part is registered to the same coordinate system, that is, registered into a complete three-dimensional point cloud model of the bullet.

[0003] For bullet point cloud data with low overlap rate (the overlap rate between the source point cloud and the target point cloud is less than 60%), the amount of point cloud data in the overlapping part is small, and the features that can be extracted are limited. Directly using the source point cloud for registration will produce many outlier subsets, and the registration process is time-consuming and prone to matching errors. Point cloud registration technology affects the quality of the 3D point cloud model. Therefore, the problem of point cloud registration with low overlap rate is a difficult point in bullet mark detection registration. Summary of the invention

[0004] In view of this, the present invention aims to provide a low-overlap bullet point cloud registration method based on line feature detection to solve the problem of poor registration accuracy of low-overlap bullet point cloud data.

[0005] To achieve the above object, the technical solution created by the present invention is implemented as follows: A low overlap bullet point cloud registration method based on line feature detection comprises the following steps: S1: voxel downsampling of source point cloud and target point cloud; S2: Fitting the central axis of the source point cloud and the target point cloud after voxel downsampling respectively, and registering the two fitted central axes to achieve coarse registration of the source point cloud and the target point cloud; S3: Linear bullet mark feature extraction is performed on the source point cloud and target point cloud after rough registration using the Hough transform method; S4: Find the optimal rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud; S5: Rotate the source point cloud to the optimal rotation angle and align it with the target point cloud to complete the precise registration of the source point cloud and the target point cloud.

[0006] Furthermore, in step S1, the specific process of performing voxel downsampling on the source point cloud and the target point cloud includes: A 3D voxel grid is constructed using a circular voxel filtering method to grid the source point cloud and the target point cloud. Each cube after gridding is a voxel, and all points in each voxel are replaced by the centroid point to complete voxel downsampling.

[0007] Furthermore, in step S2, the specific process of performing central axis fitting on the source point cloud and the target point cloud after voxel downsampling includes: S21: Layer the source point cloud and the target point cloud along the vertical Y axis, and then arbitrarily select four non-coplanar points for each layer. , , , , calculation point , , , The coordinates of the center of the circumscribed sphere , the calculation equation is as follows: ; in, r represents the radius of the circumscribed sphere; S22: Fit the sphere center coordinates calculated at each layer into a straight line, and centralize the coordinates of all points in the source point cloud and the target point cloud based on the fitted straight line to form a new point set , so that the point set The coordinates of each point in , represents the coordinates of the 𝑖th point, Indicates the coordinates of the center point of the point cloud; S23: By solving the covariance matrix The eigenvalues ​​and eigenvectors of , so that the point set exist The projection variance in the direction is maximized; where, , 𝑛 represents the point set The number of midpoint coordinates, Represents a point set The transpose of Afterwards, along Select a series of point coordinates in the direction and calculate the coordinates of each centralized point in Projection in direction , Represents a point set The coordinates of the i-th point in the maximum and minimum Evenly select a series of value, and obtain the fitted straight line corresponding to each The point coordinates of the value .

[0008] Furthermore, in step S2, the two fitted central axes are aligned, and the specific process of achieving the rough alignment of the source point cloud and the target point cloud 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, a translation vector is obtained according to the difference between the center point of the central axis of the rotated source point cloud and the center point of the center line of the target point cloud; S27: Combine the rotation matrix and the translation vector into a transformation matrix to achieve coarse alignment between the source point cloud and the target point cloud.

[0009] Furthermore, in step S3, the specific process of extracting linear bullet mark features from the roughly registered source point cloud and target point cloud using the Hough transform method includes: S31: For each point in the source point cloud and the target point cloud after voxel downsampling The value of is defined as , calculated by the following formula:

[0010] in, , , Indicate point The K-domain covariance matrix eigenvalues; If the calculated If it is greater than or equal to the set threshold τ, it is considered a point is a sharp feature point, belonging to the target point cloud Or source point cloud ; S32: Discretize the parameter space of the Hough transform into a regular icosahedron; S33: Target point cloud Or source point cloud The voting results of each sharp feature point in are mapped to each face of the regular icosahedron to form a voting matrix in the parameter space; S34: Linear bullet mark feature detection using improved Hough transform algorithm.

[0011] Furthermore, in step S34, the specific process of using the improved Hough transform algorithm to perform linear bullet mark feature detection includes: S341: In the voting matrix, the straight line is represented as parameters (ρ, θ, φ), where ρ is the shortest distance from the point to the straight line, and θ and φ are direction angles; S342: Establish an accumulator, and vote for each point in the source point cloud and the target point cloud in the accumulator according to its own normal vector or local feature. Through peak detection, find the combination of parameters (ρ, θ, φ) with a higher number of votes than the set number, and obtain the bullet mark straight line feature or bullet mark curve feature.

[0012] Furthermore, in the process of using the improved Hough transform algorithm to detect linear bullet marks, an iterative method is used to optimize the estimation of linear parameters. In each iteration, the least square method is used as the error function for linear fitting to correct the current linear parameters.

[0013] Furthermore, in step S4, the specific process of finding the optimal rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud includes: S41: Using a 3D rotation matrix Source point cloud Rotate to get the rotated source point cloud ;in, Indicates the rotation angle; S42: Based on the target point cloud With the rotated source point cloud , calculate the fitness function , the calculation formula is:

[0014] in, represents dot product, Representation vector and vector The absolute value of the dot product, vector and Belong to the target point cloud and source point cloud , and represents the parameters of the Gaussian function, and Indicates that in the target point cloud and source point cloud Feature descriptors of the corresponding point coordinates in ; 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.

[0015] Compared with the prior art, the invention can achieve the following beneficial effects: The present invention first filters the source point cloud and the target point cloud, and then realizes the coarse alignment of the source point cloud and the target point cloud by fitting the central axis. Then, the linear bullet mark features are detected by using the Hough transform method. Finally, the optimal rotation angle is found based on the detected linear bullet mark features, and the source point cloud and the target point cloud are precisely aligned based on the optimal rotation angle. Based on the above operations, the present invention can complete the precise alignment of bullet mark point cloud data when the overlap rate of the source point cloud and the target point cloud is 20%~60%. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings constituting part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation on the present invention. In the drawings: Figure 1 It is a flowchart of a method for low overlap rate bullet point cloud registration based on line feature detection according to an embodiment of the invention; Figure 2 It is a schematic diagram of the principle of a low-overlap bullet point cloud registration method based on line feature detection according to an embodiment of the invention; Figure 3 It is a schematic diagram of the principle of finding the center of a circumscribed ball according to an embodiment of the invention; Figure 4 It is a schematic diagram of a process of finding an optimal rotation angle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and do not constitute a limitation of the invention.

[0018] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0019] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0020] In the description of the invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the invention can be understood according to specific circumstances.

[0021] The following will refer to Figure 1-Figure 4 The invention is described in detail with reference to the embodiments.

[0022] like Figure 1 and Figure 2 As shown, the low overlap bullet point cloud registration method based on line feature detection provided by the present invention comprises the following steps: S1: Perform voxel downsampling on the source and target point clouds.

[0023] Since there are a lot of point cloud data collected on the surface of the bullet, the point cloud data with noise will reduce the data quality, and more point cloud data will greatly increase the running time. In order to reduce the data of the point cloud data and filter out the noise, the present invention uses a cyclic voxel filtering method to construct a three-dimensional voxel grid, grids the source point cloud and the target point cloud, and each cube after gridding is a voxel. The maximum and minimum values ​​of the X, Y, and Z axes (which are the conventional three-dimensional coordinate systems in the point cloud, the Y axis is the vertical coordinate axis, and the X axis and the Z axis are the two horizontal coordinate axes) of the source point cloud and the target point cloud are calculated, and a three-dimensional bounding box is established according to these values, and the bounding box is divided into cubes of a specified voxel size, and all the points in the cube are replaced with the centroid point of each cube to reduce the point cloud size, and complete the voxel downsampling of the source point cloud and the target point cloud.

[0024] A large amount of isolated point noise will occupy the computing and storage resources of the data processor. First, filtering the noise points of the bullet point cloud data will greatly reduce the amount of data processing calculations, prevent data blocking, and improve the efficiency of subsequent point cloud processing.

[0025] S2: Perform central axis fitting on the source point cloud and target point cloud after voxel downsampling respectively, and align the two fitted central axes to achieve coarse alignment of the source point cloud and the target point cloud.

[0026] like Figure 3 As shown in FIG. 1 , the specific process of fitting the central axis of the source point cloud and the target point cloud after voxel downsampling includes: S21: Layer the source point cloud and the target point cloud along the Y axis, and then arbitrarily select four non-coplanar points for each layer , , , , calculation point , , , The coordinates of the center of the circumscribed sphere , the calculation equation is as follows: ; in, r Represents the radius of the circumscribed sphere.

[0027] By constructing a linear equation, SVD is applied to find the coordinates of the sphere center in the least squares sense. .

[0028] S22: Fit the sphere center coordinates calculated at each layer into a straight line, and centralize the coordinates of all points in the source point cloud and the target point cloud based on the fitted straight line to form a new point set , so that the point set The coordinates of each point in , represents the coordinates of the 𝑖th point, Indicates the coordinates of the center point of the point cloud.

[0029] Taking the endpoint of the fitted straight line as the origin (0,0,0), the coordinates of all points in the source point cloud and the target point cloud are centered. In fact, the distribution of the coordinates of all points in the source point cloud and the target point cloud is moved to the coordinate system centered on the origin (0,0,0) to form a new point set. .

[0030] S23: By solving the covariance matrix The eigenvalues ​​and eigenvectors of , so that the point set exist The projection variance in the direction is maximized; where, , 𝑛 represents the point set The number of midpoint coordinates, i.e. the point set include n Point coordinates, Represents a point set The transpose of Afterwards, along Select a series of point coordinates in the direction and calculate the coordinates of each centralized point in Projection in direction , Represents a point set The coordinates of the i-th point in the maximum and minimum Evenly select a series of value, and obtain the fitted straight line corresponding to each The point coordinates of the value .

[0031] Finally, the central axis fitted by the source point cloud is aligned with the central axis fitted by the target point cloud to obtain a good rough alignment position of the source point cloud and the target point cloud with low overlap rate on the same axis. The specific process is as follows: S24: Determine the direction vectors and center points of the two central axes.

[0032] S25: Calculate the rotation matrix using the direction vectors of the two central axes.

[0033] S26: By rotating the central axis of the source point cloud, a translation vector is obtained according to the difference between the center point of the central axis of the rotated source point cloud and the center point of the center line of the target point cloud.

[0034] S27: Combine the rotation matrix and the translation vector into a transformation matrix to achieve coarse alignment between the source point cloud and the target point cloud.

[0035] S3: Use the Hough transform method to extract linear bullet mark features from the source point cloud and target point cloud after rough registration.

[0036] The specific process of step S3 includes: S31: For each point in the source point cloud and the target point cloud after voxel downsampling The value of is defined as , calculated by the following formula:

[0037] in, , , Indicate point The K-domain covariance matrix eigenvalues; If the calculated If it is greater than or equal to the set threshold τ, it is considered a point is a sharp feature point, belonging to the target point cloud Or source point cloud , if the calculated If the value is less than the set threshold value τ, it is considered as a point For smooth feature points, not attributed to the target point cloud Or source point cloud , so as to achieve the purpose of enhancing the sharp features of the bullet point cloud.

[0038] S32: Discretize the parameter space of the Hough transform into a regular icosahedron.

[0039] This step can effectively find linear bullet mark features.

[0040] S33: Target point cloud Or source point cloud The voting results of each sharp feature point in are mapped to each face of the regular icosahedron to form a voting matrix in the parameter space.

[0041] S34: Linear bullet mark feature detection using improved Hough transform algorithm.

[0042] In step S34, the specific process of using the improved Hough transform algorithm to perform linear bullet mark feature detection includes: S341: In the voting matrix, a straight line is represented as parameters (ρ, θ, φ), where ρ is the shortest distance from a point to the straight line, and θ and φ are direction angles.

[0043] S342: Establish an accumulator, and vote for each point in the source point cloud and the target point cloud in the accumulator according to its own normal vector or local feature. Find the combination of parameters (ρ, θ, φ) with a higher number of votes than the set number through peak detection, obtain the bullet mark straight line feature or bullet mark curve feature, map the bullet mark straight line feature or bullet mark curve feature back to the three-dimensional space, and complete the detection of linear bullet mark features.

[0044] In the process of linear bullet mark feature detection using the improved Hough transform algorithm, the estimation of linear parameters is optimized using an iterative method. In each iteration, the least squares method is used as the error function for linear fitting to correct the current linear parameters in order to improve the accuracy of detection.

[0045] The Hough transform method can be used to more accurately identify the subtle bullet mark features on the bullet surface.

[0046] S4: Finding the optimal rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud.

[0047] The specific process of step S4 includes: S41: Using a 3D rotation matrix Source point cloud Rotate to get the rotated source point cloud ;in, Indicates the rotation angle.

[0048] S42: Based on the target point cloud With the rotated source point cloud , calculate the fitness function , the calculation formula is:

[0049] in, represents dot product; Representation vector and vector The absolute value of the dot product; vector and Belong to the target point cloud and source point cloud , has been normalized; and Indicates that in the target point cloud and source point cloud Feature descriptors of the corresponding point coordinates in ; and Represents the parameters of the Gaussian function and adjusts the form of the function. The exponential part inside the function is the square of the difference of the vector dot product, which is the Gaussian term. It gives a certain "reward" to the vectors of the matching pairs, while having almost no effect on the vectors of the unmatched pairs.

[0050] 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.

[0051] according to Figure 4 The process shown in the figure finds the maximum fitness function within the number of iterations. The corresponding Angle is the optimal rotation angle.

[0052] S5: Rotate the source point cloud to the optimal rotation angle and align it with the target point cloud to complete the precise registration of the source point cloud and the target point cloud.

[0053] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0054] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A low overlap bullet point cloud registration method based on line feature detection, characterized in that: The steps include: S1: voxel downsampling of source point cloud and target point cloud; S2: Fitting the central axis of the source point cloud and the target point cloud after voxel downsampling respectively, and registering the two fitted central axes to achieve coarse registration of the source point cloud and the target point cloud; S3: Linear bullet mark feature extraction is performed on the source point cloud and target point cloud after rough registration using the Hough transform method; S4: Find the optimal rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud; S5: Rotate the source point cloud to the optimal rotation angle and align it with the target point cloud to complete the precise registration of the source point cloud and the target point cloud.

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

3. The low overlap bullet point cloud registration method based on line feature detection according to claim 1 is characterized in that: In step S2, the specific process of performing central axis fitting on the source point cloud and the target point cloud after voxel downsampling includes: S21: Layer the source point cloud and the target point cloud along the vertical Y axis, and then arbitrarily select four non-coplanar points for each layer. , , , , calculation point , , , The coordinates of the center of the circumscribed sphere , the calculation equation is as follows: ； in, r represents the radius of the circumscribed sphere; S22: Fit the sphere center coordinates calculated at each layer into a straight line, and centralize the coordinates of all points in the source point cloud and the target point cloud based on the fitted straight line to form a new point set , so that the point set The coordinates of each point in , represents the coordinates of the 𝑖th point, Indicates the coordinates of the center point of the point cloud; S23: By solving the covariance matrix The eigenvalues ​​and eigenvectors of , so that the point set exist The projection variance in the direction is maximized; where, , 𝑛 represents the point set The number of midpoint coordinates, Represents a point set The transpose of Afterwards, along Select a series of point coordinates in the direction and calculate the coordinates of each centralized point in Projection in direction , Represents a point set The coordinates of the i-th point in the maximum and minimum Evenly select a series of value, and obtain the fitted straight line corresponding to each The point coordinates of the value .

4. The low overlap bullet point cloud registration method based on line feature detection according to claim 1 or 3, characterized in that: In step S2, the two fitted central axes are aligned, and the specific process of achieving coarse alignment of the source point cloud and the target point cloud 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, a translation vector is obtained according to the difference between the center point of the central axis of the rotated source point cloud and the center point of the center line of the target point cloud; S27: Combine the rotation matrix and the translation vector into a transformation matrix to achieve coarse alignment between the source point cloud and the target point cloud.

5. The low overlap bullet point cloud registration method based on line feature detection according to claim 1, characterized in that: In step S3, the specific process of extracting linear bullet mark features from the roughly registered source point cloud and target point cloud using the Hough transform method includes: S31: For each point in the source point cloud and the target point cloud after voxel downsampling The value of is defined as , calculated by the following formula: in, , , Indicate point The K-domain covariance matrix eigenvalues; If the calculated If it is greater than or equal to the set threshold τ, it is considered a point is a sharp feature point, belonging to the target point cloud Or source point cloud ; S32: Discretize the parameter space of the Hough transform into a regular icosahedron; S33: Target point cloud Or source point cloud The voting results of each sharp feature point in are mapped to each face of the regular icosahedron to form a voting matrix in the parameter space; S34: Linear bullet mark feature detection using improved Hough transform algorithm.

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

7. The low overlap bullet point cloud registration method based on line feature detection according to claim 5, characterized in that: In the process of linear bullet mark feature detection using the improved Hough transform algorithm, the estimation of linear parameters is optimized using an iterative method. In each iteration, the least squares method is used as the error function for linear fitting to correct the current linear parameters.

8. The low overlap bullet point cloud registration method based on line feature detection according to claim 5, characterized in that: In step S4, the specific process of finding the optimal rotation angle based on the linear bullet mark features extracted from the source point cloud and the target point cloud includes: S41: Using a 3D rotation matrix Source point cloud Rotate to get the rotated source point cloud ;in, Indicates the rotation angle; S42: Based on the target point cloud With the rotated source point cloud , calculate the fitness function , the calculation formula is: in, represents dot product, Representation vector and vector The absolute value of the dot product, vector and Belong to the target point cloud and source point cloud , and represents the parameters of the Gaussian function, and Indicates that in the target point cloud and source point cloud Feature descriptors of the corresponding point coordinates in ; 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.

Images