Wall embedded part positioning quality evaluation method based on laser ranging

Abstract

The invention discloses a wall embedded part positioning quality evaluation method based on laser ranging, and relates to the technical field of graphic processing, and the method comprises the steps: marking the construction mode of a wall embedded part based on a wall embedded part construction drawing, and evaluating the optimal reference vector of the wall embedded part; based on the construction process point cloud data and the IMU data of the real-time wall embedded part, segmenting the point cloud data of the wall embedded part by using a clustering algorithm to obtain an embedded part point cloud cluster of a construction task of the real-time wall embedded part; construction actions in the construction process image data are marked, a construction video stream of a real-time wall embedded part construction task and a positioning point cloud cluster of the real-time wall embedded part construction task are obtained, a wall embedded part positioning multi-modal evaluation model is established, and a deviation vector of a wall embedded part is generated; and judging whether the comprehensive deviation vector of the wall embedded part is in the optimal reference vector interval of the wall embedded part or not, and if not, outputting that the deviation is too large. The method has the advantages that the embedded part positioning measurement precision is improved, the installation error is reduced, and the rework rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of graphic processing, in particular to a wall embedded part positioning quality evaluation method based on laser ranging. BACKGROUND

[0002] The wall embedded part positioning quality evaluation based on laser ranging is a process of accurately measuring the installation position, perpendicularity, levelness and other key parameters of the embedded part through laser ranging technology to ensure that it meets the design requirements. The evaluation content includes the surface flatness, dimensional accuracy of the embedded part and the connection firmness with the main structure to ensure the long-term safety and durability of the building curtain wall structure.

[0003] The existing wall embedded part positioning quality evaluation based on laser ranging mainly relies on static measurement method, that is, single data collection under fixed conditions, but there are dynamic interference factors such as vibration, temperature change, human operation error in actual construction environment, which leads to deviation of measurement results. The traditional scheme does not consider dynamic compensation mechanism; secondly, the structure stress release or concrete shrinkage after the installation of the embedded part will also cause position deviation, but static measurement cannot capture such time-varying parameters, which finally affects the positioning accuracy and the reliability of engineering acceptance. SUMMARY

[0004] To solve the above technical problems, a wall embedded part positioning quality evaluation method based on laser ranging is provided, which solves the above problems that the existing wall embedded part positioning quality evaluation based on laser ranging mainly relies on static measurement method, that is, single data collection under fixed conditions, but there are dynamic interference factors such as vibration, temperature change, human operation error in actual construction environment, which leads to deviation of measurement results. The traditional scheme does not consider dynamic compensation mechanism; secondly, the structure stress release or concrete shrinkage after the installation of the embedded part will also cause position deviation, but static measurement cannot capture such time-varying parameters, which finally affects the positioning accuracy and the reliability of engineering acceptance.

[0005] To achieve the above purposes, the technical scheme adopted by the present application is as follows:

[0006] A wall embedded part positioning quality evaluation method based on laser ranging, comprising:

[0007] Based on the wall embedded part construction drawing, the construction method of the wall embedded part is marked, the construction preference of the construction method is analyzed, and the optimal reference vector of the wall embedded part is evaluated;

[0008] Based on the real-time wall embedded part construction task, a plurality of laser ranging sensors and image sensors are deployed around the periphery, the construction task parameters of the real-time wall embedded part are collected, and the construction process time sequence data of the real-time wall embedded part is established; the construction process time sequence data includes: point cloud data, IMU data, calibration target feature points and image data;

[0009] Based on the real-time wall embedded part construction process point cloud data and IMU data, the pose transformation matrix of the laser ranging sensor of the real-time wall embedded part construction task is established, and the clustering algorithm is used to segment the wall embedded part point cloud data to obtain the embedded part point cloud cluster of the real-time wall embedded part construction task;

[0010] Based on the real-time wall embedded part construction process image data, the construction action in the construction process image data is marked to obtain the construction video stream of the real-time wall embedded part construction task and the positioning point cloud cluster of the real-time wall embedded part construction task, establish a wall embedded part positioning multi-modal evaluation model, and generate a deviation vector of the wall embedded part;

[0011] It is judged whether the comprehensive deviation vector of the wall embedded part is in the optimal reference vector interval of the wall embedded part. If yes, continue to execute, if not, output the deviation is too large.

[0012] Preferably, according to the wall embedded part construction drawing, the geometric information and attribute information in the wall embedded part construction drawing are extracted to obtain the application scenario of the wall embedded part;

[0013] According to the expected service life of the application scenario of the wall embedded part, the performance requirement of the application scenario of the wall embedded part is determined, the embedded part construction mode meeting the performance requirement is screened, and the selectable embedded part construction mode data array of the application scenario of the wall embedded part is established;

[0014] The multi-source factor parameters affecting the performance decline of the embedded part construction mode in the application scenario of the wall embedded part are marked;

[0015] According to the selectable embedded part construction mode data array of the application scenario of the wall embedded part and the multi-source factor parameters affecting the performance decline of the embedded part construction mode in the application scenario of the wall embedded part, an influence factor matrix of the selectable embedded part construction mode of the application scenario of the wall embedded part is established.

[0016] Preferably, the influence factor matrix of the selectable embedded part construction mode of the application scenario of the wall embedded part is normalized, and the influence factor weight of the selectable embedded part construction mode of the application scenario of the wall embedded part is given according to the entropy weight method;

[0017] According to the influence factor weight of the selectable embedded part construction mode of the application scenario of the wall embedded part and the multi-source factor parameters affecting the performance decline of the embedded part construction mode in the application scenario of the wall embedded part, the performance influence coefficient of the embedded part construction mode of the application scenario of the wall embedded part is calculated;

[0018] According to the ideal performance index of the selectable embedded part construction mode in each of the selectable embedded part construction mode data array of the application scenario of the wall embedded part, the performance influence coefficient of the embedded part construction mode of the application scenario of the wall embedded part is corrected to obtain the actual value of the performance of the embedded part construction mode of the application scenario of the wall embedded part;

[0019] Based on the actual value of the performance of the embedded part construction mode of the wall embedded part application scenario, the optional embedded part construction mode data array of the wall embedded part application scenario is further screened according to the performance requirements of the wall embedded part application scenario, and the selected embedded part construction mode data array of the wall embedded part application scenario is obtained.

[0020] The construction cost and construction time of each selected embedded part construction mode in the selected embedded part construction mode data array of the wall embedded part application scenario are estimated.

[0021] Based on linear programming, the unique construction mode of the wall embedded part application scenario in the selected embedded part construction mode data array of the wall embedded part application scenario is selected as the objective function, and the minimum construction cost and construction time of each selected embedded part construction mode are used as the constraint condition to construct a construction mode decision model of the wall embedded part, and an optimal reference vector of the wall embedded part is generated.

[0022] Preferably, based on the real-time wall embedded part construction process point cloud data and IMU data, the data is divided according to the time sampling rate of the data, and a real-time wall embedded part construction process point cloud sequence and an IMU pose sequence are established.

[0023] Based on the real-time wall embedded part construction process point cloud sequence and the IMU pose sequence, the least square method is used to calculate the Euclidean distance minimum value between the point cloud continuous frame ICP registration and the IMU motion estimation, and the difference optimization time offset between the real-time wall embedded part construction process point cloud sequence and the IMU pose sequence is obtained.

[0024] Based on linear interpolation, the data of the real-time wall embedded part construction process IMU pose sequence is completed, the time error of the IMU pose sequence is compensated by using the difference optimization time offset, and a point cloud-time synchronous IMU pose sequence is obtained.

[0025] The motion distortion of the real-time wall embedded part construction process point cloud sequence is corrected by using the point cloud-time synchronous IMU pose sequence, and a motion distortion compensation point cloud sequence is obtained.

[0026] Preferably, based on SIFT3D scale invariant feature transformation, the motion distortion compensation point cloud sequence and the calibration target feature points are matched, the matched feature point pairs of the real-time wall embedded part construction process are extracted, and the singular value decomposition algorithm is used to solve the rigid transformation body of the matched feature points, and the initial transformation matrix of the real-time wall embedded part construction process is obtained.

[0027] The world coordinate system transformation matrix of the synchronous point cloud-time synchronous IMU pose sequence and the motion distortion compensation point cloud sequence is established, the initial transformation matrix of the real-time wall embedded part construction process is dynamically updated, and the recursive real-time transformation matrix of the real-time wall embedded part construction process is obtained.

[0028] Preferably, based on the global point cloud data in the recursive real-time transformation matrix of the construction process of the real-time wall embedded part, the average distance and the standard deviation of a plurality of neighboring points of each point cloud are calculated, and the outlier point cloud is removed;

[0029] By using DBSCAN density clustering, the median of the Euclidean distance between each pair of points in the global point cloud data in the recursive real-time transformation matrix of the construction process of the real-time wall embedded part is taken as the point cloud division distance threshold, and the global point cloud data is divided to obtain the embedded part point cloud cluster of the construction task of the real-time wall embedded part.

[0030] Preferably, based on the construction video stream of the real-time wall embedded part construction task, the RGB-D stacked data of the construction video stream is extracted after image distortion and time sequence alignment preprocessing;

[0031] Based on the TCN time convolution network, the RGB-D stacked data is continuously input according to the sampling frame rate of the construction video stream of the real-time wall embedded part construction task, and the frame rate action category probability of the construction video stream of the real-time wall embedded part construction task is generated;

[0032] Based on the frame rate action category probability of the construction video stream of the real-time wall embedded part construction task, the average optical flow change of each frame rate action category probability of the construction video stream is calculated, the local peak frame with an average optical flow change greater than the global sampling frame and a standard deviation is screened, and the frame rate action category probability key frame of the construction video stream of the real-time wall embedded part construction task is obtained.

[0033] Based on the I3D dilated double-stream convolution network, the construction video stream frame rate action category probability key frame and the interval frame of the real-time wall embedded part construction task are taken as input, the appearance feature and the optical flow direction feature of the construction video stream frame are extracted according to the RGB stream architecture layer and the optical flow architecture layer, the global average pooling is performed by using the classification head layer, and the frame rate action classification probability of the construction video stream of the real-time wall embedded part construction task is output.

[0034] Preferably, based on the PointFusion architecture, the image encoding branch is established according to the ResNet-50, the point cloud encoding branch is established according to the PointNet, the construction video stream frame rate action classification probability of the real-time wall embedded part construction task and the embedded part point cloud cluster of the real-time wall embedded part construction task are taken as input corresponding encoding, the construction action context feature and the construction embedded part geometric feature of the real-time wall embedded part construction task are extracted;

[0035] According to the Transformer cross-modal attention, the construction action context feature of the real-time wall embedded part construction task is taken as the key, and the construction embedded part geometric feature is taken as the value for feature fusion, to generate the embedded part pose deviation vector and the embedded part construction process compliance deviation probability of the real-time wall embedded part construction task;

[0036] The embedded part pose deviation vector of the real-time wall embedded part construction task is used to calculate the confidence of the embedded part pose deviation vector of the real-time wall embedded part construction task by taking the embedded part construction process deviation probability of the real-time wall embedded part construction task as an adjustment factor, and a comprehensive deviation vector of the wall embedded part is generated.

[0037] Compared with the prior art, the beneficial effects of the present application are that:

[0038] The present application proposes a wall embedded part positioning quality evaluation scheme based on laser ranging, constructs a high-precision embedded part point cloud cluster through laser ranging point cloud and IMU pose data, combines image recognition construction action time sequence analysis (TCN+I3D), combines multi-modal fusion model (PointFusion+Transformer) to associate geometric deviation and process compliance, and finally determines the wall embedded part construction quality based on the optimal reference vector. The beneficial effects are: improving the embedded part positioning measurement accuracy; cross-modal analysis of construction action and geometric deviation realizes process-pose joint optimization, reduces installation error and reduces rework rate. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a flow chart of a wall embedded part positioning quality evaluation method based on laser ranging. DETAILED DESCRIPTION

[0040] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and those skilled in the art can think of other obvious modifications.

[0041] REFERENCE Figure 1 As shown in the figure, a wall embedded part positioning quality evaluation method based on laser ranging, comprising:

[0042] Step one, based on the wall embedded part construction drawing, marking the construction method of the wall embedded part, analyzing the construction preference of the construction method, and evaluating the optimal reference vector of the wall embedded part;

[0043] The step one includes the following contents:

[0044] Step 101, according to the wall embedded part construction drawing, extracting the geometric information and attribute information in the wall embedded part construction drawing, obtaining the application scene of the wall embedded part;

[0045] According to the expected service life of the application scene of the wall embedded part, determine the performance requirement of the application scene of the wall embedded part, select the embedded part construction method that meets the performance requirement, and establish the optional embedded part construction method data array of the application scene of the wall embedded part;

[0046] Marking the multi-source factor parameters that affect the performance decline of the embedded part construction method in the application scene of the wall embedded part;

[0047] According to the optional embedded construction mode data array of the wall embedded part application scene and the multi-source factor parameters affecting the performance decline of the embedded construction mode in the wall embedded part application scene, an influence factor matrix of the optional embedded construction mode of the wall embedded part application scene is established.

[0048]

[0049] wherein, a ij is the jth influence factor of the ith optional embedded construction mode of the wall embedded part application scene, m is the total number of optional embedded construction modes, and n is the total number of influence factors;

[0050] Step 102, normalizing the influence factor matrix of the optional embedded construction mode of the wall embedded part application scene, and assigning the influence factor weight of the optional embedded construction mode of the wall embedded part application scene according to the entropy weight method;

[0051] According to the influence factor weight of the optional embedded construction mode of the wall embedded part application scene and the multi-source factor parameters affecting the performance decline of the embedded construction mode in the wall embedded part application scene, the embedded construction mode performance influence coefficient of the wall embedded part application scene is calculated.

[0052] According to the ideal performance index of the optional embedded construction mode of each of the optional embedded construction mode data array of the wall embedded part application scene, the embedded construction mode performance influence coefficient of the wall embedded part application scene is used for correction to obtain the actual value of the embedded construction mode performance of the wall embedded part application scene.

[0053] Based on the actual value of the embedded construction mode performance of the wall embedded part application scene, the optional embedded construction mode data array of the wall embedded part application scene is subjected to secondary screening according to the performance requirement of the wall embedded part application scene to obtain the selected embedded construction mode data array of the wall embedded part application scene, in the following manner:

[0054]

[0055] wherein, Array is the selected embedded construction mode data array of the wall embedded part application scene, w ij is the jth influence factor weight of the ith optional embedded construction mode of the wall embedded part application scene, a j is the jth influence factor parameter value affecting the performance decline of the embedded construction mode in the wall embedded part application scene, is the normalized performance value of the jth influence factor of the ith optional embedded construction mode of the wall embedded part application scene, is the actual value of the ith optional embedded construction mode performance of the wall embedded part application scene, and T is the performance requirement threshold of the wall embedded part application scene.

[0056] Step 103, estimate the construction cost and construction time of each candidate construction mode of the wall embedded part application scene candidate embedded part construction mode data array;

[0057] Based on linear programming, the unique construction mode of the wall embedded part application scene in the wall embedded part application scene candidate embedded part construction mode data array is selected as the objective function, and the minimum construction cost and construction time of each candidate embedded part construction mode is used as the constraint condition to construct the construction mode decision model of the wall embedded part, and the optimal reference vector of the wall embedded part is generated;

[0058] In use, in combination with the contents in steps 101 to 103,

[0059] As further content: since the existing wall embedded part construction mode mainly depends on the personal experience of design engineers, it is easy to ignore the use scene of the wall embedded part, so that the design service life of the wall embedded part cannot meet the expected standard, making the later maintenance difficult and reducing the overall construction quality.

[0060] Therefore, by extracting the embedded part application scene and screening the initial optional construction mode, the influence of multiple source factors (such as corrosion and vibration) on the performance of the construction mode is quantified by using the entropy weight method, and the actual performance value is obtained by correcting the ideal performance index of the construction mode; Then, based on linear programming, the unique construction mode is optimized and selected under the constraints of cost and time, and finally the optimal reference vector is generated. The beneficial effects are: 1. By dynamically correcting the performance of the construction mode, the environmental adaptability of the construction mode is improved; 2. Linear programming optimization reduces the comprehensive implementation cost and shortens the construction period; 3. The reference vector is standardized output, which reduces the deviation of manual judgment.

[0061] Step two, based on the real-time wall embedded part construction task, a plurality of laser ranging sensors and image sensors are deployed around the real-time wall embedded part construction task, and the real-time wall embedded part construction task parameters are collected to form the real-time wall embedded part construction process time sequence data; The construction process time sequence data includes: point cloud data, IMU data, calibration target feature points, image data;

[0062] Step three, based on the construction process point cloud data and IMU data of the real-time wall embedded part, the pose transformation matrix of the laser ranging sensor of the real-time wall embedded part construction task is established, and the clustering algorithm is used to segment the wall embedded part point cloud data to obtain the embedded part point cloud cluster of the real-time wall embedded part construction task;

[0063] The step three includes the following contents:

[0064] Step 301, based on the real-time wall embedded construction process point cloud data and IMU data, the data is divided according to the time sampling rate of the data, and the real-time wall embedded construction process point cloud sequence and IMU pose sequence are established;

[0065] Based on the real-time wall embedded construction process point cloud sequence and IMU pose sequence, the least square method is used to calculate the Euclidean distance minimum value between the point cloud continuous frame ICP registration and the IMU motion estimation, and the difference optimization time offset between the real-time wall embedded construction process point cloud sequence and the IMU pose sequence is obtained;

[0066] Based on linear interpolation, the data of the real-time wall embedded construction process IMU pose sequence is completed, the time error of the IMU pose sequence is compensated by using the difference optimization time offset, and the point cloud-time synchronous IMU pose sequence is obtained;

[0067] The point cloud-time synchronous IMU pose sequence is used to correct the motion distortion of the real-time wall embedded construction process point cloud sequence, and the motion distortion compensation point cloud sequence is obtained;

[0068] Step 302, based on SIFT3D scale invariant feature transformation, the motion distortion compensation point cloud sequence and the calibration target feature points are matched, the matched feature point pairs of the real-time wall embedded construction process are extracted, and the singular value decomposition algorithm is used to solve the rigid transformation body of the matched feature points, and the initial transformation matrix of the real-time wall embedded construction process is obtained;

[0069] The world coordinate system transformation matrix of the synchronous point cloud-time synchronous IMU pose sequence and the motion distortion compensation point cloud sequence is established, and the initial transformation matrix of the real-time wall embedded construction process is dynamically updated, and the recursive real-time transformation matrix of the real-time wall embedded construction process is obtained;

[0070] Step 303, based on the global point cloud data in the recursive real-time transformation matrix of the real-time wall embedded construction process, the average distance and standard deviation of a plurality of neighbors of each point cloud are calculated, and the outlier point cloud is removed;

[0071] Using DBSCAN density clustering, the median of the Euclidean distance between each pair of points in the global point cloud data in the recursive real-time transformation matrix of the real-time wall embedded construction process is used as the point cloud division distance threshold, and the global point cloud data is divided to obtain the embedded point cloud cluster of the real-time wall embedded construction task;

[0072] In use, the contents in steps 301 to 303 are combined,

[0073] As a further content, due to the existing wall embedded part positioning static single-point measurement mode is difficult to capture the construction dynamic deformation (such as ±5mm displacement deviation caused by concrete pouring), secondly, there is a time synchronization error in multi-sensor data fusion, which causes the final embedded part installation position error to often exceed the engineering allowable range, seriously affecting the construction precision and acceptance qualification rate.

[0074] The scheme realizes accurate alignment of point cloud-IMU data through time synchronization and motion distortion compensation, solves the pose transformation matrix by using SIFT3D feature matching and singular value decomposition, and further combines DBSCAN clustering to segment the embedded part point cloud cluster. The beneficial effect is that the point cloud distortion caused by construction vibration is effectively overcome through dynamic pose correction, and reliable data is provided for automatic construction quality control.

[0075] Step four, based on the real-time wall embedded part construction process image data, the construction action in the construction process image data is marked, the construction video stream of the real-time wall embedded part construction task and the positioning point cloud cluster of the real-time wall embedded part construction task are obtained, the wall embedded part positioning multi-modal evaluation model is established, and the deviation vector of the wall embedded part is generated;

[0076] Step 401, based on the real-time wall embedded part construction task construction video stream, the image distortion and time sequence are preprocessed, and the RGB-D stacked data of the construction video stream is extracted;

[0077] Based on the TCN time convolution network, the RGB-D stacked data is continuously input according to the sampling frame rate of the real-time wall embedded part construction task construction video stream, and the construction video stream frame rate action category probability of the real-time wall embedded part construction task is generated;

[0078] As a further content, the internal structure of the TCN time convolution network includes: a convolution layer that maintains the time causal relationship, a dilated convolution layer, and a residual connection layer that avoids gradient disappearance, and in order to avoid the imbalance of the final action category classification, the loss function needs to introduce weighted cross entropy;

[0079] Based on the construction video stream frame rate action category probability of the real-time wall embedded part construction task, the inter-frame average optical flow change in each construction video stream frame rate action category probability is calculated, the local peak frame greater than the average optical flow change and the standard deviation of the global sampling frame is screened, and the construction video stream frame rate action category probability key frame of the real-time wall embedded part construction task is obtained.

[0080] Based on the I3D inflation double-flow convolution network, the construction video stream frame rate action category probability key frame and the interval frame of the real-time wall embedded part construction task are input, the appearance feature and the optical flow direction feature of the construction video stream frame are extracted according to the RGB flow architecture layer and the optical flow architecture layer, the global average pooling is performed by using the classification head layer, and the construction video stream frame rate action classification probability of the real-time wall embedded part construction task is output.

[0081] Step 402, based on the PointFusion architecture, the image encoding branch is established according to ResNet-50, and the point cloud encoding branch is established by PointNet, to take the construction video stream frame rate action classification probability of the real-time wall embedded construction task and the embedded point cloud cluster of the real-time wall embedded construction task as input to correspond to the encoding, and extract the construction action context features and construction embedded geometric features of the real-time wall embedded construction task;

[0082] According to the Transformer cross-modal attention, the construction action context features of the real-time wall embedded construction task are taken as the key, and the construction embedded geometric features are taken as the value for feature fusion, to generate the embedded pose deviation vector and the embedded construction process compliance deviation probability of the real-time wall embedded construction task;

[0083] Based on the embedded pose deviation vector of the real-time wall embedded construction task, the embedded construction process deviation probability of the real-time wall embedded construction task is taken as an adjustment factor to calculate the confidence of the embedded pose deviation vector of the real-time wall embedded construction task, to generate a comprehensive deviation vector of the wall embedded, in the following way:

[0084]

[0085] Where, V enhanced is the comprehensive deviation vector of the wall embedded, Δx, Δy, Δz, Δθ are the embedded pose deviation components of the real-time wall embedded construction task, φ, β are adjustment factors, p is the embedded construction process compliance deviation probability of the real-time wall embedded construction task, is the transpose.

[0086] In use, the contents in steps 401 to 402 are combined,

[0087] As further content: by fusing construction video stream (TCN time sequence segmentation + I3D action classification) and point cloud data (PointNet geometric features), combining the Transformer cross-modal attention mechanism, dynamically associating construction action semantics and embedded geometric pose, to generate a comprehensive deviation vector. The beneficial effect is that the time convolution network TCN and the dilated I3D convolution are used to extract the time sequence action features, the PointFusion architecture is used to realize the alignment of image-point cloud features, and then the construction action context is taken as the key and the geometric features are taken as the value for attention weighting, and finally the process compliance probability is used to correct the pose deviation. Realize the multi-modal real-time monitoring of the construction process, and capture the action compliance and pose accuracy; through optical flow key frame screening and attention fusion, the calculation redundancy is significantly reduced, and the embedded installation quality and efficiency are improved.

[0088] Step six, judging whether the comprehensive deviation vector of the wall embedded part is in the optimal reference vector interval of the wall embedded part, if yes, outputting continue to execute, if not, outputting the deviation is too large.

[0089] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection required by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the positioning quality of embedded wall components based on laser ranging, characterized in that, include: S1. Based on the construction drawings of wall embedded parts, mark the construction methods of wall embedded parts, analyze the construction preferences of construction methods, and evaluate the optimal reference vector of wall embedded parts. S2. Deploy multiple laser ranging sensors and image sensors around the real-time wall embedded part construction task to collect the construction task parameters of the real-time wall embedded part and construct the construction process time series data of the real-time wall embedded part; the construction process time series data includes: point cloud data, IMU data, calibration target feature points, and image data. S3. Based on the point cloud data of the construction process of wall embedded parts and IMU data, establish the pose transformation matrix of the laser ranging sensor for the construction task of wall embedded parts in real time, and use the clustering algorithm to segment the point cloud data of wall embedded parts to obtain the point cloud cluster of embedded parts for the construction task of wall embedded parts in real time. S4. Based on the construction process image data of real-time wall embedded parts, mark the construction actions in the construction process image data, obtain the construction video stream of the real-time wall embedded part construction task and the positioning point cloud cluster of the real-time wall embedded part construction task, establish a multi-modal evaluation model for wall embedded part positioning, and generate the deviation vector of the wall embedded part. S5. Determine whether the comprehensive deviation vector of the wall embedded part is within the optimal reference vector range of the wall embedded part. If yes, output to continue execution; otherwise, output that the deviation is too large.

2. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 1, characterized in that, S1 includes: Based on the construction drawings of wall embedded parts, extract the geometric and attribute information from the construction drawings of wall embedded parts to obtain the application scenarios of wall embedded parts; Based on the expected service life of the application scenarios of wall embedded parts, determine the performance requirements of the application scenarios of wall embedded parts, screen the embedded parts construction methods that meet the performance requirements, and establish a data array of optional embedded parts construction methods for the application scenarios of wall embedded parts. Multi-source factors affecting the performance degradation of embedded parts in the application scenarios of wall embedded parts; Based on the data array of optional embedded part construction methods for wall embedded parts application scenarios and the multi-source factor parameters that affect the performance degradation of embedded part construction methods in wall embedded parts application scenarios, an influencing factor matrix of optional embedded part construction methods for wall embedded parts application scenarios is established.

3. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 2, characterized in that, S1 further includes: The matrix of factors influencing the optional construction methods of wall embedded parts in the application scenarios is normalized, and the weights of the factors influencing the optional construction methods of wall embedded parts in the application scenarios are assigned according to the entropy weight method. Based on the weights of the factors influencing the optional embedding construction methods in the application scenarios of wall embedded parts and the parameters of the multi-source factors that affect the performance degradation of the embedding construction methods in the application scenarios of wall embedded parts, the performance influence coefficient of the embedding construction methods in the application scenarios of wall embedded parts is calculated. Based on the ideal performance index of each optional embedded construction method in the data array of optional embedded construction methods for the application scenario of wall embedded parts, the performance influence coefficient of the embedded construction method for the application scenario of wall embedded parts is used for correction to obtain the actual performance value of the embedded construction method for the application scenario of wall embedded parts. Based on the actual performance values ​​of the embedded parts construction methods in the application scenarios of wall embedded parts, and according to the performance requirements of the application scenarios of wall embedded parts, a second filtering is performed on the data array of optional embedded parts construction methods for the application scenarios of wall embedded parts to obtain the data array of candidate embedded parts construction methods for the application scenarios of wall embedded parts. The construction cost and construction time of each candidate embedded part construction method in the data array of candidate embedded part construction methods for the application scenarios of wall embedded parts are estimated. Based on linear programming, the objective function is the unique construction method of the wall embedded parts in the candidate embedded parts construction method data array that selects the application scenarios of wall embedded parts. The constraint is the minimization of construction cost and construction time for each candidate embedded parts construction method. A construction method decision model for wall embedded parts is constructed to generate the optimal reference vector for wall embedded parts.

4. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 1, characterized in that, S3 includes: Based on the real-time construction progress point cloud data and IMU data of wall embedded parts, the data is divided according to the time sampling rate of the data, and the real-time construction progress point cloud sequence and IMU pose sequence of wall embedded parts are established. Based on the construction process point cloud sequence and IMU pose sequence of real-time wall embedded parts, the minimum Euclidean distance between the continuous frame ICP registration of the point cloud and the IMU motion estimation is calculated using the least squares method, so as to obtain the difference optimization time offset between the construction process point cloud sequence and the IMU pose sequence of real-time wall embedded parts. Based on linear interpolation, the IMU pose sequence of the construction process of the wall embedded parts is completed. The time error of the IMU pose sequence is compensated by the difference optimization time offset to obtain the point cloud-time synchronized IMU pose sequence. The motion distortion of the point cloud sequence of the construction process of the wall embedded parts is corrected by using the point cloud-time synchronized IMU pose sequence, and the motion distortion compensated point cloud sequence is obtained.

5. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 4, characterized in that, S3 further includes: Based on SIFT3D scale-invariant feature transformation, the motion distortion compensation point cloud sequence and calibration target feature points are matched to extract the matching feature point pairs of the construction process of the real-time wall embedded parts. The rigid transformation volume of the matching feature points is solved by the singular value decomposition algorithm to obtain the initial transformation matrix of the construction process of the real-time wall embedded parts. A world coordinate system transformation matrix is ​​established between the synchronous point cloud-time synchronous IMU pose sequence and the motion distortion compensation point cloud sequence. The initial transformation matrix of the construction process of the wall embedded parts is dynamically updated to obtain the recursive real-time transformation matrix of the construction process of the wall embedded parts. Preferably, based on the global point cloud data in the recursive real-time transformation matrix of the construction process of the wall embedded parts, the average distance and standard deviation of several nearest neighbors of each point cloud are calculated, and outlier point clouds are removed. Using DBSCAN density clustering, the median Euclidean distance between each pair of points in the global point cloud data in the real-time transformation matrix of the real-time wall embedded parts construction progress is used as the point cloud segmentation distance threshold. The global point cloud data is then segmented to obtain the embedded part point cloud clusters for the real-time wall embedded parts construction task.

6. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 5, characterized in that, S4 includes: Based on the construction video stream of the real-time wall embedded parts construction task, image distortion and temporal alignment preprocessing are performed to extract the RGB-D stacked data of the construction video stream. Based on the TCN temporal convolutional network, RGB-D stacked data are continuously input according to the sampling frame rate of the construction video stream of the real-time wall embedded part construction task to generate the frame rate action category probability of the construction video stream of the real-time wall embedded part construction task. Based on the frame rate action category probability of the construction video stream for real-time wall embedded part construction task, calculate the inter-frame average optical flow change in the frame rate action category probability of each construction video stream, and filter out local peak frames that are greater than the average optical flow change and standard deviation of the global sampling frames to obtain the key frames of the frame rate action category probability of the construction video stream for real-time wall embedded part construction task. Based on the I3D dilated dual-stream convolutional network, the keyframes and interval frames of the construction video stream of the real-time wall embedded part construction task are used as input. The appearance features and optical flow direction features of the construction video stream frames are extracted according to the RGB stream architecture layer and the optical flow architecture layer. Global average pooling is performed using the classification head layer to output the frame rate and action classification probability of the construction video stream of the real-time wall embedded part construction task.

7. The method for evaluating the positioning quality of embedded wall components based on laser ranging according to claim 6, characterized in that, S4 further includes: Based on the PointFusion architecture, an image coding branch is established using ResNet-50 and a point cloud coding branch is established using PointNet. The frame rate and action classification probability of the construction video stream of the real-time wall embedded part construction task and the embedded part point cloud cluster of the real-time wall embedded part construction task are used as inputs to encode the corresponding codes, and the construction action context features and construction embedded part geometric features of the real-time wall embedded part construction task are extracted. Based on Transformer cross-modal attention, the construction action context features of the real-time wall embedded part construction task are used as the key, and the geometric features of the construction embedded part are used as the value for feature fusion to generate the embedded part pose deviation vector and the embedded part construction process compliance deviation probability of the real-time wall embedded part construction task. Based on the embedding position deviation vector of the real-time wall embedding construction task, the embedding construction process deviation probability of the real-time wall embedding construction task is used as an adjustment factor to calculate the confidence level of the embedding position deviation vector of the real-time wall embedding construction task, and generate the comprehensive deviation vector of the wall embedding.

Images