A point cloud stitching method based on measurement field and key feature constraints

Abstract

The present invention discloses a point cloud stitching method based on measurement field and key feature constraints, comprising the following steps: first, construct a high-precision measurement field and optimize the layout of ERS points in the measurement field to establish a high-precision global coordinate system; second, use the high-precision measurement field and combine the aircraft digital model to restore the fuselage coordinate system to achieve the initial alignment of multi-station point clouds; then, based on the initial point cloud alignment, realize high-precision stitching of point clouds based on multi-scale feature constraints and perform preprocessing; finally, optimize the data quality of the stitched overall point cloud through global multi-scale filtering and adaptive overlapping area removal methods, and finally obtain high-quality point clouds. The method proposed by the present invention realizes high-precision stitching of scanned point clouds of large-scale measurement objects, with higher stitching accuracy and faster stitching speed than the prior art, solving the problem that large-scale object scanning data is difficult to stitch with high precision.

Description

Technical Field

[0001] The invention belongs to the technical field of point cloud stitching, and in particular relates to a point cloud stitching method based on measurement field and key feature constraints. Background Art

[0002] When using 3D scanning equipment such as laser scanners and structured light scanners to scan large parts, a single station cannot scan the entire aircraft shape due to its large size. Therefore, the point cloud data from multiple stations must be spliced ​​and processed. The key to this process is how to accurately align the point cloud data scanned at each station.

[0003] In multi-station point cloud registration, relying on a single method alone will have limitations. The method of stitching based solely on the absolute coordinates of the measurement field space is completely determined by the accuracy of the measurement field. This method is easily affected by accidental factors such as the absolute accuracy of the point and environmental factors. Although feature-based methods are suitable for large-scale coarse registration, feature extraction is difficult and prone to mismatching, and is easily affected by noise points. There is insufficient constraint on geometric consistency, and it is easy to lose accuracy when the feature repetition is high or the structure is single. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention proposes a point cloud stitching method based on measurement field and key feature constraints, so as to achieve high-precision stitching of point clouds through measurement field and key features.

[0005] In order to achieve the above technical objectives, the present invention provides the following technical solutions:

[0006] A point cloud stitching method based on measurement field and key feature constraints, which specifically includes:

[0007] S1. Arrange the measurement field based on the location of the scanned object, and optimize the measurement field layout scheme with the purpose of minimizing the measurement error; finally obtain the measurement field coordinate system under the optimal layout scheme ;

[0008] S2, according to the key points of the scanned object in the measurement field coordinate system The coordinates of the corresponding points in the digital coordinate system The coordinates under the measurement field are aligned by the best fitting method, and the transformation matrix from the measurement field coordinate system to the digital model coordinate system is calculated; the transformation matrix includes the rotation matrix R and the translation vector t;

[0009] Then, according to the transformation matrix, the measurement field coordinate system is established And digital model coordinate system By acquiring the scanning point cloud data at each station in the measurement field, the digital point cloud data of the corresponding point in the digital coordinate system is obtained, and the multi-station splicing point cloud with preliminary alignment is obtained by integration. ;

[0010] S3. Preliminary alignment of multi-station stitching point clouds based on key feature information Preprocessing is performed, and the measured point cloud and digital model are further aligned to obtain a multi-station high-precision spliced ​​point cloud. ;

[0011] S4: The multi-station high-precision splicing point cloud obtained in step S3 Perform global optimization, including filtering out noise points and overlapping point clouds, and finally obtain the optimized complete point cloud of the scanned object .

[0012] Furthermore, step S1 specifically includes:

[0013] S11. Lay out the ground reference point ERS point of the measurement field, use the laser tracker to measure the ERS point, use the adjustment method to solve the measurement field, determine the point accuracy, and use the laser interferometer whose coordinate system completely coincides with the laser tracker coordinate system to measure the same ERS point coordinates as the true value. The formula is expressed as:

[0014] ;

[0015] in, is any point in the measurement field under the current layout scheme; is the true value measured by the laser interferometer, is the observed value measured by the laser tracker, is the measurement error between the observed value and the true value;

[0016] The least squares adjustment optimization is performed on the measurement results to optimize the weight distribution of ERS points in the solution process to obtain the minimum sum of squares of the global error terms. The optimization formula is expressed as:

[0017] ;

[0018] S12, coordinate uncertainty analysis of ERS points in the measurement field under different layout schemes is performed using the covariance matrix To evaluate the propagation of measurement errors, by continuously comparing the variances of measurement errors of different layout schemes, coordinate uncertainty analysis is performed, and the layout scheme with the smallest variance of measurement errors is selected; the covariance matrix The formula is:

[0019] ;

[0020] in, is the Jacobian matrix, which represents the sensitivity of the measurement error to each parameter. yes The transposed matrix of is the covariance matrix of measurement error;

[0021] Sensitivity-related parameters include the radial distance between the laser tracker and the measurement point. , Pitch angle and azimuth ; Jacobian matrix The formula is expressed as:

[0022] ;

[0023] Covariance matrix of measurement errors It is expressed as:

[0024] ;

[0025] in, is the variance of measurement error in three dimensions; , , The measurement error is x Axis dimensions and y The covariance between the axis dimensions, x Axis dimensions and z The covariance between the axis dimensions, y Axis dimensions and z covariance between axis dimensions;

[0026] S13, optimize the measurement field layout plan, adjust the position and layout density of ERS points, and obtain the best layout plan and measurement field coordinate system .

[0027] Furthermore, step S2 specifically includes:

[0028] S21. Measure the key points on the scanned object in the measurement field coordinate system The coordinate set below , and obtain the corresponding point on the digital model in the digital model coordinate system The coordinate set below ; Calculate the transformation relationship between the two coordinate systems based on the two coordinate sets obtained; The formula for the transformation relationship between the two coordinate systems is expressed as:

[0029] ;

[0030] in, R is the rotation matrix, t is the translation vector;

[0031] S22. For multiple different stations in the measurement field, firstly measure the ERS point at each station to locate the station; then scan the data at the station to obtain the scan point cloud at the station ,in, is the number of the current station; then according to the conversion relationship, the scanned point cloud at the station is restored to the digital-analog coordinate system to obtain the corresponding digital-analog point cloud in the digital-analog coordinate system ; The formula is:

[0032] ;

[0033] S23, change the station to perform multiple data scans until the corresponding point clouds at all stations are obtained, and then integrated into the initial aligned multi-station spliced ​​point cloud ,in is the number of stations.

[0034] More specifically, the process of calculating the conversion matrix in step S21 is as follows:

[0035] from and Select 5 pairs of corresponding points and calculate the centroid. The formula is:

[0036] ;

[0037] ;

[0038] in, , The coordinate sets , The centroid of

[0039] Then centralize and get:

[0040] ;

[0041] ;

[0042] in, The centered coordinate system in the measurement field The coordinates below, The centered digital coordinate system The coordinates below;

[0043] ;

[0044] Finally, the rotation matrix is ​​obtained by performing singular value decomposition on the covariance matrix R and translation vector t .

[0045] Furthermore, step S3 specifically includes:

[0046] S31. Preliminary alignment of multi-station stitching point clouds Preprocessing is performed by calculating the edge contour bounding box of the digital model, mapping it to the measurement field coordinate system, extracting the area of ​​interest of the measured data in the ROI area, filtering out irrelevant point clouds, and obtaining the processed point cloud ;

[0047] S32, the processed point cloud Extract key features by constructing a multi-scale adaptive local key feature perception network module as a feature extractor to locate and extract features and obtain a point cloud key feature set. , the formula is:

[0048] ;

[0049] in, represents a feature extractor;

[0050] S33, directly extract the corresponding features from the digital model for extraction and analysis to obtain the digital model feature set , for the extracted digital feature points And point cloud key feature points Perform matching to ensure the consistency of feature points;

[0051] Use distance threshold and normal similarity threshold As matching criteria; record key feature points of point cloud And digital feature points The coordinates of and , the matching conditions are:

[0052] ;

[0053] S34. Based on the corresponding features of the digital model and the point cloud, the iterative closest point algorithm is used to further align the point cloud on the basis of the initial alignment, and the rotation matrix R and translation vector are continuously optimized and updated. t , until the objective function converges, and a high-precision spliced ​​point cloud of multiple stations is obtained ;

[0054] The objective function is: .

[0055] More specifically, step S32 specifically includes:

[0056] S321, the processed point cloud Capture local details through multi-scale neighborhood feature extraction;

[0057] S322, performing a feature aggregation module to achieve deep fusion of multi-scale features;

[0058] S323, assigning weights to local features of the point cloud according to the similarity of digital and analog features, and highlighting important feature areas related to the task;

[0059] S324, parsing and normalizing the features;

[0060] S325. Finally, the point cloud key feature set is obtained by weighted summation .

[0061] Furthermore, step S4 specifically includes:

[0062] S41, high-precision stitching of point clouds at multiple locations Perform global optimization to obtain the global optimal cloud ;

[0063] S42, using multi-scale filtering technology based on global optimization, the global optimal cloud Filter out noise points to get denoised point cloud ;

[0064] S43, using the adaptive overlapping area removal method to remove the noise point cloud Remove the overlapping point clouds and finally get the complete point cloud of the scanned object .

[0065] More specifically, step S41 is as follows:

[0066] Use graph optimization method for global optimization, take each point cloud as a node in the graph, and the edges between nodes represent the error between point cloud registrations; the optimization goal is to minimize the error of the entire graph:

[0067] ;

[0068] in Point Cloud To neighboring point cloud The transformation matrix, are all the edges in the graph, representing the registration errors between nodes.

[0069] More specifically, step S43 is as follows:

[0070] Remember to remove the noise cloud Two point clouds in , ; They have been aligned through global optimization, and the overlapping area of ​​the two has a higher point cloud density in space; the spatial distance and overlap of the point clouds are used as the basis for determining the overlapping area, and the formula for the determination is expressed as:

[0071] ;

[0072] represents the union of two point clouds, Represents the intersection of two point clouds. By calculating the minimum Euclidean distance between a point in one point cloud and a point in the other point cloud, if the minimum distance exceeds the threshold, the two points are considered to be in the intersection. The formula for judging the overlap degree of two adjacent point clouds is:

[0073] ;

[0074] in is the indicator function, used to determine whether it overlaps. is the overlap threshold;

[0075] When the threshold is exceeded, it is determined that there is an overlapping area. Overlapping areas are removed, and the points to be retained are determined based on the density and normal consistency of the point cloud to obtain a complete point cloud of the scanned object. .

[0076] Further, in step S43, the points to be retained are determined based on the density and normal consistency of the points through the overlapping point cloud optimization function The judgment formula is expressed as:

[0077] ;

[0078] in, For point radius The number of inner neighborhood points of is the neighborhood volume, is the global average density, is the density threshold, for The angle between the normal and three times the neighboring normal, is the normal angle threshold; If the function value is 1, the point is removed; if it is 0, it is retained.

[0079] Based on the above technical solution, the present invention has at least the following beneficial effects:

[0080] The present invention realizes high-precision stitching of scanned point clouds of large-sized measurement objects based on measurement field and feature point constraints. Compared with the prior art, the stitching accuracy is higher and the stitching speed is faster, which solves the problem that the scanned data of large-sized objects is difficult to obtain and difficult to stitch after acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is an overall flow chart of the point cloud stitching method proposed by the present invention;

[0082] Figure 2 It is a schematic diagram of the measurement field layout of the present invention;

[0083] Figure 3 It is the multi-scale adaptive local key feature perception network module of the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0085] Although the steps in the present invention are arranged with numbers, they are not used to limit the order of the steps. Unless the order of the steps is clearly stated or the execution of a certain step requires other steps as a basis, the relative order of the steps can be adjusted. It is understood that the term "and / or" used in this article involves and covers any and all possible combinations of one or more of the associated listed items.

[0086] like Figure 1 As shown, a point cloud stitching method based on measurement field and key feature constraints proposed by the present invention is shown, which specifically includes the following steps:

[0087] S1. Arrange the measurement field based on the location of the scanned object, and optimize the measurement field layout scheme with the purpose of minimizing the measurement error; finally obtain the measurement field coordinate system under the optimal layout scheme In this embodiment, Figure 2 As shown in the figure, since the scanned object is too large, point cloud measurement needs to be performed separately from multiple stations in the measurement field, and finally the point clouds of all stations are stitched together with high precision to obtain a complete point cloud; Figure 2 In P 1- P 9 represents nine ERS points; as a preferred implementation, step S1 specifically includes:

[0088] S11, lay out the ground reference point ERS point of the measurement field, use the laser tracker to measure the ERS point, use the adjustment method to solve the measurement field, determine the point accuracy, and use the laser interferometer whose coordinate system completely coincides with the laser tracker coordinate system to measure the same ERS point coordinates as the true value. The process expression is as follows:

[0089] ;

[0090] in, is any point in the measurement field under the current layout scheme; is the true value measured by the laser interferometer, is the observed value measured by the laser tracker, is the measurement error between the observed value and the true value;

[0091] The least squares adjustment optimization is performed on the measurement results to optimize the weight distribution of ERS points in the solution process to obtain the minimum sum of squares of the global error terms. The optimization formula is expressed as:

[0092] ;

[0093] S12, coordinate uncertainty analysis of ERS points in the measurement field under different layout schemes is performed using the covariance matrix To evaluate the propagation of measurement errors, by continuously comparing the variances of measurement errors of different layout schemes, coordinate uncertainty analysis is performed, and the layout scheme with the smallest variance of measurement errors is selected; the covariance matrix The formula is:

[0094] ;

[0095] in, is the Jacobian matrix, which represents the sensitivity of the measurement error to each parameter. yes The transposed matrix of is the covariance matrix of measurement error;

[0096] Sensitivity-related parameters include the radial distance between the laser tracker and the measurement point. , Pitch angle and azimuth ; Jacobian matrix The formula is:

[0097] ;

[0098] Covariance matrix of measurement errors It is expressed as:

[0099] ;

[0100] in, is the variance of measurement error in three dimensions; , , The measurement error is x Axis dimensions and y The covariance between the axis dimensions, x Axis dimensions and z The covariance between the axis dimensions, y Axis dimensions and z covariance between axis dimensions;

[0101] S13, optimize the measurement field layout plan, adjust the position and layout density of ERS points, and obtain the best layout plan and measurement field coordinate system .

[0102] S2, according to the key points of the scanned object in the measurement field coordinate system The coordinates under and the corresponding points in the digital coordinate system The coordinates under the measurement field are aligned by the best fitting method, and the transformation matrix from the measurement field coordinate system to the digital model coordinate system is calculated; the transformation matrix includes the rotation matrix R and the translation vector t;

[0103] Then, according to the transformation matrix, the measurement field coordinate system is established And digital model coordinate system By acquiring the scanning point cloud data at each station in the measurement field, the digital point cloud data of the corresponding point in the digital coordinate system is obtained, and the multi-station splicing point cloud with preliminary alignment is obtained by integration. ;

[0104] As a preferred implementation, step S2 specifically includes:

[0105] S21. Measure the key points on the scanned object in the measurement field coordinate system The coordinate set below , and obtain the corresponding point on the digital model in the digital model coordinate system The coordinate set below ; Calculate the transformation relationship between the two coordinate systems based on the two coordinate sets obtained; The formula for the transformation relationship between the two coordinate systems is expressed as:

[0106] ;

[0107] in, R is the rotation matrix, t is the translation vector;

[0108] More specifically, the process of calculating the conversion matrix in step S21 is as follows:

[0109] from and Select 5 pairs of corresponding points and calculate the centroid. The formula is:

[0110] ;

[0111] ;

[0112] in, , The coordinate sets , The centroid of

[0113] Then centralize and get:

[0114] ;

[0115] ;

[0116] in, The centered coordinate system in the measurement field The coordinates below, After centering in the digital model coordinate system The coordinates below;

[0117] ;

[0118] Finally, the rotation matrix is ​​obtained by performing singular value decomposition on the covariance matrix R and translation vector t .

[0119] S22. For multiple different stations in the measurement field, firstly measure the ERS point at each station to locate the station; then scan the data at the station to obtain the scan point cloud at the station ,in, is the number of the current station; then according to the conversion relationship, the scanned point cloud at the station is restored to the digital-analog coordinate system to obtain the corresponding digital-analog point cloud in the digital-analog coordinate system ; The formula is:

[0120] ;

[0121] It should be noted that in the present application, a single point is measured and a point cloud is obtained by scanning; the method proposed in the present invention first obtains the conversion relationship between the measurement field coordinate system and the digital-analog coordinate system by measuring the key points (the digital-analog coordinate system is preset and known), and then converts the scanned point cloud into the corresponding digital-analog point cloud according to the conversion relationship, and then gradually realizes high-precision point cloud stitching.

[0122] S23, change the station to perform multiple data scans until the corresponding point clouds at all stations are obtained, and then integrated into the initial aligned multi-station spliced ​​point cloud ,in is the number of stations.

[0123] S3. Preliminary alignment of multi-station stitching point clouds based on key feature information Preprocessing is performed, and the measured point cloud and digital model are further aligned to obtain a multi-station high-precision spliced ​​point cloud. ;

[0124] As a preferred implementation, step S3 specifically includes:

[0125] S31. Preliminary alignment of multi-station stitching point clouds Preprocessing is performed by calculating the edge contour bounding box of the digital model, mapping it to the measurement field coordinate system, extracting the area of ​​interest of the measured data in the ROI area, filtering out irrelevant point clouds, and obtaining the processed point cloud ;

[0126] S32, the processed point cloud Extract key features by constructing a multi-scale adaptive local key feature perception network module as a feature extractor to locate and extract features and obtain a point cloud key feature set. , the formula is:

[0127] ;

[0128] in, represents a feature extractor;

[0129] More specifically, Figure 3 As shown in FIG. 1 , the process of extracting key features by the multi-scale adaptive local key feature perception network module is as follows:

[0130] S321, the processed point cloud Capture local details through multi-scale neighborhood feature extraction;

[0131] S322, performing a feature aggregation module to achieve deep fusion of multi-scale features;

[0132] S323, assigning weights to local features of the point cloud according to the similarity of digital and analog features, and highlighting important feature areas related to the task;

[0133] S324, parsing and normalizing the features;

[0134] S325. Finally, the point cloud key feature set is obtained by weighted summation .

[0135] S33, directly extract the corresponding features from the digital model for extraction and analysis to obtain the digital model feature set , for the extracted digital feature points And point cloud key feature points Perform matching to ensure the consistency of feature points;

[0136] Use distance threshold and normal similarity threshold As matching criteria; record key feature points of point cloud And digital feature points The coordinates of and , the matching conditions are:

[0137] ;

[0138] In this embodiment, the normal similarity is the angle between the normals at two points;

[0139] S34. Based on the corresponding features of the digital model and the point cloud, the iterative closest point algorithm is used to further align the point cloud on the basis of the initial alignment, and the rotation matrix R and translation vector are continuously optimized and updated. t , until the objective function converges, and a high-precision spliced ​​point cloud of multiple stations is obtained ;

[0140] The objective function is: .

[0141] S4: The multi-station high-precision splicing point cloud obtained in step S3 Perform global optimization, including filtering out noise points and overlapping point clouds, and finally obtain the optimized complete point cloud of the scanned object ;

[0142] As a preferred implementation, step S4 specifically includes:

[0143] S41, high-precision stitching of point clouds at multiple locations Perform global optimization to obtain the global optimal cloud ;

[0144] In this embodiment, step S41 is specifically as follows:

[0145] Use graph optimization method for global optimization, take each point cloud as a node in the graph, and the edges between nodes represent the error between point cloud registrations; the optimization goal is to minimize the error of the entire graph:

[0146] ;

[0147] in Point Cloud To neighboring point cloud The transformation matrix, are all the edges in the graph, representing the registration errors between nodes;

[0148] S42, using multi-scale filtering technology based on global optimization, the global optimal cloud Filter out noise points to get denoised point cloud ;

[0149] S43, using the adaptive overlapping area removal method to remove the noise point cloud Remove the overlapping point clouds and finally get the complete point cloud of the scanned object ;

[0150] In this embodiment, step S43 is specifically as follows: record the denoised point cloud Two point clouds in , ; They have been aligned through global optimization, and the overlapping area of ​​the two has a higher point cloud density in space; the spatial distance and overlap of the point clouds are used as the basis for determining the overlapping area, and the formula for the determination is expressed as:

[0151] ;

[0152] represents the union of two point clouds, Represents the intersection of two point clouds. By calculating the minimum Euclidean distance between a point in one point cloud and a point in the other point cloud, if the minimum distance exceeds the threshold, the two points are considered to be in the intersection. The formula for judging the overlap degree of two adjacent point clouds is:

[0153] ;

[0154] in is the indicator function, used to determine whether it overlaps. is the overlap threshold;

[0155] When the threshold is exceeded, it is determined that there is an overlapping area. Overlapping areas are removed, and the points to be retained are determined based on the density and normal consistency of the point cloud to obtain a complete point cloud of the scanned object. .

[0156] The points to be retained are determined based on the density and normal consistency of the points. The function is optimized by overlapping the point cloud. The judgment formula is expressed as:

[0157] ;

[0158] in, For point radius The number of inner neighborhood points of is the neighborhood volume, is the global average density, is the density threshold, for The angle between the normal and three times the neighboring normal, is the normal angle threshold; If the function value is 1, the point is removed; if it is 0, it is retained.

[0159] In summary, the present invention realizes high-precision stitching of scanned point clouds of large-size measurement objects based on measurement field and feature point constraints. Compared with the existing technology, the stitching accuracy is higher and the stitching speed is faster, which solves the problem of difficult stitching of scanned data of large-size objects.

[0160] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0161] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A point cloud stitching method based on measurement field and key feature constraints, characterized in that: The specific steps include: S1. Arrange the measurement field based on the location of the scanned object, and optimize the measurement field layout scheme with the purpose of minimizing the measurement error; finally obtain the measurement field coordinate system under the optimal layout scheme S2, according to the key points in the scanned object in the measurement field coordinate system The coordinates under and the corresponding points in the digital coordinate system The coordinates under the measurement field are aligned by the best fitting method, and the transformation matrix from the measurement field coordinate system to the digital model coordinate system is calculated; the transformation matrix includes the rotation matrix R and the translation vector t; Then, according to the transformation matrix, the measurement field coordinates are established And digital model coordinate system By acquiring the scanning point cloud data at each station in the measurement field, the digital point cloud data of the corresponding point in the digital coordinate system is obtained, and the multi-station splicing point cloud P is obtained by integration. o ; S3, based on the key feature information, the preliminarily aligned multi-station splicing point cloud P o Preprocessing is performed, and the measured point cloud and the digital model are further aligned to obtain a multi-station high-precision spliced ​​point cloud P. E ; Step S3 specifically includes: S31, for the preliminarily aligned multi-station spliced ​​point cloud P o Preprocessing is performed by calculating the edge contour bounding box of the digital model and mapping it to the measurement field coordinate system. The ROI area is extracted to extract the area of ​​interest of the measured data, and irrelevant point clouds are filtered out to obtain the processed point cloud P. os ; S32, the processed point cloud P os To extract key features, we construct a multi-scale adaptive local key feature perception network module as a feature extractor to locate and extract features, and obtain the point cloud key feature set F os , the formula is: F os =FeatureExtractor(P os ); Wherein, FeatureExtractor(·) represents a feature extractor; the specific process of step S32 is: S321, the processed point cloud P os Capture local details through multi-scale neighborhood feature extraction; S322, performing a feature aggregation module to achieve deep fusion of multi-scale features; S323, assigning weights to local features of the point cloud according to the similarity of digital and analog features, and highlighting important feature areas related to the task; S324, parsing and normalizing the features; S325, finally obtain the point cloud key feature set F through weighted summation os ; S33, directly extract the corresponding features from the digital model for extraction and analysis, and obtain the digital model feature set F m , for the extracted digital feature points And point cloud key feature points Perform matching to ensure the consistency of feature points; Use the distance threshold γ and the normal similarity threshold τ as the matching criteria; record the key feature points of the point cloud And digital feature points The coordinates of and The matching conditions are: S34. Based on the corresponding features of the digital model and the point cloud, the iterative closest point algorithm is used to further align the point cloud on the basis of the initial alignment, and the rotation matrix R and the translation vector t are continuously optimized and updated until the objective function converges to obtain the multi-station high-precision splicing point cloud P E ; The objective function is: S4, the multi-station high-precision splicing point cloud P obtained in step S3 E Perform global optimization, including filtering out noise points and overlapping point clouds, and finally obtain the optimized complete point cloud P of the scanned object. F .

2. A point cloud stitching method based on measurement field and key feature constraints according to claim 1, characterized in that: Step S1 specifically includes: S11. Lay out the ground reference point ERS point of the measurement field, use the laser tracker to measure the ERS point, use the adjustment method to solve the measurement field, determine the point accuracy, and use the laser interferometer whose coordinate system completely coincides with the laser tracker coordinate system to measure the same ERS point coordinates as the true value: the formula is expressed as: in, is any point in the measurement field under the current layout scheme; is the true value measured by the laser interferometer, is the observed value measured by the laser tracker, ∈ i is the measurement error between the observed value and the true value; The least squares adjustment optimization is performed on the measurement results to optimize the weight distribution of ERS points in the solution process to obtain the minimum sum of squares of the global error terms. The optimization formula is expressed as: S12. Coordinate uncertainty analysis is performed on the ERS points of the measurement field under different layout schemes. The covariance matrix C is used to evaluate the propagation of the measurement error. By continuously comparing the variance of the measurement error of different layout schemes, coordinate uncertainty analysis is performed and the layout scheme with the smallest variance of the measurement error is selected. The formula of the covariance matrix C is expressed as: C=J T ·P·J; Among them, J is the Jacobian matrix, which represents the sensitivity of measurement error to each parameter, J T is the transposed matrix of J, P is the covariance matrix of measurement error; The sensitivity-related parameters include the radial distance r between the laser tracker and the measurement point, the pitch angle θ and the azimuth angle The formula of Jacobian matrix J is expressed as: The covariance matrix P of the measurement error is expressed as: in, is the variance of the measurement error in three dimensions; σ xy , σ xz , σ yz is the covariance of the measurement error between the x-axis dimension and the y-axis dimension, the covariance between the x-axis dimension and the z-axis dimension, and the covariance between the y-axis dimension and the z-axis dimension; S13, optimize the measurement field layout plan, adjust the position and layout density of ERS points, and obtain the best layout plan and measurement field coordinate system 3. The point cloud stitching method based on measurement field and key feature constraints according to claim 1 is characterized in that: Step S2 specifically includes: S21. Measure the key points on the scanned object in the measurement field coordinate system The coordinate set below And obtain the corresponding point on the digital model in the digital model coordinate system The coordinate set below The transformation relationship between the two coordinate systems is calculated based on the two obtained coordinate sets; the formula for the transformation relationship between the two coordinate systems is expressed as: Among them, R is the rotation matrix and t is the translation vector; S22. For multiple different stations in the measurement field, firstly measure the ERS point at each station to locate the station; then scan the data at the station to obtain the scan point cloud at the station Among them, j is the number of the current station; then according to the conversion relationship, the scanned point cloud at the station is restored to the digital-analog coordinate system to obtain the corresponding digital-analog point cloud in the digital-analog coordinate system The formula is: S23, change the station to perform multiple data scans until the corresponding point clouds at all stations are obtained, and then integrated into the initial aligned multi-station spliced ​​point cloud Where S is the number of stations.

4. The point cloud stitching method based on measurement field and key feature constraints according to claim 3 is characterized in that: The process of calculating the conversion matrix in step S21 is specifically as follows: From P CK and P MK Select 5 pairs of corresponding points respectively and calculate the centroid. The formula is expressed as: Among them, C C , C M The coordinate sets P are CK , P MK The centroid of Then centralize and get: in, The centered coordinate system in the measurement field The coordinates below, After centering in the digital model coordinate system The coordinates below; Then calculate the covariance matrix H and get: Finally, by performing singular value decomposition on the covariance matrix, the rotation matrix R and translation vector t are obtained.

5. The point cloud stitching method based on measurement field and key feature constraints according to claim 1 is characterized in that: Step S4 specifically includes: S41, high-precision stitching of multi-station point cloud P E Perform global optimization and obtain the global optimal cloud P EP ; S42, using the multi-scale filtering technology based on global optimization, the global optimal point cloud P EP Filter out the noise points to get the denoised point cloud P EF ; S43, using the adaptive overlapping area removal method, the denoised point cloud P EF Remove the overlapping point clouds and finally get the complete point cloud P of the scanned object. F .

6. The point cloud stitching method based on measurement field and key feature constraints according to claim 5, characterized in that: Step S41 is specifically as follows: Use graph optimization method for global optimization, take each point cloud as a node in the graph, and the edges between nodes represent the error between point cloud registrations; the optimization goal is to minimize the error of the entire graph: Where T m is the point cloud P m To the adjacent point cloud P n The transformation matrix of , E is all the edges in the graph, representing the registration error between nodes.

7. The point cloud stitching method based on measurement field and key feature constraints according to claim 5, characterized in that: Step S43 is specifically as follows: Note the denoised point cloud P EF Two point clouds P1 and P2 in; They have been aligned through global optimization, and the overlapping area of ​​the two has a higher point cloud density in space; the spatial distance and overlap of the point clouds are used as the basis for determining the overlapping area, and the formula for the determination is expressed as: P1∪P2 represents the union of two point clouds, and P1∩P2 represents the intersection of two point clouds. By calculating the minimum Euclidean distance between a point in one point cloud and a point in the other point cloud, if the distance exceeds the threshold, the two points are points in the intersection. The formula for judging the overlap degree of two adjacent point clouds is: in is the indicator function used to determine whether there is overlap, and δ is the overlap threshold; When the threshold is exceeded, it is determined that there is an overlapping area. EF The overlapping areas are removed, and the points to be retained are determined based on the density and normal consistency of the point cloud to obtain the complete point cloud P of the scanned object. F .

8. The point cloud stitching method based on measurement field and key feature constraints according to claim 7, characterized in that: The process of deciding which points to keep based on point density and normal consistency is as follows: By optimizing the overlapping point cloud function Remove(p k ) is used for judgment, and its formula is expressed as: Among them, N n (p k ,r) is point p k The number of inner neighborhood points of radius r, V n (r) is the neighborhood volume, ρ global is the global average density, ξ is the density threshold, θ k,l For p k The angle between the normal and three times the neighboring normal, is the normal angle threshold; if the Remove(·) function value is 1, the point is removed, and if it is 0, it is retained.

Images