A three-dimensional rapid scanning measurement system and method

Abstract

The application discloses a kind of three-dimensional fast scanning measurement system and method, system includes fuselage base, transparent scanning platform, LED lighting unit, depth camera array, data processing module and data storage module;Transparent scanning platform is set to fuselage base;Depth camera array includes upper camera group and lower camera group, for synchronous acquisition depth map and RGB chart and transmission to data processing module;LED lighting unit is used to the uniform light compensation of scanning area;Data processing module is used to the data processing of acquisition and the result is output to data storage module.The application is aimed at the problem of the shaking of the measured object scanning, through the synchronous acquisition technology of multiple depth cameras and the anti-shaking reconstruction algorithm, realize non-invasive rapid detection and improve the result accuracy, and effectively reduce the number of repeated scanning, greatly improve detection efficiency.

Description

Technical Field

[0001] This invention relates to the technical field of three-dimensional scanning and image processing, and specifically to a three-dimensional rapid scanning measurement system and method. Background Technology

[0002] 3D scanning and surface reconstruction technologies have wide applications in various fields such as industrial inspection, reverse engineering, biometric recognition, and digital modeling. Compared with traditional 2D measurement methods, 3D scanning can acquire complete spatial geometric information of a target object, providing multi-dimensional parameters such as length, width, curvature, and symmetry, which is of great value for accurate evaluation and digital modeling of target objects. Therefore, establishing a fast and accurate 3D scanning measurement method has an urgent application need and broad development prospects.

[0003] Currently, there are various methods for acquiring and reconstructing the surface morphology of 3D target objects, such as: 1) Contact measurement methods, such as using calipers, measuring tools, etc., to measure the target object point by point. This method is simple to operate and low in cost, but the measurement results are easily affected by factors such as the operator's technique and measuring force, resulting in poor measurement accuracy and repeatability. Furthermore, it can only acquire limited discrete point data and cannot reflect the complete 3D morphological features of the target object. 2) Handheld 3D scanner acquisition. The operator moves around the target object with a handheld 3D scanning device to acquire surface point cloud data through continuous shooting. However, handheld scanners rely on the operator's movement speed and stability, resulting in a long scanning cycle; and the scanning accuracy is significantly affected by the operator's own operation. When the target object moves during the scanning process, severe motion artifacts are produced. Studies have shown that handheld 3D imaging systems have systematic biases in dynamic target measurement, generating a large amount of poor-quality measurement data in applications involving moving targets. 3) Fixed multi-camera array scanning system: This system acquires images of the surface of the target object simultaneously through multiple cameras. However, the system structure is complex and the cost is high. It also has extremely high requirements for camera synchronization accuracy and calibration accuracy, making it difficult to deploy flexibly in practical applications.

[0004] Although the above methods have been applied in practice, they still have the following insurmountable drawbacks: 1) Dynamic targets (such as living organisms, operating machinery, robots, etc.) are difficult to keep still during scanning, and frequent displacement and shaking make image acquisition difficult; in addition, handheld scanners rely on the operator to move around the target to collect images, and the scanning cycle is long, which further increases the interference caused by the target movement; 2) Target movement can easily cause point cloud misalignment and model loss between consecutive frames, which seriously affects the quality of 3D reconstruction and measurement accuracy, and often requires repeated scanning, reducing detection efficiency; 3) Traditional contact measurement methods cannot establish a complete digital 3D model, making it difficult to achieve digital archiving and long-term tracking analysis of target objects.

[0005] Therefore, there is an urgent need for a high-precision 3D scanning and detection system that can complete 3D data acquisition in a very short time, has anti-motion interference capabilities, and can perform anti-shake compensation. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to propose a three-dimensional rapid scanning measurement system and method. Addressing issues such as camera shake during three-dimensional scanning of dynamic targets, the system utilizes multi-depth camera synchronous acquisition technology and an anti-shake reconstruction algorithm to achieve non-invasive rapid detection and improve result accuracy. Furthermore, it effectively reduces the number of repeated scans and increases detection efficiency.

[0007] This was achieved through the following technical solutions: A three-dimensional rapid scanning measurement system includes a base, a transparent scanning platform, an LED lighting unit, a depth camera array, a data processing module, and a data storage module. The transparent scanning platform is mounted on the base and supports the object being measured. The depth camera array includes an upper camera group and a lower camera group. The upper camera group is positioned above the transparent scanning platform, and the lower camera group is positioned below the transparent scanning platform. Both the upper and lower camera groups are communicatively connected to the data processing module to synchronously acquire depth maps and RGB images of the target area of ​​the object being measured and transmit them to the data processing module. The LED lighting unit is arranged on the base around the scanning area of ​​the transparent scanning platform to provide uniform illumination to the scanning area. The data processing module is connected to the data storage module and processes the acquired data, outputting the results to the data storage module.

[0008] Optionally, the transparent scanning platform is made of tempered glass or optical glass; the transparent scanning platform is used to support the object being measured, the upper camera group collects data from above the transparent scanning platform downwards, and the lower camera group collects data from below the transparent scanning platform upwards.

[0009] Optionally, the upper camera group includes four depth cameras: a first depth camera for capturing a first region of the target area; a second depth camera for capturing a second region of the target area; a third depth camera for capturing a third region of the target area; and a fourth depth camera for capturing a fourth region of the target area. The first, second, third, and fourth depth cameras are all positioned above the transparent scanning platform.

[0010] Optionally, the lower camera group includes two depth cameras: a fifth depth camera for acquiring the left rear region of the target area and a sixth depth camera for acquiring the right rear region of the target area. Both the fifth and sixth depth cameras are positioned below the transparent scanning platform and acquire data upwards through the transparent scanning platform.

[0011] Optionally, the LED lighting unit is a ring-shaped LED light source, which is arranged around the scanning area of ​​the transparent scanning platform to provide uniform illumination and eliminate shadows.

[0012] Optionally, the data processing module is configured to perform the following operations: control multi-camera synchronous triggering, enabling each depth camera in the depth camera array to complete synchronous exposure within the same time window; perform point cloud fusion, converting the point cloud data collected by each depth camera to a unified coordinate system for fusion; perform 3D reconstruction, generating a complete model based on the fused point cloud data; perform anti-shake compensation, compensating for point cloud offsets caused by target area shaking; perform parameter calculation, automatically calculating target parameters based on the complete 3D reconstructed model; and store each piece of data generated in point cloud fusion, 3D reconstruction, and anti-shake compensation.

[0013] Optionally, the data processing module performs anti-shake compensation in the following stages: In the first stage, synchronous acquisition is triggered, and each depth camera in the depth camera array is controlled to complete the exposure within the same time window. The second stage is key feature point extraction, which extracts multiple target points from the target area from the data collected by each depth camera to form a feature network. In the third stage, motion estimation and compensation, feature points are first extracted from the source and target point clouds using Fast Feature Histogram (FPFH). Coarse corresponding point pairs are found based on the Euclidean distance of the histograms. The initial rotation matrix R0 and initial translation vector t0 are then solved using the RANSAC random sampling consensus algorithm. Next, fine registration is performed using the ICP iterative nearest point algorithm. The centroids of the two sets of point clouds are calculated in a decentralized manner, the covariance matrix H is constructed, and the optimal rotation matrix R and translation vector t are solved using singular value decomposition (SVD). The X-axis flip angle, Y-axis pitch angle, and Z-axis yaw angle are extracted from the optimal rotation matrix R to obtain six degrees of freedom motion parameters, including three translational degrees of freedom and three rotational degrees of freedom. Each point cloud is then transformed to a unified coordinate system based on these six degrees of freedom motion parameters. The fourth stage, fusion and reconstruction, involves fusing the point clouds mapped to a unified reference coordinate system to generate a complete model in the unified coordinate system.

[0014] Optionally, the data processing module operates as follows during the fourth stage of fusion and reconstruction: Prepare standard templates corresponding to the target parts in advance; perform initial coarse registration and alignment between the standard templates and each point cloud to reduce position and pose differences; after completing the initial coarse registration and alignment, use the non-rigid ICP algorithm to calculate the energy function E to measure the degree of alignment; E = E_data + λ × E_smooth, where E_data is the fit term and is used to ensure that the template vertices move closer to the target point cloud, E_smooth is the smoothness constraint term and is used to constrain the deformation smoothness of neighboring vertices, and λ is the balance weight; After initial coarse registration, multiple alignments and registrations are performed for refinement to generate a topologically consistent complete model.

[0015] Optionally, the data storage module establishes a long-term digital archive database, which manages archives, automatically archives, and performs historical queries for each scanned object based on retrieval parameters, including name, number, or date.

[0016] Secondly, a three-dimensional rapid scanning measurement method is proposed, which is operated using the aforementioned three-dimensional rapid scanning measurement system and includes the following steps: S1. Establish a test subject profile: Enter the individual information of the test subject and store the profile data in the data storage module; S2. Place the object under test and start scanning: Place the target part of the object under test flat in the scanning area of ​​the transparent scanning platform; then use the data processing module to control the six depth cameras in the depth camera array to expose synchronously. S3. Data Acquisition: Simultaneously acquire point cloud data of the target area using a depth camera array. The point cloud data includes a depth map and an RGB map. Align the depth map and the RGB map, calculate the 3D coordinates of each pixel in the depth map by back-projecting the coordinates using the camera intrinsic parameter matrix, and assign the color of the corresponding pixel in the RGB map to the 3D point to generate 3D point cloud data. S4. Execute the anti-shake compensation algorithm: The data processing module sequentially executes four stages: synchronous trigger acquisition, key feature point extraction, motion estimation and compensation, and fusion reconstruction. In the motion estimation and compensation and fusion reconstruction stages, FPFH feature matching combined with RANSAC algorithm is first used to solve the initial transformation. Then, ICP algorithm is used to solve the optimal rotation matrix and translation vector through SVD decomposition and iteration to obtain six degrees of freedom motion parameters. After compensating the point cloud offset according to the six degrees of freedom motion parameters, the model is fused and reconstructed in a unified coordinate system to generate a complete model. S5. Execute the 3D reconstruction algorithm: The data processing module uses a standard template and performs multiple alignment and registration operations through initial coarse registration and non-rigid ICP algorithm to generate a topologically consistent complete 3D model. S6. Calculation parameters: The data processing module automatically calculates the target parameters based on the 3D point cloud data of the complete 3D model; S7. Data storage: The data storage module stores and archives any data generated by the data processing module.

[0017] The beneficial effects of this invention compared to the prior art are: 1) By synchronously triggering the six-depth camera hardware, the data acquisition of the target part is completed quickly. Then, based on ICP registration and feature matching, the motion parameters of the six degrees of freedom are calculated for motion compensation, which effectively solves the problems of point cloud misalignment, model loss and repeated scanning caused by the inability of the tested object to remain still for a long time.

[0018] 2) Six cameras work together to cover all areas of the target area, avoiding omissions caused by handheld scanning. They can comprehensively and easily calculate multiple parameters of the target area, reducing the tediousness and errors of manual operation. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the architecture of a three-dimensional rapid scanning measurement system; Figure 2 This is a flowchart of a rapid three-dimensional scanning measurement method. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] like Figure 1 The diagram shown is an architectural schematic of a three-dimensional rapid scanning measurement system. Figure 1 As shown, this system mainly includes a base, a transparent scanning platform, a depth camera array, an LED lighting unit, a data processing module, and a data storage module, which together form the scanning host to overcome problems such as shaking of the target body during scanning. In particular, through multi-depth camera synchronous acquisition technology and anti-shake reconstruction algorithm, it achieves non-invasive rapid detection and improves result accuracy, effectively reducing the number of repeated scans and improving detection efficiency. This system can be widely used in 3D scanning and measurement scenarios for various target objects, including but not limited to human bodies (such as infants, adults, etc.), animals (such as pets, laboratory animals, etc.), robots, and industrial products. The target body can be the head, torso, limbs, or other specific areas of the target object that require 3D morphological acquisition and measurement.

[0022] In this embodiment, a transparent scanning platform is provided in the central area of ​​the top of the base. The transparent scanning platform is made of high-transmittance tempered glass or optical glass with a transmittance of ≥92%, and is used to support the target part of the object being measured. In practical applications, depending on the type of target object, the target part is placed above the transparent scanning platform in an appropriate posture. For example, when the target object is an infant or adult, the target part is the head, and the object being measured lies flat above the transparent scanning platform with the head placed in the scanning area; when the target object is an animal, the target part can also be the head or torso, and the target part is placed in the scanning area while the animal is sedated or anesthetized; when the target object is a robot or industrial product, the target part can be the robot's head, joint components, or other structural parts to be tested, and it can be stably placed in the scanning area.

[0023] The depth camera array is configured with six depth cameras, each pointing towards a target area on the transparent scanning platform. The center of each depth camera is aligned with the center of the target area on the platform. Specifically, four depth cameras are positioned above the platform as the upper camera group: the first depth camera captures the first region of the target area, the second depth camera captures the second region, the third depth camera captures the third region, and the fourth depth camera captures the fourth region. Additionally, two depth cameras are positioned below the platform as the lower camera group: the fifth depth camera captures the left rear region of the target area, and the sixth depth camera captures the right rear region. Each depth camera in both the upper and lower camera groups is communicatively connected to a data processing module, enabling it to simultaneously acquire corresponding depth and RGB images of the target area and transmit them to the data processing module.

[0024] It should be noted that the first, second, third, and fourth regions mentioned above are divided according to the actual shape of the target part to ensure comprehensive coverage. Taking the target part as the head as an example, the first region can be the left side (including the left ear), the second region can be the right side (including the right ear), the third region can be the top region, and the fourth region can be the front region (including the face). Taking the target part as a robot's mechanical joint as an example, the first to fourth regions can correspond to the four side regions of the joint, respectively. The two depth cameras below collect data from the bottom left rear region and bottom right rear region of the target part from below through the transparent scanning platform, thereby achieving 360° all-round coverage of the target part. In actual deployment, the installation angle and position of each depth camera can be adjusted according to the typical size range of the target part to ensure that the field of view completely covers the target part.

[0025] LED lighting units surround the transparent scanning platform and are used to arrange ring-shaped LED light sources in the scanning area, thereby providing uniform illumination, eliminating shadows, facilitating the acquisition of high-quality images during shooting, and improving the quality of subsequent 3D reconstruction.

[0026] The data processing module is connected to the depth camera array, LED lighting unit and data storage module respectively. It is used to perform multi-camera synchronous triggering, point cloud fusion, 3D reconstruction, anti-shake compensation and parameter calculation, and output the processing results to the data storage module for storage.

[0027] In this embodiment, after simultaneous exposure by six depth cameras, corresponding point cloud data are acquired. The point cloud data includes a depth map (recording the distance of each pixel to the camera) and an RGB map (recording color information). The depth map and the RGB map are aligned, that is, the pixels in the depth map and the pixels in the RGB map correspond precisely in space, and color point cloud data is generated.

[0028] When generating color point cloud data, the system first reads the depth image of the depth map and the color image of the RGB image. Then, it converts the depth values ​​to meters based on the pixel value conversion factor of the depth image (e.g., 1000) and sets a depth truncation threshold (e.g., 3 meters) to filter distant noise. Next, it sets the camera's intrinsic parameter matrix, including image width, height, focal length (fx, fy), and optical center coordinates (cx, cy). The system uses the parameters of the intrinsic parameter matrix to create an RGBD image object and generates the corresponding color point cloud data from the RGBD image through the interface of the 3D vision library. The specific process is as follows: it iterates through each pixel (u, v) of the depth map, where u and v represent the coordinate values ​​(horizontal and vertical coordinates) of any pixel in the depth map, obtains its depth value d, and calculates the 3D coordinates of each pixel in the camera coordinate system using the formulas X=(u-cx)×d / fx, Y=(v-cy)×d / fy, and Z=d. At the same time, it assigns the color of the corresponding pixel in the RGB image to the 3D point to complete the color fusion. Finally, the generated color point cloud data is saved as a PLY format file and transferred to the data processing module for further processing.

[0029] In this embodiment, the data processing module is configured to perform the following operations: control multi-camera synchronous triggering, enabling each depth camera in the depth camera array to complete synchronous exposure within the same time window; perform point cloud fusion, converting the point cloud data collected by each depth camera to a unified coordinate system for fusion; perform 3D reconstruction, generating a complete model based on the fused point cloud data; perform anti-shake compensation, compensating for point cloud offsets caused by target area shaking; perform parameter calculation, automatically calculating target parameters based on the complete 3D reconstructed model; and store each piece of data generated in point cloud fusion, 3D reconstruction, and anti-shake compensation.

[0030] Anti-sway compensation is implemented in four main stages.

[0031] Phase 1: Synchronous Triggered Acquisition. All depth cameras complete exposure within the same time window via hardware triggering, with a time error ≤5ms, to capture the same instantaneous image. Before acquisition begins, the intrinsic parameters (focal length, distortion, etc.) and extrinsic parameters (relative position and rotation) of each camera are precisely calculated using a rigid, fixed calibration object (such as a checkerboard pattern). This synchronization mechanism ensures that regardless of any slight movement at the target location, all cameras acquire the surface state at the same moment, providing a data foundation for subsequent precise registration and compensation.

[0032] The second stage involves key feature point extraction. Key feature points of the target area are extracted from the data from each camera to form a feature network. Specifically, the selection of key feature points depends on the geometry of the target area: when the target area has obvious anatomical or structural landmarks (such as the ear protrusions, the highest point on the top, and the front and rear protrusions of the head), these landmarks are directly extracted as key feature points; when the target area is a relatively regular industrial part or robot component, its corner points, curvature extrema, and endpoints of the axis of symmetry are extracted as key feature points. The feature network consists of these key feature points and their spatial topological relationships, providing a stable reference benchmark for subsequent motion estimation.

[0033] The third stage is motion estimation and compensation. First, the Fast Point Feature Histogram (FPFH) is used to extract feature points from the source and target point clouds. Based on the Euclidean distance of the histogram, a coarse corresponding point pair is found. The initial rotation matrix R0 and the initial translation vector t0 are solved using the RANSAC random sampling consensus algorithm. Then, the ICP iterative nearest point algorithm is used for fine registration. The centroids of the two sets of point clouds are calculated in a decentralized manner, the covariance matrix H is constructed, and the optimal rotation matrix R and translation vector t are solved using singular value decomposition (SVD). The flip angle around the X-axis, the pitch angle around the Y-axis, and the yaw angle around the Z-axis are extracted from the optimal rotation matrix R to obtain six degrees of freedom motion parameters containing three translational degrees of freedom and three rotational degrees of freedom, including the X-axis translation ΔX, the Y-axis translation ΔY, the Z-axis translation ΔZ, the flip angle around the X-axis Rx, the pitch angle around the Y-axis Ry, and the yaw angle around the Z-axis Rz.

[0034] In the third stage, the initial position is not relied upon to prevent ICP from getting trapped in local optima. A Fast Point Feature Histogram (FPFH) is used to quantify the spatial angular differences (yaw and pitch angles between normals) between the query point and its neighbors, forming a 33-dimensional histogram as the geometric fingerprint of that point. Feature points are extracted from the source and target point clouds, and coarse corresponding point pairs are found based on the Euclidean distance of the histograms. Combined with RANSAC (Random Sample Consensus) random sampling, the transformation is repeatedly solved, and the model with the most interior points is selected to obtain the initial value (R0, t0).

[0035] After obtaining the initial values, ICP iteration continues to reduce errors and output the final motion parameters. The iteration includes the following process: First, establish nearest neighbor pairs. After transforming the source point cloud P according to (R0,t0), traverse each point p_i and use KD-Tree to search for the point q_i with the closest Euclidean distance in the target point cloud Q, where i is a positive integer. At the same time, set a maximum distance threshold to remove incorrect pairs that are too far apart. Second, solve for the optimal transformation, that is, minimize the objective function in the current iteration: E(R,t)=Σ||q_i-(R×p_i+t)||²; then find the closed-form solution through singular value decomposition (SVD), including: decentralization, calculating the centroids μ_p and μ_q of the two sets of point clouds, and moving the point clouds to the origin p_i'=p_i-μ_p, q_i'=q_i-μ_q; then construct the covariance matrix H=Σ(p_i'×q_i'^T), which is a 3 A 3×3 matrix is ​​then formed. Next, SVD decomposition is performed on H, resulting in H = U × Σ × V^T, where U is the left singular matrix, Σ is the singular value diagonal matrix, V is the right singular matrix, and T is the transpose. Then, the rotation matrix R = V × U^T is calculated. A critical check is performed on the rotation matrix: if det(R) < 0 (reflection), the third column of V needs to be inverted, i.e., R = V × diag(1,1,-1) × U^T. Finally, the translation vector is calculated: t = μ_q - R × μ_p. The third step is iteration and convergence. The obtained (R,t) is applied to the source point cloud to update the position. These three steps are repeated until either of the following conditions is met: the error change is less than a set threshold (e.g., 1e-6) or the maximum number of iterations (e.g., 30 iterations) is reached.

[0036] After the iteration terminates, the final t(tx,ty,tz) is directly the translational degree of freedom. The rotational degree of freedom needs to be extracted from the final matrix R: the roll angle around the X-axis (Roll), α=atan2(R[2,1],R[2,2]); the pitch angle around the Y-axis (Pitch), β=atan2(-R[2,0],sqrt(R[2,1]²+R[2,2]²)); and the yaw angle around the Z-axis (Yaw), γ=atan2(R[1,0],R[0,0]). This gives the complete six-degree-of-freedom (tx,ty,tz,α,β,γ) motion compensation parameters for the target part. Then, based on the estimated six-degree-of-freedom motion compensation parameters, the corresponding point clouds are transformed to the same reference coordinate system (such as the first frame), thereby eliminating the point cloud misalignment caused by the translation and rotation of the target part within the scanning window, and completing the motion compensation.

[0037] The fourth stage, fusion and reconstruction, involves fusing the point clouds mapped to a unified reference coordinate system to generate a complete model in the unified coordinate system.

[0038] In 3D fusion reconstruction, a template-based non-rigid deformation method can be used. Since similar target parts of similar objects have high structural similarity (e.g., the head structures of animals of the same species are highly similar, and the geometric shapes of robot components of the same model are highly consistent), a standard high-precision template corresponding to the target part is prepared in advance. Non-rigid deformation is then used to precisely fit the template to the scanned color 3D point cloud. This method can repair holes (such as occluded areas, reflective areas, etc.) and output a topologically consistent model. Specifically, initial alignment is first performed, coarsely registering the template model with the point cloud to reduce positional and pose differences. Then, the core algorithm uses non-rigid ICP, which adds regularization constraints to the traditional ICP. Its energy function is E = E_data + λ × E_smooth, where E_data is the fitting term used to ensure that the template vertices converge towards the target point cloud, E_smooth is the smoothing constraint term used to constrain the deformation smoothness of neighboring vertices (preventing distortion), and λ is the balancing weight. For the balancing weight λ, it can be set from a relatively large value (e.g., 1.5, empirical weight), observe the registration results, and aim to minimize E each time. If the resulting deformation does not meet the user's preset requirements (e.g., insufficient fit, i.e., the average distance between the template vertex and the target point cloud is greater than the preset threshold), then decrease λ. If the deformation is too large (excessive distortion, local overstretching, etc.), then increase λ. Each decrease or increase of λ is adjusted according to the preset step size (e.g., 0.1 to 0.2) until the preset maximum number of iterations is reached or until the requirements are met (e.g., the fit meets the average distance requirement, and each change of E is within t%, t≤5).

[0039] It should be noted that when making multiple adjustments to λ, a multi-level refinement strategy from coarse to fine can be adopted: first, perform overall contour matching with a larger smoothing weight (e.g., 1.5), then gradually decrease λ (or reduce the smoothing constraint term) by step size to release local deformation capability, thereby sequentially fitting the detailed areas of the target part. After each level of refinement is completed, the deformation result with the lowest current energy function E value (i.e., the best overall fit and smoothness) is selected as the initial template for the next level of refinement, realizing phased and progressive fine reconstruction; the degree of multi-level refinement is determined by the preset model integrity. For example, if the preset requirement is that the final model integrity reaches more than 90%, then refinement can stop when this requirement is met.

[0040] For example, after completing the 3D reconstruction, the data processing module can automatically calculate the target parameters of the target area based on the 3D point cloud data of the complete model. The specific types and calculation methods of the target parameters are determined according to the type of the target object and the shape of the target area: 1) Perimeter parameter: The accurate perimeter is obtained by fitting the point cloud profile curve on the specified measurement plane. For example, when the target part is the head, the head circumference (HC) can be calculated, that is, the perimeter of the head profile is calculated on the horizontal plane with the largest anteroposterior diameter.

[0041] 2) Length parameter: Calculate the maximum straight-line distance along a specified direction on a specified measurement plane. For example, when the target part is the head, the head length (HL), i.e., the maximum straight-line distance in the front-to-back direction, can be calculated; when the target part is an industrial component, the maximum axial length of the component can be calculated.

[0042] 3) Width parameter: Calculates the maximum straight-line distance in the direction perpendicular to the length direction on the specified measurement plane. For example, when the target part is the head, the head width (HW) can be calculated, which is the maximum straight-line distance in the left and right directions.

[0043] 4) Inclined diameter parameter: On the specified measurement plane, calculate the maximum distance from the center of the target part to the edge in the direction of the preset angle (such as 30°) with respect to the central symmetry plane. The left inclined diameter (LD) and the right inclined diameter (RD) can be calculated separately.

[0044] 5) Ratio index: The ratio is calculated based on the length and width parameters, such as CR index = (width / length) × 100%, which is used to describe the morphological characteristics of the target part.

[0045] 6) Asymmetry Index: The degree of asymmetry is calculated based on the left and right sloping diameters. For example, CVAI index = (|left sloping diameter - right sloping diameter| / longer sloping diameter) × 100%. The larger the value, the more asymmetric the asymmetry.

[0046] 7) Volume parameters: Perform three-dimensional integral calculations on the target parts above the specified plane. The model can be divided into multiple quadrants (such as left front, right front, left rear, right rear) and the volume can be calculated separately.

[0047] 8) Symmetry index: Evaluate the volume ratio or shape difference, such as the front symmetry ratio = left front volume / right front volume. The closer the ratio is to 1, the better the symmetry.

[0048] It should be noted that the target parameters mentioned above are merely illustrative examples. In practical applications, the required measurement parameters can be flexibly defined and calculated according to the different target objects and target parts. For example, for robot joint components, target parameters may include industrial measurement indicators such as diameter, roundness, cylindricity, and surface roughness; for animal bodies, target parameters may include body shape evaluation indicators such as body length, body height, and body circumference.

[0049] In this embodiment, the data storage module establishes a long-term digital archive database, creating an independent digital archive for each tested object (one file per object), storing the 3D model, parameter calculation results, and related test data from each test, and storing them in chronological order. The system supports quick retrieval of historical data by name, number, or date, facilitating long-term tracking of the same target object and comparative analysis of multiple test data.

[0050] Combination Figure 2 As shown in the flowchart, this embodiment also provides a three-dimensional rapid scanning measurement method. Using the above-described system for rapid scanning measurement, the method specifically includes the following steps: S1. Establish a test object file: Enter the individual information of the test object (such as name, number, type, test date, etc.) and store the file data in the data storage module. The type of test object can be human body, animal, robot, or industrial product, etc., and the individual information should be entered accordingly based on the object type.

[0051] S2. Place the object under test and start: Place the target part of the object under test in the scanning area of ​​the transparent scanning platform to ensure that the target part is within the acquisition field of view of each depth camera; the data processing module controls the six depth cameras in the depth camera array to expose synchronously.

[0052] S3. Data Acquisition Completed: Within 0.5-1 second, the depth camera array synchronously acquires the depth map and RGB image of the target area and generates color point cloud data. The specific calculation process is as follows: First, data preparation is performed by acquiring a depth map recording the distance of each pixel to the camera and an RGB image recording color information. The two images must be aligned, meaning the pixels in the depth map and the pixels in the RGB image correspond precisely in space. Next, camera calibration is performed by acquiring the camera's intrinsic parameter matrix, including focal lengths fx and fy and optical center coordinates cx and cy. This is crucial for mapping 2D pixels to 3D space. Then, coordinate transformation, i.e., back projection calculation, is performed. Each pixel (u, v) in the depth map is traversed to obtain its depth value d. The 3D coordinates (X, Y, Z) of this point in the camera coordinate system are calculated using the following formulas: X = (u - cx) × d / fx; Y = (v - cy) × d / fy; Z = d. Finally, color fusion is performed, assigning the color of the corresponding pixel (u,v) in the RGB image to the calculated 3D point (X,Y,Z) to generate color point cloud data.

[0053] S4. Execute the anti-sway compensation algorithm: The data processing module executes the following four stages sequentially: The first phase, synchronous trigger acquisition, ensures that the timestamps of the six cameras are strictly synchronized, and that all depth cameras complete their exposures within the same time window, with a time error ≤5ms. Before starting, the intrinsic and extrinsic parameters of each camera are precisely calculated using rigidly fixed calibration objects.

[0054] The second stage is key feature point extraction: key feature points of the target area are extracted from the data collected by each camera to form a feature network. The type of key feature points depends on the geometry of the target area, including but not limited to protrusions, curvature extrema, and endpoints of the axis of symmetry.

[0055] The third stage, motion estimation and compensation, involves calculating the six-degree-of-freedom motion parameters (ΔX, ΔY, ΔZ, Rx, Ry, Rz) using the ICP registration algorithm and feature matching algorithm. This stage is divided into two steps: coarse registration and fine registration.

[0056] Coarse registration based on feature matching is used to calculate the initial rotation matrix R0 and the initial translation vector t0, preventing ICP from getting trapped in local optima. Fast Point Feature Histogram (FPFH) is employed to quantify the spatial angular differences (yaw and pitch angles between normals) between the query point and its neighbors, forming a 33-dimensional histogram as the point's "geometric fingerprint." Feature points are extracted from the source and target point clouds, and coarse corresponding point pairs are found based on the Euclidean distance of the histograms. RANSAC (Random Sample Consensus) algorithm is used for random sampling, and the transformation is repeatedly solved to select the model with the most interior points as the initial value (R0, t0).

[0057] Fine-grained registration employs the ICP algorithm to determine the accurate rotation matrix R and translation vector t. The first step involves establishing nearest-neighbor pairs: after transforming the source point cloud P according to R0 and t0, each point p_i is traversed, and the nearest point q_i in the target point cloud Q is searched using a KD-Tree algorithm based on its Euclidean distance. A maximum distance threshold is set to eliminate incorrect pairings. The second step involves solving for the optimal transformation (core SVD calculation): a closed-form solution is obtained through Singular Value Decomposition (SVD) to minimize the objective function E(R,t)=Σ||q_i-(R×p_i+t)||². The specific calculations are as follows: (1) Decentralization: Calculate the centroids μ_p and μ_q of the two sets of point clouds, and move the point clouds to the origin: p_i'=p_i-μ_p, q_i'=q_i-μ_q; (2) Construct the covariance matrix H=Σ(p_i'×q_i'^T), which is a 3x3 matrix; (3) SVD decomposition: Decompose H to obtain H=U×Σ×V^T; (4) Calculate the rotation matrix R=V×U^T. Key verification: If det(R)<0 (reflection), the third column of V needs to be inverted, i.e., R=V×diag(1,1,-1)×U^T; (5) Calculate the translation vector t=μ_q-R×μ_p. The third step is iteration and convergence: Apply the obtained (R,t) to the source point cloud update position, and repeat the above steps until the error change is less than the set threshold (e.g., 1e-6) or the maximum number of iterations (e.g., 30 times). After the iteration terminates, the final t(tx,ty,tz) represents the translational degrees of freedom. The rotational degrees of freedom are extracted from the final matrix R: the flip angle α around the X-axis = atan2(R[2,1],R[2,2]); the pitch angle β around the Y-axis = atan2(-R[2,0],sqrt(R[2,1]²+R[2,2]²)); and the yaw angle γ around the Z-axis = atan2(R[1,0],R[0,0]). This yields the complete six-degree-of-freedom motion compensation parameters for the target part. Based on these six-degree-of-freedom motion parameters, all point clouds are mapped to a unified reference coordinate system through an inverse transformation, eliminating point cloud offsets caused by shaking.

[0058] The fourth stage, fusion and reconstruction: transform all the motion-compensated point clouds into a unified coordinate system to generate a complete model.

[0059] S5. Execute the 3D Reconstruction Algorithm: Since similar target parts of similar target objects have high similarity in overall structure, this step prepares a standard high-precision template corresponding to the target part in advance. Non-rigid deformation is used to precisely fit it to the scanned point cloud to repair holes and output a topologically consistent model. First, initial alignment is performed, coarsely registering the template model with the point cloud to reduce positional and pose differences. Then, the core algorithm, non-rigid ICP, is used, adding regularization constraints to the traditional ICP. Its energy function is E=E_data+λ×E_smooth, where E_data ensures that the template vertices move closer to the target point cloud (fit), E_smooth constrains the deformation of adjacent vertices to be as smooth as possible (preventing distortion), and λ is used to balance the two. Finally, multi-level refinement is performed, deforming from coarse to fine stages. First, the overall contour is matched, and then the local deformation capability is gradually released to fit the detailed areas, ultimately achieving a model integrity of over 90%.

[0060] S6. Calculate Target Parameters: Based on the 3D point cloud data of the complete model, the data processing module automatically calculates the corresponding target parameters according to the type of target object and target part. Target parameters may include, but are not limited to, perimeter, length, width, slant diameter, ratio index, asymmetry index, volume, and symmetry index, etc. The specific parameter types and calculation formulas are determined according to the actual application scenario.

[0061] S7. Data Storage: The data storage module stores and archives any data generated by the data processing module (including raw point cloud data, 3D reconstruction models, target parameter calculation results, etc.). The data storage module establishes a long-term digital archive database, creating an independent digital archive for each measured object, supporting quick retrieval and historical queries by name, number, or date.

[0062] In summary, this invention, through simultaneous hardware triggering of six depth cameras, rapidly completes omnidirectional data acquisition of the target area. Then, based on ICP registration and feature matching, it calculates six-degree-of-freedom motion parameters for motion compensation. This effectively solves the problems of point cloud misalignment, model gaps, and repeated scanning caused by the inability of dynamic target objects (such as living organisms, operating machinery, and robots) to remain stationary for extended periods. The six cameras collaboratively cover all areas of the target area, avoiding omissions from handheld scanning. It allows for comprehensive and convenient calculation of multiple parameters of the target area, reducing the tediousness and errors of manual operation. Furthermore, the establishment of a database for data archiving and management provides data support for the long-term digital tracking and management of various target objects, demonstrating broad application prospects and significant advancements.

[0063] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A three-dimensional rapid scanning measurement system, characterized in that, The system includes a base, a transparent scanning platform, an LED lighting unit, a depth camera array, a data processing module, and a data storage module. The transparent scanning platform, mounted on the base, supports the object being measured. The depth camera array consists of an upper camera group and a lower camera group. The upper camera group is positioned above the transparent scanning platform, and the lower camera group is positioned below it. Both the upper and lower camera groups are communicatively connected to the data processing module to simultaneously acquire depth maps and RGB images of the target area of ​​the object being measured and transmit them to the data processing module. The LED lighting unit is arranged on the base around the scanning area of ​​the transparent scanning platform to provide uniform illumination to the scanning area. The data processing module is connected to the data storage module and is used to process the collected data and output the results to the data storage module; The data processing module is configured to perform anti-shake compensation; the anti-shake compensation includes at least motion estimation and compensation and fusion reconstruction; The motion estimation and compensation process includes: first, using Fast Point Feature Histogram (FPFH) to extract feature points from the source and target point clouds; finding coarse corresponding point pairs based on the Euclidean distance of the histograms; and then solving for the initial rotation matrix R0 and initial translation vector t0 using the RANSAC random sampling consensus algorithm. Next, fine registration is performed using the ICP iterative nearest point algorithm. This involves calculating the centroids of the two sets of point clouds in a decentralized manner, constructing the covariance matrix H, and solving for the optimal rotation matrix R and translation vector t0 using SVD decomposition. Reflection verification is then performed; if det(R) < If 0, the third column of V is inverted, where V is the right singular matrix corresponding to the covariance matrix H; iterate until the error change is less than 1e-6 or the maximum number of iterations of 30 is reached; the error change is the error change of the ICP iteration; extract the flip angle around the X-axis, the pitch angle around the Y-axis, and the yaw angle around the Z-axis from the final R to obtain the motion compensation parameters of the target part with six degrees of freedom; transform each point cloud to the same reference coordinate system according to the motion compensation parameters of the target part with six degrees of freedom to eliminate the point cloud misalignment caused by the translation and rotation of the target part within the scanning window; The fusion reconstruction includes: using a standard template, initial alignment and coarse registration are performed first, followed by a non-rigid ICP algorithm with an energy function E = E_data + λ × E_smooth; E_data is used to ensure the template vertices closely fit the target point cloud, E_smooth is used to constrain the deformation of neighboring vertices to be as smooth as possible, and λ is used to balance the two; λ starts from 1.5, with the goal of minimizing E. If the fit is insufficient, λ is reduced; if the deformation is too large, λ is increased, with a step size of 0.1 to 0.2; a multi-level refinement from coarse to fine is adopted. First, a larger λ is used for overall contour matching, and then λ is gradually reduced to release local deformation. The deformation result with the lowest E value is selected as the initial template for the next level, until the model integrity reaches more than 90%.

2. The three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The transparent scanning platform is made of tempered glass or optical glass. The transparent scanning platform is used to support the object being measured. The upper camera group collects data from above the transparent scanning platform, while the lower camera group collects data from below through the transparent scanning platform.

3. The three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The upper camera group includes four depth cameras: a first depth camera for capturing the first region of the target area; a second depth camera for capturing the second region of the target area; a third depth camera for capturing the third region of the target area; and a fourth depth camera for capturing the fourth region of the target area. The first, second, third, and fourth depth cameras are all positioned above the transparent scanning platform.

4. The three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The lower camera group includes two depth cameras: a fifth depth camera, used to acquire data of the left rear region of the target area; and a sixth depth camera, used to acquire data of the right rear region of the target area. Both the fifth and sixth depth cameras are located below the transparent scanning platform and acquire data upwards through the transparent scanning platform.

5. A three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The LED lighting unit is a ring-shaped LED light source, which is arranged around the scanning area of ​​the transparent scanning platform to provide uniform illumination and eliminate shadows.

6. The three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The data processing module is configured to perform the following operations: control the synchronous triggering of multiple cameras, so that each depth camera in the depth camera array completes synchronous exposure within the same time window; and perform point cloud fusion, converting the point cloud data collected by each depth camera to a unified coordinate system for fusion. Perform 3D reconstruction and generate a complete model based on the fused point cloud data; Anti-shake compensation compensates for point cloud offset caused by target area shaking; parameter calculation automatically calculates target parameters based on the complete 3D reconstruction model; and stores every piece of data generated in point cloud fusion, 3D reconstruction, and anti-shake compensation.

7. A three-dimensional rapid scanning measurement system according to claim 6, characterized in that, The data processing module performs anti-shake compensation in the following stages: In the first stage, synchronous acquisition is triggered, and each depth camera in the depth camera array is controlled to complete the exposure within the same time window. The second stage is key feature point extraction, which extracts multiple target points from the target area from the data collected by each depth camera to form a feature network. In the third stage, motion estimation and compensation, feature points are first extracted from the source and target point clouds using Fast Feature Histogram (FPFH). Coarse corresponding point pairs are found based on the Euclidean distance of the histograms. The initial rotation matrix R0 and initial translation vector t0 are then solved using the RANSAC random sampling consensus algorithm. Next, fine registration is performed using the ICP iterative nearest point algorithm. The centroids of the two sets of point clouds are calculated in a decentralized manner, the covariance matrix H is constructed, and the optimal rotation matrix R and translation vector t are solved using singular value decomposition (SVD). The X-axis flip angle, Y-axis pitch angle, and Z-axis yaw angle are extracted from the optimal rotation matrix R to obtain six degrees of freedom motion parameters, including three translational degrees of freedom and three rotational degrees of freedom. Each point cloud is then transformed to a unified coordinate system based on these six degrees of freedom motion parameters. The fourth stage, fusion and reconstruction, involves fusing the point clouds mapped to a unified reference coordinate system to generate a complete model in the unified coordinate system.

8. A three-dimensional rapid scanning measurement system according to claim 7, characterized in that, The data processing module operates as follows during the fourth phase of fusion and reconstruction: Prepare standard templates corresponding to the target parts in advance; perform initial coarse registration and alignment between the standard templates and each point cloud to reduce position and pose differences; after completing the initial coarse registration and alignment, use the non-rigid ICP algorithm to calculate the energy function E to measure the degree of alignment; E = E_data + λ × E_smooth, where E_data is the fit term and is used to ensure that the template vertices move closer to the target point cloud, E_smooth is the smoothness constraint term and is used to constrain the deformation smoothness of neighboring vertices, and λ is the balance weight; After initial coarse registration, multiple alignments and registrations are performed for refinement to generate a topologically consistent complete model.

9. A three-dimensional rapid scanning measurement system according to claim 1, characterized in that, The data storage module has established a long-term digital archive database. Based on the retrieval parameters, it performs archive management, automatic archiving, and historical querying for each scanned object. The retrieval parameters include name, number, or date.

10. A three-dimensional rapid scanning measurement method, wherein the three-dimensional rapid scanning measurement system as described in any one of claims 1-9 is used, characterized in that, Includes the following steps: S1. Establish a test subject profile: Enter the individual information of the test subject and store the profile data in the data storage module; S2. Place the object under test and start scanning: Place the target part of the object under test flat in the scanning area of ​​the transparent scanning platform; then use the data processing module to control the six depth cameras in the depth camera array to expose synchronously. S3. Data Acquisition: Simultaneously acquire point cloud data of the target area using a depth camera array. The point cloud data includes a depth map and an RGB image. Align the depth map and the RGB image, calculate the 3D coordinates of each pixel in the depth map by back-projecting the coordinates using the camera intrinsic parameter matrix, and assign the color of the corresponding pixel in the RGB image to the 3D point to generate 3D point cloud data. S4. Execute the anti-shake compensation algorithm: The data processing module sequentially executes four stages: synchronous trigger acquisition, key feature point extraction, motion estimation and compensation, and fusion reconstruction. In the motion estimation and compensation and fusion reconstruction stages, FPFH feature matching combined with RANSAC algorithm is first used to solve the initial transformation. Then, ICP algorithm is used to solve the optimal rotation matrix and translation vector through SVD decomposition and iteration to obtain six degrees of freedom motion parameters. After compensating the point cloud offset according to the six degrees of freedom motion parameters, the model is fused and reconstructed in a unified coordinate system to generate a complete model. S5. Execute the 3D reconstruction algorithm: The data processing module uses a standard template and performs multiple alignment and registration operations through initial coarse registration and non-rigid ICP algorithm to generate a topologically consistent complete 3D model. S6. Calculation parameters: The data processing module automatically calculates the target parameters based on the 3D point cloud data of the complete 3D model; S7. Data storage: The data storage module stores and archives any data generated by the data processing module.

Images