Three-dimensional measurement and simulation analysis system and method based on multi-sensor fusion

Abstract

The invention relates to the technical field of three-dimensional model construction, in particular to a three-dimensional measurement and simulation analysis system and method based on multi-sensor fusion, and the method comprises the steps: obtaining the point cloud data of a target area through I sensors; performing denoising processing on the point cloud data by using a filtering algorithm to remove sensor noise and abnormal values; point cloud data collected by different sensors are converted into a unified coordinate system through a feature matching algorithm. Different weights are given to the point cloud data collected by each sensor according to the performance of the sensors correspondingly collected by the point cloud data and the corresponding environmental factors, then the data are fused, the different weights are given to the data of each sensor, weighted fusion is carried out, and the data fusion speed is improved. According to the method, the accuracy and the reliability of data fusion can be remarkably improved, the robustness and the adaptability of the system are enhanced, and when a sensor breaks down or outputs an abnormal value, the influence of error data on a final result can be reduced by adopting weighted fusion.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional model construction, and in particular to a three-dimensional measurement and simulation analysis system and method based on multi-sensor fusion. Background Art

[0002] The multi-sensor fusion 3D measurement system aims to integrate multiple sensors and achieve more accurate 3D measurement and simulation analysis by fusing different sensor data. This technology can be applied to many fields such as automated manufacturing, robot navigation, and intelligent transportation. The existing multi-sensor fusion system still faces several technical shortcomings in practice.

[0003] For example, the patent with publication number CN117807782A discloses a method for realizing a three-dimensional simulation model. The specific steps of the implementation method are as follows: (1) collecting and processing real scene images through sensors and cameras; (2) using the processed data to construct a training adversarial network; (3) constructing a three-dimensional simulation model through the adversarial network; (4) storing the constructed model in the form of a blockchain and optimizing the three-dimensional simulation model; (5) real-time monitoring of the operating efficiency of the three-dimensional simulation model and optimizing memory management; (6) displaying the three-dimensional simulation model and performing feedback evaluation; the invention does not require manual intervention or the use of expensive scanning equipment, improves the generation efficiency of the three-dimensional model, can learn the distribution of real data, and at the same time reduce the cost of generating the three-dimensional model, can make the generation of the simulation model more intelligent, and is conducive to generating high-quality and realistic simulation models. It can also explore the potential optimal solution of the model and improve the overall performance of the model.

[0004] For the above applications, multiple sensors are used to collect data (whether image data or point cloud data) and perform three-dimensional construction, but a common problem is that the response time of the sensors may be different, which will lead to inconsistency in the time dimension of the sensor data. This time mismatch may affect the accuracy and reliability of the final construction result. If there is no precise data synchronization mechanism, the three-dimensional model after data fusion may produce errors. Secondly, different sensors have different performance and reliability under different environmental conditions. Simply fusing the data may ignore the impact of sensor performance and environmental factors, resulting in inaccurate and unreliable results. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a three-dimensional measurement and simulation analysis system and method based on multi-sensor fusion.

[0006] The present invention adopts the following technical solutions: Three-dimensional measurement and simulation analysis methods based on multi-sensor fusion, including: Use I sensors to obtain point cloud data of the target area, I ≥ 3; Use filtering algorithms to denoise point cloud data and remove sensor noise and outliers; The point cloud data collected by different sensors are converted into a unified coordinate system through feature matching algorithms; Collect in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; Based on historical correction training data, a machine learning model is trained to predict correction point cloud data, real-time comprehensive collection data is collected, and correction point cloud data is predicted based on the trained machine learning model; According to the performance of the sensors corresponding to the point cloud data and the corresponding environmental factors, different weights are assigned to the point cloud data collected by each sensor, the data are fused, and the voxel method is used to construct the fused data in three dimensions.

[0007] As a further description of the above technical solution: the method for converting the point cloud data collected by different sensors into a unified coordinate system is: Step S01: downsampling: downsampling the point cloud data using a Voxel Grid filter, and marking the sampled point cloud data as feature data to reduce the computational burden; Step S02: feature extraction, using normal estimation, FPFH or SHOT method to extract point cloud features of feature data; Step S03: feature matching, using a k-nearest neighbor-based matching algorithm to find feature corresponding points between different data sets; Step S04: coordinate transformation. Calculate the coordinate transformation matrix through the matched feature points, apply the calculated transformation matrix T to the sensor data, and transform it into a unified coordinate system. For each point p=(x, y, z), apply the change matrix T to obtain the transformed coordinates. , , update the coordinates of all points so that they are in a unified coordinate system.

[0008] As a further description of the above technical solution: the step S03 further includes using the RANSAC algorithm to eliminate wrong matches; Among them, the method of using the RANSAC algorithm to eliminate wrong matches includes: Select the homography matrix as the model type to be fitted, randomly select some point pairs from the matched feature points, and use the selected point pairs to calculate the model parameters; For all feature matching points, the calculated model parameters are used to evaluate the degree of conformity of each point pair. If the degree of conformity does not meet the standard, the matching feature points are removed.

[0009] As a further description of the above technical solution: the method of using the calculated model parameters to evaluate the degree of conformity of each point pair, and if the degree of conformity does not meet the standard, then removing the matching feature points includes: For each feature matching point pair, the target point is predicted according to the model and the reprojection error is calculated: ; In the formula, is the target point position calculated by the model, is the target point position, is the reprojection error; Preset the reprojection error threshold Emax, when ≥Emax, the feature matching point is eliminated, otherwise it is retained.

[0010] As a further description of the above technical solution: Based on the historical correction training data, the method for training a machine learning model for predicting the correction point cloud data includes: Convert the collected historical corrected training data into a corresponding set of feature vectors; Each group of feature vectors is used as the input of the machine learning model, and the machine learning model takes the corrected point cloud data corresponding to each group of comprehensive acquisition data as output, and takes the corrected point cloud data actually corresponding to each group of comprehensive acquisition data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training goal; training is stopped when the loss function value of the machine learning model is less than or equal to the preset target loss value.

[0011] As a further description of the above technical solution: the method of assigning different weights to the point cloud data collected by each sensor includes: Obtain sensor performance metrics, including resolution, update frequency, and signal-to-noise ratio; Get indicators of environmental factors, including light intensity and shade area; Use the formula to calculate the weight of each sensor. The calculation formula is: ; In the formula, is the total weight of the i-th sensor, i=1, 2, ..., I, , , , and is the weight coefficient of the corresponding indicator, is the i-th sensor resolution, is the update frequency of the ith sensor, is the signal-to-noise ratio of the ith sensor, is the light intensity and is the shaded area.

[0012] As a further description of the above technical solution: the method for fusing data includes: According to the output data of each sensor and its weight, the weighted average is calculated for fusion. The calculation formula is: ; In the formula, is the output data of the i-th sensor, is the total weight of the ith sensor, represents the fused data, is the sum of the weights of all sensors.

[0013] The 3D measurement and simulation analysis system based on multi-sensor fusion includes: The data acquisition module uses I sensors to obtain point cloud data of the target area, where I≥3; The data processing module uses a filtering algorithm to denoise the point cloud data and remove sensor noise and outliers; The data conversion module converts the point cloud data collected by different sensors into a unified coordinate system through feature matching algorithms; A data collection module collects in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; The data correction module trains a machine learning model for predicting corrected point cloud data based on historical corrected training data, collects real-time comprehensive collection data, and predicts corrected point cloud data based on the trained machine learning model; The three-dimensional construction module assigns different weights to the point cloud data collected by each sensor according to the performance of the sensor corresponding to the point cloud data and the corresponding environmental factors, fuses the data, and uses the voxel method to perform three-dimensional construction on the fused data.

[0014] An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the three-dimensional measurement and simulation analysis method based on multi-sensor fusion by calling the computer program stored in the memory.

[0015] A computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the three-dimensional measurement and simulation analysis method based on multi-sensor fusion.

[0016] Beneficial effects: The three-dimensional measurement and simulation analysis method based on multi-sensor fusion provided by the present invention collects historical correction training data required for predicting and correcting point cloud data, then trains a machine learning model for predicting and correcting point cloud data, collects real-time comprehensive collection data, and predicts the corrected point cloud data based on the trained machine learning model, that is, corrects the point cloud data collected by each sensor at different timestamps, automatically corrects the time error of the same sensor data, and reduces the synchronization problem caused by time delay. Secondly, according to the performance of the sensor corresponding to the point cloud data and the corresponding environmental factors, different weights are assigned to the point cloud data collected by each sensor, and then the data is fused. By assigning different weights to the data of each sensor and performing weighted fusion, the accuracy and reliability of data fusion can be significantly improved, and the robustness and adaptability of the system can be enhanced. That is, when the sensor fails or outputs abnormal values, weighted fusion can be used to reduce the impact of these erroneous data on the final result. In weighted fusion, the noise level in the overall data can be reduced by giving higher weights to sensors with less noise.

[0017] Furthermore, by aligning the data from different sensors to the same coordinate system, the spatial consistency of the data is ensured, errors and confusion caused by different coordinate systems are avoided, and the accuracy of the three-dimensional model is enhanced. The aligned data can be more easily fused, reflecting the advantages of using different sensors (such as the accuracy of lidar and the color information of RGB cameras) to generate richer three-dimensional models. Before feature extraction, the point cloud data is first downsampled. After downsampling, the amount of point cloud data is reduced, which can significantly speed up subsequent processing (such as feature extraction, matching, and registration). Voxel Grid filters are used for downsampling. Voxel Grid filters can retain the overall shape and geometric features of the point cloud by selecting the center point of each voxel, reduce redundant data, and maintain important information. In summary, the feature matching algorithm and Voxel Grid filters jointly improve the accuracy, efficiency, and usability of three-dimensional model construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further explained below in conjunction with the accompanying drawings and embodiments: Figure 1 The process of the three-dimensional measurement and simulation analysis method based on multi-sensor fusion provided by the embodiment of the present invention; Figure 2A flow chart of a method for converting different point cloud data into a unified coordinate system provided by an embodiment of the present invention; Figure 3 A module connection diagram of a three-dimensional measurement and simulation analysis system based on multi-sensor fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, objectives and effects of the present invention easy to understand, the present invention is further described below with reference to specific diagrams. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0020] Example 1 See also Figure 1 and Figure 2 The embodiment of the present invention provides a technical solution: a three-dimensional measurement and simulation analysis method based on multi-sensor fusion, comprising: Use I sensors to obtain point cloud data of the target area, I ≥ 3; It should be noted that the sensors include lidar sensors, RGB-D cameras, and ultrasonic sensors, which send laser pulses or infrared light through the sensors and measure the return time to calculate the distance to the object and generate point cloud data; Point cloud data is the most commonly used data representation in 3D model construction. Each point is represented by its 3D coordinates (x, y, z) in space and usually contains other information, such as: Color information: RGB value, indicating the color of the point (obtained from an RGB camera or a depth camera); Normal vector: surface normal, indicating the surface direction of a point; Intensity value: In LiDAR, the reflection intensity at each point.

[0021] Use filtering algorithms (such as median filtering and Gaussian filtering) to denoise point cloud data and remove sensor noise and outliers; It should be noted that Gaussian filtering is a linear filtering technique that smoothes data by weighted averaging; Median filtering is a nonlinear filtering technique used to remove noise from data, especially effective for salt-and-pepper noise. The basic steps are as follows: Neighborhood selection: For each pixel point, extract the pixel values ​​(neighborhood) within a certain range around it; Sorting: Sort the pixel values ​​in the neighborhood; Median calculation: Take the middle value after sorting as the new value of the current pixel.

[0022] The point cloud data collected by different sensors are converted into a unified coordinate system through feature matching algorithms; Coordinate transformation: calculate the coordinate transformation matrix through the matched feature points, apply the calculated transformation matrix T to the sensor data, and convert it to a unified coordinate system. For each point p=(x, y, z), apply the change matrix T to obtain the transformed coordinates. , , update the coordinates of all points so that they are in a unified coordinate system.

[0023] It should be noted that the method of calculating the coordinate transformation matrix through the matched feature points includes: Least squares method: Calculate the optimal transformation matrix by minimizing the error between matching points; SVD (Singular Value Decomposition): Use SVD to decompose the covariance matrix of the point cloud data and obtain the optimal rotation matrix.

[0024] Collect in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; It should be noted that resolution refers to the smallest detail that the sensor can distinguish. For example, the depth image resolution of an RGB-D camera affects the ability to capture details. The update frequency refers to the number of frames per second (FPS) that the sensor can collect, which affects real-time processing capabilities. A high update frequency helps track changes in dynamic scenes. The environmental data refers to whether the environment in which the data is collected is a dynamic scene or a static scene.

[0025] Example: Point cloud data with timestamp is represented as , where n is the timestamp, corrected point cloud data ; Based on historical correction training data, a machine learning model is trained to predict correction point cloud data, real-time comprehensive collection data is collected, and correction point cloud data is predicted based on the trained machine learning model; Methods for training a machine learning model for predicting corrected point cloud data based on historical corrected training data include: Convert the collected historical corrected training data into a corresponding set of feature vectors; Each group of feature vectors is used as the input of the machine learning model, and the machine learning model takes the corrected point cloud data corresponding to each group of comprehensive acquisition data as output, and takes the corrected point cloud data actually corresponding to each group of comprehensive acquisition data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training goal; training is stopped when the loss function value of the machine learning model is less than or equal to the preset target loss value.

[0026] The loss function value of the machine learning model is the mean square error, which is one of the commonly used loss functions. Minimization is used as the goal to train the model so that the machine learning model better fits the data, thereby improving the performance and accuracy of the model; In the loss function is the loss function value of the machine learning model, is the feature vector group number; is the number of eigenvector groups; For the The predicted and corrected point cloud data corresponding to the group feature vector, For the The corrected point cloud data actually corresponding to the group feature vector; Other model parameters of the machine learning model, target loss value, optimization algorithm, training set test set validation set ratio, and loss function optimization are all achieved through actual engineering implementation and continuous experimental tuning.

[0027] According to the performance of the sensors corresponding to the point cloud data and the corresponding environmental factors, different weights are assigned to the point cloud data collected by each sensor, the data are fused, and the voxel method is used to construct the fused data in three dimensions.

[0028] Different weights are given to the point cloud data collected by each sensor. The methods for fusing the data include: Obtain sensor performance indicators including resolution, update frequency, and signal-to-noise ratio; Indicators for obtaining environmental factors include light intensity and shadow area; Use the formula to calculate the weight of each sensor. The calculation formula is: ; In the formula, is the total weight of the i-th sensor, i=1, 2, ..., I, , , , and is the weight coefficient of the corresponding indicator, is the i-th sensor resolution, is the update frequency of the ith sensor, is the signal-to-noise ratio of the ith sensor, is the light intensity and is the shaded area; It should be noted that the size of the weight coefficient is a specific value obtained by quantifying each data to facilitate subsequent comparison. The size of the weight coefficient depends on the number of comprehensive parameters and the preliminary setting of the corresponding weight coefficient for each set of comprehensive parameters by technical personnel in this field.

[0029] According to the output data of each sensor and its weight, the weighted average is calculated for fusion. The calculation formula is: ; In the formula, is the output data of the i-th sensor, is the total weight of the ith sensor, represents the fused data, is the sum of the weights of all sensors.

[0030] In this embodiment, by collecting historical correction training data required for predicting and correcting point cloud data, and then training a machine learning model for predicting and correcting point cloud data, real-time comprehensive collection data is collected, and the corrected point cloud data is predicted based on the trained machine learning model, that is, the point cloud data collected by each sensor at different timestamps is corrected, and the time error of the same sensor data is automatically corrected to reduce the synchronization problem caused by time delay. Secondly, according to the performance of the sensor corresponding to the point cloud data and the corresponding environmental factors, different weights are assigned to the point cloud data collected by each sensor, and then the data is fused. By assigning different weights to the data of each sensor and performing weighted fusion, the accuracy and reliability of data fusion can be significantly improved, and the robustness and adaptability of the system can be enhanced. That is, when the sensor fails or outputs abnormal values, weighted fusion can be used to reduce the impact of these erroneous data on the final result. In weighted fusion, the noise level in the overall data can be reduced by giving higher weights to sensors with less noise.

[0031] Example 2 Reference Figure 2 ,The method of converting point cloud data collected by different sensors into a unified coordinate system is: Downsampling: Use Voxel Grid filter to downsample the point cloud data and mark the sampled point cloud data as feature data to reduce the computational burden; Feature extraction, using normal estimation, FPFH (Fast Point Feature Histograms) or SHOT (Signature of Histograms of Orientations) methods to extract point cloud features of feature data; Feature matching uses a k-nearest neighbor-based matching algorithm to find feature correspondence points between different data sets; the RANSAC algorithm (random sampling consensus algorithm) is used to eliminate false matches and improve the reliability of matching results; Methods for using RANSAC to remove false matches include: Select homography matrix as the model type to be fitted, randomly select a small number of point pairs from the matched feature points, and use the selected point pairs to calculate model parameters, such as using the least squares method to calculate the homography matrix; For all feature matching points, the calculated model parameters are used to evaluate the degree of conformity of each point pair. If the degree of conformity does not meet the standard, the matching feature points are removed.

[0032] The specific method of using the calculated model parameters to evaluate the degree of conformity of each point pair includes: For each feature matching point pair, the target point is predicted according to the model and the reprojection error is calculated: ; In the formula, is the target point position calculated by the model, is the target point position, is the reprojection error; Preset the reprojection error threshold Emax, when ≥Emax, the feature matching point is removed, otherwise the feature matching point is retained. In this embodiment, by aligning data from different sensors to the same coordinate system, the spatial consistency of the data is ensured, errors and confusion caused by different coordinate systems are avoided, and the accuracy of the three-dimensional model is enhanced. The aligned data can be more easily fused, reflecting the advantages of using different sensors (such as the accuracy of the lidar and the color information of the RGB camera) to generate a richer three-dimensional model. Before feature extraction, the point cloud data is first downsampled. After downsampling, the amount of point cloud data is reduced, which can significantly speed up subsequent processing (such as feature extraction, matching, and registration). The Voxel Grid filter is used for downsampling. The Voxel Grid filter can retain the overall shape and geometric features of the point cloud by selecting the center point of each voxel, reduce redundant data, and maintain important information. In summary, the feature matching algorithm and the Voxel Grid filter jointly improve the accuracy, efficiency, and usability of three-dimensional model construction.

[0033] Example 3 Reference Figure 3 , a three-dimensional measurement and simulation analysis system based on multi-sensor fusion, including: The data acquisition module uses I sensors to obtain point cloud data of the target area, where I≥3; The data processing module uses a filtering algorithm to denoise the point cloud data and remove sensor noise and outliers; The data conversion module converts the point cloud data collected by different sensors into a unified coordinate system through feature matching algorithms; A data collection module collects in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; The data correction module trains a machine learning model for predicting corrected point cloud data based on historical corrected training data, collects real-time comprehensive collection data, and predicts corrected point cloud data based on the trained machine learning model; The three-dimensional construction module assigns different weights to the point cloud data collected by each sensor according to the performance of the sensor corresponding to the point cloud data and the corresponding environmental factors, fuses the data, and uses the voxel method to perform three-dimensional construction on the fused data.

[0034] Example 4 An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the three-dimensional measurement and simulation analysis method based on multi-sensor fusion by calling the computer program stored in the memory.

[0035] Since the electronic device introduced in this embodiment is an electronic device used to implement the three-dimensional measurement and simulation analysis method based on multi-sensor fusion in the embodiment of this application, based on the three-dimensional measurement and simulation analysis method based on multi-sensor fusion introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the three-dimensional measurement and simulation analysis method based on multi-sensor fusion in the embodiment of this application, it belongs to the scope of protection of this application.

[0036] Example 5 A computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the three-dimensional measurement and simulation analysis method based on multi-sensor fusion.

[0037] The basic principles, main features and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and the description in the specification are only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A three-dimensional measurement and simulation analysis method based on multi-sensor fusion, characterized in that: include: Use I sensors to obtain point cloud data of the target area, I ≥ 3; Use filtering algorithms to denoise point cloud data and remove sensor noise and outliers; The point cloud data collected by different sensors are converted into a unified coordinate system through feature matching algorithms; Collect in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; Based on historical correction training data, a machine learning model is trained to predict correction point cloud data, real-time comprehensive collection data is collected, and correction point cloud data is predicted based on the trained machine learning model; According to the performance of the sensors corresponding to the point cloud data and the corresponding environmental factors, different weights are assigned to the point cloud data collected by each sensor, the data are fused, and the voxel method is used to construct the fused data in three dimensions.

2. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 1 is characterized in that: The method for converting point cloud data collected by different sensors into a unified coordinate system is: Step S01: downsampling: downsampling the point cloud data using a Voxel Grid filter, and marking the sampled point cloud data as feature data; Step S02: feature extraction, using normal estimation, FPFH or SHOT method to extract point cloud features of feature data; Step S03: feature matching, using a k-nearest neighbor-based matching algorithm to find feature corresponding points between different data sets; Step S04: coordinate transformation. Calculate the coordinate transformation matrix through the matched feature points, apply the calculated transformation matrix T to the sensor data, and transform it into a unified coordinate system. For each point p=(x, y, z), apply the change matrix T to obtain the transformed coordinates. , , update the coordinates of all points.

3. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 2 is characterized in that: The step S03 also includes using a RANSAC algorithm to eliminate false matches; Among them, the method of using the RANSAC algorithm to eliminate wrong matches includes: Select the homography matrix as the model type to be fitted, randomly select some point pairs from the matched feature points, and use the selected point pairs to calculate the model parameters; For all feature matching points, the calculated model parameters are used to evaluate the degree of conformity of each point pair. If the degree of conformity does not meet the standard, the matching feature points are removed.

4. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 3 is characterized in that: The method of using the calculated model parameters to evaluate the degree of conformity of each point pair and removing the matching feature points if the degree of conformity does not meet the standard includes: For each feature matching point pair, the target point is predicted according to the model and the reprojection error is calculated: ； In the formula, is the target point position calculated by the model, is the target point position, is the reprojection error; Preset the reprojection error threshold Emax, when ≥Emax, the feature matching point is eliminated, otherwise it is retained.

5. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 1 is characterized in that: Methods for training a machine learning model for predicting corrected point cloud data based on historical corrected training data include: Convert the collected historical corrected training data into a corresponding set of feature vectors; Each group of feature vectors is used as the input of the machine learning model, and the machine learning model takes the corrected point cloud data corresponding to each group of comprehensive acquisition data as output, and takes the corrected point cloud data actually corresponding to each group of comprehensive acquisition data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training goal; training is stopped when the loss function value of the machine learning model is less than or equal to the preset target loss value.

6. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 1 is characterized in that: The method of assigning different weights to the point cloud data collected by each sensor includes: Obtain sensor performance metrics, including resolution, update frequency, and signal-to-noise ratio; Get indicators of environmental factors, including light intensity and shade area; Use the formula to calculate the weight of each sensor. The calculation formula is: ； In the formula, is the total weight of the i-th sensor, i=1, 2, ..., I, , , , and is the weight coefficient of the corresponding indicator, is the i-th sensor resolution, is the update frequency of the ith sensor, is the signal-to-noise ratio of the ith sensor, is the light intensity and is the shaded area.

7. The three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to claim 6 is characterized in that: The method for fusing data includes: According to the output data of each sensor and its weight, the weighted average is calculated for fusion. The calculation formula is: ； In the formula, is the output data of the i-th sensor, is the total weight of the ith sensor, represents the fused data, is the sum of the weights of all sensors.

8. The three-dimensional measurement and simulation analysis system based on multi-sensor fusion is characterized by: include: The data acquisition module uses I sensors to obtain point cloud data of the target area, where I≥3; The data processing module uses a filtering algorithm to denoise the point cloud data and remove sensor noise and outliers; The data conversion module converts the point cloud data collected by different sensors into a unified coordinate system through feature matching algorithms; A data collection module collects in advance historical correction training data required for predicting and correcting point cloud data, wherein the historical correction training data is collected when the three-dimensional model is constructed by point cloud data to meet the standards, and the historical correction training data includes comprehensive collection data and correction point cloud data, wherein the comprehensive collection data includes sensor type, resolution, update frequency, point cloud data with timestamp, and environmental data; The data correction module trains a machine learning model for predicting corrected point cloud data based on historical corrected training data, collects real-time comprehensive collection data, and predicts corrected point cloud data based on the trained machine learning model; The three-dimensional construction module assigns different weights to the point cloud data collected by each sensor according to the performance of the sensor corresponding to the point cloud data and the corresponding environmental factors, fuses the data, and uses the voxel method to perform three-dimensional construction on the fused data.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the three-dimensional measurement and simulation analysis method based on multi-sensor fusion according to any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes the three-dimensional measurement and simulation analysis method based on multi-sensor fusion as claimed in any one of claims 1 to 7.

Images