CN-122023483-A - Complex part point cloud registration method and device for blue light scanning detection of small five-axis machine tool

CN122023483ACN 122023483 ACN122023483 ACN 122023483ACN-122023483-A

Abstract

The invention relates to the field of three-dimensional measurement and point cloud data processing, and discloses a complex part point cloud registration method and device for blue light scanning detection of a small five-axis machine tool, wherein the method comprises the steps of acquiring a measured point cloud and CAD digital modulus, and establishing multi-coordinate system association; the method comprises the steps of carrying out initial alignment based on the prior of machine tool pose, carrying out noise suppression and structure maintenance downsampling on point cloud, constructing a self-adaptive multi-scale neighborhood according to point spacing reference and local stability index, extracting features, constructing point-to-surface soft correspondence by taking a CAD surface as a reference, calculating confidence coefficient weight, carrying out weighting screening, introducing matching coverage degree constraint to optimize correspondence set distribution, and finally carrying out point-to-surface fine registration of confidence coefficient weight and outputting registration results and deviation analysis information. According to the invention, registration drift caused by noise, shielding and uneven distribution is effectively inhibited by fusing pose priori, CAD guide soft correspondence, confidence weighting and coverage constraint, and the registration stability, precision and result repeatability are improved.

Inventors

Ren Huaidao
ZHAO GUOJUN
FANG TIANQUAN
XIE TIANPEI
Sun Gaoxu
ZHANG BEI
TIAN YUNLONG

Assignees

武汉软件工程职业学院(武汉开放大学)

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. A complex part point cloud registration method for blue light scanning detection of a small five-axis machine tool is characterized by comprising the following steps: s1, acquiring measurement point cloud data acquired by a blue light scanner and CAD data module data of a part to be tested, and establishing a coordinate association relation among a scanning coordinate system, a machine tool coordinate system and a digital-analog coordinate system; S2, generating initial pose estimation of the measurement point cloud data relative to the CAD digital-analog data based on pose information output by the small five-axis machine tool, and carrying out initial alignment on the measurement point cloud based on the initial pose estimation; S3, performing noise suppression and structure maintenance downsampling on the initially aligned measurement point cloud to obtain a sampling point set; S4, calculating a point spacing reference and a local stability index according to the sampling point set, constructing a self-adaptive multi-scale neighborhood for different areas based on the point spacing reference and the local stability index, and extracting multi-scale geometric features from the neighborhood to form a candidate feature set; S5, constructing a point-to-surface soft corresponding relation for the measurement points in the candidate feature set by taking the CAD digital-analog surface as a reference to form a candidate corresponding set; s6, calculating the confidence coefficient weight of each corresponding point in the candidate corresponding set, and carrying out weighted screening on the candidate corresponding set based on the confidence coefficient weight to obtain a weighted corresponding set; s7, introducing a matching coverage constraint, and shaping and constraining the weighted corresponding set to obtain an optimized corresponding set meeting the preset coverage requirement; And S8, based on the optimized corresponding set, performing point-to-surface fine registration weighted by confidence coefficient, solving the final rigid transformation, and outputting a registration result and deviation analysis information for detection.
2. The complex part point cloud registration method for blue light scanning detection of a small five-axis machine tool as set forth in claim 1, wherein the step S2 comprises the following steps: S21, acquiring installation external parameters of the blue light scanner, and combining the installation external parameters with pose information output by the small five-axis machine tool to obtain the transformation of a measuring point cloud from a scanning coordinate system to a machine tool coordinate system; s22, based on the clamping reference or the preset reference characteristic of the part to be tested, determining the transformation from a machine tool coordinate system to a digital-analog coordinate system of the CAD digital-analog; S23, combining the transformation obtained in the steps S21 and S22 to obtain an initial transformation matrix from the measurement point cloud to the digital-analog coordinate system, and using the initial transformation matrix as the initial pose estimation to initially align the measurement point cloud.
3. The method for registering the complex part point cloud for blue light scanning detection of the small five-axis machine tool according to claim 1, wherein the structure maintaining downsampling in the step S3 comprises the following steps: S31, identifying areas with abrupt change of apertures, edges, chamfers, steps and curvatures in the measurement point cloud after initial alignment, and setting high retention rate for the areas, wherein the high retention rate refers to retention rate larger than a preset threshold value; S32, identifying large-area flat areas in the measurement point cloud after initial alignment, and setting low retention rate for the areas, wherein the low retention rate means that the retention rate does not exceed a preset threshold value; s33, downsampling the point cloud according to the retention rate set in the steps S31 and S32, and retaining the characteristic points of the required key detection structure area while reducing the total amount of the point cloud.
4. The complex part point cloud registration method for blue light scanning detection of a small five-axis machine tool as set forth in claim 1, wherein the step S4 includes: S41, calculating the nearest neighbor distance of each point in the sampling point set, and determining a point distance benchmark of the point cloud through statistics; S42, multiplying a plurality of preset scale coefficients based on the point spacing reference to generate a multi-scale neighborhood radius set; S43, calculating local stability indexes of each point, wherein the local stability indexes comprise normal consistency, normal variance and local fitting residual errors; S44, for the area with high local stability, a fine-scale neighborhood is preferentially adopted and high participation weight is given in subsequent feature extraction and matching, and for the area with low local stability, a steady large-scale neighborhood is preferentially adopted and low participation weight is given.
5. The method for registering the complex part point cloud for blue light scanning detection of the small five-axis machine tool according to claim 11, wherein the constructing the point-to-surface soft correspondence in the step S5 comprises the following steps: S51, for each measuring point in the candidate feature set, calculating the nearest projection point of each measuring point on the CAD digital-analog surface as a corresponding point, and acquiring the surface normal at the corresponding point; S52, calculating the distance from the measuring point to the corresponding CAD surface as a point-to-plane residual error, and calculating an included angle between the normal direction of the measuring point and the normal direction of the corresponding point; and S53, if the included angle between the point-to-surface residual errors or the normal direction exceeds a preset threshold, marking the correspondence of the measurement point as low-confidence correspondence or eliminating the low-confidence correspondence from the candidate correspondence set.
6. The complex part point cloud registration method for blue light scanning detection of a small five-axis machine tool as claimed in claim 1, wherein the confidence weight calculated in the step S6 is determined by the following steps: S61, calculating a first weight factor related to the point-to-plane residual error, wherein the larger the residual error is, the smaller the first weight factor is; S62, calculating a second weight factor related to the consistency of the normal direction of the measuring point and the normal direction of the CAD surface, wherein the larger the normal included angle is, the smaller the second weight factor is; S63, calculating a visibility score related to the scanning visual angle, the shielding judgment and the digital-analog observability, and taking the visibility score as a third weight factor; s64, combining the first weight factor, the second weight factor and the third weight factor to obtain the comprehensive confidence weight of the corresponding point.
7. The method for registering the complex part point cloud for blue light scanning detection of the small five-axis machine tool according to claim 1, wherein the step S7 of introducing the matching coverage constraint comprises the following steps: s71, dividing the CAD digital-analog surface into a plurality of areas; s72, counting the distribution condition of high-confidence corresponding points with confidence coefficient higher than a threshold value in each region in the weighted corresponding set; s73, judging whether the high confidence corresponding coverage of each area is balanced and sufficient; And S74, if the area with unbalanced coverage or insufficient coverage exists, performing at least one operation of supplementing corresponding points from the area with insufficient coverage, resampling partial corresponding points of the area with excessive concentration, reducing the weight of the corresponding points of the area with excessive concentration or eliminating partial low-confidence corresponding points of the area with excessive concentration on the weighted corresponding set to improve the distribution balance.
8. The method for registering the complex part point cloud for blue light scanning detection of the small five-axis machine tool according to claim 1, wherein the confidence weighted point-to-face fine registration in the step S8 comprises the following steps: s81, constructing a point-to-surface distance error objective function weighted by a robust kernel function, wherein the error term of each corresponding point is weighted by the confidence weight of the error term; S82, taking the optimized corresponding set and the initial pose as inputs, and solving a rotation matrix and a translation vector which minimize the objective function by adopting an iterative optimization algorithm; S83, periodically updating the point-to-surface soft corresponding relation and the confidence weight of each corresponding point in the iterative optimization process; S84, when the iteration meets the convergence condition or reaches the maximum iteration number, outputting final rigid body transformation parameters, and synchronously outputting detection information comprising deviation statistics, coverage indexes and unobservable region marks.
9. A storage medium is characterized in that the storage medium stores instructions and data for realizing the complex part point cloud registration method for blue light scanning detection of a small five-axis machine tool according to any one of claims 1-8.
10. The complex part point cloud registration device for the blue light scanning detection of the small five-axis machine tool is characterized by comprising a processor and a storage medium, wherein the processor loads and executes instructions and data in the storage medium to realize the complex part point cloud registration method for the blue light scanning detection of the small five-axis machine tool according to any one of claims 1-8.

Description

Complex part point cloud registration method and device for blue light scanning detection of small five-axis machine tool Technical Field The invention relates to the technical field of three-dimensional measurement and point cloud data processing, in particular to a complex part point cloud registration method and device for blue light scanning detection of a small five-axis machine tool. Background In the application scene of blue light scanning detection of a small five-axis machine tool, point cloud registration is a key link for realizing unified reference of 'measuring point cloud-machine tool coordinates-CAD digital-analog', and the accuracy and repeatability of subsequent point-to-surface deviation calculation, geometric tolerance assessment and key feature size extraction are directly determined by the result. Because structural features such as free curved surfaces, tiny chamfers/orifices, multiple reference surfaces and the like commonly exist in complex parts, blue light scanning is easily influenced by shielding, reflection and edge shaking to generate noise and missing, if registration errors or local drift are not effectively restrained, the whole deviation and local errors of a deviation cloud picture are amplified or covered, misjudgment or missed judgment is further caused, and closed loop adjustment and on-site quality control of a working procedure are influenced. Therefore, the point cloud registration method with high robustness, high precision and engineering feasibility is constructed aiming at the scene, and is a foundation and a premise for improving the on-site detection reliability of the small five-axis machine tool. The technical difficulty of point cloud registration in the application scene is mainly characterized in that complex parts often contain high-frequency geometric structures such as thin walls, sharp edges, orifices, chamfers, free curved surfaces and the like at the same time, local scale differences are obvious, so that feature extraction and matching of fixed neighborhood scales are difficult to achieve both detail fidelity and noise-resistant stability, meanwhile, burrs, outliers and local jitter are easy to occur in the edges and high-reflection areas in blue light scanning, regional defects occur due to machine tool posture, shielding and view angle limitation, the corresponding relation is incomplete, mismatching is easy to mix, and therefore registration optimization is easy to sink into local optimum or posture drift occurs. In addition, the detection task emphasizes the overall stability of the deviation field after registration, namely even if the local matching precision is higher, if the spatial distribution of the matching points is uneven and excessively concentrated on a certain plane or local area, the numerical condition of the rigid body transformation solution is poor, global alignment offset and error map distortion are caused, and the requirements of on-site detection on calculation efficiency, repeatability and process controllability are overlapped, so that the realization of stable high-precision registration under the condition of coexistence of noise, missing and high-density data is more challenging. At present, in the point cloud registration of blue light scanning detection of a small five-axis machine tool, a technical route of rough registration-fine registration is generally adopted: In the rough registration stage, candidate corresponding relations are established mainly through matching of geometric key points and local descriptors, and error correspondence is removed through combination of consistency test (such as random sampling consistency estimation), so that reliable initial pose is obtained; in order to improve the registrability under the conditions of shielding, insufficient overlapping or structural repetition, a matching strategy based on local geometric constraint (such as introducing normal consistency, local curved surface type consistency, primitive geometric invariants and the like) is provided to improve the discrimination and robustness of the corresponding screening, reduce mismatching and enlarge the initial range of convergence; In the fine registration stage, the common practice is to perform iterative optimization by adopting a point-to-surface or weighted point-to-surface error model, inhibit high residual points by matching with a robust kernel function so as to improve the adaptability to outlier and measurement noise and improve final alignment precision, and the improvement of hierarchical sampling, area balance point selection, multi-scale processing and the like for improving the efficiency and result stability of detection application, so that the calculation amount is reduced, the key structural features are maintained, the excessive concentration of matching points is avoided, and the repeatability and the interpretability of registration results in deviation analysis and size assessment are improve