A three-dimensional size synchronous measurement method based on multi-view feature fusion

Abstract

The application provides a three-dimensional size synchronous measurement method based on multi-view feature fusion, and relates to the technical field of machine vision three-dimensional measurement.The application obtains unified three-dimensional feature data by acquiring multi-source data of a workpiece to be measured collected by multiple cameras for timing synchronization calibration and normalization coding;the unified three-dimensional feature data is subjected to layered processing, and each feature layer is subjected to enhancement processing to obtain multi-layer enhanced feature data;three-dimensional coordinate mapping relationships corresponding to the collection angles of the multiple cameras are established, the multi-layer enhanced feature data is subjected to cross-angle fusion fitting, and three-dimensional space points corresponding to the workpiece to be measured are obtained;the three-dimensional space points are subjected to incremental updating, multiple three-dimensional size parameters of the workpiece to be measured are determined, three-dimensional size measurement results to be calibrated are obtained, a size error constraint model is established according to the error relationship between the three-dimensional size measurement results to be calibrated and preset reference size parameters, the three-dimensional size measurement results to be calibrated are calibrated, and standardized three-dimensional size detection results are output.

Description

Technical Field

[0001] This invention relates to the field of machine vision 3D measurement technology, and more specifically, to a method for synchronous 3D dimension measurement based on multi-view feature fusion. Background Technology

[0002] With the rapid development of intelligent manufacturing technology, the requirements for accuracy and efficiency in online inspection of precision workpieces in industrial production processes are constantly increasing. Machine vision-based 3D dimension inspection technology, due to its non-contact, high efficiency, and high degree of automation, is gradually being applied to production scenarios such as 3C components, precision machined parts, and plastic workpieces. Among them, 3D dimension measurement methods based on multi-camera vision systems can acquire workpiece surface images and depth information from multiple perspectives, enabling 3D dimension inspection of workpieces with complex structures.

[0003] Existing multi-camera vision-based 3D dimension measurement methods typically acquire image data from different viewpoints using multiple cameras and combine this data with camera calibration, point cloud registration, and 3D reconstruction algorithms to obtain the workpiece's spatial dimensions. However, in practical applications, the following problems still exist:

[0004] First, traditional single-camera 3D measurement methods rely on a single viewpoint to obtain workpiece information. When the workpiece has occluded areas or complex structures, visual information is easily lost. At the same time, it is affected by perspective distortion, which leads to a decrease in the measurement accuracy of dimensional parameters such as workpiece length, width, height, and hole diameter.

[0005] Secondly, existing multi-camera 3D measurement systems typically use a multi-device independent acquisition method, which leads to time synchronization errors between different cameras. Data from different perspectives may have timestamp offsets, resulting in spatial deviations during subsequent point cloud registration and 3D coordinate reconstruction, affecting the consistency of batch workpiece inspection results.

[0006] Furthermore, existing 3D dimension measurement methods typically process the acquired image information, depth point cloud information, and texture identification information as a whole, lacking a layered extraction and independent optimization mechanism for different types of 3D features. This makes geometric contour features, depth structure features, and texture positioning features susceptible to noise interference, resulting in error accumulation during the dimension fitting process, which makes it difficult to meet the high-precision inspection requirements of precision workpieces.

[0007] In addition, the existing multi-view 3D reconstruction and size calculation process usually adopts a serial processing method, which requires the completion of all image acquisition before unified matching and size calculation. This results in a long inspection time for a single workpiece, making it difficult to meet the real-time inspection requirements of high-speed automated production lines.

[0008] Meanwhile, during continuous detection, static structural information such as tooling and fixed benchmarks usually need to be repeatedly used in 3D feature extraction and coordinate calculation. The lack of a caching and reuse mechanism for static benchmark data leads to repeated consumption of computing resources and increases the deployment cost of edge computing devices.

[0009] In summary, existing multi-view 3D size measurement technologies typically optimize point cloud registration, size fitting, or a single calculation step, making it difficult to simultaneously address issues such as inconsistent timing of multi-camera data, feature noise interference, size calculation delays, and wasted computing resources.

[0010] In view of the above, this application is hereby submitted. Summary of the Invention

[0011] This invention provides a method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion to improve at least one of the above-mentioned technical problems, comprising the following steps: S1: Acquire multi-source data of the workpiece under test collected by multiple cameras, perform time-series synchronization calibration on the multi-source data according to the acquisition time of each camera, and normalize and encode the multi-source data to obtain unified three-dimensional feature data; wherein, the multi-source data includes two-dimensional image data, depth point cloud data and texture reference marker data; Preferably, the plurality of cameras are arranged around the workpiece to form two or more different acquisition perspectives; Furthermore, the timing synchronization calibration includes: A multi-camera time synchronization matrix is ​​established based on the acquisition time of multiple cameras. The time deviation between the data acquired by different cameras is determined based on the multi-camera time synchronization matrix. Time compensation is performed on the multi-source data acquired by different cameras based on the time deviation, so that the multi-source data acquired by different cameras correspond to the same time reference. The normalized encoding includes: The two-dimensional image data, depth point cloud data, and texture reference marker data are respectively subjected to scale normalization, data format conversion, and feature dimension unification processing to obtain corresponding data feature codes, and the unified three-dimensional feature data is generated based on the data feature codes.

[0012] S2: Perform layered processing on the unified three-dimensional feature data to obtain a geometric contour feature layer, a depth point cloud feature layer, and a texture reference feature layer, and perform enhancement processing on each feature layer to obtain multi-layer enhanced feature data. Furthermore, the layering process includes: Based on the gradient magnitude information corresponding to different spatial locations in the unified three-dimensional feature data, the feature response intensity of each spatial region is determined, and the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer are obtained based on the feature response intensity. The enhancement process includes: Noise suppression, feature enhancement, and feature compensation are performed on the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer respectively to enhance the three-dimensional structural information corresponding to different feature layers.

[0013] S3: Establish a three-dimensional coordinate mapping relationship corresponding to the acquisition viewpoints of multiple cameras; perform cross-viewpoint fusion fitting on the multi-layer enhanced feature data according to the three-dimensional coordinate mapping relationship to obtain the three-dimensional spatial point position corresponding to the workpiece under test; construct a static benchmark cache; incrementally update the three-dimensional spatial point position to obtain the optimized three-dimensional spatial point position; and determine multiple three-dimensional dimension parameters of the workpiece under test based on the optimized three-dimensional spatial point position to obtain the three-dimensional dimension measurement result to be calibrated composed of the multiple three-dimensional dimension parameters. Furthermore, the three-dimensional coordinate mapping relationship is determined based on the internal and external parameters corresponding to multiple cameras; The internal parameters include camera focal length parameters and imaging center parameters, and the external parameters include positional relationship parameters and attitude relationship parameters between cameras; Based on the aforementioned three-dimensional coordinate mapping relationship, the three-dimensional feature data corresponding to different camera acquisition perspectives are transformed into a unified spatial coordinate system; The static reference cache includes the three-dimensional feature data, spatial coordinate information, and reference size parameters corresponding to the static reference object; The static reference object is a reference structure whose spatial position remains fixed in the detection scene or a reference object whose size is known. The incremental update includes: Based on the three-dimensional feature data and spatial coordinate information in the static reference cache, it is matched with the currently obtained three-dimensional spatial points to determine the dynamic change area corresponding to the workpiece to be tested, and only the three-dimensional spatial points corresponding to the dynamic change area are updated to obtain the optimized three-dimensional spatial points. Determining multiple three-dimensional dimensional parameters of the workpiece to be tested based on the optimized three-dimensional spatial points includes: calculating the distance based on the spatial positional relationship between the optimized three-dimensional spatial points to obtain multiple three-dimensional dimensional parameters corresponding to the workpiece to be tested; Furthermore, the plurality of three-dimensional dimensional parameters include at least one of the length dimension, width dimension, and height dimension.

[0014] S4. Establish a size error constraint model based on the error relationship between the three-dimensional size measurement result to be calibrated and the preset reference size parameters, and calibrate the three-dimensional size measurement result to be calibrated based on the size error constraint model, and output a standardized three-dimensional size detection result. Furthermore, the size error constraint model includes a size error function, a size gradient smoothing constraint, and a timing synchronization constraint. Based on the size error constraint model, the three-dimensional size measurement results to be calibrated are iteratively calibrated to obtain standardized three-dimensional size detection results that meet the size error constraint conditions.

[0015] Compared with the prior art, the present invention has the following beneficial effects: By synchronizing the acquisition time of multiple cameras, a unified time reference for multiple cameras is established, which reduces the impact of acquisition time sequence deviation of different cameras on the 3D data fusion accuracy and improves the consistency of multi-view measurement. By performing layered processing on unified 3D feature data, geometric contour information, depth point cloud information and texture reference information are respectively enhanced to reduce mutual interference between different types of data and improve the ability to extract 3D structural information of complex workpieces. By establishing a 3D coordinate mapping relationship corresponding to the viewpoints of multiple cameras, the fusion and fitting of 3D feature data from different viewpoints is realized. Incremental updates of 3D spatial points are performed by combining static benchmark cache, reducing redundant calculations and improving online measurement efficiency. By calculating dimensional parameters based on optimized 3D spatial points and calibrating them using a dimensional error constraint model, the accuracy and stability of 3D dimensional measurement results are improved. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some specific embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for synchronous 3D dimension measurement based on multi-view feature fusion.

[0018] Figure 2 This is a schematic diagram of a three-dimensional dimension synchronous measurement system based on multi-view feature fusion. Detailed Implementation

[0019] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Example 1 See Figure 1 The first embodiment of the present invention provides a method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion, comprising the following steps: S1: Acquire multi-source data of the workpiece under test collected by multiple cameras; Preferably, the plurality of cameras are arranged around the workpiece to form two or more different acquisition perspectives; The multi-source data includes two-dimensional image data, depth point cloud data, and texture reference marker data; Let the first Each camera at the time of data collection The acquired multi-source data of the workpiece under test is represented as follows: ; In the formula, It is two-dimensional image data. For deep point cloud data, Use the data to mark the texture reference.

[0021] Since there are differences in the acquisition time among multiple cameras, the multi-source data is time-synchronized and calibrated according to the acquisition time of each camera, and the multi-source data is normalized and encoded to obtain unified three-dimensional feature data. Specifically, the timing synchronization calibration includes: Establish a multi-camera time synchronization matrix based on the acquisition times of multiple cameras; Let the first The acquisition time for each camera is , No. The acquisition time for each camera is... The time difference between the two cameras for: ; According to the time difference Constructing timing synchronization coefficients : ; In the formula, This is a timing decay parameter used to control the rate at which the synchronization weight decreases as the time difference increases; The preset synchronization time threshold; As an indicator function, when the time difference between two cameras... Less than the preset synchronization time threshold The value is 1 if the condition is met, otherwise it is 0. Timing synchronization coefficients between multiple cameras Constructing a multi-camera time synchronization matrix: ; In the formula, The number of cameras participating in the timing synchronization calibration (i.e., the total number of cameras). The time deviation between the data collected by different cameras is determined based on the multi-camera time synchronization matrix, and time compensation is performed on the multi-source data collected by different cameras based on the time deviation, so that the multi-source data collected by different cameras corresponds to the same time reference. The normalized encoding is specifically as follows: The two-dimensional image data, depth point cloud data, and texture reference marker data are respectively subjected to scale normalization, data format conversion, and feature dimension unification to obtain the corresponding data feature codes. Data feature encoding corresponding to two-dimensional image data: ; In the formula, For two-dimensional image data normalization encoding function, These are the two-dimensional image features obtained after encoding; Data feature encoding corresponding to deep point cloud data: ; In the formula, For the normalization encoding function of deep point cloud data, These are the encoded depth space features; Data feature encoding corresponding to texture baseline marker data: ; In the formula, For the normalization encoding function of texture reference marker data, This is the encoded texture baseline feature.

[0022] Furthermore, different types of data feature codes are uniformly represented to generate the unified three-dimensional feature data: ; In the formula, For a moment Pixel coordinates Unified three-dimensional feature tensor; Encoding the data features corresponding to two-dimensional image data. Encoding the data features corresponding to deep point cloud data. Encode the data features corresponding to the texture reference marker data. For multi-camera timing synchronization matrix, A unified encoding mapping function for multi-source data; The unified three-dimensional feature data is a three-dimensional feature representation formed by normalizing and encoding two-dimensional image data, depth point cloud data, and texture reference marker data.

[0023] In one alternative implementation, to further reduce the impact of acquisition noise, spatiotemporal continuity constraint processing can also be performed based on unified three-dimensional feature data: ; In the formula, For the spatiotemporal joint constraint loss function, Constrained by changes in time direction. For time continuity constraints, For spatial gradient constraints, and These are constraint parameters; Spatiotemporal continuity constraint processing is used to measure the degree of change of uniform 3D feature data in the temporal and spatial dimensions. By optimizing each constraint term, temporal disturbances and spatial noise during multi-camera acquisition are reduced.

[0024] S2: Perform layered processing on the unified three-dimensional feature data to obtain a geometric contour feature layer, a depth point cloud feature layer, and a texture reference feature layer, and perform enhancement processing on each feature layer to obtain multi-layer enhanced feature data. The layered processing includes: Based on the gradient magnitude information corresponding to different spatial locations in the unified three-dimensional feature data, the feature response intensity of each spatial region is determined, and the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer are obtained based on the feature response intensity. The gradient magnitude information is calculated based on the unified three-dimensional feature data as follows: ; In the formula, , These represent the gradient changes of the unified three-dimensional feature data along the spatial coordinate directions, respectively. This represents the characteristic response intensity at the corresponding spatial location.

[0025] Pre-set gradient stratification threshold: ; Based on the relationship between the gradient magnitude information and the gradient layering threshold, the unified three-dimensional feature data is divided: ; ; ; In the formula, This is a geometric contour feature layer used to characterize the edge contour and structural change areas of the workpiece under test; This is a depth point cloud feature layer used to characterize the spatial depth variation information of the workpiece under test; This is the texture reference feature layer, used to characterize the workpiece texture markings and reference positioning information; Through the above layered processing, the unified three-dimensional feature data is decomposed into different types of feature layers; Furthermore, enhancement processing is performed on each feature layer: Noise suppression, feature enhancement, and feature compensation are performed on the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer respectively to enhance the three-dimensional structural information corresponding to different feature layers; For geometric contour feature layer To improve the continuity and stability of the workpiece edge contour information, contour enhancement processing is performed: ; In the formula, For the enhanced geometric contour features, Preserve weights for the original contour information. To enhance the weights for gradient enhancement, To smooth the constraint weights, Let the first norm of the contour gradient be denoted as . This is a higher-order smoothing constraint term for the contour features; The above processing improves the feature response of the geometric contour boundary region and reduces the impact of local noise on the fitting of unified three-dimensional feature data.

[0026] For deep point cloud feature layers Considering that depth point cloud data is easily affected by acquisition noise and local depth fluctuations, consistency enhancement processing is performed: ; In the formula, To enhance the features of the deep point cloud, Preserve weights for deep features For depth smoothing constraint weights, This represents the depth consistency constraint coefficient. For the second-order difference smoothing term of the depth feature, , This represents the depth value of adjacent spatial sampling points.

[0027] The above processing maintains depth continuity between adjacent spatial points and reduces the impact of outliers on the calculation of unified 3D feature data.

[0028] For texture reference feature layer To improve the positioning accuracy of the reference marker, texture feature enhancement is performed: ; In the formula, For enhanced texture baseline features, To preserve the weights for texture baseline information, To enhance the weights of the spatiotemporal joint gradient, This represents the coupled gradient in the time and space directions.

[0029] The above processing improves the stability of texture reference features, providing reference information for subsequent cross-view coordinate mapping and size calibration.

[0030] After the above three-layer feature layering and enhancement processing, multi-layer enhanced feature data is obtained. : ; S3: Establish a three-dimensional coordinate mapping relationship corresponding to the acquisition viewpoints of multiple cameras; perform cross-viewpoint fusion fitting on the multi-layer enhanced feature data according to the three-dimensional coordinate mapping relationship to obtain the three-dimensional spatial point position corresponding to the workpiece under test; construct a static benchmark cache; incrementally update the three-dimensional spatial point position to obtain the optimized three-dimensional spatial point position; and determine multiple three-dimensional dimension parameters of the workpiece under test based on the optimized three-dimensional spatial point position to obtain the three-dimensional dimension measurement result to be calibrated composed of the multiple three-dimensional dimension parameters. Furthermore, the plurality of three-dimensional dimensional parameters include length, width, and height. In one alternative implementation, the plurality of three-dimensional dimensional parameters may further include one or more of the following: inner hole diameter, fitting clearance, and spatial angle.

[0031] Furthermore, the three-dimensional coordinate mapping relationship is determined based on the internal and external parameters corresponding to multiple cameras; The internal parameters include camera focal length parameters and imaging center parameters, and the external parameters include positional relationship parameters and attitude relationship parameters between cameras; Based on the aforementioned three-dimensional coordinate mapping relationship, the three-dimensional feature data corresponding to different camera viewpoints are transformed into a unified spatial coordinate system: ; In the formula, To obtain the three-dimensional spatial points through fusion, For the first The coordinate mapping matrix corresponding to each camera is used to represent the camera's intrinsic and extrinsic parameters and coordinate transformation relationships. For the first Multi-layered augmented feature data output from each camera.

[0032] To reduce the impact of changes in the fixed detection environment on the measurement accuracy of three-dimensional dimensional parameters, a static reference cache is pre-established. The static reference cache includes the three-dimensional feature data, spatial coordinate information and reference dimensional parameters corresponding to the static reference object. The static reference object is a reference structure whose spatial position remains fixed in the detection scene or a reference object whose size is known. The static benchmark cache Represented as: ; In the formula, The three-dimensional feature data corresponding to the static reference object. This refers to the spatial coordinate information corresponding to the static reference object. These are the reference dimension parameters corresponding to the static reference object; During the dimensional measurement process, the static reference cache is invoked to match the currently obtained three-dimensional spatial points and determine the dynamic change area corresponding to the workpiece to be measured.

[0033] Based on the dynamically changing region, the corresponding three-dimensional spatial points are incrementally updated. Specifically, the 3D spatial points obtained through cross-view fusion are used as the initial points: ; Based on the 3D feature data, spatial coordinate information, and reference size parameters in the static reference cache, spatial point optimization constraints are established, and the 3D spatial points are iteratively updated: ; In the formula, For the first Three-dimensional spatial points in the round iteration process For the first The updated three-dimensional spatial points To iteratively update the step size, To jointly optimize the loss function.

[0034] In one optional implementation, the joint optimization loss function includes spatial point consistency constraints, depth continuity constraints, and temporal stability constraints: ; In the formula, For the first The initial 3D spatial points after camera mapping For the optimized three-dimensional spatial point positions, , These are the corresponding constraint weights; Used to constrain the consistency of spatial point locations under different camera perspectives; Used to constrain the continuity of the depth space; Used to constrain the stability of depth changes during continuous data acquisition; After iterative optimization, the optimized three-dimensional spatial point locations are obtained: .

[0035] Based on the optimized three-dimensional spatial point positions : To determine multiple three-dimensional dimensional parameters of the workpiece to be measured; Specifically, distance calculations are performed based on the optimized spatial relationships between the three-dimensional spatial points to obtain the three-dimensional dimensional parameters of the workpiece to be measured. For example, for any two spatial points: and ; Spatial distance for: ; Based on the geometric relationship between multiple spatial points, the 3D dimension measurement results to be calibrated can be obtained simultaneously.

[0036] In one alternative implementation, to improve the accuracy of three-dimensional dimension parameter calculation, a hierarchical dimension fitting loss function is established based on the optimized three-dimensional spatial points. : ; In the formula, The number of iterations for size fitting. For actual three-dimensional spatial points, These are the model points obtained by fitting geometric features based on the three-dimensional spatial points. It is the Frobenius norm. For spatial smoothing constraint weights, For depth smoothing constraint weights, For the gradient change of points in three-dimensional space, This represents the second-order difference transformation of depth data.

[0037] By minimizing the hierarchical size fitting loss function : ; The optimized geometric fitting results are obtained, and the three-dimensional dimensional parameters are redetermined based on the geometric fitting results.

[0038] S4. Based on the three-dimensional dimension measurement results to be calibrated Compared with preset reference size parameters Establish a dimensional error constraint model based on the error relationship between them. ; Furthermore, the dimensional error constraint model includes a dimensional error function and dimensional constraint conditions; In one alternative implementation, the size error constraint model for: ; In the formula, For the gradient of size parameter changes, The size gradient smoothing constraint coefficient is... These are the timing synchronization constraint coefficients. The gradient of spatial position change. This is the multi-camera timing synchronization constraint matrix; in, This is a dimensional error function used to characterize the deviation between the measured dimension and the reference dimension; This is a dimension smoothing constraint term used to limit fluctuations in dimensional parameters; This is a timing synchronization constraint term, used to reduce the impact of timing deviations on the dimensional calibration results by combining the synchronization relationship of multiple cameras.

[0039] Based on the size error constraint model The three-dimensional dimension measurement results to be calibrated are iteratively calibrated to obtain calibrated three-dimensional dimension results that meet the dimension error constraint conditions. : ; By minimizing the dimensional error constraint model, the calibrated dimensional parameters meet the preset accuracy requirements; The calibrated three-dimensional dimensions Output as standardized 3D dimension inspection results.

[0040] The above calibration process reduces the deviation between the calculated three-dimensional dimensions to be calibrated and the standard reference dimensions, thereby improving the accuracy and stability of the three-dimensional dimension detection results.

[0041] By adopting the above technical solution, the following technical effects can be achieved in this embodiment: This method achieves the fusion processing of multi-view 3D information of the workpiece under test, improving the consistency and integrity of 3D data under different acquisition perspectives. At the same time, by combining static reference information to optimize spatial points and calibrating the dimensional measurement results based on an error constraint model, the accuracy and stability of 3D dimensional measurement are improved. Compared with traditional single-view measurement methods, it can reduce measurement errors caused by viewpoint occlusion, data deviation, and changes in the detection environment, and realize the synchronous calculation of multiple types of 3D dimensional parameters of the workpiece, improving the reliability and detection efficiency of online dimensional inspection of complex workpieces.

[0042] Example 2 Please see Figure 2 The second embodiment of the present invention provides a three-dimensional size synchronous measurement system based on multi-view feature fusion, comprising: The multi-source data acquisition module is used to acquire multi-source data of the workpiece under test collected by multiple cameras. The multi-source data includes two-dimensional image data, depth point cloud data and texture reference mark data. The multi-source data is time-series synchronized and calibrated according to the acquisition time of each camera, and the multi-source data is normalized and encoded to obtain unified three-dimensional feature data. The feature layer enhancement module is used to perform layer processing on the unified three-dimensional feature data to obtain a geometric contour feature layer, a depth point cloud feature layer, and a texture reference feature layer, and to perform enhancement processing on each feature layer to obtain multi-layer enhanced feature data. The cross-view fusion measurement module is used to establish a three-dimensional coordinate mapping relationship corresponding to the viewpoints of multiple cameras, perform cross-view fusion fitting on the multi-layer enhanced feature data according to the three-dimensional coordinate mapping relationship, and obtain the three-dimensional spatial point position corresponding to the workpiece under test; construct a static benchmark cache, incrementally update the three-dimensional spatial point position to obtain the optimized three-dimensional spatial point position, and determine multiple three-dimensional dimension parameters of the workpiece under test based on the optimized three-dimensional spatial point position to obtain the three-dimensional dimension measurement result to be calibrated; The size error calibration module establishes a size error constraint model based on the error relationship between the three-dimensional size measurement result to be calibrated and the preset reference size parameters, and calibrates the three-dimensional size measurement result to be calibrated based on the size error constraint model, and outputs a standardized three-dimensional size detection result.

[0043] Example 3 The third embodiment of the present invention provides a three-dimensional size synchronous measurement device based on multi-view feature fusion, including a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement a three-dimensional size synchronous measurement method based on multi-view feature fusion as described in the first embodiment.

[0044] Example 4 The fourth embodiment of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute a three-dimensional size synchronous measurement method based on multi-view feature fusion as described in Embodiment 1.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion, characterized in that, Includes the following steps: S1: Acquire multi-source data of the workpiece under test collected by multiple cameras, perform time-series synchronization calibration on the multi-source data according to the acquisition time of each camera, and normalize and encode the multi-source data to obtain unified three-dimensional feature data; wherein, the multi-source data includes two-dimensional image data, depth point cloud data and texture reference marker data; S2: Perform layered processing on the unified three-dimensional feature data to obtain a geometric contour feature layer, a depth point cloud feature layer, and a texture reference feature layer, and perform enhancement processing on each feature layer to obtain multi-layer enhanced feature data. S3: Establish a three-dimensional coordinate mapping relationship corresponding to the acquisition viewpoints of multiple cameras; perform cross-viewpoint fusion fitting on the multi-layer enhanced feature data according to the three-dimensional coordinate mapping relationship to obtain the three-dimensional spatial point position corresponding to the workpiece under test; construct a static benchmark cache; incrementally update the three-dimensional spatial point position to obtain the optimized three-dimensional spatial point position; and determine multiple three-dimensional dimension parameters of the workpiece under test based on the optimized three-dimensional spatial point position to obtain the three-dimensional dimension measurement result to be calibrated composed of the multiple three-dimensional dimension parameters. S4: Establish a size error constraint model based on the error relationship between the three-dimensional size measurement result to be calibrated and the preset reference size parameters, and calibrate the three-dimensional size measurement result to be calibrated based on the size error constraint model, and output a standardized three-dimensional size detection result.

2. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S1, the multiple cameras are set up around the workpiece to form two or more different acquisition perspectives; The timing synchronization calibration includes: A multi-camera time synchronization matrix is ​​established based on the acquisition time of multiple cameras. The time deviation between the data acquired by different cameras is determined based on the multi-camera time synchronization matrix. Time compensation is performed on the multi-source data acquired by different cameras based on the time deviation, so that the multi-source data acquired by different cameras correspond to the same time reference.

3. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S1, the normalization encoding includes: The two-dimensional image data, depth point cloud data, and texture reference marker data are respectively subjected to scale normalization, data format conversion, and feature dimension unification processing to obtain corresponding data feature codes, and the unified three-dimensional feature data is generated based on the data feature codes.

4. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S2, the layering process includes: Based on the gradient magnitude information corresponding to different spatial locations in the unified three-dimensional feature data, the feature response intensity of each spatial region is determined, and the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer are obtained based on the feature response intensity.

5. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S2, the enhancement process includes: Noise suppression, feature enhancement, and feature compensation are performed on the geometric contour feature layer, depth point cloud feature layer, and texture reference feature layer respectively to enhance the three-dimensional structural information corresponding to different feature layers.

6. The method for synchronous three-dimensional size measurement based on multi-view feature fusion according to claim 1, characterized in that, In step S3, the three-dimensional coordinate mapping relationship is determined based on the internal and external parameters corresponding to multiple cameras; The internal parameters include camera focal length parameters and imaging center parameters, and the external parameters include positional relationship parameters and attitude relationship parameters between cameras; Based on the aforementioned three-dimensional coordinate mapping relationship, the three-dimensional feature data corresponding to different camera acquisition perspectives are transformed into a unified spatial coordinate system.

7. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S3, the static reference cache includes the three-dimensional feature data, spatial coordinate information, and reference size parameters corresponding to the static reference object; The static reference object is a reference structure whose spatial position remains fixed in the detection scene or a reference object whose size is known.

8. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S3, the incremental update includes: Based on the three-dimensional feature data and spatial coordinate information in the static reference cache, the three-dimensional spatial points are matched with the currently obtained three-dimensional spatial points to determine the dynamic change area corresponding to the workpiece to be tested, and only the three-dimensional spatial points corresponding to the dynamic change area are updated to obtain the optimized three-dimensional spatial points.

9. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S3, multiple three-dimensional dimension parameters of the workpiece to be tested are determined based on the optimized three-dimensional spatial points, including: calculating the distance according to the spatial positional relationship between the optimized three-dimensional spatial points to obtain multiple three-dimensional dimension parameters corresponding to the workpiece to be tested; The plurality of three-dimensional dimensional parameters include at least one of length, width, and height.

10. The method for synchronous measurement of three-dimensional dimensions based on multi-view feature fusion according to claim 1, characterized in that, In step S4, the size error constraint model includes a size error function, a size gradient smoothing constraint, and a timing synchronization constraint. Based on the size error constraint model, the three-dimensional size measurement results to be calibrated are iteratively calibrated to obtain standardized three-dimensional size detection results that meet the size error constraint conditions.

Images