System and method for hole detection and feature measurement using dense point clouds

Abstract

A method for optical detection and measurement of features in a workpiece includes: generating, by an imaging device, an initial point cloud representing a workpiece having one or more features; down-sampling the initial point cloud to generate a downsampled point cloud; extracting, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece; cropping, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud; extracting, using the cropped Rol, boundaries and clusters representing the feature; and fitting, using the clusters from the cropped Rol, a shape representing the feature.

Description

SYSTEM AND METHOD FOR HOLE DETECTION AND FEATURE MEASUREMENT USING DENSE POINT CLOUDSCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This PCT International Patent Application claims the benefit of U.S. Provisional Patent Application Serial No. 63 / 572,452, filed April 1, 2024, the contents of which is incorporated herein by reference in its entirety.FIELD100021 The present disclosure relates generally to image processing for measuring features in a workpiece.BACKGROUND10003] Detection and measurement of features, such as holes, in a workpiece, using point clouds is a critical task with applications in various fields, including robotics, quality control, and object recognition. Traditional approaches, while effective, often suffer from requirement significant computational resources and / or time, especially when dealing with high-density point clouds.

[0004] Traditional hole detection methodology may include the following steps: input point cloud: the algorithm starts with a dense point cloud obtained through scanning; boundary extraction: the boundary of the point cloud is extracted to identify potential hole locations; cluster extraction: clusters are identified within the boundary, representing potential holes; circle fitting: circles are fitted to these clusters to estimate the center and radii of the holes; and visualization: the clusters and corresponding radii are visualized for further analysis.

[0005] Traditional techniques for feature detection have several drawbacks. Such drawbacks include computational intensity, and lack of precision, resulting in an inherentcontradiction. Running traditional algorithms for feature detection using high-density point clouds is computationally intensive and time-consuming. The precision of measurements suffers due to the sparse nature of points in clusters for low-density point clouds. Traditional techniques for feature detection are characterized by a typical contradiction: to get higher accuracy of hole measurement a more dense point cloud is required, but a more dense point cloud causes higher computational intensity.SUMMARY

[0006] The present disclosure provides a method for optical detection and measurement of features in a workpiece. The method includes: generating, by an imaging device, an initial point cloud representing a workpiece having one or more features; down-sampling the initial point cloud to generate a downsampled point cloud; extracting, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece; cropping, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud; extracting, using the cropped Rol, boundaries and clusters representing the feature; and fitting, using the clusters from the cropped Rol, a shape representing the feature.

[0007] The present disclosure also provides a system for optical detection and measurement of features in a workpiece. The system includes: an imaging device configured to generate an initial point cloud representing a workpiece having one or more features; and a controller. The controller is configured to: receive, from the imaging device, the initial point cloud; down-sample the initial point cloud to generate a downsampled point cloud; extract, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece; crop, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud; extract, using the cropped Rol, boundaries andclusters representing the feature; and fit, using the clusters from the cropped Rol, a shape representing the feature.

[0008] The present disclosure provides a method for optical detection and measurement of features in a workpiece, using a dense point cloud. The method includes: dividing the dense point cloud into a plurality of voxels of a predetermined size; down-sampling, in parallel, each voxel of the plurality of voxels to generate a downsampled point cloud; determining, in parallel, approximate boundaries and clusters using the downsampled point cloud; selecting and cropping a cropped region of interest (Rol) from the dense point cloud and using measurements of the approximate boundaries and clusters; determining, in parallel, measurements of precise boundaries and clusters using the cropped Rol; determining, in parallel, circles fitting the measurements of the precise boundaries and clusters.[0009| These and other aspects of the present disclosure are disclosed in the following detailed description of the embodiments, the appended claims, and the accompanying figuresBRIEF DESCRIPTION OF THE DRAWINGS

[0010] Further details, features and advantages of designs of the invention result from the following description of embodiment examples in reference to the associated drawings.

[0011] FIG. 1 shows a schematic block diagram of a system for optical detection and analysis of features in a workpiece, in accordance with the present disclosure.

[0012] FIG. 2 shows a schematic diagram of the system for optical detection and analysis of features in a workpiece, in accordance with the present disclosure.

[0013] FIG. 3 shows a workflow diagram including graphic illustrations of the workpiece at various steps during a process for optical detection and analysis of features in the workpiece.(0014] FIG. 4 shows a listing of steps in a first method for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure.

[0015] FIG. 5 shows a flow chart of steps in a second method for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure.

[0016] FIG. 6 shows a curved workpiece for optical detection and measurement by a system and method of the present disclosure.(0017] FIGS. 7A-7B show graphic illustrations of the curved workpiece during the process for optical detection and analysis of features in the curved workpiece.(0018] FIG. 8 shows a flow chart of steps in a third method for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure.(0019] FIG. 9 shows an example of a workpiece illustrating domain decomposition, in accordance with some embodiments of the present disclosure.10020] FIG. 10 shows an example of a workpiece illustrating voxelization, in accordance with some embodiments of the present disclosure.DETAILED DESCRIPTION(0021] Referring to the drawings, the present invention will be described in detail in view of following embodiments. The present disclosure provides a novel methodology is designed to optimize hole detection, addressing the shortcomings of previous algorithms.(0022] This advanced methodology for precise hole detection in dense point clouds is specifically tailored for applications in advanced computer vision. The proposed technique buildsupon traditional algorithms by incorporating down-sampling and region of interest (ROI) cropping, resulting in significantly improved efficiency and accuracy. We present a detailed comparison with conventional approaches, highlighting the drawbacks addressed by our novel methodology.|0023| The present disclosure provides a two-step process involving down-sampling and region of interest (ROI) cropping. The methods and systems of the present disclosure provide several advantages over conventional techniques, including improvements in efficiency, precision, and resolution of the contradiction. Down-sampling significantly reduces the computational load, making the process of the present disclosure more efficient than conventional techniques. By cropping a region of interest around holes or features of interest in the original dense cloud, the proposed process maintains a higher point density for more precise measurements. The process of the present disclosure improves resolution of the contradiction by including a high density point cloud that is limited to only the areas of interest, which typically account for only a small fraction of the overall part surface area.

[0024] The process of the present disclosure is flexible and can be adapted for variations, including different down-sampling rates, alternative boundary extraction algorithms, and varied region of interest thresholding. The process of the present disclosure addresses computational intensity and precision drawbacks of traditional hole detection algorithms. The novel approach of down-sampling and region of interest cropping demonstrates improved efficiency and accuracy. The methodology's flexibility allows for customization to different scenarios, making it a valuable advancement in the field of computer vision.(0025] FIG. 1 shows a schematic block diagram of a system 10 for optical detection and analysis of features in a workpiece, 12 defining a plurality of holes 14.(0026] The system 10 includes a controller 20 having a first processor 22 operably connected to a first storage memory 24. The first storage memory 24 stores instructions, such as program code for execution by the first processor 22. The first storage memory 24 also holds data to be used by the first processor 22. An imaging device 30 is functionally connected to the controller 30 and configured to capture one or more images of the workpiece 12. The imaging device 30 may include one or more cameras, scanners, or other image capturing hardware. The imaging device 30 may include one or more lenses, illumination sources, and / or other hardware and / or software for capturing and processing images of the workpiece 12. The imaging device 30 may capture images of the workpiece in the visual light spectrum and / or in invisible spectra, such as infrared or ultraviolet.

[0027] The system 10 also includes a server 50, which may include one or more computers, and which are located remotely from controller 20. The server 50 includes a second processor 52 operably coupled to a second storage memory 54. The server 50 is configured to communicate with the controller 20 via one or more data networks 56.

[0028] FIG. 2 shows a schematic diagram of the system 10 for optical detection and analysis of features in a workpiece 12. As shown in FIG. 2, the imaging device 30 includes a 3- dimensional (3D) scanner 32 which is configured to produce data in the form of a text file 34 containing coordinates of points on the workpiece 12. The 3D scanner may include a stereoscopic camera, a LIDAR sensor or any other hardware capable of measuring 3D coordinates of the workpiece 12. The imaging device 30 is configured to transfer the text file 34 to the controller 20, which runs an algorithm to obtain precise measurements of features of the workpiece 12, such as locations and sizes of the holes 14.(0029] FIG. 3 shows a workflow diagram including graphic illustrations of the workpiece 12 at various steps during a process for optical detection and analysis of features in the workpiece 12. An initial (high-resolution) image is used to create a downsampled image 60 of the workpiece 12. The downsampled image 60 shows features, such as the holes 14 in white on a black background. Regions of interest (Rol) are cropped from around the detected features on the initial (high-resolution) image 62, as indicated by the square sections around the holes. These cropped Rol may have any shape, such as circular, elliptic, rectangular, or cylindrical sections. Precise measurements, such as radii and center coordinate measurements are determined using the cropped Rol, and are shown graphically as colored outlines on an output image 64.[0030| FIG. 4 shows a listing of steps in a first method 100 for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure. As can be appreciated in light of the disclosure, the order of operation within the method is not limited to the sequential execution as illustrated in FIG. 4, but may be performed in one or more varying orders as applicable and in accordance with the present disclosure.

[0031] The first method 100 includes generating, at step 102, a scanned point cloud representing a workpiece having one or more features. For example, the 3D scanner 32 may scan the workpiece 12 to generate the text file 34 containing the scanned point cloud as data denoting coordinates of points on the workpiece 12. Additionally or alternatively, the scanned point cloud may be stored in a different output format, such as in a *.ply file format.[0(132] The first method 100 also includes reading, at step 104, the scanned point cloud stored in a text file 34. For example, the controller 20 may receive and process the text file 34 to input the scanned point cloud.(0033] The first method 100 also includes down-sampling the point cloud at step 106. For example, the processor 22 may execute instructions to generate a downsampled point cloud having a fraction of the resolution of a scanned point cloud. Step 106 may include down-sampling the point cloud by a tunable factor, which may be based on a problem statement, such as particular requirements for fidelity and / or for a given processing time requirement.(0034] The first method 100 also includes applying boundary extraction algorithm on the downsampled point cloud at step 108. For example, the processor 22 may execute instructions to determine boundaries or edges of features in the downsampled point cloud.(0035] The first method 100 also includes extracting clusters from the boundary of downsampled point cloud at step 110. For example, the processor 22 may execute instructions to extract the clusters, which may represent a size and / or shape of the features in the downsampled point cloud.|0036| The first method 100 also includes fitting circles and determining approximate measurements from the extracted clusters at step 112. For example, the processor 22 may execute instructions to fit circles with a size and shape based on the extracted clusters and to determine an approximate size of a feature of the workpiece, such as a hole corresponding to the clusters.The first method 100 also includes cropping regions of interest (Rol) from the original dense point cloud, using the approximate measurements, at step 114. For example, the processor 22 may execute instructions to isolate one or more Rol from the scanned point cloud and containing the fitted circles produced at step 112. Step 114 may include using a tunable threshold to determine features, such as size and / or shape of the Rol. For example, the controller 20 may be configured to crop each Rol a given number of pixels larger than an approximated size of the corresponding feature, where the given number of pixels is a tunable threshold.(0037] The first method 100 also includes extracting a boundary from the cropped point cloud at step 116. For example, the processor 22 may execute instructions to determine a boundary representing an edge of the feature in the cropped Rol. Because the source cropped Rol contains high resolution data, the extracted boundary more accurately represents the actual feature of the workpiece when compared with the lower-resolution boundary determined at step 108.

[0038] The first method 100 also includes extracting clusters from the boundary of the cropped point cloud, at step 118. For example, the processor 22 may execute instructions to extract the clusters, which may represent a size and / or shape of the features in the cropped Rol and with higher accuracy than the clusters extracted at step 110.

[0039] The first method 100 also includes fitting circles and determining precise measurements from the extracted clusters at step 120. For example, the processor 22 may execute instructions to fit circles with a size and shape based on the extracted clusters in the cropped Rol and to determine precise measurements of the feature of the workpiece, such as a hole corresponding to the clusters.

[0040] FIG. 5 shows a flow chart of steps in a second method 200 for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure. As can be appreciated in light of the disclosure, the order of operation within the method is not limited to the sequential execution as illustrated in FIG. 5, but may be performed in one or more varying orders as applicable and in accordance with the present disclosure.

[0041] The second method 200 includes generating, at step 202, by an imaging device, an initial point cloud representing a workpiece having one or more features. For example, the 3D scanner 32 may scan the workpiece 12 to generate the text file 34 containing the scanned pointcloud as data denoting coordinates of points on the workpiece 12. Additionally or alternatively, the scanned point cloud data may be stored and maintained in computer memory.

[0042] The second method 200 also includes down-sampling, at step 204, the initial point cloud to generate a downsampled point cloud. For example, the processor 22 may execute instructions to generate a downsampled point cloud having a fraction of the resolution of the initial point cloud. Step -204 may include down-sampling the point cloud by a tunable factor, which may be based on a problem statement, such as particular requirements for fidelity and / or for a given processing time requirement.(0043) The second method 200 also includes extracting, at step 206, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece. For example, the processor 22 may execute instructions to determine boundaries and / or clusters, which may represent a size and / or shape of the features in the downsampled point cloud.(0044) The second method 200 also includes cropping, at step 208, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud. For example, the processor 22 may execute instructions to isolate one or more Rol from the scanned point cloud around and based on the extracted boundaries and / or clusters from step 206. Step 208 may include using a tunable threshold to determine features, such as size and / or shape of the Rol. For example, the controller 20 may be configured to crop each Rol a given number of pixels larger than an approximated size of the corresponding feature, where the given number of pixels is a tunable threshold.(0045) The second method 200 also includes extracting, at step 210, using the croppedRol, boundaries and clusters representing the feature. For example, the processor 22 may executeinstructions to determine a boundary representing an edge of the feature in the cropped Rol and / or to extract clusters representing a size and / or shape of the features in the cropped Rol. Because the cropped Rol contains high resolution data, the extracted boundary more accurately represents the actual feature of the workpiece when compared with the lower-resolution boundaries and / or clusters determined at step 206.

[0046] The second method 200 also includes fitting, at step 212, using the clusters from the cropped Rol, a shape representing the feature. For example, the processor 22 may execute instructions to fit a circle to a feature that is known to be a circular-shaped hole. Precise measurements, such as center location, and radius length, of the fitted circle can be determined. However, the shape may be different for a given feature. For example, a rectangle or a pentagon may be fit to another feature known to have a corresponding rectangular or pentagon-shaped profile.

[0047] The method of the present disclosure provides more accurate measurement of the features when compared with alternative techniques, such as techniques that use only a low-density point cloud (PC), while also providing efficiency gains over alternative techniques that use an initial (i.e. high-density) point cloud throughout.

[0048] The workpiece 12 may be a calibrated bent part having a square shape 200 mm x200 mm. with circular holes having sizes and center locations as set forth in Table 1, below:Table 1

[0049] FIG. 6 shows a curved workpiece for optical detection and measurement by a system and method of the present disclosure. The curved workpiece has a 3-dimensional (3D) shape and may also be called an angled workpiece. The curved workpiece includes three holes having radius 25mm, 20mm, and 15mm.|0050] FIGS. 7A-7B show graphic illustrations of the curved workpiece during the process for optical detection and analysis of features in the workpiece. Table 2, below, shows comparison data for the second experiment including results from each of the method of the present disclosure and from an alternative method using a low-density angled / curved point cloud (PC). As shown, the cropped method of the present disclosure provides substantially smaller fitting errors than the low-density angled PC method.Table 2(0051 ] The proposed methodology includes: Down-sampling: The dense point cloud is downsampled to reduce computational complexity while maintaining critical features. Boundary and Cluster Extraction (Approximate): The algorithm is applied to the downsampled cloud to extract boundaries and clusters, providing approximate center and radii measurements. Region of Interest Cropping: Using the approximate measurements, a region of interest is cropped from the original dense point cloud. Boundary and Cluster Extraction (Precise): The algorithm is reapplied to the cropped point cloud to extract precise boundaries and clusters. Circle Fitting (Precise): Circles are fitted to these precise clusters, resulting in accurate center and radii measurements.(0052] FIG. 8 shows a flow chart of steps in a third method 300 for optical detection and measurement of features in a workpiece, in accordance with some embodiments of the present disclosure. As can be appreciated in light of the disclosure, the order of operation within the method is not limited to the sequential execution as illustrated in FIG. 8, but may be performed in one or more varying orders as applicable and in accordance with the present disclosure.(0053] In order to significantly reduce processing time for feature extraction from dense point clouds, the third method 300 of the present disclosure provides a two-step process for 3D point cloud processing. The two-step process of the third method 300 incorporates domain decomposition, voxelization, down-sampling, and region of interest (Rol) cropping into the previously mentioned feature extraction process. One or more steps of the third method 300 may leverage parallel processing using the OpenMP compiler. However, one or more steps of the third method 300 may employ one or more distributive techniques, such as message passing interface (MPI):(0054] The third method 300 includes domain-decomposition at step 302. To achieve significant parallelism, step 302 included dividing the dense point cloud into a plurality ofsubdomains, each of the subdomains being smaller than the dense point cloud. By distributing the processing of these subdomains across multiple available OpenMP computing cores, computation times can be significantly reduced. This parallelization enables efficient handling of very large dense point clouds.

[0055] The third method 300 also includes voxelization at step 304. Step 304 includes dividing the dense point cloud into a plurality of grid cells, or voxels, each having a predetermined size. Step 304 may further enhance parallel processing. Each voxel stores the points that fall into it, effectively transforming the dense point cloud into a 3D grid. This parallelized voxelization process improves the overall efficiency of the algorithm. In some embodiments, step 304 may be performed to divide each of the subdomains into a corresponding plurality of voxels.

[0056] The third method 300 also includes down-sampling at step 306. Step 306 includes performing parallelized down-sampling using a voxelized grid, such as the plurality of voxels generated at step 304. Step 306 may maintain critical features while reducing computational complexity. Step 306 may ensure that the processing remains efficient, even with a reduced number of points.

[0057] The third method 300 also includes approximate boundary and cluster extraction at step 308. Step 308 includes applying a boundary and cluster extraction algorithm to the voxelized and downsampled point cloud in parallel to extract approximate boundaries and clusters. This parallelized extraction process provides approximate center and radii measurements, improving the overall speed and accuracy of the feature extraction.

[0058] The third method 300 also includes Rol cropping at step 310. Step 310 includes selecting and cropping of an Rol from the original dense point cloud, using the approximate measurements obtained from step 308. This Rol is chosen based on the approximatemeasurements, further optimizing the subsequent processing steps. Step 310 may include selecting and / or cropping for a plurality of different regions of interest, in parallel.

[0059] The third method 300 also includes precise boundary and cluster extraction at step 312. Step 312 includes reapplying the boundary and cluster extraction algorithm to the original, high density point cloud, using the cropped point cloud in parallel to extract precise boundaries and clusters. This parallelized extraction process enhances the accuracy of the measurements, ensuring a more precise representation of the features.

[0060] The third method 300 also includes precise circle fitting at step 314. Step 314 includes parallelized fitting of circles to the precise clusters, resulting in accurate center and radii measurements. This parallelized circle fitting step further improves the efficiency of the overall feature extraction process.[00611 The third method 300, as shown in FIG. 8, offers a comprehensive approach to 3D point cloud processing, combining parallel processing techniques for domain decomposition, voxelization, down-sampling, boundary and cluster extraction, Rol cropping, and circle fitting. While the present description is based on the circle fitting, it can be generalized to other geometrical shapes describing other typical features found in manufactured parts. This parallelization significantly reduces processing time and enhances the efficiency of the feature extraction algorithm.

[0062] FIG. 9 shows the workpiece 12 illustrating domain decomposition. As shown, the workpiece 12 is divided into four quadrants. However, the domain decomposition may include any number of domains. The domain decomposition allows for parallel processing of the large point cloud and can reduce computational time significantly.(0063] FIG. 10 shows the workpiece 12 illustrating voxelization, in accordance with some embodiments of the present disclosure. Voxelization is the process of converting a set of points or other geometric data into a regular grid of voxels (volumetric pixels / cubes). Each voxel may represent a volume in 3D space. The methods and systems of the present disclosure may use voxelization to down-sample data and for efficient processing of extracting the approximate boundary points. The domain decomposition illustrated on FIG. 9 is a kind of 3D box, and voxelization is a 3D cube, so they are topologically similar. Voxelization may be used to downsample data and for efficient processing of extracting the approximate boundary points.(0064] The system, methods and / or processes described above, and steps thereof, may be realized in hardware, software or any combination of hardware and software suitable for a particular application. The hardware may include a general purpose computer and / or dedicated computing device or specific computing device or particular aspect or component of a specific computing device. The processes may be realized in one or more microprocessors, microcontrollers, embedded microcontrollers, programmable digital signal processors or other programmable device, along with internal and / or external memory. The processes may also, or alternatively, be embodied in an application specific integrated circuit, a programmable gate array, programmable array logic, or any other device or combination of devices that may be configured to process electronic signals. It will further be appreciated that one or more of the processes may be realized as a computer executable code capable of being executed on a machine readable medium.

[0065] The computer executable code may be created using a structured programming language such as C, an object oriented programming language such as C++, or any other high- level or low-level programming language (including assembly languages, hardware descriptionlanguages, and database programming languages and technologies) that may be stored, compiled or interpreted to run on one of the above devices as well as heterogeneous combinations of processors, processor architectures, combinations of different hardware and software, or any other machine capable of executing program instructions.|0066| Thus, in one aspect, each method described above and combinations thereof may be embodied in computer executable code that, when executing on one or more computing devices performs the steps thereof. In another aspect, the methods may be embodied in systems that perform the steps thereof, and may be distributed across devices in a number of ways, or all of the functionality may be integrated into a dedicated, standalone device or other hardware. In another aspect, the means for performing the steps associated with the processes described above may include any of the hardware and / or software described above. All such permutations and combinations are intended to fall within the scope of the present disclosure.10067] The foregoing description is not intended to be exhaustive or to limit the disclosure. Individual elements or features of a particular embodiment are generally not limited to that particular embodiment, but, where applicable, are interchangeable and can be used in a selected embodiment, even if not specifically shown or described. The same may also be varied in many ways. Such variations are not to be regarded as a departure from the disclosure, and all such modifications are intended to be included within the scope of the disclosure.

Claims

CLAIMSWhat is claimed is:

1. A method for optical detection and measurement of features in a workpiece, the method comprising: generating, by an imaging device, an initial point cloud representing a workpiece having one or more features; down-sampling the initial point cloud to generate a downsampled point cloud; extracting, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece; cropping, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud; extracting, using the cropped Rol, boundaries and clusters representing the feature; and fitting, using the clusters from the cropped Rol, a shape representing the feature.

2. The method of Claim 1, further comprising: determining, based on the shape representing the feature, a measurement of the feature.

3. The method of Claim 1, wherein the feature includes a hole in the workpiece.

4. The method of Claim 3, wherein the hole in the workpiece space has a circular cross-section and the shape represents a radius and a location of the hole.

5. The method of Claim 1, wherein cropping the cropped Rol from the initial point cloud further includes determining a shape of the cropped Rol based on a shape of the approximate measurements of the feature.

6. The method of Claim 1, wherein cropping the cropped Rol from the initial point cloud further includes using a tunable threshold to determine at least one of a size and a shape of the cropped Rol.

7. The method of Claim 6, wherein cropping the cropped Rol from the initial point cloud further includes determining the cropped Rol based on a location of the approximate measurements and extended by a distance in accordance with the tunable threshold.

8. A system for optical detection and measurement of features in a workpiece, the system comprising: an imaging device configured to generate an initial point cloud representing a workpiece having one or more features; and a controller configured to: receive, from the imaging device, the initial point cloud; down-sample the initial point cloud to generate a downsampled point cloud; extract, using the downsampled point cloud, boundaries and clusters to determine approximate measurements of a corresponding feature of the workpiece; crop, using the approximate measurements, a cropped region of interest (Rol) from the initial point cloud;extract, using the cropped Rol, boundaries and clusters representing the feature; and fit, using the clusters from the cropped Rol, a shape representing the feature.

9. The system of Claim 8, wherein the controller is further configured to determine, based on the shape representing the feature, a measurement of the feature.

10. The system of Claim 8, wherein the feature includes a hole in the workpiece.

11. The system of Claim 10, wherein the hole in the workpiece space has a circular cross-section and the shape represents a radius and a location of the hole.

12. The system of Claim 8, wherein the imaging device includes a 3-dimensional (3D) scanner configured to produce data representing coordinates of points on the workpiece.

13. The system of Claim 12, wherein the 3D scanner includes at least one of a stereoscopic camera or a LIDAR sensor.

14. A method for optical detection and measurement of features in a workpiece, using a dense point cloud the method comprising: voxelization, in parallel, dividing the dense point cloud into a plurality of voxels of a predetermined size; down-sampling, in parallel, each voxel of the plurality of voxels to generate a downsampled point cloud;determining, in parallel, approximate boundaries and clusters using the downsampled point cloud; selecting and cropping a cropped region of interest (Rol) from the dense point cloud and using measurements of the approximate boundaries and clusters; determining, in parallel, measurements of precise boundaries and clusters using the cropped Rol; determining, in parallel, circles fitting the measurements of the precise boundaries and clusters.

15. The method of Claim 14, further comprising dividing the dense point cloud into a plurality of subdomains, each smaller than the dense point cloud, and wherein dividing the dense point cloud into the plurality of voxels includes dividing each subdomain of the plurality of subdomains into a corresponding plurality of voxels.

Images