A real-time reverse scanning guidance method

Abstract

The present invention discloses a real-time reverse scanning guidance method, which belongs to the technical field of 3D point cloud processing. The method comprises collecting point clouds, establishing a 3D master model and a W-model model, aligning the two models, calculating the alignment error, sending the rendered data to VR, and formulating a re-scanning plan. The method solves the technical problems of time sequence synchronization, precise spatial registration and error compensation, and efficient block cache rendering of multi-period dynamic point clouds during scanning modeling. The method realizes precise spatial alignment and time-weighted error compensation of multi-period point clouds, effectively solves the problems of posture drift and time asynchrony, adopts spatial block cache and multi-layer fusion rendering technology, realizes efficient management of large-scale dynamic point clouds and VR real-time visualization, automatically identifies missed scanning areas and plans paths based on point cloud coverage, and greatly improves re-scanning efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional point cloud processing, and in particular relates to a real-time reverse scanning guidance method. Background Art

[0002] With the rapid development of 3D laser scanning technology and virtual reality (VR) technology, point cloud-based 3D modeling and dynamic visualization have become important research directions in the field of reverse development of large sheet metal parts such as car shells.

[0003] Current traditional technologies rely on handheld laser or structured light scanners to collect point clouds, and then use offline point cloud stitching, noise filtering and CAD modeling. It is difficult for operators to visually check the scanning progress and quality on site, and it is prone to missed scans, repeated scans and posture errors, resulting in inefficient subsequent surface fitting and assembly verification. Summary of the Invention

[0004] The purpose of the present invention is to provide a real-time reverse scanning guidance method, which solves the technical problems of time synchronization of multi-period dynamic point clouds during scanning modeling, precise spatial alignment and error compensation, and efficient block cache rendering.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A real-time reverse scanning guidance method comprises the following steps:

[0007] Step 1: Use a handheld 3D scanner to scan the workpiece surface and collect the spatial coordinate information and posture information of the point cloud in real time;

[0008] Each point cloud data records its corresponding acquisition time, and a time tag is added to each point according to the acquisition time to construct the original point cloud dataset;

[0009] Step 2: After obtaining the original point cloud dataset and preprocessing it, the point cloud in the original point cloud dataset is processed using a voxel grid filtering algorithm + a triangular mesh reconstruction algorithm to generate a 3D master model with a continuous surface and output it;

[0010] Step 3: By fusing the real-time 6-DOF pose of the handheld 3D scanner with the original point cloud data, the point cloud frames are unified into the world coordinate system using a homogeneous transformation matrix. A set of standard time tags is constructed using a preset time interval, and the timestamp of each point is aligned to the nearest neighbor standard tag to generate an enhanced point cloud dataset. The enhanced point cloud dataset is processed into a time-block structure, and the subsets in the block structure are divided into subsets in chronological order to generate a W-model with time tags.

[0011] Step 4: Use the geometric center alignment method and ICP algorithm to spatially align the 3D master model and the W-model. Calculate the alignment error by building an error compensation function, convert the alignment error into a visual color, obtain a color code, and output the aligned 3D master model and W-model along with the color code.

[0012] Step 5: Divide the aligned W-model into blocks and cache them;

[0013] Render the W-model and the 3D main model, and push the rendered layer results to the VR device;

[0014] Step 6: Compare the W-model with the preset CAD model of the workpiece. Identify the missed scanning areas based on the coverage of the point cloud on the surface mesh of the CAD model. Generate path planning based on the location of the missed scanning areas using the cost function. Render the lines of the path planning into the 3D main model according to the rendering method in step 5 and send it to the VR device.

[0015] Preferably, when executing step 1, the following steps are specifically included:

[0016] Step 1-1: Use a laser scanner to scan the point cloud data of the workpiece surface, record the spatial coordinates and acquisition time of each point, and generate a point cloud set for each frame based on the spatial coordinates of each point , add a time tag to each point cloud according to the acquisition time ;

[0017] Step 1-2: Based on point cloud collection Add a time stamp to each sampling point , generate the original point cloud dataset :

[0018] ;

[0019] Where M is the total number of frames, represents the three-dimensional coordinates of the j-th point in the i-th frame, , ,N represents the number of point clouds in the i-th frame.

[0020] Preferably, when executing step 2, the following steps are specifically included:

[0021] Step 2-1: Original point cloud dataset Perform preprocessing to remove invalid points, isolated points and abnormal noise;

[0022] Step 2-2: Sparse the point cloud using a voxel grid filtering algorithm;

[0023] Step 2-3: Reconstruct the sparse point cloud using a triangular mesh reconstruction algorithm Perform structural modeling to generate a 3D master model of continuous surfaces ;

[0024] Step 2-4: Export the 3D master model .

[0025] Preferably, when executing step 3, the following steps are specifically included:

[0026] Step 3-1: Obtain the original point cloud dataset , and obtain posture information at the same time;

[0027] Step 3-2: Obtain the six-degree-of-freedom pose of each frame captured by the handheld 3D scanner through the spatial tracking system, generate a homogeneous transformation matrix, and transform each point in the homogeneous transformation matrix into a unified world coordinate system;

[0028] Step 3-3: Construct a standard time tag set, in which a standard time tag is defined at a fixed time interval;

[0029] Step 3-4: Calculate the standard time label close to the time label of each point to generate a new enhanced point cloud dataset;

[0030] Step 3-5: Construct the enhanced point cloud dataset into a time block structure consisting of multiple subsets, sort the subsets in chronological order, and obtain a structured point cloud model with time labels, namely the W-model model.

[0031] Preferably, when executing step 4, the following steps are specifically included:

[0032] Step 4-1: Obtain the 3D master model, W-model, homogeneous transformation matrix and standard time tag set;

[0033] Step 4-2: Calculate the geometric centers of all vertices of the 3D main model and the center of all points of the W-model ; By calculation and The difference between , translate all points of the W-model to the 3D main model for center alignment;

[0034] Step 4-3: Using the weighted ICP algorithm, based on the processing results of step 4-2, find the optimal matching result;

[0035] Step 4-4: Construct an error compensation function to calculate the alignment error of each point, convert the error value into a visual color through the color mapping function ColorMap, and generate a color code;

[0036] Step 4-5: Output the aligned W-model and the 3D main model, and output the alignment error of each point and its corresponding color coding.

[0037] Preferably, when executing step 5, the following steps are specifically included:

[0038] Step 5-1: Divide the aligned W-model into blocks and cache them. When the W-model is cached in blocks, its structure is as follows:

[0039] [Meta information] + [Time tag index] + [Spatial data block collection] + [Operation log];

[0040] The [spatial data block set] includes multiple [spatial blocks], and the structure of each [spatial block] is as follows:

[0041] [Block unique identifier] + [Point cloud data] + [Time stamp] + [Version tracking identifier] + [Alignment error];

[0042] Step 5-2: Use multi-layer management to render the W-model and the 3D main model separately. Through transparency overlay and depth sorting, the two models are superimposed and fused. The point cloud deviation is displayed in the layer according to color coding to generate the rendered layer result.

[0043] Step 5-3: Push the rendered layer results to the VR device.

[0044] When executing step 6, the specific steps include:

[0045] Step 6-1: Count the coverage of the point cloud on the surface grid of the CAD model, obtain the coverage of each grid unit, and count the number of point clouds actually falling into each grid unit. When the coverage of a grid is less than the preset threshold, the area where the grid is located is determined to be a missed scan area;

[0046] Step 6-2: Overlay a highlight color mask on the grid cells in the missed area;

[0047] Step 6-3: Plan the next scanning path based on the location of the missed scanning area and generate a path plan. When planning the next scanning path, first extract the geometric centers of all missed scanning grid cells to form a region set.

[0048] Combined with the posture of current handheld 3D scanner , taking the current posture position as the starting point, using the cost function to calculate the path point sequence, the cost function As shown below:

[0049] ;

[0050] ;

[0051] Among them, the cost function It represents the optimal path; It represents the sequence of path points, that is, starting from the current scanner position, it passes through several path points in sequence to finally cover the missed scanning area; n represents the total number of path points, Represents two consecutive points;

[0052] Step 6-4: Generate path lines based on the path planning, render the path lines onto the 3D main model according to the rendering method in step 5, and push them to the VR device.

[0053] The real-time reverse scanning guidance method described in the present invention solves the technical problems of time synchronization, precise spatial registration and error compensation, and efficient block cache rendering of multi-period dynamic point clouds during scanning modeling. The present invention introduces a dynamic point cloud model (W-model) with time series markers, and combines the scanner posture to achieve precise spatial alignment and time-weighted error compensation of multi-period point clouds, effectively solving the problems of posture drift and time asynchrony. It adopts spatial block caching and multi-layer fusion rendering technology to achieve efficient management of large-scale dynamic point clouds and VR real-time visualization, and automatically identifies missed scanning areas and plans paths based on point cloud coverage, greatly improving the efficiency of re-scanning. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is the main flow chart of the present invention;

[0055] Figure 2 is a flow chart of step 1 of the present invention;

[0056] Figure 3 is a flow chart of step 2 of the present invention;

[0057] Figure 4 is a flow chart of step 3 of the present invention;

[0058] Figure 5 is a flow chart of step 4 of the present invention;

[0059] Figure 6 is a flow chart of step 5 of the present invention;

[0060] Figure 7 It is a flow chart of step 6 of the present invention. DETAILED DESCRIPTION

[0061] A real-time reverse scanning guidance method shown in FIG1 to FIG7 includes the following steps:

[0062] Step 1: Use a handheld 3D scanner to scan the workpiece surface and collect the spatial coordinate information and posture information of the point cloud in real time;

[0063] Each point cloud data records its corresponding acquisition time, and a time tag is added to each point according to the acquisition time to construct the original point cloud dataset;

[0064] When executing step 1, the specific steps include:

[0065] Step 1-1: Use a laser scanner to scan the point cloud data of the workpiece surface, record the spatial coordinates and acquisition time of each point, and generate a point cloud set for each frame based on the spatial coordinates of each point , add a time tag to each point cloud according to the acquisition time ;

[0066] Handheld 3D scanner with fixed frequency Continuously scan and sample at a fixed frequency The number of laser points emitted per second by a handheld 3D scanner can be calculated ;

[0067] Point cloud collection collected in the i-th frame Expressed as:

[0068] ;

[0069] Attach a time tag to each sampling point (i.e. point cloud) :

[0070] ;

[0071] in, Indicates the starting sampling time of the i-th frame, represents the delay of the sampling time of the jth point relative to the starting sampling time, .

[0072] Step 1-2: Based on point cloud collection Add a time stamp to each sampling point , generate the original point cloud dataset :

[0073] ;

[0074] Where M is the total number of frames, represents the three-dimensional coordinates of the j-th point in the i-th frame, , ,N represents the number of point clouds in the i-th frame.

[0075] In this embodiment, the original point cloud dataset It contains the spatial coordinates and precise time information of the point cloud, which can provide a data source for the subsequent modeling of the 3D main model and W-model.

[0076] Step 2: After obtaining the original point cloud dataset and preprocessing it, the point cloud in the original point cloud dataset is processed using a voxel grid filtering algorithm + a triangular mesh reconstruction algorithm to generate a 3D master model with a continuous surface.

[0077] When executing step 2, the specific steps include:

[0078] Step 2-1: Original point cloud dataset Perform preprocessing to remove invalid points, isolated points and abnormal noise;

[0079] In this embodiment, the statistical outlier filtering method is used to filter the original point cloud data set. Preprocessing is performed, including counting the average distance μ and standard deviation σ of all points in its k neighborhood for each point. If the average distance of a point exceeds If the range is less than , it will be identified as an abnormal outlier and removed;

[0080] k represents the number of points within the neighborhood radius, ranging from 10 to 50. Indicates the outlier determination tolerance coefficient, with a value of 1 to 2.

[0081] The statistical outlier filtering method is an existing technology and will not be described in detail.

[0082] Step 2-2: Sparse the point cloud using a voxel grid filtering algorithm;

[0083] Specifically, the space is divided into a three-dimensional voxel grid with a side length of r, and the centroid of all points in each voxel is retained and used as the representative point. The calculation formula is as follows:

[0084] ;

[0085] in, Represents the coordinates of the representative point of voxel v; represents the number of points falling into voxel v, represents the jth point in the voxel.

[0086] The result after sparse processing is represented as a sparse point cloud :

[0087] ;

[0088] r represents the length of the voxel side.

[0089] Step 2-3: Reconstruct the sparse point cloud using a triangular mesh reconstruction algorithm Perform structural modeling to generate a 3D master model of continuous surfaces ;

[0090] In this embodiment, a Poisson reconstruction method is specifically used to generate a closed mesh surface, and then generate a three-dimensional main model of a continuous surface. The Poisson reconstruction method is an existing technology and will not be described in detail.

[0091] Step 2-4: Export the 3D master model .

[0092] In this embodiment, the three-dimensional main model Expressed as:

[0093] ;

[0094] in, , V represents the vertex set, and the vertices are generated by the representative points of voxel v; , F represents the face set, Represents a triangular patch, and nn and mm both represent the total number.

[0095] Step 3: By fusing the real-time 6-DOF pose of the handheld 3D scanner with the original point cloud data, the point cloud frames are unified into the world coordinate system using a homogeneous transformation matrix. A set of standard time tags is constructed using a preset time interval, and the timestamp of each point is aligned to the nearest neighbor standard tag to generate an enhanced point cloud dataset. The enhanced point cloud dataset is processed into a time-block structure, and the subsets in the block structure are divided into subsets in chronological order to generate a W-model with time tags.

[0096] When executing step 3, the specific steps include:

[0097] Step 3-1: Obtain the original point cloud dataset , and obtain posture information at the same time;

[0098] Step 3-2: Obtain the six-degree-of-freedom pose of each frame captured by the handheld 3D scanner through the spatial tracking system, generate a homogeneous transformation matrix, and transform each point in the homogeneous transformation matrix into a unified world coordinate system;

[0099] In this embodiment, after the handheld 3D scanner scans the surface of the workpiece, the current posture of the workpiece in the global coordinate system is obtained in real time. :

[0100] ;

[0101] in, represents the rotation matrix at the time of acquisition of the i-th frame, Indicates the translation vector when the i-th frame is captured.

[0102] Current posture The six-degree-of-freedom pose.

[0103] The handheld 3D scanner is synchronized with the spatial tracker's clock.

[0104] Obtain the current posture of each frame of the handheld 3D scanner through the spatial tracking system , re- Represented as a homogeneous transformation matrix :

[0105] ;

[0106] in, represents the transformation from the handheld 3D scanner coordinate system to the global world coordinate system; It is a three-dimensional special Euclidean group that represents all possible rotation + translation combinations in three-dimensional space, ensuring that the shape and size remain unchanged.

[0107] The original point cloud dataset Each point in is transformed into a unified world coordinate system. The specific formula is as follows:

[0108] ;

[0109] Represents a point in the unified world coordinate system.

[0110] Step 3-3: Construct a standard time tag set, in which a standard time tag is defined at a fixed time interval;

[0111] In this embodiment, in order to unify the time sequence structure, a standard time tag set is introduced to normalize the time tags. The standard time tag set As shown below:

[0112] ;

[0113] in, represents a standard time label, k ranges from 0 to nn. In this embodiment, the standard time label is a discrete time slice, specifically, a standard time label is defined every 50 ms.

[0114] Step 3-4: Calculate the standard time label close to the time label of each point to generate a new enhanced point cloud dataset;

[0115] In this embodiment, the time label of each point is calculated by the following formula: Nearby standard time stamp :

[0116] ;

[0117] Summarize the calculated results to form a new enhanced point cloud dataset :

[0118] ;

[0119] in, is the world coordinate of the point, is the nearest standard time label obtained after normalization, is the homogeneous transformation matrix, representing the real-time posture.

[0120] Step 3-5: Construct the enhanced point cloud dataset into a time-blocked structure consisting of multiple subsets, sort the subsets in chronological order, and obtain a structured point cloud model with time labels, namely the W-model model;

[0121] In this embodiment, all points are marked with standard time tags. Clustering to form a time block structure :

[0122] ;

[0123] in, To enhance point cloud dataset The subset in , k represents the subset number, and nn represents the total number of subsets.

[0124] ;

[0125] After sorting the subsets in chronological order, the W-model is obtained.

[0126] In this embodiment, the purpose of constructing the structured point cloud model W-model with time tags is to preserve the temporal evolution characteristics of the point cloud during the scanning process.

[0127] The model not only retains the spatial position of each point, but also synchronously records its acquisition time and the corresponding scanner posture information, supporting subsequent operations such as time correction, error compensation, and path tracing.

[0128] Step 4: Use the geometric center alignment method and ICP algorithm to spatially align the 3D master model and the W-model. Calculate the alignment error by building an error compensation function, convert the alignment error into a visual color, obtain a color code, and output the aligned 3D master model and W-model along with the color code.

[0129] When executing step 4, the specific steps include:

[0130] Step 4-1: Obtain the 3D master model, W-model, homogeneous transformation matrix and standard time tag set; specifically the 3D master model , W-model model, homogeneous transformation matrix and a standard set of time tags ;

[0131] Step 4-2: Calculate the geometric centers of all vertices of the 3D main model and the center of all points of the W-model ; By calculation and The difference between , translate all points of the W-model to the 3D main model for center alignment;

[0132] In this embodiment, the three-dimensional main model The geometric center of all vertices Calculated by the following formula:

[0133] ;

[0134] Where N represents the total number of vertices, Represents the 3D main model The vertices in .

[0135] The center of all points in the W-model Calculated by the following formula:

[0136] ;

[0137] in, Represents the total number of subsets of the W-model model, nk represents the number of point clouds in the k-th subset, represents the coordinates of the jth point in the kth subset.

[0138] Translation compensation vector The calculation formula is as follows:

[0139] ;

[0140] Step 4-3: Use the weighted ICP algorithm to find the optimal matching result based on the processing results of step 4-2 ,in, Indicates rotation, Indicates translation;

[0141] In this embodiment, the formula of the weighted ICP algorithm is as follows:

[0142] ;

[0143] in, is the i-th point in the W-model; 3D master model and The nearest corresponding point; For weight:

[0144] ;

[0145] in, Yes Time tag, is the current reference time, is the time tolerance window.

[0146] Step 4-4: Construct an error compensation function to correct the alignment error, and convert the error value into a visual color through the color mapping function ColorMap;

[0147] The specific formula of the error compensation function is as follows:

[0148] ;

[0149] in, is the residual error of the i-th point, which represents the spatial deviation between the W-model and the corresponding point of the 3D main model; Represents the three-dimensional coordinates of the i-th point in the W-model; Represents The three-dimensional main model corresponding to the point The point in .

[0150] Step 4-5: Output the aligned W-model and 3D main model, and output the error value of each point and their corresponding color codes.

[0151] Step 5: Divide the aligned W-model into blocks and cache them;

[0152] Render the W-model and the 3D main model, and push the rendered layer results to the VR device;

[0153] When executing step 5, the specific steps include:

[0154] Step 5-1: Divide the aligned W-model into blocks and cache them;

[0155] In this embodiment, when the W-model is cached in blocks, its structure is as follows:

[0156] [Meta information] + [Time tag index] + [Spatial data block collection] + [Operation log];

[0157] The [spatial data block set] includes multiple [spatial blocks], and the structure of each [spatial block] is as follows:

[0158] [Block unique identifier] + [Point cloud data] + [Time stamp] + [Version tracking identifier] + [Alignment error];

[0159] Among them, [meta information] is the basic information of the block, which is the name or number of the subset;

[0160] [Time Tag Index] is an index created using time tags. In this embodiment, it can quickly locate the corresponding spatial block by time range, facilitating historical point cloud backtracking, time series animation, and dynamic error compensation.

[0161] [Operation log] records the user operation history of each space block, such as user ID, operation time, operation block ID, and operation type (measurement, annotation, etc.).

[0162] [Block unique identifier] is the ID of the space block;

[0163] [Point cloud data] is an enhanced point cloud dataset A subset of

[0164] [Version tracking identifier] is the version tracking identifier for multi-user collaboration.

[0165] Step 5-2: Use multi-layer management to render the W-model and the 3D main model separately. Through transparency overlay and depth sorting, the two models are superimposed and fused. The point cloud deviation is displayed in the layer according to color coding to generate the rendered layer result.

[0166] Step 5-3: Push the rendered layer results to the VR device.

[0167] Step 6: Compare the W-model with the preset workpiece CAD model. Identify the missed scanning areas based on the coverage of the point cloud on the CAD model surface mesh. Generate a path plan based on the location of the missed scanning areas. Render the path plan lines into the 3D master model using the rendering method in step 5 and send it to the VR device.

[0168] When executing step 6, the specific steps include:

[0169] Step 6-1: Count the coverage of the point cloud on the surface grid of the CAD model, obtain the coverage of each grid unit, and count the number of point clouds actually falling into each grid unit. When the coverage of a grid is less than the preset threshold, the area where the grid is located is determined to be a missed scan area;

[0170] Step 6-2: Overlay a highlight color mask on the grid cells in the missed area;

[0171] In this embodiment, in the rendering layer of the three-dimensional main model, a highlight color mask is added to all meshes in the missed-scan area, and then the surface of the missed-scan area is covered with red or other eye-catching colors to achieve highlight display.

[0172] Step 6-3: Plan the next scanning path based on the location of the missed scanning area and generate the path plan;

[0173] In this embodiment, when planning the next scanning path, the geometric centers of all missed grid cells are first extracted to form a region set;

[0174] Combined with the posture of current handheld 3D scanner , taking the current posture position as the starting point, using the cost function to calculate the path point sequence, the cost function As shown below:

[0175] ;

[0176] ;

[0177] Among them, the cost function It represents the optimal path; It represents the sequence of path points, that is, starting from the current scanner position, it passes through several path points in sequence to finally cover the missed scanning area; n represents the total number of path points, Represents two consecutive points;

[0178] Step 6-4: Generate path lines based on the path planning, render the path lines onto the 3D main model according to the rendering method in step 5, and push them to the VR device.

[0179] In this embodiment, a system architecture of main server + VR + spatial tracking system + handheld scanner can be adopted. The main server is responsible for core computing tasks such as processing and storage of large-scale point cloud data, W-model construction, error alignment, path planning, etc. to ensure data consistency and processing efficiency; VR equipment is mainly responsible for real-time rendering and interactive display of three-dimensional models, and the handheld scanner is responsible for data collection and uploading the original point cloud and posture information to the main server; the spatial tracking system Tracker is used to obtain the six-degree-of-freedom posture information of the handheld three-dimensional scanner in space in real time, and data is transmitted in real time between the Tracker and the main server via WLAN.

[0180] Each device can communicate with each other through the local area network. The main server centrally processes and coordinates the workflow of each module and pushes the processing results (such as rendering layers and path guidance) to the VR device.

[0181] The real-time reverse scanning guidance method described in the present invention solves the technical problems of time synchronization, precise spatial registration and error compensation, and efficient block cache rendering of multi-period dynamic point clouds during scanning modeling. The present invention introduces a dynamic point cloud model (W-model) with time series markers, and combines the scanner posture to achieve precise spatial alignment and time-weighted error compensation of multi-period point clouds, effectively solving the problems of posture drift and time asynchrony. It adopts spatial block caching and multi-layer fusion rendering technology to achieve efficient management of large-scale dynamic point clouds and VR real-time visualization, and automatically identifies missed scanning areas and plans paths based on point cloud coverage, greatly improving the efficiency of re-scanning.

Claims

1. A real-time reverse scanning guidance method, characterized by: The steps include: Step 1: Use a handheld 3D scanner to scan the workpiece surface and collect the spatial coordinate information and posture information of the point cloud in real time. Each point cloud data is recorded with its corresponding acquisition time. A time tag is added to each point based on the acquisition time to construct the original point cloud dataset. Step 2: After obtaining the original point cloud dataset and preprocessing it, the point cloud in the original point cloud dataset is processed using a voxel grid filtering algorithm + a triangular mesh reconstruction algorithm to generate a 3D master model with a continuous surface and output it; Step 3: By fusing the real-time 6-DOF pose of the handheld 3D scanner with the original point cloud data, the point cloud of each frame is unified to the world coordinate system using a homogeneous transformation matrix; A set of standard time labels is constructed using preset time intervals, and the timestamp of each point is aligned to the nearest neighbor standard label to generate an enhanced point cloud dataset; The enhanced point cloud dataset is processed into a time block structure, and the subsets in the block structure are divided in chronological order to generate a W-model with time labels. Step 4: Use the geometric center alignment method and ICP algorithm to spatially align the 3D master model and the W-model. Calculate the alignment error by building an error compensation function, convert the alignment error into a visual color, obtain a color code, and output the aligned 3D master model and W-model along with the color code. Step 5: Divide the aligned W-model into blocks and cache them; Render the W-model and the 3D main model, and push the rendered layer results to the VR device; Step 6: Compare the W-model with the preset CAD model of the workpiece. Identify the missed scanning areas based on the coverage of the point cloud on the surface mesh of the CAD model. Generate path planning based on the location of the missed scanning areas using the cost function. Render the lines of the path planning into the 3D main model according to the rendering method in step 5 and send it to the VR device.

2. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 1, the specific steps include: Step 1-1: Use a laser scanner to scan the point cloud data of the workpiece surface, record the spatial coordinates and acquisition time of each point, and generate a point cloud set for each frame based on the spatial coordinates of each point , add a time tag to each point cloud according to the acquisition time ; Step 1-2: Based on point cloud collection Add a time stamp to each sampling point , generate the original point cloud dataset : ； Where M is the total number of frames, represents the three-dimensional coordinates of the j-th point in the i-th frame, , ,N represents the number of point clouds in the i-th frame.

3. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 2, the specific steps include: Step 2-1: Original point cloud dataset Perform preprocessing to remove invalid points, isolated points and abnormal noise; Step 2-2: Sparse the point cloud using a voxel grid filtering algorithm; Step 2-3: Reconstruct the sparse point cloud using a triangular mesh reconstruction algorithm Perform structural modeling to generate a 3D master model of continuous surfaces ; Step 2-4: Export the 3D master model .

4. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 3, the specific steps include: Step 3-1: Obtain the original point cloud dataset , and obtain posture information at the same time; Step 3-2: Obtain the six-degree-of-freedom pose of each frame captured by the handheld 3D scanner through the spatial tracking system, generate a homogeneous transformation matrix, and transform each point in the homogeneous transformation matrix into a unified world coordinate system; Step 3-3: Construct a standard time tag set, in which a standard time tag is defined at a fixed time interval; Step 3-4: Calculate the standard time label close to the time label of each point to generate a new enhanced point cloud dataset; Step 3-5: Construct the enhanced point cloud dataset into a time block structure consisting of multiple subsets, sort the subsets in chronological order, and obtain a structured point cloud model with time labels, namely the W-model model.

5. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 4, the specific steps include: Step 4-1: Obtain the 3D master model, W-model, homogeneous transformation matrix and standard time tag set; Step 4-2: Calculate the geometric centers of all vertices of the 3D main model and the center of all points of the W-model ; By calculation and The difference between , translate all points of the W-model to the 3D main model for center alignment; Step 4-3: Using the weighted ICP algorithm, based on the processing results of step 4-2, find the optimal matching result; Step 4-4: Construct an error compensation function to calculate the alignment error of each point, convert the error value into a visual color through the color mapping function ColorMap, and generate a color code; Step 4-5: Output the aligned W-model and the 3D main model, and output the alignment error of each point and its corresponding color coding.

6. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 5, the specific steps include: Step 5-1: Divide the aligned W-model into blocks and cache them. When the W-model is cached in blocks, its structure is as follows: [Meta information] + [Time tag index] + [Spatial data block collection] + [Operation log]; The [spatial data block set] includes multiple [spatial blocks], and the structure of each [spatial block] is as follows: [Block unique identifier] + [Point cloud data] + [Time stamp] + [Version tracking identifier] + [Alignment error]; Step 5-2: Use multi-layer management to render the W-model and the 3D main model separately. Through transparency overlay and depth sorting, the two models are superimposed and fused. The point cloud deviation is displayed in the layer according to color coding to generate the rendered layer result. Step 5-3: Push the rendered layer results to the VR device.

7. The real-time reverse scanning guidance method according to claim 1, characterized in that: When executing step 6, the specific steps include: Step 6-1: Count the coverage of the point cloud on the surface grid of the CAD model, obtain the coverage of each grid unit, and count the number of point clouds actually falling into each grid unit. When the coverage of a grid is less than the preset threshold, the area where the grid is located is determined to be a missed scan area; Step 6-2: Overlay a highlight color mask on the grid cells in the missed area; Step 6-3: Plan the next scanning path based on the location of the missed scanning area and generate a path plan. When planning the next scanning path, first extract the geometric centers of all missed scanning grid cells to form a region set. Combined with the posture of current handheld 3D scanner , taking the current posture position as the starting point, using the cost function to calculate the path point sequence, the cost function As shown below: ； ； Among them, the cost function It represents the optimal path; It represents the sequence of path points, that is, starting from the current scanner position, it passes through several path points in sequence to finally cover the missed scanning area; n represents the total number of path points, Represents two consecutive points; Step 6-4: Generate path lines based on the path planning, render the path lines onto the 3D main model according to the rendering method in step 5, and push them to the VR device.

Images