A point cloud registration method and system for gaps in automobile panels

Abstract

The present invention discloses a point cloud registration method for automobile cover gaps, including: using a line structured light camera to shoot automobile cover gaps, and obtaining multiple automobile cover gap point cloud data sets after shooting; using a small arc nested loop fitting algorithm based on the Bessel equation to achieve small arc fitting in the curve; using a CPD algorithm to pre-process multiple groups of automobile cover gap point cloud data sets to obtain an optimal conversion matrix; using the CPD algorithm to register two groups of pre-processed data to obtain a complete automobile cover gap profile; rotating, translating, and scaling matrices according to the registration results, and performing coordinate transformation on the original data. Compared with the existing gap measurement method, the gap profile of the present invention after registration is more complete, more accurate for gap measurement, and the entire measurement process does not require human intervention, and the measurement efficiency is higher.

Description

Technical Field

[0001] The invention relates to the technical field of visual inspection and measurement, and in particular to a point cloud registration method and system for automobile cover gaps. Background Art

[0002] With the implementation of "Smart Manufacturing 2025", people's material living standards continue to improve, and consumers have higher and higher requirements for the quality of automobile appearance. Among them, the quality of the gap and flushness of automobile covering parts is an important aspect of the quality evaluation of the whole vehicle, which directly affects the appearance, sealing, vibration resistance, wind and rain resistance and other issues of the whole vehicle. Therefore, the majority of car owners and car companies have higher and higher requirements for the accuracy of gap and flushness. Machine vision online measurement technology has been increasingly used as a means of detecting the quality of body matching, and has gradually become the mainstream trend of measurement. Due to the camera field of view and the complex structure of the part to be tested, the gap contour of the automobile covering part cannot be obtained in a single scan. The traditional method of multi-angle measurement combined with coherent point drift registration has the problem of poor accuracy and cannot meet the requirements of body measurement.

[0003] At present, common gap and flushness inspection methods include: (1) Traditional contact measurement is manual measurement using simple tools such as feeler gauges and flushness gauges, which generally have problems such as large measurement errors, low efficiency, and poor repeatability; (2) Advanced non-contact measurement using visual cameras has the advantages of high accuracy and easy automation, and has gradually become an important method for automobile quality assessment and has been widely used. Nowadays, there are more and more cases of using machine vision technology for high-precision inspection in engineering. However, in actual engineering applications, since the matching gap of automobile panels is usually less than 8 mm, when using a single-angle camera to shoot, there will be a problem of incomplete collection of the internal contour of the gap. Therefore, a robot carrying a camera is often used to perform multi-angle scanning and use point cloud registration to integrate multi-angle shooting data to piece together relatively complete contour structure data to expand the camera point collection range and improve measurement accuracy.

[0004] In terms of camera function selection, 3D structured light cameras have a wide range of applications, but they are expensive, sensitive to light, and complex to configure and calibrate. The gaps in the automotive panels that this study focuses on have a single shape, so a 2D camera with a low price, less point cloud, good stability, and easy calibration and identification can meet the shooting needs. Although the total number of pixels in a 3D structured light camera is large, the number of pixels per row is often less than that of a 2D line laser camera, so the point sampling accuracy is lower in the scenarios that this study focuses on. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the purpose of the present invention is to provide a point cloud registration method and system for automobile cover gaps to solve the problems of large errors, low efficiency, poor repeatability in traditional contact measurement and incomplete collection of the internal contour of the gap in non-contact measurement with visual cameras.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, an embodiment of the present invention provides a point cloud registration method for automobile cover gaps, comprising: photographing automobile cover gaps with a line structured light camera, and acquiring a plurality of automobile cover gap point cloud data sets after the photographing;

[0010] The small arc fitting in the curve is realized by using the small arc nested loop fitting algorithm based on Bessel equation;

[0011] The CPD algorithm is used to preprocess the multiple sets of automobile cover gap point cloud data sets to obtain an optimal conversion matrix, and the two sets of preprocessed data are aligned to obtain a complete automobile cover gap contour.

[0012] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: the small arc nested loop fitting algorithm based on the Bessel equation includes:

[0013] The positions of the first and last points of the small arc can be limited to ensure that the fitting curve is always smooth;

[0014] If the left and right limiting slopes of the contact point when the fitted curve straight line is tangent to the arc are approximately equal, then the point where each segment connects is the tangent point;

[0015] If there is discontinuity or the left and right limit slopes at the contact point when the curve straight line and the arc are tangent are not approximately equal, it proves that the position of the tangent point here is not accurate or the arc and straight line fitting is wrong;

[0016] By changing the nested loop scheme of the step size, the temporary tangent point, i.e. the starting point q0 of the arc to be fitted and the end point q1 of the arc to be fitted, is optimized to solve the least squares fitting equation, so as to efficiently fit the small arc and obtain the tangent point.

[0017] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: in the outer loop, a point located on the arc is found as the initial tangent point q0, and a small arc is fitted by the inner loop and a straight line is fitted by the least squares method to determine whether the tangent point meets the requirement of equal limit slopes;

[0018] If the requirements are met, all loops are terminated, otherwise point q0 is updated and re-judged whether point q0 meets the requirement of equal limit slope;

[0019] In the inner loop, the optimal fitting shape control point q1 is obtained by fixing the tangent point q0 and the end point q2, and the least squares fitting equation is substituted as follows:

[0020]

[0021] For q 1x ,q 1y Find the partial derivatives. When the partial derivatives are all 0, the desired result is obtained. The Bezier curve at this time is the Bezier curve closest to the given source data.

[0022] In the inner loop, we only need to determine the optimal q 1x value, the best small arc fitting curve can be obtained;

[0023] Use iterative search to determine q 1x The value of , and the step-by-step iteration to select q 1x The value of

[0024] The root mean square error is used as a criterion for evaluating the fitting quality of the small arc nested loop fitting algorithm based on the Bessel equation.

[0025] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: a transfer clustering algorithm based on neighborhood search is used to separate the double "J-shaped" curve contours of the automobile cover gap and remove most of the noise points, reduce the influence of noise points and dry dyeing line segments, and provide high-quality conditions for post-processing;

[0026] The tangent point positioning algorithm based on curve fitting is used to accurately locate the tangent point of the "J-shaped" curve of the gap between automobile panels, overcoming the problems of inaccurate traditional tangent point positioning and the inability to reasonably connect segmented fitting.

[0027] A feature extension algorithm based on point cloud clipping is used to expand the target features of the point cloud contour for CPD algorithm to discover;

[0028] The CPD algorithm solves the optimal transformation matrix by calculating the probability that each point in the point cloud set A to be registered exists in the reference point cloud set A'.

[0029] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: the transfer clustering algorithm based on neighborhood search includes:

[0030] Take any endpoint in the point cloud set Ω as the starting center D0, and establish a circular search area according to the search radius r. C Search for the point set Ω in the circle I And included in the cluster point set Ω C ;

[0031] In the circle point set Ω I Search for the point D farthest from the center D0 F , the farthest point D F Take D0 as the new circle center and update the point set Ω inside the circle I and cluster point set Ω C ;

[0032] Constantly update the center D0 and the point set Ω inside the circle I and cluster point set Ω C Until the point cloud can no longer be selected;

[0033] The value range of r should satisfy the following formula:

[0034] max(σ*FOV)≤r <R

[0035] Among them, R is the estimated radius of the arc segment of the gap between the automobile cover parts, and σ is the camera resolution.

[0036] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: the tangent point positioning algorithm based on curve fitting includes:

[0037] A small arc nested loop fitting algorithm based on Bessel equation and the least squares straight line fitting method are used to select a point as a temporary tangent point;

[0038] The left straight line is fitted using the least squares straight line fitting method, and the right straight line is fitted using the small arc nested loop fitting algorithm based on the Bessel equation, with the accurate tangent point position being found based on the condition that the slopes at the contact points when the straight line and the arc are tangent are nearly equal.

[0039] As a point cloud registration method for automobile cover gaps described in the present invention, wherein: the feature expansion algorithm based on point cloud clipping includes:

[0040] Crop the registration area, take the tangent point as the core, limit the length of the straight line segment, and expand the target features of the point cloud contour from a "tangent point" to multiple point-line features, shape features and spatial relationship features including the tangent point, starting point, registration segment straight line, arc and their positional relationships for CPD algorithm to explore;

[0041] Crop the non-registration area and crop the redundant non-registration area to reduce the interference to the registration process.

[0042] In a second aspect, an embodiment of the present invention provides a point cloud registration system for gaps of automobile covering parts, comprising:

[0043] The acquisition module uses a line structured light camera to shoot the gaps between automobile panels and obtains multiple point cloud data sets of gaps between automobile panels after shooting;

[0044] The fitting module uses a small arc nested loop fitting algorithm based on the Bessel equation to achieve small arc fitting in the curve;

[0045] The registration module uses the CPD algorithm to preprocess the multiple sets of automobile cover gap point cloud data sets to obtain an optimal transformation matrix, and registers the two sets of preprocessed data to obtain a complete automobile cover gap contour.

[0046] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0047] Memory and processor;

[0048] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement a point cloud registration method for automobile cover gaps as described in any embodiment of the present invention.

[0049] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the point cloud registration method for gaps in automobile covering parts.

[0050] Beneficial effects of the present invention: The present invention proposes a point cloud registration scheme and system for the gap contour of automobile covering parts. By taking photos of the gap of automobile covering parts to be measured and extracting point cloud data, a series of preprocessing processes are performed to highlight more target features that can be selected by the registration algorithm, and the CPD algorithm is further used for registration to obtain a complete contour. Compared with the existing gap measurement method, the gap contour after registration is more complete and more accurate for gap measurement. The entire measurement process does not require human intervention, and the measurement efficiency is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0052] Figure 1 A schematic diagram of a shooting platform constructed by a slide module and a camera combination of a point cloud registration method and system for automobile cover gaps of the present invention.

[0053] Figure 2 The present invention is a schematic diagram of a point cloud registration method and system for automobile panel gaps for multiple shots of automobile panel gaps.

[0054] Figure 3 The present invention is an algorithm flow chart of a point cloud registration method and system for automobile cover gaps.

[0055] Figure 4 It is a schematic diagram of a Bezier curve when reducing the step distance of a point cloud registration method and system for automobile cover gaps of the present invention.

[0056] Figure 5 It is a schematic diagram of the root mean square error of a point cloud registration method and system for automobile cover gaps when reducing the step distance of the present invention.

[0057] Figure 6 It is a schematic diagram of binary iteration of a point cloud registration method and system for automobile cover gaps of the present invention.

[0058] Figure 7 The present invention is a schematic diagram of a clustering process of a point cloud registration method and system for automobile cover gaps.

[0059] Figure 8 It is a schematic diagram of the "J-type" curve structure of a point cloud registration method and system for automobile cover gaps of the present invention.

[0060] Fig. 9 A schematic diagram of a gap between automobile panels of a point cloud registration method and system for gap between automobile panels of the present invention.

[0061] Fig.10 This is a comparison diagram of slopes at wrong tangent points of a point cloud registration method and system for automobile cover gaps according to the present invention.

[0062] Fig.11 This is a comparison diagram of the slopes at the correct tangent points of a point cloud registration method and system for gaps in automobile covering parts of the present invention.

[0063] Fig.12 The present invention is a schematic diagram of standard parts of a point cloud registration method and system for automobile cover gaps.

[0064] Fig.13 The present invention is a schematic diagram of a point cloud registration method and system for automobile cover gaps according to the present invention.

[0065] Fig.14 This is a diagram of the root mean square error convergence of a point cloud registration method and system for automobile cover gaps according to the present invention.

[0066] Fig.15 It is a 95% confidence interval diagram of a point cloud registration method and system for automobile cover gaps of the present invention. DETAILED DESCRIPTION

[0067] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0068] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0069] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0070] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0071] Example 1

[0072] Reference Figures 1 to 11 , which is an embodiment of the present invention, provides a point cloud registration method and system for automobile cover gaps, including:

[0073] S1: Use a line structured light camera to shoot the gaps between automobile panels, and obtain multiple point cloud datasets of gaps between automobile panels after shooting. It should be noted that:

[0074] The operating environment of the selected algorithm is a portable computer with Windows 11 system, the computing environment is Matlab 2019, and the acquisition tool is a GOCATOR 2520 line laser camera. The position of the slide module is fixed according to the camera depth of field parameters, and the camera is fixed on the slide module base through a rotating connector. The slide module base is about 60 mm away from the measuring piece.

[0075] Among them, the camera depth of field parameters are camera shooting depth 47.5mm~72.5mm, shooting width -16.25mm~16.25mm.

[0076] like Figure 1 As shown, a shooting platform constructed by combining a slide module and a line laser camera is designed. By shooting the gaps between automobile covers multiple times, multiple automobile cover gap point cloud data sets are obtained after shooting.

[0077] When using a shooting platform built by combining a slide module and a line laser camera, fix the position of the slide module according to the camera depth of field parameters, fix the camera on the slide module base through a rotating connector, control the slide module platform to move to the left, rotate the platform clockwise, find a clear image position, and shoot the first set of data.

[0078] Control the slide module platform to move right, rotate the rotating platform counterclockwise, find the clear image position, and take the second set of data.

[0079] The requirements for multiple shots of the gaps between automotive panels include: Figure 2 As shown, multiple shots must cover the same outline or area of ​​the object; the point cloud data obtained by shooting must have a large proportion of repeated areas and as many target features as possible. The shooting mode of the slide module combined with the line laser camera limits the camera to movement along the x-axis and rotation along the z-axis.

[0080] The algorithm flow provided in this embodiment is as follows Figure 3 As shown, the details are as follows:

[0081] S2: Use the small arc nested loop fitting algorithm based on the Bessel equation to achieve the small arc fitting in the curve. It should be noted that:

[0082] The positions of the first and last points of the fitted small arc can be limited to ensure that the fitted curve is always smooth.

[0083] According to the geometric characteristics of the tangent point, the left and right limit slopes at the contact point when the straight line and the arc are tangent are approximately equal. The specific formula is as follows:

[0084]

[0085] in, is the point cloud set of the fitted line, D is the point cloud set of the arc segment after fitting. T is the tangent point, is the local slope of the straight line about the point, is the local slope of the small arc part about the point, and It is the position closest to the tangent point between the fitted straight line and the small arc.

[0086] If the left and right limiting slopes of the contact point when the fitted curve straight line is tangent to the arc are approximately equal, then the point where each segment connects is the tangent point.

[0087] If there is a discontinuity or the left and right limiting slopes at the contact point when the curve straight line and the arc are tangent are not approximately equal, it proves that the position of the tangent point here is not accurate or the arc and straight line fitting is wrong.

[0088] The traditional second-order Bessel equation is as follows:

[0089] B=(1-t) 2 q0+2t(1-t)q1+t 2 q2, (t∈[0,1])

[0090] Among them, when fitting a small arc, q0 is the starting point of the arc after fitting, q2 is the end point of the arc after fitting, q1 is the fitting shape control point, and t is the point cloud position control point, which is related to the number of point clouds after fitting, and the values ​​are as follows:

[0091]

[0092] Among them, a is a point on q0 and q1, b is a point on q1 and q2, and a is a point on c and b.

[0093] In order to ensure that the fitted small arc is perfectly connected with the adjacent curve or straight line, the following requirements exist:

[0094] q2 is the end point of the arc to be fitted or the starting point of the connected curve.

[0095] q0 is the starting point of the arc to be fitted or the end point of the connected curve.

[0096] To ensure that the fitting is an arc segment rather than an irregular curve, according to the properties of the arc, q1 must be located on the perpendicular bisector of the line connecting q0 and q2.

[0097] The idea of ​​least squares method is introduced to ensure the accuracy of fitting by minimizing the variance between the point cloud fitted by the Bezier curve and the original point cloud, while considering x and y, that is:

[0098]

[0099] In order to efficiently fit the small arc and obtain the tangent point, a nested loop scheme with a changed step size is used to optimize the values ​​of q0 and q1 to solve the above equation.

[0100] In the outer loop, find a point on the arc as the initial tangent point q0, and use the small arc fitted by the inner loop and the straight line fitted by the least squares method to determine whether the tangent point meets the requirement of equal limiting slope.

[0101] If the requirements are met, all loops are terminated, otherwise the q0 point is updated and the tangent point is re-judged to see whether it meets the requirement of equal limiting slope.

[0102] In the inner loop, the optimal fitting curve parameter q1 is calculated by fixing the tangent point q0 and the end point q2. The specific steps are as follows:

[0103] The optimal fitting shape control points are obtained by fixing the tangent points as follows:

[0104] The best fitting curve should be determined by q0, q1, q2 and t. In a single inner loop, q0 is the temporary tangent point (the starting point of the arc to be fitted), q2 is the end point of the arc to be fitted, t is any value between [0, 1], and only q1 is unknown; therefore, the solution to q1 determines the quality of the small arc fitting in a single inner loop.

[0105] For the convenience of calculation, this embodiment sets:

[0106] t1=(1-t) 2 , t2=2t(1-t), t3=t 2

[0107] q1=(q 1x ,q 1y )

[0108] Substitute into the least squares fitting equation:

[0109]

[0110] For q 1x ,q 1y Find the partial derivatives. When the partial derivatives are all 0, they are the desired results, such as Figure 4 and Figure 5 As shown, the Bezier curve at this time is the Bezier curve closest to the given source data.

[0111] The partial derivative results are as follows:

[0112]

[0113] Separate the unknown term to one side and resolve it into the common form AX=B:

[0114]

[0115] Dissolved into common forms:

[0116]

[0117] At this point, we have solved q 1x ,q 1y The relationship between q0, q2, and t; From the above, we know that q1 must be located on the perpendicular bisector of q0 and q1, and the perpendicular bisector equation of q0 and q1 is solved y=kx+b, so q 1y =kq 1x +b.

[0118] It is deduced that, through the temporary determination of q0, q2, and t values ​​in each outer loop, only the optimal q needs to be determined in the inner loop. 1x value, the best small arc fitting curve can be obtained.

[0119] The position of q1 in different small arc fittings is different. In order to ensure the accuracy of small arc fitting, it is necessary to use an iterative search method to determine q 1x The value of .

[0120] The usual method is to start from the limit value and select a minimum value to determine the moving step q 1x , and then continuously fit to find a method with the best fitting effect; however, this method has a slow convergence speed, a long time, and its accuracy is affected by the selected step size.

[0121] Therefore, in order to save the calculation time and convergence speed of the inner and outer loops, this embodiment refers to the principle of stepping, such as Figure 6 As shown, set the initial q 1x The value of q 2x ; By continuously moving q 1x The position (moving distance v) fits the arc Ω R , until two fitting results are located on both sides of the actual point cloud, that is Reduce q 1x Movable range, update q 1x Initial position, reduce the moving distance and repeat the above process until the accuracy requirement is met and the iteration ends.

[0122] The best fit is determined as follows:

[0123] In engineering machinery, the root mean square error of the overlap is usually required to be less than 1 wire and meet the minimum engineering requirements, and 10 -2millimeters or less; therefore, the small arc nested loop fitting algorithm based on the Bessel equation in this embodiment uses the root mean square error as the criterion for evaluating the quality of the fitting; at the same time, the smaller the root mean square error threshold is set, the higher the accuracy is, but the slower the convergence speed is and the longer the calculation time is; the higher the threshold is set, the lower the accuracy is, but the faster the convergence speed is and the shorter the calculation time is, but it must be less than 10 -2 .

[0124] In the inner loop, in order to balance accuracy, time consumption, and convergence speed, considering that the point cloud data set has certain fluctuations, the root mean square error is not too small. Therefore, when the root mean square error reaches 5×10 -3 In this embodiment, the repeatability of the camera in the Z direction is 0.4 microns, so when the minimum moving step is less than 4×10 -4 Stop the inner loop.

[0125] S3: Use the CPD algorithm to preprocess multiple sets of automobile cover gap point cloud data sets to obtain the optimal transformation matrix, and align the two sets of preprocessed data to obtain a complete automobile cover gap profile. It should be noted that:

[0126] The transfer clustering algorithm based on neighborhood search is used to separate the double "J-shaped" curve contours of the gap between automobile panels and remove most of the noise points, reduce the influence of noise points and dry dyeing line segments, and provide high-quality conditions for post-processing as follows:

[0127] Among them, considering the missing point cloud during camera shooting and the need to remove outliers, the algorithm field search radius r is calculated:

[0128] max(σ*FOV)≤r <R

[0129] Where R is the estimated radius of the arc segment of the gap between the automobile cover parts, and σ is the camera resolution.

[0130] In this embodiment, r is selected to be 0.5 mm.

[0131] Input two sets of data in sequence, and the operation program takes any endpoint in the point cloud set Ω as the starting circle center D0, and establishes a circular search area according to the search radius r. C Search for the point set Ω in the circle I And Ω I Included in the cluster point set Ω C .

[0132] In Ω I Search for the point D farthest from the center D0 F , the farthest point D F Take D0 as the new circle center and update the point set Ω inside the circle I and cluster point set Ω C .

[0133] Constantly update the center D0 and the point set Ω inside the circle I and cluster point set Ω C , that is, iterate continuously until no more point clouds can be selected. The iterative search process is as follows Figure 7 As shown; during the iterative process, noise points and dry-dyed line segments are reduced.

[0134] The tangent point location algorithm based on curve fitting is adopted, such as Figure 8 and Fig. 9 As shown, the precise positioning of the tangent point of the "J-shaped" curve of the gap between automobile panels is completed, overcoming the problems of inaccurate traditional tangent point positioning and the inability to reasonably connect segmented fitting.

[0135] The small arc nested loop fitting algorithm based on Bessel equation and the least square fitting method are used to select a point as a temporary tangent point.

[0136] The left straight line is fitted using the least squares method, and the right straight line is fitted using the small arc nested loop fitting algorithm based on the Bessel equation proposed in the present invention, and the accurate tangent point position is found with the condition that the slope at the contact point when the straight line and the arc are tangent is almost equal as the core.

[0137] Locate the initial tangent point, construct the straight line equation of the first and last points of the point cloud set, traverse all point sets, find the point with the largest distance, and use this point as the initial tangent point D q ; Use this point as the segmentation point for segmentation. By default, the left half is a straight line point cloud set Ω r The right half is the arc point cloud set Ω r .

[0138] Fitting the line point cloud set Ω using the least squares method l Get Ω L , the equation of the line is as follows:

[0139]

[0140] Among them, x i ,y i is the horizontal and vertical coordinates of the original point cloud, a+bx i is the equation of the fitted straight line.

[0141] Take partial derivatives of equations a and b and rearrange them to get the equation:

[0142]

[0143] Solving the above system of equations gives the best estimates of the line parameters a and b as follows:

[0144]

[0145]

[0146] Through the above process, the relationship between the coefficients of the fitted straight line equation and the original point cloud can be obtained; the fitting equation can be obtained using the original point cloud.

[0147] Through the above process, the fitted straight line and arc are obtained, and the right limit slope of the straight line is calculated. and the left limiting slope of the arc like Fig.10 and Fig.11 As shown in the figure, when the two limit slopes are almost equal, this point is considered to be the desired tangent point; when the two limit slopes differ greatly, this point is considered not to be a tangent point, and the next point is traversed as the new D q Repeat the above process until the limiting slopes are almost equal, and the tangent point D is obtained. q .

[0148] Using the feature extension algorithm based on point cloud clipping, the target features of the extended point cloud contour are used for CPD algorithm to explore as follows:

[0149] The registration area is cropped, with the tangent point as the core, the length of the "J-type" curve registration segment is determined to make the registration segment lengths of the two sets of data the same, and the length of the straight line segment is limited. The target features of the point cloud contour are expanded from a "tangent point" to multiple point and line features, shape features and spatial relationship features including the tangent point, starting point, registration segment straight line, arc and their positional relationships for CPD algorithm to explore.

[0150] Crop the non-registration area and crop the redundant non-registration area <5mm to reduce the interference to the registration process.

[0151] Since the CPD algorithm is a probabilistic point set registration algorithm based on the Gaussian mixture model (GMM), and the core of GMM is to use the probabilistic statistical method to convert the traditional eigenvalue registration into probabilistic registration, it can reflect the feature points of the point cloud set macroscopically to a certain extent and use them to find corresponding points. That is, the CPD algorithm solves the optimal transformation matrix by calculating the probability that each point in the point cloud set A to be registered exists in the reference point cloud set A'.

[0152] Select the CPD algorithm registration mode as rigid registration, the maximum number of iterations as 100 and the minimum error limit as 10 -6 , set the noise weight to:

[0153]

[0154] Among them, X and Y are two sets of point cloud data sets.

[0155] Since the CPD algorithm only uses the corresponding points it obtains to derive the optimal transformation matrix, and does not directly perform registration or the registered point cloud is not a complete point cloud, it is necessary to convert the rotation, translation, and scaling optimal transformation matrices obtained based on the registration results and perform coordinate transformation on the original data.

[0156] This embodiment also provides a point cloud registration system for automobile cover gaps, including:

[0157] The acquisition module uses a line structured light camera to shoot the gaps between automobile panels and obtains multiple automobile panel gap point cloud data sets after shooting.

[0158] The fitting module uses a small arc nested loop fitting algorithm based on the Bessel equation to achieve small arc fitting in the curve.

[0159] The registration module uses the CPD algorithm to preprocess multiple sets of automobile cover gap point cloud data sets to obtain the optimal transformation matrix, and aligns the two sets of preprocessed data to obtain a complete automobile cover gap contour.

[0160] This embodiment further provides a computing device, which is applicable to a point cloud registration method for gaps in automobile panels, and includes:

[0161] A memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a point cloud registration method for automobile cover gaps as proposed in the above embodiment.

[0162] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0163] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a point cloud registration method for gaps between automobile covers is implemented as proposed in the above embodiment.

[0164] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0165] Example 2

[0166] Reference Figure 12 to Figure 15 , which is another embodiment of the present invention, provides a point cloud registration method and system verification test for the gap between automobile covering parts, and verifies and explains the technical effects used in the method.

[0167] Considering the large manufacturing error of automobile panels, the existing measurement methods of gap and face difference cannot guarantee the accuracy, etc., this study uses Fig.12 The high-precision machining standard parts shown are used to verify the effectiveness and practicality of the developed algorithm; this standard part contains six different types of matching modes, with gaps ranging from 1mm to 5mm, and the gap depth, shape, and chamfer size are also different, simulating the matching situations that often occur in engineering; therefore, this embodiment switches different angles from the six gaps to shoot 600 sets of data for experimental verification; in order to verify the accuracy of the tangent point, the "H-shaped" gaps on both sides are specially designed, and the accuracy of the tangent point positioning algorithm based on curve fitting is verified according to its precise tangent point position; the operating environment of the algorithm of the present invention is a portable computer equipped with win11 system, the computing environment is Matlab2019, and the acquisition tool is GOCATOR2520 line laser camera.

[0168] In order to verify the small arc fitting accuracy of the small arc nested loop fitting algorithm based on the Bessel equation, this embodiment extracts 100 groups of small arc point clouds of 30 degrees, 60 degrees, and 90 degrees from the 600 groups of data taken, and respectively uses the least squares method, the RANSAC fitting circle method and the algorithm in this paper to fit the small arc; and the root mean square error (RMSE) is used to evaluate the fitting effect; the root mean square error expresses the quality of the fitting by measuring the degree of deviation of the point set before and after fitting. The root mean square error is very sensitive to the extremely large or extremely small errors in a set of measurements. Therefore, the root mean square error can well reflect the precision between the two groups of point sets. The experimental results are shown in Table 1 below:

[0169] Table 1: Comparison of the root mean square error of three algorithms for fitting small arcs.

[0170]

[0171] It can be seen from the table that the smaller the angle of the arc, the worse the fitting effect of the least square method and RANSAC method, while the nested loop fitting algorithm of the arc based on the Bessel equation can basically remain stable and reach 10 -3 , meeting the fitting requirements; for small arc fitting at the same angle, the algorithm of the present invention also has significant improvement in fitting effect.

[0172] In addition, the arc nested loop fitting algorithm based on the Bessel equation can limit the positions of the first and last points of the arc, which has higher usability. Although it will also be affected by the fluctuation of the first and last points, it has a lower overall impact than the least squares method and RANSAC method, which use all point clouds for fitting, and it is easier to get accurate results.

[0173] In order to verify the effect of the transitive clustering algorithm based on neighborhood search, this section verifies the clustering effect and denoising effectiveness of the clustering algorithm of the present invention by comparing three methods: k-means clustering, hierarchical clustering and the transitive clustering based on neighborhood search proposed in the present invention.

[0174] This section will use the following common clustering algorithm evaluation indicators: time consumption (T), accuracy (ACC, the ratio of correctly clustered samples to the total number of samples), Jaccard coefficient (JC, the similarity and difference between finite sample sets, the larger the coefficient, the higher the sample similarity), FM index (FMI, the set mean of pairwise precision and recall, ranging from [0, 1], the closer to 1, the better the clustering effect) and adjusted Rand index (ARI, ranging from [-1, 1], the closer to 1, the better the algorithm clustering performance).

[0175] Through analysis, the research object is composed of two relative "J-shaped" curves. In addition, there may be some noise points and outliers located inside the gap; the purpose of clustering in the present invention is to separate the above components.

[0176] In the experiment, 100 groups were randomly sampled from the 600 groups of data collected, and the clustering results were calculated using three methods: k-means clustering, hierarchical clustering, and transitive clustering based on neighborhood search proposed in this paper. The experimental results are shown in Table 2.

[0177] Table 2: Clustering performance comparison.

[0178] algorithm T(ms) ACC(%) JC FMI ARI k-means clustering 550.40 93.9 0.929 0.936 0.934 Hierarchical clustering 706.53 91.4 0.853 0.918 0.861 Transitive Clustering Based on Neighborhood Search 433.49 99.3 0.955 0.975 0.968

[0179] Compared with the three clustering methods, the time consumption of hierarchical clustering and K-means clustering is significantly higher than that of transitive clustering based on neighborhood search, and the accuracy is lower when applied to the "J-type" curve; the Jaccard coefficient, FM index, and adjusted Rand index (ARI) of the three methods are compared. As can be seen from the table, the Jaccard coefficient (JC) of transitive clustering based on neighborhood search can reach 0.955, the FM index (FMI) can reach 0.975, and the adjusted Rand index (ARI) can reach 0.968; hierarchical clustering and K-means clustering can achieve a certain clustering effect, but the clustering performance is not good, which is significantly lower than that of transitive clustering based on neighborhood search.

[0180] In order to verify the accuracy of the tangent point positioning method based on curve fitting, this part of the algorithm takes the least square method and the small arc nested loop fitting algorithm based on the Bessel equation as the core, and verifies the accuracy of the tangent point positioning algorithm based on straight line fitting by measuring the tangent point position of the standard part; Fig.13 As shown, the theoretical distance from the tangent point to the rightmost horizontal plane is 12 mm ± 0.05.

[0181] When shooting data, it is required that the tangent point and the rightmost plane must exist in the data at the same time, and the shooting area must be larger than the area included in the theoretical distance; therefore, 60 sets of data are shot at different angles in this area and randomly divided into 3 groups, recorded as A, B, and C, and the calculated average value is taken as the final result. The results are shown in Table 3.

[0182] The experimental results show that the error of the algorithm proposed in this study can be controlled below 1mm, and the fluctuation range can be controlled within 0.01mm, which fully meets the accuracy requirements of this experiment for tangent point positioning, proving the accuracy of this algorithm.

[0183] Table 3: Tangent point location measurement results.

[0184] Shooting angle classification Standard value / mm Calculated value / mm Average error / mm Error fluctuation range / mm A 12 11.9967 0.0033 0.008 B 12 11.9574 0.0426 0.009 C 12 12.0025 0.0025 0.006

[0185] In order to verify the accuracy of the registration algorithm in this paper, this embodiment switches different angles from the six standard gaps to shoot 100 sets of data each, a total of 600 sets of data are used for verification experiments, and the registration root mean square error and time are recorded.

[0186] The efficiency and error of point cloud registration are important indicators for evaluating the effect of point cloud registration. The registration efficiency is evaluated by the registration time. In order to conduct experiments, this embodiment selects the average value of the root mean square error as the evaluation standard. The experimental results are shown in Table 4 below.

[0187] Table 4: Comparison of the root mean square error of the registration with and without adding the preprocessing algorithm in this paper.

[0188]

[0189] It can be seen from the table that the algorithm proposed in this paper can achieve good registration results in the six gap conditions; it can be seen from the table that the average root mean square error between the point clouds after registration will be reduced to 1.26×10 -3 Compared with the direct registration using CPD algorithm, the registration algorithm of the present invention reduces the average RMS error by 98.0586%, but the registration time is relatively increased by 3.5963s. In addition, it is affected by the registration environment and gap type, such as missing or fluctuating point clouds caused by reflections, complex gap contours, etc. In a few cases, the RMS error and time of the registration results may fluctuate to a certain extent, but the fluctuation range is usually not greater than 0.01, which is acceptable.

[0190] like Fig.14 As shown in the figure, it is a schematic diagram of the convergence of the E gap registration process with and without preprocessing. The CPD algorithm performs registration by continuously iterating corresponding points. The more correct corresponding points mean the closer to the correct registration result. It can be seen from the figure that the convergence speed of the point cloud set that has been preprocessed is higher than that without preprocessing, and the registration error of the point cloud set that has been preprocessed is lower.

[0191] Through the above experiments, the registration results of the classic point cloud data and the pre-processed point cloud data are analyzed and compared respectively. The results show that the algorithm proposed in the present invention has high registration accuracy on the gap of automobile cover parts.

[0192] In this experiment, 20 samples were collected from six gaps, with 30 groups of data in each sample. The sample distribution obeyed the normal distribution. The interval was estimated at a confidence interval of 95%. The confidence interval obtained was as follows: Fig.15 As shown, the experimental results obtained by the experimental algorithm have a probability of meeting the preset requirements of more than 95%.

[0193] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A point cloud registration method for gaps in automobile panels, characterized in that: include: A line structured light camera is used to shoot the gaps between automobile panels, and multiple point cloud data sets of gaps between automobile panels are obtained after shooting; The CPD algorithm is used to pre-process the plurality of sets of automobile cover gap point cloud data sets to obtain an optimal transformation matrix, and the two sets of pre-processed data are aligned to obtain a complete automobile cover gap profile; The CPD algorithm is used to pre-process the plurality of automobile cover gap point cloud data sets, including: Adopting the transfer clustering algorithm based on neighborhood search, the double "J-shaped" curve contour of the gap between automobile panels is separated and most of the noise points are removed, which reduces the influence of noise points and dry dyeing line segments and provides high-quality conditions for post-processing. The tangent point positioning algorithm based on curve fitting is used to accurately locate the tangent point of the "J-shaped" curve of the gap of the automobile cover, overcoming the problems of inaccurate traditional tangent point positioning and the inability to reasonably connect segmented fitting. A feature extension algorithm based on point cloud clipping is used to expand the target features of the point cloud contour for CPD algorithm to discover; The CPD algorithm solves the optimal transformation matrix by calculating the probability that each point in the point cloud set A to be registered exists in the reference point cloud set A'; The transitive clustering algorithm based on neighborhood search includes: Take any endpoint in the point cloud set Ω as the starting center D0, and establish a circular search area according to the search radius r. C Search for the point set Ω in the circle I And included in the cluster point set Ω C ; In the circle point set Ω I Search for the point D farthest from the center D0 F , the farthest point D F Take D0 as the new circle center and update the point set Ω inside the circle I and cluster point set Ω C ; Constantly update the center D0 and the point set Ω inside the circle I and cluster point set Ω C Until the point cloud can no longer be selected; The value range of r should satisfy the following formula: max(σ*FOV)≤r <R Among them, R is the estimated radius of the arc segment of the gap between automobile covers, and σ is the camera resolution; The tangent point positioning algorithm based on curve fitting includes: A small arc nested loop fitting algorithm based on Bessel equation and the least squares straight line fitting method are used to select a point as a temporary tangent point; The left straight line is fitted by the least squares straight line fitting method, and the right straight line is fitted by the small arc nested loop fitting algorithm based on the Bessel equation, and the accurate tangent point position is found based on the condition that the slopes at the contact points when the straight line and the arc are tangent are nearly equal; The small arc nested loop fitting algorithm based on Bessel equation includes: The positions of the first and last points of the small arc can be limited to ensure that the fitting curve is always smooth; If the left and right limiting slopes of the contact point when the fitted curve straight line is tangent to the arc are approximately equal, then the point where each segment connects is the tangent point; If there is discontinuity or the left and right limit slopes at the contact point when the curve straight line and the arc are tangent are not approximately equal, it proves that the position of the tangent point here is not accurate or the arc and straight line fitting is wrong; By changing the nested loop scheme of the step size, the temporary tangent point, i.e. the starting point q0 of the arc to be fitted and the end point q2 of the arc to be fitted, is optimized to solve the least squares fitting equation, so as to efficiently fit the small arc and obtain the tangent point. In the outer loop, find a point on the arc as the initial tangent point, i.e., the starting point q0 of the arc to be fitted, and use the small arc fitted by the inner loop and the straight line fitted by the least squares method to determine whether the tangent point, i.e., the starting point q0 of the arc to be fitted, meets the requirement of equal limit slope. If the requirements are met, all loops are terminated. Otherwise, the starting point q0 of the arc to be fitted is updated, and the tangent point, i.e., the starting point q0 of the arc to be fitted, is re-judged to see whether it meets the requirement of equal limit slope. In the inner loop, the optimal fitting shape control point q1 is calculated through the starting point q0 and the end point q2 of the arc to be fitted, and is substituted into the least squares fitting equation as follows: For q 1x ,q 1y Find the partial derivatives. When the partial derivatives are all 0, the desired result is obtained. The Bezier curve at this time is the Bezier curve closest to the given source data. In the inner loop, we only need to determine the optimal q 1x value, the best small arc fitting curve can be obtained; Use iterative search to determine q 1x The value of , and the step-by-step iteration to select q 1x The value of The root mean square error is used as a criterion for evaluating the fitting quality of the small arc nested loop fitting algorithm based on the Bessel equation; The feature expansion algorithm based on point cloud clipping includes: Crop the registration area, take the tangent point as the core, limit the length of the straight line segment, and expand the target features of the point cloud contour from one "tangent point" to multiple point-line features, shape features and spatial relationship features including the tangent point, starting point, registration segment straight line, arc and their positional relationships for CPD algorithm to explore; Crop the non-registration area and crop the redundant non-registration area to reduce the interference to the registration process.

2. A system for the point cloud registration method for automobile cover gaps as claimed in claim 1, characterized in that: include, The acquisition module uses a line structured light camera to shoot the gaps between automobile panels and obtains multiple point cloud data sets of gaps between automobile panels after shooting; The fitting module uses a small arc nested loop fitting algorithm based on the Bessel equation to achieve small arc fitting in the curve; The registration module uses the CPD algorithm to preprocess the multiple sets of automobile cover gap point cloud data sets to obtain an optimal transformation matrix, and registers the two sets of preprocessed data to obtain a complete automobile cover gap contour.

3. A computing device comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the point cloud registration method for automobile cover gaps described in claim 1 are implemented.

4. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a point cloud registration method for automobile cover gaps as described in claim 1.

Images