A high-precision gear visual multi-dimensional detection method and system

Abstract

The application relates to the technical field of gear detection, and provides a high-precision gear visual multi-dimensional detection method and system, which comprises the following steps: acquiring original three-dimensional point cloud data of a gear to be detected, multispectral optical reflection image information of a surface area of the gear to be detected and a standard geometric structure model of the gear to be detected; based on the multispectral optical reflection image information, fitting processing is performed on the original three-dimensional point cloud data and the standard geometric structure model of the gear to be detected, so that intermediate three-dimensional point cloud data of the gear to be detected are obtained; according to infrared reflection image information in the multispectral optical reflection image information, the intermediate three-dimensional point cloud data of the gear teeth of the gear to be detected are corrected, so that target three-dimensional point cloud data of the gear to be detected are obtained; and quality detection is performed on the gear to be detected based on the target three-dimensional point cloud data of the gear to be detected. Through the scheme, the accuracy of gear quality detection can be improved.

Description

Technical Field

[0001] This application relates to the field of gear inspection technology, and in particular to a high-precision visual multidimensional inspection method and system for gears. Background Technology

[0002] In the field of modern intelligent manufacturing, the production of high-precision gears places extremely stringent requirements on quality control. This necessitates not only precise measurement of gear dimensions but also a comprehensive inspection of their surface quality. Traditional inspection methods, such as single-sensor or manual sampling, struggle to simultaneously capture multiple features of the gear, including its contours, pitch deviations, and minute defects. Therefore, advanced vision inspection systems are widely adopted. These systems typically employ optical devices with multiple viewing angles, combined with multi-spectral light imaging and 3D reconstruction techniques, to simultaneously acquire and compare all gear features, overcoming the limitations of traditional inspection methods under complex working conditions.

[0003] However, as downstream applications place increasingly stringent demands on gear performance, such as in aerospace or high-speed transmission fields, gear manufacturing processes have incorporated special functional coating deposition steps to enhance their wear resistance and corrosion resistance. This coating possesses strong specular reflection and translucency, and its thickness and optical properties exhibit unpredictable, minute variations across different parts of the gear and between different products. This non-uniformity leads to randomly appearing overexposed areas of varying location and intensity on the gear surface during image acquisition, as well as image blurring caused by subsurface scattering.

[0004] Specifically, the highlight areas appear as large areas of overexposed white spots in the camera image, causing sensor signal saturation and loss of detail. This makes it impossible for 3D reconstruction methods that rely on identifying surface feature points to find effective information in these highlight areas, hindering feature matching between viewpoints. The resulting 3D data contains numerous voids and missing data in key curvature sections of the tooth surface, making accurate measurement of the tooth profile impossible. Furthermore, during surface defect detection, the system incorrectly identifies these highlight patches caused by specular reflection as "surface anomalies," frequently misreporting them as large-area surface defects, leading to a sharp increase in the false positive rate.

[0005] More problematic is that this coating exhibits a degree of translucency under certain wavelengths of light. Light not only reflects off the surface but also penetrates the coating, scattering at the interface between the coating and the metal substrate before refracting back and being captured by the camera. This subsurface scattering effect blurs the true, minute geometric contours of the gear surface. For example, tiny machining marks or grinding textures that existed before coating deposition and were on the edge of acceptable quality, which should have been detected, are now "softened" and masked by light scattering within the coating, rendering the system unable to effectively identify them and resulting in significant missed detections. Multiple spectral analysis methods originally designed to differentiate between different material defects are now completely ineffective because they cannot distinguish whether signal changes originate from actual surface imperfections or from the complex light propagation paths within the coating. Summary of the Invention

[0006] This application provides a high-precision gear visual multidimensional inspection method and system, which aims to solve the problems in the prior art where, after the deposition of special functional coatings on the surface of high-precision gears, the strong specular reflection and semi-transparent characteristics of the coatings lead to overexposed areas of highlights and image blurring caused by subsurface scattering, resulting in missing 3D reconstruction data, false alarms of surface defects, and missed detections.

[0007] The technical solution of this application is as follows: In a first aspect, this application discloses a high-precision gear visual multidimensional inspection method, comprising the following steps: acquiring the original three-dimensional point cloud data of the gear to be inspected, the multispectral optical reflection image information of the surface region of the gear to be inspected, and the standard geometric structure model of the gear to be inspected; fitting the original three-dimensional point cloud data with the standard geometric structure model of the gear to be inspected based on the multispectral optical reflection image information to obtain the intermediate three-dimensional point cloud data of the gear to be inspected; correcting the intermediate three-dimensional point cloud data of the gear teeth to be inspected based on the infrared reflection image information in the multispectral optical reflection image information to obtain the target three-dimensional point cloud data of the gear to be inspected; and performing quality inspection on the gear to be inspected based on the target three-dimensional point cloud data of the gear to be inspected.

[0008] This technical solution can comprehensively utilize three-dimensional point cloud data and multispectral optical reflection image information. Through multidimensional data fusion and correction, it can effectively solve the detection challenges posed by high-precision gear surface coatings, thereby improving the accuracy and reliability of detection.

[0009] Furthermore, when acquiring the original 3D point cloud data of the gear to be inspected, multiple image acquisition devices are arranged around the gear to be inspected. Specifically, this includes: sequentially projecting multiple sinusoidal stripe patterns onto the gear to be inspected, and controlling the multiple image acquisition devices to acquire original images to obtain multiple sets of original images; the stripe frequencies of the multiple sinusoidal stripe patterns are different, and the multiple sets of original images correspond one-to-one with the multiple stripe frequencies. The original image sets include the original images acquired by each of the multiple image acquisition devices; and determining the original 3D point cloud data of the gear to be inspected based on the multiple sets of original images.

[0010] This technical solution, employing multi-frequency sinusoidal fringe projection and multi-angle image acquisition, can effectively obtain high-precision raw 3D point cloud data, laying the foundation for subsequent detection.

[0011] Based on this, the original 3D point cloud data of the gear to be detected is determined according to multiple sets of original images. Specifically, this includes: for each set of original images, performing Fourier transform and phase calculation on multiple original images in the set to obtain the wrapped phase; combining multiple wrapped phases based on the multi-frequency heterodyne principle to calculate the absolute phase of the surface of the gear to be detected; and determining the coordinates of each pixel in 3D space based on the absolute phase of the surface of the gear to be detected, based on the triangulation principle, to obtain the original 3D point cloud data of the gear to be detected.

[0012] This technical solution utilizes Fourier transform, phase calculation, and multi-frequency heterodyne principle to accurately calculate the absolute phase of the surface of the gear to be tested, and achieves high-precision three-dimensional point cloud reconstruction through triangulation principle, overcoming the limitations of traditional methods in complex surface reconstruction.

[0013] Based on the above, this application further proposes to obtain intermediate three-dimensional point cloud data of the gear under test by fitting the original three-dimensional point cloud data with the standard geometric structure model of the gear under test based on multispectral optical reflection image information. Specifically, this includes: determining the reflection intensity value of each region on the surface of the gear under test based on multispectral optical reflection image information; normalizing the reflection intensity value of each region to obtain the optical interference value of each region on the surface of the gear under test; and fitting the original three-dimensional point cloud data with the standard geometric structure model of the gear under test based on the optical interference value of each region on the surface of the gear under test to obtain intermediate three-dimensional point cloud data of the gear under test.

[0014] This technical solution introduces multispectral optical reflection image information. By analyzing the reflection intensity value and optical interference value, it is possible to effectively identify and quantify the impact of the coating on the three-dimensional point cloud data, thereby enabling targeted fitting processing and reducing errors caused by specular and subsurface scattering.

[0015] More specifically, in some implementations, the reflection intensity value of each region of the surface of the gear to be tested is determined based on multispectral optical reflection image information, including: for each optical reflection image in the multispectral optical reflection image information, determining the sub-reflection intensity value of each region of the surface of the gear to be tested in the optical reflection image; and taking the average of the multiple sub-reflection intensity values ​​of each region of the surface of the gear to be tested as the reflection intensity value of each region of the surface of the gear to be tested.

[0016] This technical solution effectively reduces the impact of single image acquisition errors and improves the accuracy and robustness of reflection intensity value determination by averaging the sub-reflection intensity values ​​of multiple optical reflection images.

[0017] Preferably, based on the optical interference values ​​of each region on the surface of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain intermediate three-dimensional point cloud data of the gear to be tested. This includes: for each region on the surface of the gear to be tested, determining W_i = exp(-k × O_i); W_i is the first weight value of the i-th region, O_i is the optical interference value of the i-th region, and k is a preset coefficient; based on the first weight values ​​of each region in the standard geometric structure model of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain intermediate three-dimensional point cloud data of the gear to be tested.

[0018] This technical solution introduces a weighting factor based on optical interference values, which can adaptively adjust the weights in the fitting process according to the degree of influence of the coating on different areas, making the fitting results more accurately reflect the true geometry of the gear and effectively suppressing errors in the highlight area.

[0019] Based on the above, this application further proposes to perform fitting processing on the original three-dimensional point cloud data and the standard geometric structure model of the gear under test according to the first weight value of each region in the standard geometric structure model of the gear under test, so as to obtain the intermediate three-dimensional point cloud data of the gear under test. Specifically, this includes: for each region on the surface of the gear under test, taking the difference between 1 and the first weight value of the region as the second weight value of the region in the original three-dimensional point cloud data of the gear under test; and performing nonlinear fitting on the original three-dimensional point cloud data and the standard geometric structure model according to the first weight value and the second weight value of each region to obtain the intermediate three-dimensional point cloud data of the gear under test.

[0020] This technical solution employs a nonlinear fitting method, combined with a first weight value and a second weight value, to more precisely balance the contributions of the original 3D point cloud data and the standard geometric structure model in the fitting process, thereby obtaining more accurate intermediate 3D point cloud data and effectively compensating for the lack of highlights in the original data.

[0021] Furthermore, this application proposes to correct the intermediate three-dimensional point cloud data of the teeth of the gear to be detected based on the infrared reflection image information in the multispectral optical reflection image information to obtain the target three-dimensional point cloud data of the gear to be detected. Specifically, this includes: extracting the two-dimensional edge contour information of the teeth of the gear to be detected from the infrared reflection image information using an edge detection algorithm; and correcting the intermediate three-dimensional point cloud data of the teeth of the gear to be detected based on the two-dimensional edge contour information of the teeth to be detected to obtain the target three-dimensional point cloud data of the gear to be detected.

[0022] This technical solution utilizes infrared reflection image information to correct the edge of the gear teeth, effectively overcoming the edge blurring problem caused by the semi-transparency of the coating under visible light, accurately extracting the true edge contour of the gear teeth, and improving the accuracy of gear tooth geometric parameter measurement.

[0023] Furthermore, based on the two-dimensional edge contour information of the teeth of the gear to be tested, the intermediate three-dimensional point cloud data of the teeth of the gear to be tested is corrected to obtain the target three-dimensional point cloud data of the gear to be tested. This includes: projecting the two-dimensional edge contour information of the teeth of the gear to be tested onto the intermediate three-dimensional point cloud data of the teeth of the gear to be tested; and correcting the edge contour information in the intermediate three-dimensional point cloud data of the teeth of the gear to be tested to two-dimensional edge contour information to obtain the target three-dimensional point cloud data of the gear to be tested.

[0024] This technical solution accurately projects and corrects two-dimensional edge contour information into three-dimensional point cloud data, effectively correcting the blurring and deformation of tooth edges caused by subsurface scattering, so that the final target three-dimensional point cloud data more accurately reflects the true geometry of the tooth.

[0025] Secondly, this application also discloses a high-precision gear vision multidimensional inspection system, comprising: an acquisition device and a processing device; the acquisition device is used to acquire the original three-dimensional point cloud data of the gear to be inspected, the multispectral optical reflection image information of the surface area of ​​the gear to be inspected, and the standard geometric structure model of the gear to be inspected; the processing device is used to fit the original three-dimensional point cloud data and the standard geometric structure model of the gear to be inspected based on the multispectral optical reflection image information to obtain the intermediate three-dimensional point cloud data of the gear to be inspected; the processing device is used to correct the intermediate three-dimensional point cloud data of the gear teeth to be inspected based on the infrared reflection image information in the multispectral optical reflection image information to obtain the target three-dimensional point cloud data of the gear to be inspected; the processing device is used to perform quality inspection on the gear to be inspected based on the target three-dimensional point cloud data of the gear to be inspected. Beneficial effects

[0026] This application discloses a high-precision visual multidimensional inspection method for gears. It acquires the original 3D point cloud data, multispectral optical reflectance image information, and a standard geometric structure model of the gear to be inspected. Based on the multispectral optical reflectance image information, the original 3D point cloud data and the standard geometric structure model are fitted to obtain intermediate 3D point cloud data. This step effectively utilizes multispectral information to identify and quantify optical interference caused by the gear surface coating, such as overexposure of highlights and subsurface scattering. Through weighted fitting and other methods, it compensates for the lack and errors in the highlight areas of the original 3D point cloud data, thereby overcoming the problems of incomplete and inaccurate 3D reconstruction data caused by coating characteristics in existing technologies.

[0027] Furthermore, this application corrects the intermediate three-dimensional point cloud data of the gear teeth to be tested based on the infrared reflection image information in the multispectral optical reflection image information, thereby obtaining the target three-dimensional point cloud data. Infrared light has good penetration into the coating, which can effectively avoid the edge blurring problem caused by the semi-transparency of the coating under visible light, thereby accurately extracting the true edge contour of the gear teeth and correcting the geometric distortion of the gear teeth caused by subsurface scattering in the intermediate three-dimensional point cloud data.

[0028] Finally, by performing quality inspection on the gear under test based on the target 3D point cloud data, high-precision and high-completeness 3D geometric data and surface defect information of the gear can be obtained. This method effectively solves the detection problem caused by the surface coating of high-precision gears in the prior art, avoids false alarms and missed detections in the detection of complex coated surfaces by traditional methods, and significantly improves the accuracy and reliability of gear quality inspection. It is especially suitable for fields such as aerospace where gear performance requirements are extremely high. Attached Figure Description

[0029] Figure 1 A flowchart illustrating a high-precision gear visual multidimensional inspection method provided in this application; Figure 2 A flowchart illustrating another high-precision gear visual multidimensional inspection method provided in this application; Figure 3 This is a schematic diagram of the structure of a high-precision gear vision multidimensional inspection system provided in this application. Detailed Implementation

[0030] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0032] In the field of modern intelligent manufacturing, the production of high-precision gears places extremely stringent requirements on quality control. This necessitates not only precise measurement of gear dimensions but also a comprehensive inspection of their surface quality. Traditional inspection methods, such as single-sensor or manual sampling, struggle to simultaneously capture multiple features of the gear, including its contours, pitch deviations, and minute defects. Therefore, advanced vision inspection systems are widely adopted. These systems typically employ optical devices with multiple viewing angles, combined with multi-spectral light imaging and 3D reconstruction techniques, to simultaneously acquire and compare all gear features, overcoming the limitations of traditional inspection methods under complex working conditions.

[0033] However, as downstream applications place increasingly stringent demands on gear performance, such as in aerospace or high-speed transmission fields, gear manufacturing processes have incorporated special functional coating deposition steps to enhance their wear resistance and corrosion resistance. This coating possesses strong specular reflection and translucency, and its thickness and optical properties exhibit unpredictable, minute variations across different parts of the gear and between different products. This non-uniformity leads to randomly appearing overexposed areas of varying location and intensity on the gear surface during image acquisition, as well as image blurring caused by subsurface scattering.

[0034] Specifically, the highlight areas appear as large areas of overexposed white spots in the camera image, causing sensor signal saturation and loss of detail. This makes it impossible for 3D reconstruction methods that rely on identifying surface feature points to find effective information in these highlight areas, hindering feature matching between viewpoints. The resulting 3D data contains numerous voids and missing data in key curvature sections of the tooth surface, making accurate measurement of the tooth profile impossible. Furthermore, during surface defect detection, the system incorrectly identifies these highlight patches caused by specular reflection as "surface anomalies," frequently misreporting them as large-area surface defects, leading to a sharp increase in the false positive rate.

[0035] More problematic is that this coating exhibits a degree of translucency under certain wavelengths of light. Light not only reflects off the surface but also penetrates the coating, scattering at the interface between the coating and the metal substrate before refracting back and being captured by the camera. This subsurface scattering effect blurs the true, minute geometric contours of the gear surface. For example, tiny machining marks or grinding textures that existed before coating deposition and were on the edge of acceptable quality, which should have been detected, are now "softened" and masked by light scattering within the coating, rendering the system unable to effectively identify them and resulting in significant missed detections. Multiple spectral analysis methods originally designed to differentiate between different material defects are now completely ineffective because they cannot distinguish whether signal changes originate from actual surface imperfections or from the complex light propagation paths within the coating.

[0036] In this regard, such as Figure 1 As shown, this application proposes a high-precision visual multidimensional inspection method for gears, including: S101. Acquire the original three-dimensional point cloud data of the gear to be tested, the multispectral optical reflection image information of the surface area of ​​the gear to be tested, and the standard geometric structure model of the gear to be tested.

[0037] S102. Based on multispectral optical reflection image information, the original three-dimensional point cloud data is fitted with the standard geometric structure model of the gear to be tested to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

[0038] S103. Based on the infrared reflection image information in the multispectral optical reflection image information, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected to obtain the target three-dimensional point cloud data of the gear to be detected.

[0039] S104. Perform quality inspection on the gear under test based on the target three-dimensional point cloud data of the gear under test.

[0040] This method can effectively solve the problems of high-gloss overexposure and subsurface scattering caused by coatings on the surface of high-precision gears, and significantly improve detection accuracy and reliability.

[0041] The "raw 3D point cloud data" mentioned in this application refers to the set of discrete points on the surface of the gear to be inspected, directly acquired by a 3D scanning device. These points contain the 3D coordinate information of each point, used to describe the initial geometry of the gear. "Multispectral optical reflectance image information" refers to the set of reflectance images of the gear surface acquired at different wavelengths (e.g., visible light, near-infrared, etc.). These images reflect the optical properties of the gear surface under different spectra. "Standard geometric structure model" refers to the ideal design model of the gear to be inspected, usually existing in the form of a CAD model or high-precision scanning data, serving as the benchmark for inspection. "Intermediate 3D point cloud data" is data that has been preliminarily fitted to the standard geometric structure model based on the raw 3D point cloud data, aiming to eliminate some initial errors. "Infrared reflectance image information" refers to specific bands in the multispectral optical reflectance image information, typically used to penetrate certain surface coatings and reveal subsurface features. "Target 3D point cloud data" is the final 3D point cloud data corrected by the infrared reflectance image information, possessing higher accuracy and reliability, used for final quality inspection.

[0042] This application provides a high-precision visual multidimensional inspection method for gears, the main steps of which include: First, the original three-dimensional point cloud data of the gear to be tested, the multispectral optical reflection image information of the surface area of ​​the gear to be tested, and the standard geometric structure model of the gear to be tested are obtained.

[0043] Acquiring raw 3D point cloud data can be achieved in several ways. For example, a laser scanner can be used to scan the gear to be inspected. The laser scanner emits a laser beam and receives the reflected light. By measuring the time of flight of light or using triangulation principles, the 3D coordinates of points on the gear surface are calculated, thereby generating raw 3D point cloud data. Another method is to use a structured light projection system to project a known pattern (such as stripes or a dot matrix) onto the gear surface, capture the deformed pattern with a camera, and then use image processing algorithms to reconstruct the 3D shape of the gear.

[0044] Acquiring multispectral optical reflectance image information can be achieved by configuring a multispectral camera system. This system can include multiple filters or tunable light sources to illuminate and image the gear surface at different wavelengths, thereby acquiring a series of reflectance images in different spectral bands. For example, images in the visible light band and near-infrared band can be acquired.

[0045] The standard geometric model of the gear to be tested is usually pre-stored CAD model data, which accurately describes the ideal geometry and size of the gear.

[0046] Secondly, based on multispectral optical reflection image information, the original three-dimensional point cloud data is fitted with the standard geometric structure model of the gear to be tested to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

[0047] This fitting process aims to optimize the matching between the original 3D point cloud data and the standard geometric model by utilizing the surface optical properties contained in multispectral optical reflection image information. For example, different weights can be assigned to points in the original 3D point cloud data based on the reflection intensity or color information of different regions in the multispectral image. During the fitting process, regions with higher reflection intensity (possibly corresponding to highlight regions) can be assigned lower weights to reduce their impact on the overall fitting result, thereby reducing the error caused by highlight regions. The fitting algorithm can employ the Iterative Closest Point (ICP) algorithm or its variants, continuously adjusting the position and orientation of the original 3D point cloud data to make it coincide as closely as possible with the standard geometric model, while considering the weights provided by the optical reflection image information.

[0048] Next, based on the infrared reflection image information in the multispectral optical reflection image information, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected to obtain the target three-dimensional point cloud data of the gear to be detected.

[0049] Infrared reflection image information has advantages in penetrating certain translucent coatings, revealing the true geometric features beneath. For example, the sharpness of tooth edges in infrared reflection images can be used to correct tooth edges blurred by subsurface scattering in intermediate 3D point cloud data. Specifically, edge detection can be performed on the infrared reflection image to extract the 2D edge contours of the teeth. These 2D edge contours are then mapped onto the intermediate 3D point cloud data, and the geometry of the corresponding regions in the point cloud data is adjusted according to the mapping relationship to align with the more realistic edge information in the infrared image. This correction helps restore tooth details distorted by optical effects, improving the accuracy of the 3D data.

[0050] Finally, the quality of the gear under test is inspected based on the target three-dimensional point cloud data of the gear under test.

[0051] After obtaining high-precision 3D point cloud data of the target, multifaceted quality inspection can be performed. For example, geometric parameters such as gear tooth profile deviation, pitch deviation, and tooth direction deviation can be calculated and compared with standard geometric structure models to assess whether they are within tolerance ranges. Simultaneously, surface defect detection can be performed on the target 3D point cloud data to identify surface defects such as scratches, pits, and burrs. Because the target 3D point cloud data has been corrected to eliminate interference from specular highlights and subsurface scattering, the inspection results will be more accurate and reliable.

[0052] The core innovation of the high-precision gear visual multidimensional inspection method proposed in this application lies in the staged fine processing of three-dimensional point cloud data by combining multispectral optical reflectance image information. Traditional methods, when dealing with gears with strong specular reflection and semi-transparent coatings, suffer from numerous holes, missing parts, and blurring in the three-dimensional reconstruction data due to high-light overexposure and subsurface scattering effects, which seriously affects the inspection accuracy.

[0053] This application introduces multispectral optical reflectance image information and first uses it to fit the original 3D point cloud data with a standard geometric structure model. Specifically, by analyzing the reflection intensity values ​​of different regions and normalizing them to optical interference values, a first weight value for each region is determined. These weight values ​​are used during the fitting process to adjust the degree of matching between the original 3D point cloud data and the standard geometric structure model, effectively reducing the negative impact of highlight regions on the fitting results, thereby obtaining more accurate intermediate 3D point cloud data. This step is significantly superior to the simple fitting of 3D point cloud data in existing technologies because it can intelligently identify and weaken the influence of optical interference regions, reducing geometric distortion caused by highlights.

[0054] Furthermore, this application utilizes infrared reflection image information from multispectral optical reflection image information to correct the intermediate three-dimensional point cloud data of the gear teeth to be inspected. Infrared light has strong penetrating power and can penetrate a semi-transparent coating to reveal the true geometric contour beneath it. By extracting the two-dimensional edge contour information of the teeth from the infrared reflection image and mapping it onto the intermediate three-dimensional point cloud data for correction, this application effectively solves the problem of tooth edge blurring caused by subsurface scattering. This correction mechanism enables the final target three-dimensional point cloud data to more accurately reflect the true geometry of the gear, including minute machining marks and defects, thereby avoiding missed detections caused by optical effects in traditional methods.

[0055] Through the refined processing described in the two stages, the method of this application overcomes the challenges posed by special coatings on the surface of high-precision gears, significantly improving the accuracy and reliability of 3D point cloud data. Compared with existing technologies, this application not only effectively handles overexposed areas with high highlights, reducing data gaps and missing data, but also uses infrared information to penetrate the semi-transparent coating, restoring the hidden true geometric details, thereby achieving a more comprehensive and accurate quality inspection of high-precision gears. This multi-dimensional information fusion and step-by-step correction strategy demonstrates significant progress and innovation in the detection of complex surface features.

[0056] This application proposes an optimized method for acquiring raw 3D point cloud data. By using multi-view, multi-frequency structured light projection and acquisition, it aims to overcome the limitations of traditional methods and ensure the comprehensiveness and high accuracy of the acquired data.

[0057] like Figure 2 As shown, in some embodiments of this application, multiple image acquisition devices are arranged around the gear to be inspected to acquire the original three-dimensional point cloud data of the gear to be inspected, including: S201. Project multiple sinusoidal stripe patterns onto the gear to be tested in sequence, and control multiple image acquisition devices to acquire original images to obtain multiple sets of original images.

[0058] The multiple sinusoidal stripe patterns have different stripe frequencies, and the multiple original image sets correspond one-to-one with the multiple stripe frequencies. The original image sets include the original images acquired by each of the multiple image acquisition devices.

[0059] S202. Determine the original 3D point cloud data of the gear to be detected based on multiple sets of original images.

[0060] Specifically, multiple image acquisition devices are arranged around the gear to be inspected. The purpose is to acquire images of the gear surface from different perspectives, thereby effectively avoiding occlusion problems from a single perspective and ensuring complete information about the surface of the gear to be inspected. The image acquisition devices can be understood as any device capable of capturing optical images, such as a CCD camera or a CMOS camera.

[0061] In practical applications, projecting multiple sinusoidal fringe patterns sequentially onto the gear under test involves using a structured light projection device to project sinusoidal fringe patterns with specific spatial frequencies and phases onto the gear surface. These patterns deform on the gear surface due to its three-dimensional morphology, and the three-dimensional information of the gear surface can be retrieved by analyzing the deformed fringe images. Furthermore, the multiple sinusoidal fringe patterns have different fringe frequencies. This aims to utilize the multi-frequency heterodyne principle to solve the phase ambiguity problem that may occur when projecting single-frequency fringe patterns onto complex surfaces, thereby achieving accurate measurement of large-scale depth variations. Thus, multiple sets of original images correspond one-to-one with multiple fringe frequencies, and each set of original images includes original images acquired by each of the multiple image acquisition devices. This ensures the systematic and complete nature of data acquisition, providing sufficient image information for subsequent phase calculation and three-dimensional point cloud data determination.

[0062] This application's solution effectively solves the occlusion and phase blurring problems inherent in traditional single-viewpoint or single-frequency acquisition methods when obtaining high-precision 3D point cloud data of gears by setting up multiple image acquisition devices around the gear under test and combining them with multi-frequency sinusoidal fringe projection technology. Specifically, multiple image acquisition devices acquire images synchronously or asynchronously from different angles, capturing deformation fringes in various regions of the gear surface, thus avoiding local occlusion caused by the complex geometry of the gear and ensuring comprehensive data acquisition. Simultaneously, multiple sinusoidal fringe patterns of different frequencies are projected sequentially, and the multiple sets of original images formed by these patterns can be used for phase calculation based on the multi-frequency heterodyne principle. This method can combine wrapping phases of different frequencies to generate an equivalent low-frequency wrapping phase, effectively expanding the measurement range and solving the 2π phase jump blurring problem that may occur under single frequency, ensuring accurate calculation of the absolute phase. Finally, through the principle of triangulation combined with precise absolute phase information, the coordinates of each pixel on the surface of the gear under test in 3D space can be determined with high precision, resulting in high-quality, high-precision original 3D point cloud data.

[0063] Through the above technical solution, this application overcomes the problems of occlusion, insufficient accuracy, and phase ambiguity inherent in traditional 3D point cloud data acquisition methods when dealing with high-precision gears. Specifically, the surround setup of multiple image acquisition devices significantly improves the integrity of data acquisition, reducing blind spots and information loss. The projection and acquisition of multi-frequency sinusoidal fringe patterns, combined with subsequent phase calculation, greatly enhances the measurement accuracy and robustness of 3D point cloud data, especially when processing gears with complex curved surfaces and large depth variations, enabling the acquisition of more accurate absolute phase information. Therefore, this application can acquire high-quality, high-precision original 3D point cloud data of the gear under inspection, providing reliable basic data for subsequent precision inspection, thereby effectively improving the overall performance and reliability of the high-precision gear visual multidimensional inspection method.

[0064] Specifically, the original 3D point cloud data of the gear to be detected, determined from multiple sets of original images, includes: For each set of multiple original images, Fourier transform and phase calculation are performed on the multiple original images in the set to obtain the wrapped phase. Based on the multi-frequency heterodyne principle, the multiple wrapped phases are combined to calculate the absolute phase of the surface of the gear to be detected. Based on the triangulation principle, the coordinates of each pixel in three-dimensional space are determined according to the absolute phase of the surface of the gear to be detected, thus obtaining the original three-dimensional point cloud data of the gear to be detected.

[0065] The Fourier transform is a mathematical tool that converts an image from the spatial domain to the frequency domain, aiming to extract phase information from a periodic sinusoidal fringe image. Phase decomposition refers to extracting the wrapped phase from the spectrum obtained by the Fourier transform using a specific algorithm. This wrapped phase typically varies periodically within the range of [-π, π] or [0, 2π], reflecting the local positional information of the fringe pattern in the image.

[0066] Furthermore, the multi-frequency heterodyne principle is a technique for phase unwrapping. By combining the wrapped phase from fringe patterns of different frequencies, it effectively eliminates phase ambiguity, thereby obtaining the absolute phase of the surface of the gear under test. The absolute phase is continuous, non-abrupt phase information that can accurately reflect the depth information of the object's surface, providing a reliable foundation for subsequent 3D reconstruction.

[0067] Furthermore, triangulation is a method for calculating the distance to a target point using a known baseline length and two observation angles. In this application, by utilizing the geometric relationship between the projection device and the image acquisition device, combined with the calculated absolute phase, the coordinates of each pixel on the surface of the gear under test in three-dimensional space can be accurately calculated, thereby constructing the original three-dimensional point cloud data of the gear under test.

[0068] The proposed solution first performs Fourier transform and phase calculation on the images in each original image set to extract preliminary phase information, i.e., the wrapped phase, from the periodic stripe pattern. However, due to their periodicity, these wrapped phases exhibit 2π ambiguity and cannot be directly used for 3D reconstruction. Therefore, this application further introduces the multi-frequency heterodyne principle, cleverly combining wrapped phases from stripe patterns of different frequencies to effectively eliminate phase ambiguity, thereby obtaining a unique and continuous absolute phase for each point on the surface of the gear to be inspected. Because of this accurate absolute phase, combined with the pre-calibrated geometric parameters between the projection system and the image acquisition system, and using the principle of triangulation, the precise coordinates of each pixel on the surface of the gear to be inspected in 3D space can be accurately calculated. Thus, by calculating the coordinates of all pixels, high-precision original 3D point cloud data of the gear to be inspected is finally constructed.

[0069] Through the above technical solution, this application overcomes the phase ambiguity problem that easily occurs in traditional single-frequency fringe projection technology when processing complex surfaces or objects with large depth ranges, significantly improving the reconstruction accuracy and reliability of 3D point cloud data. This method achieves robust phase unfolding through the multi-frequency heterodyne principle, ensuring the accuracy of the absolute phase, thereby making the 3D coordinates obtained based on the triangulation principle more precise. Thus, it provides a high-precision, high-reliability foundation of original 3D point cloud data for subsequent gear quality inspection, effectively improving the accuracy of the inspection results.

[0070] Specifically, in some implementations of the aforementioned high-precision gear visual multidimensional detection method, based on multispectral optical reflection image information, the original three-dimensional point cloud data is fitted with the standard geometric structure model of the gear to be detected to obtain intermediate three-dimensional point cloud data of the gear to be detected. This process can be further refined into the following steps: Based on multispectral optical reflectance image information, the original 3D point cloud data is fitted to the standard geometric structure model of the gear to be inspected to obtain intermediate 3D point cloud data of the gear to be inspected, including: The reflection intensity values ​​of each region on the surface of the gear to be tested are determined based on the multispectral optical reflection image information; the reflection intensity values ​​of each region are normalized to obtain the optical interference values ​​of each region on the surface of the gear to be tested; based on the optical interference values ​​of each region on the surface of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

[0071] Specifically, determining the reflection intensity value of each region on the surface of the gear under test refers to quantifying the degree of reflection of different wavelengths of light by different regions on the surface of the gear under test by analyzing multispectral optical reflection image information. This can be achieved by processing the pixel values ​​in the multispectral image, for example, extracting intensity information of specific bands to reflect the optical properties of the surface material or coating.

[0072] The reflection intensity values ​​of each region are normalized to eliminate the influence of differences in the measurement environment or equipment, making the reflection intensity values ​​of different regions comparable. The normalized values ​​are defined as the optical interference values ​​of each region on the surface of the gear to be inspected. These values ​​can objectively characterize the uniformity or degree of defects in the surface optical properties. For example, a high optical interference value may indicate the presence of anomalies or defects on the surface.

[0073] In practical applications, the original 3D point cloud data is fitted to the standard geometric model of the gear based on the optical interference values ​​of various regions on the surface of the gear under test. This means that the optical interference values ​​are used as weights or constraints during the geometric fitting process. For example, regions with high optical interference values ​​can be assigned lower weights during the fitting process, or they can be designated as regions requiring focused correction to ensure that the fitting results more accurately reflect the actual surface conditions, thereby obtaining more refined and reliable intermediate 3D point cloud data.

[0074] This application's solution introduces multispectral optical reflectance image information and converts it into optical interference values, enabling the fitting process between the original 3D point cloud data and the standard geometric model to fully consider the optical characteristics of the gear surface under test. Specifically, by determining the reflection intensity values ​​of each region, the optical response of different regions can be quantified. Subsequently, normalizing these reflection intensity values ​​eliminates the influence of ambient lighting or sensor differences, thereby obtaining optical interference values ​​that objectively reflect the surface characteristics. These optical interference values ​​are used to guide the fitting of point cloud data with the standard model, making the fitting process no longer solely dependent on geometric shape matching, but incorporating surface material and defect information, thus improving the accuracy and reliability of the fitting.

[0075] By employing the aforementioned technical solution, multispectral optical reflection image information of the gear surface under inspection can be effectively utilized when fitting the original 3D point cloud data to a standard geometric structure model. This allows the fitting process to consider not only the macroscopic geometry of the gear but also its microscopic surface characteristics, such as surface roughness, coating uniformity, or potential surface defects. Consequently, the resulting intermediate 3D point cloud data can more accurately reflect the true surface condition of the gear under inspection, providing a more reliable and comprehensive data foundation for subsequent quality inspection and significantly improving the accuracy and precision of high-precision gear inspection.

[0076] This application further proposes determining the reflection intensity values ​​of various regions on the surface of the gear to be inspected based on multispectral optical reflection image information, specifically including: For each optical reflection image in the multispectral optical reflection image information, determine the sub-reflection intensity value of each region on the surface of the gear to be detected in the optical reflection image; take the average of the multiple sub-reflection intensity values ​​of each region on the surface of the gear to be detected as the reflection intensity value of each region on the surface of the gear to be detected.

[0077] Specifically, multispectral optical reflection image information typically contains multiple optical reflection images acquired at different wavelengths, polarization states, or angles. When determining the reflection intensity value of each region on the surface of the gear to be inspected, the reflection intensity of each region on the surface of the gear to be inspected in each optical reflection image in the multispectral optical reflection image information can be calculated separately. These separately calculated reflection intensity values ​​are called "sub-reflection intensity values".

[0078] For example, the sub-reflection intensity value can be obtained by analyzing the brightness or grayscale value of each pixel. Then, the multiple sub-reflection intensity values ​​obtained from different optical reflection images of the same area on the surface of the gear to be inspected are averaged to obtain the final reflection intensity value for that area. This averaging process can employ various methods such as arithmetic averaging and weighted averaging, aiming to eliminate random errors or local anomalies that may exist in a single image, thereby obtaining a more representative and stable reflection intensity value.

[0079] This application's solution utilizes multiple optical reflection images from multispectral optical reflection image information and averages the sub-reflection intensity values ​​of the same region in each image. This effectively reduces the impact of random noise, local illumination fluctuations, or sensor errors that may be introduced during single image acquisition on the calculation of reflection intensity values. Through the fusion and statistical processing of multi-source information, the determined reflection intensity values ​​are more stable and accurate, thus providing more reliable input data for subsequent normalization processing and fitting of the original 3D point cloud data with a standard geometric structure model.

[0080] The above technical solution significantly improves the accuracy and robustness of determining the reflection intensity values ​​of various regions on the surface of the gear under inspection. Compared to relying solely on a single optical reflection image, this solution effectively suppresses noise and local interference by averaging multiple sub-reflection intensity values, making the determined reflection intensity values ​​more accurately reflect the optical characteristics of the gear surface. This provides a more precise basis for subsequent fitting processing, thereby improving the quality of the generated intermediate 3D point cloud data and ultimately contributing to the overall accuracy and reliability of high-precision gear visual multidimensional inspection.

[0081] This application further proposes a method for optimizing the fitting process by introducing weight values ​​to accurately reflect the impact of optical interference on the fitting process, thereby improving the robustness and accuracy of the fitting.

[0082] Specifically, based on the optical interference values ​​of various regions on the surface of the gear under test, the original 3D point cloud data is fitted to the standard geometric model of the gear under test to obtain intermediate 3D point cloud data of the gear under test. This process includes: For each region on the surface of the gear to be tested, W_i = exp(-k × O_i) is determined; W_i is the first weight value of the i-th region, O_i is the optical interference value of the i-th region, and k is a preset coefficient; based on the first weight value of each region in the standard geometric structure model of the gear to be tested, the original three-dimensional point cloud data and the standard geometric structure model of the gear to be tested are fitted to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

[0083] The optical interference value O_i can be understood as an index reflecting the local quality or defect degree of the gear surface to be inspected. The larger the value, the worse the surface quality or the greater the interference in that area. The first weight value W_i is calculated using the exponential function exp(-k×O_i), which aims to convert the optical interference value O_i into a weight coefficient between 0 and 1. When the optical interference value O_i increases, the first weight value W_i will decrease exponentially, meaning that the proportion of the original 3D point cloud data in that area during the fitting process will decrease. Conversely, when the optical interference value O_i decreases, the first weight value W_i will increase, giving the original 3D point cloud data in that area a higher degree of confidence in the fitting process. The preset coefficient k is used to adjust the sensitivity of the optical interference value to the weight value, and can be adjusted according to the actual application scenario and gear material characteristics.

[0084] This application's solution effectively addresses the problem of traditional fitting methods being overly sensitive to local surface defects or noise by introducing a weighting mechanism based on optical interference values. Specifically, when local defects or optical interference exist on the surface of the gear to be inspected, the corresponding optical interference value O_i will be higher. The first weight value W_i, calculated using the exponential function W_i=exp(-k×O_i), will be significantly reduced. In subsequent fitting processing, these lower weight values ​​reduce the fitting algorithm's dependence on the original measurement data when processing the original 3D point cloud data of that region, and instead refer more to the standard geometric structure model, thereby effectively suppressing errors introduced by surface defects or noise. Conversely, for regions with good surface quality and low optical interference values ​​O_i, the first weight value W_i will be higher, allowing the fitting algorithm to fully utilize high-quality original 3D point cloud data and accurately capture the actual geometric features of the gear. It is precisely because of this adaptive weight adjustment mechanism that the fitting process can intelligently adjust the fitting strength between the original data and the standard model according to the quality status of different regions of the gear surface, thereby obtaining more accurate and robust intermediate 3D point cloud data.

[0085] Through the above technical solution, this application can adaptively adjust the fitting weights between the original 3D point cloud data and the standard geometric structure model according to the degree of optical interference in different areas of the surface of the gear to be inspected. This significantly improves the robustness and accuracy of the fitting process, especially when there are optical interferences such as local defects, scratches, or material inhomogeneity on the surface of the gear to be inspected. Therefore, it can effectively avoid the distortion of fitting results caused by local surface defects, ensuring that the intermediate 3D point cloud data can more realistically and accurately reflect the actual geometry of the gear, providing a more reliable data foundation for subsequent quality inspection, thereby improving the overall accuracy and reliability of the high-precision gear visual multidimensional inspection method.

[0086] This application further proposes a high-precision gear visual multidimensional detection method, wherein, according to the first weight value of each region in the standard geometric structure model of the gear to be detected, the original three-dimensional point cloud data is fitted with the standard geometric structure model of the gear to be detected to obtain intermediate three-dimensional point cloud data of the gear to be detected, including: For each region on the surface of the gear to be tested, the difference between 1 and the first weight value of the region is used as the second weight value of the region in the original three-dimensional point cloud data of the gear to be tested; based on the first weight value and the second weight value of each region, the original three-dimensional point cloud data and the standard geometric structure model are nonlinearly fitted to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

[0087] Specifically, the aforementioned second weight value refers to a parameter used to measure the influence of the original 3D point cloud data on the fitting results in a specific region. It is calculated by subtracting 1 from the first weight value of the region. The first weight value reflects the confidence level of the standard geometric model in that region, while the second weight value reflects the relative importance of the original 3D point cloud data in that region or the degree to which it needs to be preserved. For example, when the optical interference value of a region is high, its first weight value is low, and the corresponding second weight value will be high. This means that when fitting in that region, the original 3D point cloud data should be given greater influence to avoid over-reliance on a potentially inaccurate standard model.

[0088] The aforementioned nonlinear fitting can be understood as a mathematical fitting method capable of handling complex data relationships and nonlinear characteristics. Compared to linear fitting, nonlinear fitting can better capture subtle changes and irregularities in the geometric shape of the gear surface being tested. Specifically, nonlinear fitting can be implemented using various algorithms, such as the least squares method and the Levenberg-Marquardt algorithm. Its purpose is to find an optimal nonlinear function or model while simultaneously considering the first and second weight values, enabling a more accurate fusion of the original 3D point cloud data and the standard geometric structure model, thereby generating intermediate 3D point cloud data that better reflects the actual situation.

[0089] This application's solution effectively addresses the aforementioned limitations by introducing a second weight value and employing nonlinear fitting. Specifically, the first weight value primarily measures the contribution of the standard geometric structure model to the fitting process; a higher value indicates less optical interference in the region and greater reference value of the standard model. However, in regions with significant optical interference, the first weight value is lower. If the standard model is weighted solely based on the first weight value, the actual geometric information contained in the original 3D point cloud data may be ignored. Therefore, this application introduces a second weight value and applies it to the original 3D point cloud data. When the first weight value is lower, the second weight value is higher. This means that in these regions with significant optical interference, the original 3D point cloud data is given a higher weight, thereby ensuring that even if noise exists in the original data, its true geometric features can be fully considered and preserved during the fitting process.

[0090] Furthermore, by employing a nonlinear fitting method, this application can handle the more complex and nonlinear relationship between the original 3D point cloud data and the standard geometric model. Nonlinear fitting allows the model to have greater flexibility during the fitting process, better adapting to the complex curvature changes and local features of the surface of the gear under test, rather than simply being a linear superposition. Thus, the first and second weight values ​​can participate in the fitting process in a more refined and dynamic way, enabling the fitting results to not only fully utilize the prior knowledge of the standard model but also effectively integrate the actual measurement information of the original point cloud data. Especially in areas with significant optical interference, by balancing the influence of both, the bias caused by a single weight is avoided.

[0091] Through the above technical solutions, this application can significantly improve the accuracy and robustness of fitting the original 3D point cloud data to the standard geometric structure model. Introducing a second weight value allows the original 3D point cloud data to receive higher weight in areas with significant optical interference and low first weight values, effectively avoiding detail loss or fitting deviations caused by over-reliance on the standard model. Simultaneously, employing a nonlinear fitting method allows the fitting process to better adapt to the complex geometric features and nonlinear changes of the gear surface under inspection, thereby generating more accurate and realistic intermediate 3D point cloud data. This combination of dual weighting and nonlinear processing enables the final intermediate 3D point cloud data to more accurately reflect the actual geometry of the gear under inspection, providing a higher-quality data foundation for subsequent quality inspection, and thus improving the overall accuracy and reliability of the inspection method.

[0092] Specifically, in the aforementioned high-precision gear visual multidimensional detection method, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected based on the infrared reflection image information in the multispectral optical reflection image information to obtain the target three-dimensional point cloud data of the gear to be detected. This can include the following steps: The two-dimensional edge contour information of the gear teeth to be detected is extracted from the infrared reflection image information by an edge detection algorithm; based on the two-dimensional edge contour information of the gear teeth to be detected, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected to obtain the target three-dimensional point cloud data of the gear to be detected.

[0093] Edge detection algorithms aim to identify regions with significant brightness variations in an image, thereby determining the boundaries of objects. Specifically, various classic edge detection methods can be employed, such as the Canny operator, Sobel operator, Prewitt operator, or Laplacian operator. These algorithms locate the boundaries of gear teeth in an image by calculating the gradient or second derivative of image pixels. Infrared reflectance image information is particularly suitable for detecting the teeth of high-precision gears due to its sensitivity to material surface characteristics and defects. Infrared imaging can effectively highlight the geometric features and potential surface defects of the teeth, providing a clear data foundation for subsequent edge extraction. Two-dimensional edge contour information refers to the set of boundary lines or curves of the gear teeth to be detected, identified by edge detection algorithms in infrared reflectance images. This contour information accurately depicts the geometry and dimensions of the teeth on a two-dimensional plane.

[0094] The intermediate 3D point cloud data of the gear teeth under test is corrected based on the 2D edge contour information of the teeth. Specifically, this involves mapping or projecting the precise 2D edge contour information extracted from the infrared reflection image onto the intermediate 3D point cloud data. This mapping corrects potential tooth edge deviations or inaccuracies in the intermediate 3D point cloud data, making the 3D point cloud data more closely resemble the actual geometry of the tooth edge region. For example, the 2D edge contour information can be used as a constraint to locally optimize or adjust the intermediate 3D point cloud data, ensuring accurate reconstruction of the tooth edges in 3D space.

[0095] The proposed solution first utilizes an edge detection algorithm to extract the two-dimensional edge contour information of the gear teeth from infrared reflection image information. This step accurately captures the geometric boundaries of the teeth on a two-dimensional plane. Subsequently, this precise two-dimensional edge contour information is used as a correction basis to correct the intermediate three-dimensional point cloud data of the gear teeth. This correction mechanism allows the three-dimensional point cloud data to more accurately reflect the actual geometry in the tooth edge region, thereby compensating for the potential accuracy deficiencies of traditional three-dimensional point cloud data at edge details.

[0096] By employing the aforementioned technical solution and utilizing infrared reflection image information for edge detection, the extraction accuracy of gear tooth edge information can be effectively improved, as infrared images are highly sensitive to surface details and defects. Therefore, applying precise two-dimensional edge contour information to correct intermediate three-dimensional point cloud data can significantly improve the geometric accuracy and detail restoration capability of the three-dimensional point cloud data of the gear teeth under inspection, especially in sharp edges and complex curved surface areas, thus providing a more reliable and accurate data foundation for subsequent quality inspection.

[0097] This application further proposes a specific method for correcting the intermediate three-dimensional point cloud data of the gear teeth based on the two-dimensional edge contour information of the gear teeth to be tested, so as to obtain more accurate target three-dimensional point cloud data of the gear to be tested.

[0098] Specifically, based on the two-dimensional edge contour information of the teeth of the gear to be inspected, the intermediate three-dimensional point cloud data of the teeth of the gear to be inspected is corrected to obtain the target three-dimensional point cloud data of the gear to be inspected, including: The two-dimensional edge contour information of the gear teeth to be tested is projected onto the intermediate three-dimensional point cloud data of the gear teeth to be tested; the edge contour information in the intermediate three-dimensional point cloud data of the gear teeth to be tested is corrected into two-dimensional edge contour information to obtain the target three-dimensional point cloud data of the gear to be tested.

[0099] The phrase "projecting the two-dimensional edge contour information of the gear teeth to be detected onto the intermediate three-dimensional point cloud data of the gear teeth" refers to the need to first establish a spatial correspondence between the two-dimensional image coordinate system and the three-dimensional point cloud coordinate system. This is typically achieved through camera calibration parameters and projection transformation matrices. Specifically, using known camera intrinsic and extrinsic parameters, each pixel on the two-dimensional edge contour extracted from the infrared reflection image can be projected back into three-dimensional space, forming a series of three-dimensional rays or curves. These three-dimensional rays or curves represent the precise position of the two-dimensional edge contour in three-dimensional space.

[0100] Furthermore, "correcting the edge contour information in the intermediate 3D point cloud data of the gear teeth to be inspected to 2D edge contour information" refers to adjusting the regions in the intermediate 3D point cloud data of the gear teeth to be inspected that are close to the aforementioned projected 3D edge contour in 3D space. Specifically, points in the intermediate 3D point cloud data that belong to the tooth edges can be identified, and their positions can be adjusted to align with the projected 3D edge contour. For example, the distance from each edge point to the nearest 3D edge contour can be calculated, and the point can be moved onto that contour, or the original edge points can be replaced with new points generated along the 3D edge contour. Thus, the edge portions in the intermediate 3D point cloud data are corrected with high-precision 2D edge contour information, resulting in more accurate target 3D point cloud data of the gear to be inspected.

[0101] This application's solution effectively integrates high-precision two-dimensional edge contour information into three-dimensional point cloud data, solving the problem of insufficient accuracy in edge regions that may exist in traditional three-dimensional point cloud data. Infrared reflection image information can typically provide clear edge information that is less affected by surface texture, and its two-dimensional edge contour has high accuracy. By projecting this precise two-dimensional edge contour information into three-dimensional space and using it as a benchmark to correct the edge portions in intermediate three-dimensional point cloud data, deviations in the three-dimensional point cloud data at the edges can be effectively corrected, allowing the three-dimensional model to more accurately reflect the actual geometric boundaries of the gear. It is precisely because of this precise edge correction mechanism that the final target three-dimensional point cloud data has higher reliability and accuracy in the critical tooth edge regions.

[0102] Through the above technical solution, the three-dimensional point cloud data of the middle teeth of the gear under test can be corrected with high precision, significantly improving the accuracy of the three-dimensional point cloud data in the edge region. This corrected target three-dimensional point cloud data can more accurately represent the actual geometry of the gear, especially in the tooth edge parts that are crucial to gear precision. Therefore, in subsequent quality inspection stages, measurements and analyses can be performed based on more accurate three-dimensional data, thereby improving the sensitivity and reliability of gear defect detection and ensuring the quality control level of high-precision gears.

[0103] This application also discloses a high-precision gear vision multidimensional inspection system, comprising: an acquisition device and a processing device; the acquisition device is used to acquire the original three-dimensional point cloud data of the gear to be inspected, the multispectral optical reflection image information of the surface area of ​​the gear to be inspected, and the standard geometric structure model of the gear to be inspected; the processing device is used to fit the original three-dimensional point cloud data and the standard geometric structure model of the gear to be inspected based on the multispectral optical reflection image information to obtain the intermediate three-dimensional point cloud data of the gear to be inspected; the processing device is used to correct the intermediate three-dimensional point cloud data of the gear teeth to be inspected based on the infrared reflection image information in the multispectral optical reflection image information to obtain the target three-dimensional point cloud data of the gear to be inspected; the processing device is used to perform quality inspection on the gear to be inspected based on the target three-dimensional point cloud data of the gear to be inspected.

[0104] This system aims to solve the problems of missing or blurred 3D reconstruction data, as well as false alarms and missed detections, caused by overexposure of high-brightness light and subsurface scattering effects when inspecting special functional coatings on the surface of traditional high-precision gears. By acquiring multi-dimensional data and performing staged fine processing using acquisition and processing devices, this system can generate high-precision and high-reliability target 3D point cloud data, thereby achieving comprehensive and accurate quality inspection of high-precision gears.

[0105] The above embodiments have already described the specific methods for acquiring the original three-dimensional point cloud data of the gear to be tested, the multispectral optical reflectance image information of the surface area of ​​the gear to be tested, and the standard geometric structure model of the gear to be tested, as well as the subsequent fitting processing, correction, and quality inspection, which will not be repeated here. It should be emphasized that the system proposed in this application realizes the above-mentioned detection method through the coordinated work of the acquisition device and the processing device.

[0106] Specifically, the acquisition device can be understood as a collection of hardware modules used to collect various types of data. For example, the acquisition device may include one or more 3D scanners for acquiring raw 3D point cloud data of the gear to be inspected; simultaneously, the acquisition device may also include one or more multispectral cameras for acquiring multispectral optical reflectance image information of the surface area of ​​the gear to be inspected. Furthermore, the acquisition device may include a data interface module for receiving a pre-stored standard geometric model of the gear to be inspected. In some embodiments, the acquisition device may be an integrated multifunctional sensor unit capable of simultaneously acquiring 3D point cloud data and multispectral image information. However, such an integrated device may have limitations in flexibility or acquisition performance in specific bands; for example, a single sensor may struggle to simultaneously capture all the necessary raw images from different viewpoints, or to efficiently acquire data at different fringe frequencies.

[0107] A processing device can be understood as a computing unit used to perform data processing and analysis. For example, a processing device can be a high-performance industrial computer, internally configured with a central processing unit, graphics processing unit, memory, and various input / output interfaces. This computer runs specialized inspection software containing algorithm modules for performing point cloud fitting, data correction, and quality inspection. The processing device can also be a distributed computing system, where different processing tasks are performed by different computing nodes. In some implementations, the processing device can be an embedded system optimized for specific inspection tasks. However, such a general-purpose or embedded processing device may require additional optimization or specific algorithms to improve processing efficiency and accuracy when processing complex multi-frequency image data.

[0108] The core innovation of the high-precision gear vision multidimensional inspection system proposed in this application lies in the organic combination of the acquisition device and the processing device, which effectively overcomes the problems of high light overexposure and subsurface scattering encountered by traditional inspection methods when dealing with gears with strong specular reflection and semi-transparent coatings.

[0109] Unlike existing technologies that typically employ a single sensor or general-purpose processing system, this system can simultaneously acquire multi-dimensional raw data through its acquisition device, including high-precision 3D point cloud data and multispectral optical reflectance image information, providing a comprehensive data foundation for subsequent refined processing. Based on this, the processing device employs a phased intelligent processing strategy. First, it uses multispectral optical reflectance image information to fit the raw 3D point cloud data with a standard geometric structure model, effectively mitigating the negative impact of highlight areas on geometric reconstruction and obtaining more accurate intermediate 3D point cloud data. Further, the processing device uses infrared reflectance image information to correct the intermediate 3D point cloud data of the gear teeth, utilizing the penetrating characteristics of infrared light to restore the true geometric details blurred by subsurface scattering.

[0110] Through the aforementioned system architecture and processing flow, this system can significantly improve the accuracy and reliability of 3D point cloud data, avoiding data gaps, missing data, false alarms, and missed detections caused by optical effects in traditional methods. This system design, which integrates multi-dimensional information fusion and intelligent correction, demonstrates significant progress and innovation in the high-precision inspection of gears with complex surface features, providing a more reliable quality control solution for the modern intelligent manufacturing field.

[0111] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A high-precision visual multidimensional inspection method for gears, characterized in that, include: Acquire the original three-dimensional point cloud data of the gear to be tested, the multispectral optical reflection image information of the surface area of ​​the gear to be tested, and the standard geometric structure model of the gear to be tested; Based on the multispectral optical reflection image information, the original three-dimensional point cloud data is fitted with the standard geometric structure model of the gear to be detected to obtain the intermediate three-dimensional point cloud data of the gear to be detected. Based on the infrared reflection image information in the multispectral optical reflection image information, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected to obtain the target three-dimensional point cloud data of the gear to be detected. Quality inspection of the gear under test is performed based on the target three-dimensional point cloud data of the gear under test.

2. The high-precision gear visual multidimensional inspection method according to claim 1, characterized in that, Multiple image acquisition devices are arranged around the gear to be inspected to acquire the original 3D point cloud data of the gear, including: Multiple sinusoidal stripe patterns are sequentially projected onto the gear to be inspected, and multiple image acquisition devices are controlled to acquire original images, resulting in multiple sets of original images. The stripe frequencies of the multiple sinusoidal stripe patterns are different, and the multiple sets of original images correspond one-to-one with the multiple stripe frequencies. The sets of original images include the original images acquired by each of the multiple image acquisition devices. The original 3D point cloud data of the gear to be detected is determined based on multiple sets of original images.

3. The high-precision gear visual multidimensional inspection method according to claim 2, characterized in that, The original 3D point cloud data of the gear to be detected is determined based on multiple sets of original images, including: For each of the multiple sets of original images, perform Fourier transform and phase calculation on the multiple original images in the original image set to obtain the wrapped phase; Based on the principle of multi-frequency heterodyne, multiple wrapped phases are combined to calculate the absolute phase of the surface of the gear to be tested. Based on the principle of triangulation, the coordinates of each pixel in three-dimensional space are determined according to the absolute phase of the surface of the gear to be detected, thus obtaining the original three-dimensional point cloud data of the gear to be detected.

4. The high-precision gear visual multidimensional inspection method according to claim 1, characterized in that, Based on the multispectral optical reflection image information, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be detected to obtain intermediate three-dimensional point cloud data of the gear to be detected, including: The reflection intensity value of each region on the surface of the gear to be tested is determined based on the multispectral optical reflection image information; The reflection intensity values ​​of each region are normalized to obtain the optical interference values ​​of each region on the surface of the gear to be tested. Based on the optical interference values ​​of each region on the surface of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

5. The high-precision gear visual multidimensional inspection method according to claim 4, characterized in that, The reflection intensity values ​​of each region on the surface of the gear to be inspected are determined based on the multispectral optical reflection image information, including: For each optical reflection image in the multispectral optical reflection image information, determine the sub-reflection intensity value of each region on the surface of the gear to be detected in the optical reflection image; The average of multiple sub-reflection intensity values ​​of each region on the surface of the gear to be tested is taken as the reflection intensity value of each region on the surface of the gear to be tested.

6. The high-precision gear visual multidimensional inspection method according to claim 4, characterized in that, Based on the optical interference values ​​of various regions on the surface of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain intermediate three-dimensional point cloud data of the gear to be tested, including: For each region on the surface of the gear to be inspected, we determine W_i = exp(-k×O_i); W_i is the first weight value of the i-th region, O_i is the optical interference value of the i-th region, and k is a preset coefficient; Based on the first weight value of each region in the standard geometric structure model of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain the intermediate three-dimensional point cloud data of the gear to be tested.

7. The high-precision gear visual multidimensional inspection method according to claim 6, characterized in that, Based on the first weight value of each region in the standard geometric structure model of the gear to be tested, the original three-dimensional point cloud data is fitted to the standard geometric structure model of the gear to be tested to obtain intermediate three-dimensional point cloud data of the gear to be tested, including: For each region on the surface of the gear to be inspected, the difference between 1 and the first weight value of the region is used as the second weight value of the region in the original three-dimensional point cloud data of the gear to be inspected. Based on the first weight value and the second weight value of each region, nonlinear fitting is performed on the original 3D point cloud data and the standard geometric structure model to obtain the intermediate 3D point cloud data of the gear to be detected.

8. The high-precision gear visual multidimensional inspection method according to claim 1, characterized in that, Based on the infrared reflection image information in the multispectral optical reflection image information, the intermediate three-dimensional point cloud data of the gear teeth to be detected is corrected to obtain the target three-dimensional point cloud data of the gear to be detected, including: Two-dimensional edge contour information of the gear teeth to be detected is extracted from the infrared reflection image information using an edge detection algorithm; Based on the two-dimensional edge contour information of the gear teeth to be tested, the intermediate three-dimensional point cloud data of the gear teeth to be tested is corrected to obtain the target three-dimensional point cloud data of the gear to be tested.

9. The high-precision gear visual multidimensional inspection method according to claim 8, characterized in that, Based on the two-dimensional edge contour information of the gear teeth to be inspected, the intermediate three-dimensional point cloud data of the gear teeth to be inspected is corrected to obtain the target three-dimensional point cloud data of the gear to be inspected, including: The two-dimensional edge contour information of the gear teeth to be tested is projected onto the three-dimensional point cloud data of the middle of the gear teeth to be tested. The edge contour information in the middle three-dimensional point cloud data of the gear teeth to be tested is corrected to two-dimensional edge contour information to obtain the target three-dimensional point cloud data of the gear to be tested.

10. A high-precision gear vision multidimensional inspection system, characterized in that, include: Acquisition device and processing device; The acquisition device is used to acquire the original three-dimensional point cloud data of the gear to be tested, the multispectral optical reflection image information of the surface area of ​​the gear to be tested, and the standard geometric structure model of the gear to be tested; The processing device is used to fit the original three-dimensional point cloud data with the standard geometric structure model of the gear to be detected based on the multispectral optical reflection image information, so as to obtain the intermediate three-dimensional point cloud data of the gear to be detected. The processing device is used to correct the intermediate three-dimensional point cloud data of the gear teeth of the gear to be detected based on the infrared reflection image information in the multispectral optical reflection image information, so as to obtain the target three-dimensional point cloud data of the gear to be detected. The processing device is used to perform quality inspection on the gear under inspection based on the target three-dimensional point cloud data of the gear under inspection.

Images