Risk survey method and device based on digital twinning

Abstract

The invention discloses a risk survey method and device based on digital twinning. According to the scheme, a scene image is obtained through various data acquisition devices; performing preprocessing and adaptive fusion on the scene image; based on a digital twinning technology, reconstructing a scene image after preprocessing and adaptive fusion to obtain a three-dimensional structure of the scene to be subjected to risk survey; and inputting the three-dimensional structure into a risk assessment model, performing risk assessment on the three-dimensional structure by the risk assessment model according to a risk assessment rule of a target industry for the scene, and generating a risk survey report of the scene to be subjected to risk survey. Different from high-precision data acquisition equipment adopted in the prior art, the method has the advantages that the requirement on the data acquisition equipment is reduced, the scene images acquired by various data acquisition equipment are adaptively fused, and the three-dimensional structure of the scene is precisely restored by adopting a digital twinning technology. Compared with the prior art, the method has obvious advantages.

Description

Technical Field

[0001] The present application relates to the technical field of risk survey, and in particular to a risk survey method and device based on digital twins. Background Art

[0002] Current on-site risk survey technology is gradually developing from traditional manual operations to automation, remote monitoring and precise data collection, using advanced technologies such as sensors, drones, AI and the Internet of Things to make the risk survey process more accurate and efficient. The main methods of on-site risk surveys include technologies such as drones, sensors and geographic information systems (GIS). When drones are surveying, they can quickly cover a large area, which is particularly suitable for high-risk and difficult-to-access sites. They can generate high-definition images, 3D models and thermal imaging data; through long-term deployed sensors, continuous real-time data collection and monitoring are carried out, which is suitable for detecting structural health, environmental changes and equipment failures; GIS and LiDAR provide high-precision three-dimensional maps and geographic information, which are particularly suitable for risk analysis in complex terrain. The data obtained after the on-site survey obtains a large amount of complex on-site information, which can be used for intuitive risk prediction and assessment, and support more effective risk management, early warning and emergency response.

[0003] Although current on-site risk survey technology has made significant progress in many areas, the technology cost is high, especially in the case of long-term deployment and large-scale application. The procurement, maintenance, calibration and operation costs of high-precision drones, sensors, GIS systems and lidar are high, and these technologies usually require operators with professional skills and technical support teams, which further increases the complexity and cost of implementation. Therefore, how to reduce the implementation cost of on-site risk surveys and ensure the accuracy of risk surveys is a technical problem that needs to be solved urgently. Summary of the invention

[0004] Based on the above problems, the present application provides a risk survey method and device based on digital twins, aiming to ensure the accuracy of risk survey with low implementation cost.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] The first aspect of the present application provides a risk survey method based on digital twins, the method comprising:

[0007] Acquire scene images through a variety of data acquisition devices; the data acquisition devices include drones, cameras, and mobile phones; the scene images include photos or video data of the scene to be risk surveyed;

[0008] Preprocessing and adaptively fusing the scene image; the adaptive fusion at least includes fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions;

[0009] Based on the digital twin technology, the three-dimensional structure of the scene to be risk surveyed is obtained by reconstructing the scene images after preprocessing and adaptive fusion;

[0010] The three-dimensional structure is input into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scene, and generates a risk survey report for the scene to be risk surveyed.

[0011] In an optional implementation, the method for obtaining the adaptive weighted result based on data quality includes:

[0012] By using an adaptive fusion algorithm, the data collected by each data collection device is weighted according to quality indicators, and the scene image is weighted based on the weight distribution result; the quality indicators include noise level, resolution and viewing angle coverage.

[0013] In an optional implementation, the method for obtaining the spatiotemporal adaptive weighting result based on scene feature points includes:

[0014] The weights of static objects and dynamic objects in the scene image are allocated through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, and the weights allocated to the dynamic objects are adjusted through time-synchronous filtering, and the scene image is weighted based on the adjusted weight allocation results.

[0015] In an optional implementation, the method for obtaining the adaptive weighted result based on multiple resolutions includes:

[0016] The Laplace pyramid multi-scale representation method is used to decompose the scene image into different scales according to different resolutions, and weights are adaptively allocated to the scene images of different scales; an adaptive resolution enhancement algorithm based on regional complexity adjusts the weight allocation according to the complexity of the scene region, and weights the scene image based on the adjusted weight allocation result.

[0017] In an optional implementation, the method of reconstructing the scene image after preprocessing and adaptive fusion includes:

[0018] The preprocessed and adaptively fused scene image is input into a Gaussian mixture model or a neural rendering module for reconstruction.

[0019] In an optional implementation, the preprocessed and adaptively fused scene image is input into a Gaussian mixture model for reconstruction, including:

[0020] The Gaussian mixture model adaptively reconstructs the three-dimensional structure of the scene to be risk surveyed based on the preprocessed and adaptively fused scene images, including adaptive adjustment of the number of Gaussian distributions, adaptive optimization of the Gaussian mixture model parameters, and adaptive enhancement of the robustness of the Gaussian mixture model.

[0021] In an optional implementation, the adaptive adjustment of the number of Gaussian distributions includes:

[0022] The geometric complexity of the scene is monitored based on an adaptive complexity estimator, and the number of required Gaussian distributions is adaptively adjusted according to the geometric complexity of the scene through an adaptive clustering algorithm.

[0023] In an optional implementation, the adaptive optimization of the Gaussian mixture model parameters includes:

[0024] The parameters of the Gaussian mixture model are adaptively optimized according to the characteristics of the area covered by each Gaussian distribution by using an adaptively extended expectation-maximization algorithm; a dynamic update mechanism is introduced to dynamically adjust the weight of each Gaussian distribution according to the local characteristics of the data distribution during each iteration of expectation-maximization; the parameters include mean and covariance matrix parameters; the local characteristics include gradient changes and texture complexity.

[0025] In an optional implementation, the adaptive enhancement of the robustness of the Gaussian mixture model includes:

[0026] A robust Gaussian filtering mechanism is introduced to automatically adjust the accuracy of the Gaussian distribution according to the noise level in the scene image, and combined with an adaptive noise detection mechanism to automatically detect the noise level of the scene image and perform adaptive adjustment.

[0027] In an optional implementation, the preprocessed and adaptively fused scene image is input into a neural rendering module for reconstruction, including:

[0028] The neural rendering module predicts a scene depth map according to the preprocessed and adaptively fused scene image, generates a three-dimensional point cloud of the scene based on the predicted scene depth map, and converts the three-dimensional point cloud into a mesh model to form a complete three-dimensional structure.

[0029] The second aspect of the present application provides a risk survey device based on digital twins, the device comprising:

[0030] A scene image acquisition module, used to acquire scene images through a variety of data acquisition devices; the data acquisition devices include drones, cameras, and mobile phones; the scene images include photos or video data of the scene to be risk surveyed;

[0031] A scene image adaptive fusion module, used for preprocessing and adaptively fusing the scene image; the adaptive fusion at least includes fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions;

[0032] A scene three-dimensional structure reconstruction module is used to reconstruct the three-dimensional structure of the scene to be risk surveyed based on the digital twin technology using the scene images after preprocessing and adaptive fusion;

[0033] The scenario risk assessment module is used to input the three-dimensional structure into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scenario, and generates a risk survey report for the scenario to be risk surveyed.

[0034] Compared with the prior art, this application has the following beneficial effects:

[0035] The present application provides a risk survey method based on digital twins. Compared with the high-precision data acquisition equipment in the prior art, the present application reduces the requirements for data acquisition equipment. Through a variety of low-cost data acquisition devices such as cameras, mobile phones or drones, scene images to be surveyed are obtained, and adaptive fusion is performed based on the characteristics of the data contained in the scene images, such as data quality, spatiotemporal characteristics and resolution. This method can ensure that the fused scene image is not affected by low-quality data under the condition of low implementation cost and different data quality, thereby providing data support for accurately restoring the three-dimensional structure of the scene.

[0036] This application also uses digital twin technology to realistically reproduce the three-dimensional structure of the scene to be risk surveyed. This method can simulate the impact of disasters in a virtual environment, identify potential risk points in advance, and improve the accuracy of risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1A flowchart of a risk survey method based on digital twins provided in an embodiment of the present application;

[0039] Figure 2 A flowchart of a risk survey method based on digital twins provided in an embodiment of the present application;

[0040] Figure 3 A flowchart of another risk survey method based on digital twins provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the structure of a risk survey device based on digital twins provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] As described above, current field risk survey technology has made significant progress in many areas, but the technology cost is relatively high, especially in the case of long-term deployment and large-scale application. The procurement, maintenance, calibration and operation costs of high-precision drones, sensors, GIS systems and lidar are high. Moreover, these technologies usually require operators with professional skills and technical support teams, which further increases the complexity and cost of implementation.

[0043] In view of the above problems, the inventors have proposed a risk survey method and device based on digital twins after research, which obtains scene images through multiple data acquisition devices; preprocesses and adaptively fuses the scene images; based on digital twin technology, reconstructs the three-dimensional structure of the scene to be risk surveyed using the scene images after preprocessing and adaptive fusion; inputs the three-dimensional structure into a risk assessment model, and the risk assessment model conducts risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scene, and generates a risk survey report for the scene to be risk surveyed.

[0044] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] See also Figure 1 , which is a flow chart of a risk survey method based on digital twins provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0046] S101. Acquire scene images through a variety of data acquisition devices.

[0047] Wherein, the data acquisition equipment includes drones, cameras and mobile phones; the scene images include photos or video data of the scene to be risk surveyed.

[0048] Drones are suitable for large-scale high-altitude photography and can cover wide areas; cameras are suitable for fixed-point or mobile photography and can obtain detailed images of specific areas; mobile phones are convenient and easy to carry and can shoot and upload data in real time.

[0049] The scene image data obtained by these data acquisition devices is the basis for reconstructing the three-dimensional structure of the scene, and it also helps to promptly discover and deal with existing risk points, so as to facilitate subsequent risk assessment and management of the scene.

[0050] S102: Preprocess and adaptively fuse the scene image.

[0051] The adaptive fusion includes at least fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions.

[0052] The preprocessing includes data cleaning, data calibration and format conversion of the collected scene images.

[0053] The method for obtaining the adaptive weighted results based on data quality includes allocating weights to static objects and dynamic objects in the scene image through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, adjusting the weights allocated to the dynamic objects through time-synchronous filtering, and weighting the scene image based on the adjusted weight allocation results.

[0054] High-altitude data from drones, 360-degree camera views, and local details from mobile phone cameras have different resolutions, view angles, and precisions. Adaptive weighting based on data quality is used to evaluate the quality of scene images collected by each data acquisition device, such as noise level, resolution, and view angle coverage, so as to dynamically adjust the weights of scene images collected by different data acquisition devices in the reconstruction process. This method can ignore low-quality scene image data according to the actual situation of the data acquisition device to ensure the accuracy and consistency of scene reconstruction.

[0055] The method for obtaining the spatiotemporal adaptive weighted results based on scene feature points includes allocating weights to static objects and dynamic objects in a scene image through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, adjusting the weights allocated to the dynamic objects through time-synchronous filtering, and weighting the scene image based on the adjusted weight allocation results.

[0056] Data acquisition devices collect scene images in different time and space coordinate systems. In particular, the data from drones and ground equipment may have perspective deviation and time asynchrony. Adopting spatiotemporal adaptive weighting based on scene feature points ensures the spatiotemporal consistency between data acquisition devices.

[0057] The method for obtaining the adaptive weighted result based on multi-resolution includes using the Laplace pyramid multi-scale representation method to decompose the scene image into different scales according to different resolutions, and adaptively allocating weights to the scene images of different scales; an adaptive resolution enhancement algorithm based on regional complexity adjusts the weight allocation according to the complexity of the scene region, and weights the scene image based on the adjusted weight allocation result.

[0058] The data collected by drones usually covers a large area but has a low resolution. On the contrary, the local images collected by ground data collection equipment have a high resolution but a limited viewing angle. The multi-resolution adaptive mechanism is adopted to dynamically adjust the weight distribution of high and low resolution scene image data by combining multi-resolution information to ensure the accuracy of scene reconstruction.

[0059] By adaptively fusing scene images collected by multiple data acquisition devices, ignoring the influence of low-quality scene images, and automatically aligning multiple scene image data at different times, spaces, and resolutions, high-precision scene reconstruction can be achieved.

[0060] S103. Based on the digital twin technology, the three-dimensional structure of the scene to be risk surveyed is reconstructed using the scene images after preprocessing and adaptive fusion.

[0061] The preprocessed and adaptively fused scene images are input into a Gaussian mixture model or a neural rendering module for reconstruction to accurately restore the spatial structure and environmental details of the survey scene.

[0062] S104, inputting the three-dimensional structure into a risk assessment model, and having the risk assessment model perform risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scene, and generating a risk survey report for the scene to be risk surveyed.

[0063] Through digital twin technology, potential risk events can be rehearsed, simulated, and calculated in a virtual environment, and possible risks and their impacts can be assessed in advance. This process significantly improves the accuracy of risk identification and assists in generating more accurate risk prevention measures through digital means. Ultimately, this method provides reliable scientific data support and decision-making basis, making risk management and prevention more targeted and effective.

[0064] A risk survey method based on digital twins provided in this embodiment reduces the requirements for data acquisition equipment. Scene images to be surveyed are acquired through a variety of low-cost data acquisition devices such as cameras, mobile phones or drones. According to the characteristics of the data contained in the scene images, such as data quality, spatiotemporal characteristics or resolution, corresponding adaptive fusion methods are used for fusion. This method can ensure that the fused scene image is not affected by low-quality data under the condition of low implementation cost and different data quality, thereby providing data support for accurately restoring the three-dimensional structure of the scene.

[0065] This method also uses digital twin technology to realistically reproduce the three-dimensional structure of the scene to be risk surveyed. This approach can simulate the impact of disasters in a virtual environment, identify potential risk points in advance, and improve the accuracy of risk assessment.

[0066] The following method embodiment is used to describe the risk survey method based on digital twins in the insurance industry. Insurance companies use digital twin technology to conduct comprehensive risk surveys on multiple target buildings, optimize policy pricing, and reduce potential losses.

[0067] See also Figure 2 , which is a flow chart of another risk survey method based on digital twins provided in an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0068] S201. Acquire scene images through a variety of data acquisition devices.

[0069] The insurance company arranged a data collection team to conduct a detailed survey of the target building and the surrounding environment using a variety of data collection equipment. Drones were used to fly around the building to capture high-altitude panoramic videos, with a particular focus on structures such as roofs, exterior walls, and drainage systems. Panoramic cameras and ordinary cameras were used to take high-definition pictures and videos from multiple angles on the ground to capture building facades and key areas such as doors, windows, sewers, and parking lots. On-site surveyors used mobile phones to take videos and pictures inside the target building to obtain internal information and equipment conditions.

[0070] S202: Preprocess and adaptively fuse the scene image.

[0071] Preprocess the collected scene images, including filtering out blurry and redundant data, calibrating the scene image data collected by drones, cameras and mobile phones through multi-view fusion technology, and converting scene image data in different formats into a standard format for reconstructing three-dimensional structures. This method can ensure that the collected scene images are complete and that all scene image data formats are the same.

[0072] By adaptively fusing scene images collected by multiple data acquisition devices, ignoring the influence of low-quality scene images, and automatically aligning multiple scene image data at different times, spaces, and resolutions, high-precision scene reconstruction can be achieved.

[0073] S203. Based on the digital twin technology, the scene image after preprocessing and adaptive fusion is input into the Gaussian mixture model for reconstruction to obtain the three-dimensional structure of the scene to be risk surveyed.

[0074] The Gaussian mixture model adaptively reconstructs the three-dimensional structure of the scene to be risk surveyed based on the preprocessed and adaptively fused scene images, including adaptive adjustment of the number of Gaussian distributions, adaptive optimization of the Gaussian mixture model parameters, and adaptive enhancement of the robustness of the Gaussian mixture model.

[0075] The adaptive adjustment of the Gaussian distribution quantity includes:

[0076] The geometric complexity of the scene is monitored based on an adaptive complexity estimator, and the number of required Gaussian distributions is adaptively adjusted according to the geometric complexity of the scene through an adaptive clustering algorithm.

[0077] The number of Gaussian distributions required will vary in areas of different scene complexity. For example, flat areas require fewer Gaussian distributions, while areas with complex structures or rich details require more Gaussian distributions to accurately describe the geometric details. Through adaptive clustering algorithms (such as the density-based clustering method DBSCAN), the number of Gaussian distributions required is adaptively adjusted according to the complexity of the scene. In simple areas, the number of components of the Gaussian mixture model is reduced to reduce computational overhead; in complex areas, the number of Gaussian distributions is dynamically increased to improve accuracy. An adaptive complexity estimator is introduced to monitor the geometric complexity of the scene in real time and dynamically adjust the number of components of the Gaussian mixture model to ensure high-quality scene reconstruction even with limited resources.

[0078] The adaptive optimization of the Gaussian mixture model parameters comprises:

[0079] The parameters of the Gaussian mixture model are adaptively optimized according to the characteristics of the area covered by each Gaussian distribution by using an adaptively extended expectation-maximization algorithm; a dynamic update mechanism is introduced to dynamically adjust the weight of each Gaussian distribution according to the local characteristics of the data distribution during each iteration of expectation-maximization; the parameters include mean and covariance matrix parameters; the local characteristics include gradient changes and texture complexity.

[0080] In different scenes and regions, parameters such as the mean and covariance matrix of the Gaussian distribution need to be optimized according to the characteristics of the local scene. For example, for objects or structures with sharp edges, the covariance matrix needs to reflect its asymmetry. Using the adaptive extension of the expectation maximization (EM) algorithm, the mean and covariance matrix parameters are adaptively optimized according to the characteristics of the area covered by each Gaussian distribution. For example, in smooth areas, the covariance matrix is ​​adaptively reduced to simplify calculations; while in complex areas, the covariance matrix is ​​adaptively enlarged to cover more details. A dynamic weight update mechanism is introduced. During each EM iteration, the weight of each Gaussian distribution is dynamically adjusted according to the local characteristics of the data distribution (such as gradient changes, texture complexity) to ensure the optimal model fit in different scenes.

[0081] The adaptive enhancement of the robustness of the Gaussian mixture model includes:

[0082] A robust Gaussian filtering mechanism is introduced to automatically adjust the accuracy of the Gaussian distribution according to the noise level in the scene image, and combined with an adaptive noise detection mechanism to automatically detect the noise level of the scene image and perform adaptive adjustment.

[0083] In the data collected by various data acquisition devices, there may be different degrees of noise (such as image blur caused by drone jitter and inconsistency caused by lighting changes). By introducing a robust Gaussian filtering mechanism, in areas with high noise, the fineness of the Gaussian distribution is adaptively reduced to prioritize fitting the general outline; while in areas with low noise, the high accuracy of the Gaussian mixture model is maintained. Combined with an adaptive noise detection mechanism, the dynamic change trend of the scene is used to automatically detect the noise level and perform adaptive adjustments, so that high reconstruction accuracy can be maintained under scene image data of different qualities.

[0084] The adaptability of the Gaussian mixture model improves its flexibility in complex scenarios, and the parameters of the Gaussian mixture model can be dynamically adjusted according to the geometric characteristics of the scene, so that complex areas have higher accuracy and simple areas maintain computational efficiency. By adaptively adjusting the parameters and distribution number of the Gaussian mixture model, the method can maintain robustness and stability when processing data with different noise levels and data incompleteness, and adapt to a variety of application scenarios.

[0085] S204: Input the three-dimensional structure into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the insurance industry for the scene, and generates a risk survey report for the scene to be risk surveyed.

[0086] Based on the three-dimensional structure of the scene generated by the Gaussian mixture model, the insurance company combines it with the risk assessment rules of the insurance industry for analysis. By analyzing the physical structure, roof condition, drainage system, etc. of the building, possible weaknesses are identified, such as roofs that are prone to water accumulation and aging window seals. Typhoon and flood simulators are used to test the three-dimensional structure of the scene for natural disasters of different intensities to observe the potential degree and scope of damage. Based on the simulation results, the economic losses that the building may suffer under different disaster scenarios are evaluated, the repair costs are estimated, the risk probability is calculated, and the final analysis report and recommendations are generated. The risk points of the building are listed in detail, including the simulation results of disasters such as typhoons and floods, and the estimated loss amount and possible damaged parts are estimated. Improvement suggestions are put forward, such as strengthening the roof, improving the drainage system, and installing flood control facilities, to help customers reduce risks. Insurance companies use the risk assessment results to optimize policies, adjust premiums, and design targeted insurance products to provide customers with more targeted protection plans.

[0087] Through digital twin technology, insurance companies can simulate the impact of disasters in a virtual environment, identify potential risk points in advance, and optimize insurance decisions. This not only improves the accuracy of risk assessment, but also reduces potential losses, providing customers with a more scientific risk management solution.

[0088] See also Figure 3 , which is a flow chart of another risk survey method based on digital twins provided in an embodiment of the present application. Figure 3 As shown, the method comprises the following steps:

[0089] S301. Acquire scene images through various data acquisition devices.

[0090] Wherein, the data acquisition equipment includes drones, cameras and mobile phones; the scene images include photos or video data of the scene to be risk surveyed.

[0091] S302: Preprocess and adaptively fuse the scene image.

[0092] The adaptive fusion includes at least fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions.

[0093] The preprocessing includes data cleaning, data calibration and format conversion of the collected scene images.

[0094] S303. Based on the digital twin technology, the scene image after preprocessing and adaptive fusion is input into the neural rendering module for reconstruction to obtain the three-dimensional structure of the scene to be risk surveyed.

[0095] The neural rendering module predicts a scene depth map according to the preprocessed and adaptively fused scene image, generates a three-dimensional point cloud of the scene based on the predicted scene depth map, and converts the three-dimensional point cloud into a mesh model to form a complete three-dimensional structure.

[0096] The neural rendering module includes a neural network model, such as Monodepth2, MiDaS, etc., and the fused scene image is predicted by the trained neural network model. Converting the predicted scene depth map into a three-dimensional point cloud includes converting each pixel point in the predicted scene depth map into a coordinate point in three-dimensional space, and then combining all the converted coordinate points into a three-dimensional point cloud. Processing the three-dimensional point cloud and converting it into a mesh model includes generating a mesh model using a Poisson surface reconstruction algorithm or generating a mesh model using a Marching Cubes algorithm.

[0097] By reconstructing the three-dimensional structure of the scene through neural rendering technology, the spatial structure and environmental information of the scene can be accurately restored, which helps insurance companies conduct risk assessments.

[0098] S304: Input the three-dimensional structure into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the insurance industry for the scene, and generates a risk survey report for the scene to be risk surveyed.

[0099] The implementation methods of S301, S302 and S304 in the above embodiments have been introduced in the previous text. Please refer to the description of S201, S202 and S204 in the previous embodiments, which will not be repeated here.

[0100] Based on the risk survey method based on digital twins introduced in the previous embodiments, the present application also provides a risk survey device based on digital twins. Figure 4 Figure 1 is a schematic diagram of the structure of the device. Figure 4 As shown, the risk survey device based on digital twin includes:

[0101] The scene image acquisition module 401 is used to acquire scene images through various data acquisition devices;

[0102] A scene image adaptive fusion module 402, used for preprocessing and adaptively fusion of the scene image;

[0103] The scene three-dimensional structure reconstruction module 403 is used to reconstruct the three-dimensional structure of the scene to be risk surveyed by using the scene images after preprocessing and adaptive fusion based on the digital twin technology;

[0104] The scenario risk assessment module 404 is used to input the three-dimensional structure into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scenario, and generates a risk survey report for the scenario to be risk surveyed.

[0105] In an optional implementation, the data acquisition device of the scene image acquisition module includes a drone, a camera, and a mobile phone; the scene image includes a photo or video data of the scene to be risk surveyed;

[0106] In an optional implementation, the scene image adaptive fusion includes fusing at least two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions;

[0107] In an optional implementation, the preprocessing includes data cleaning, data calibration and format conversion of the collected scene images.

[0108] In an optional implementation, the method for obtaining the adaptive weighted results based on data quality includes allocating weights to static objects and dynamic objects in the scene image through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, and then adjusting the weights allocated to the dynamic objects through time-synchronous filtering, and weighting the scene image based on the adjusted weight allocation results.

[0109] In an optional implementation, the method for obtaining the spatiotemporal adaptive weighted results based on scene feature points includes allocating weights to static objects and dynamic objects in the scene image through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, adjusting the weights allocated to the dynamic objects through time-synchronous filtering, and weighting the scene image based on the adjusted weight allocation results.

[0110] In an optional implementation, the method for obtaining the adaptive weighted result based on multi-resolution includes using a Laplace pyramid multi-scale representation to decompose a scene image into different scales according to different resolutions, and adaptively allocating weights to scene images of different scales; an adaptive resolution enhancement algorithm based on regional complexity adjusts the weight allocation according to the complexity of the scene region, and weights the scene image based on the adjusted weight allocation result.

[0111] In an optional implementation, the preprocessed and adaptively fused scene image is input into a Gaussian mixture model or a neural rendering module for reconstruction.

[0112] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0113] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A risk survey method based on digital twins, characterized in that: include: Acquire scene images through a variety of data acquisition devices; the data acquisition devices include drones, cameras, and mobile phones; the scene images include photos or video data of the scene to be risk surveyed; Preprocessing and adaptively fusing the scene image; the adaptive fusion at least includes fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions; Based on the digital twin technology, the three-dimensional structure of the scene to be risk surveyed is obtained by reconstructing the scene images after preprocessing and adaptive fusion; The three-dimensional structure is input into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scene, and generates a risk survey report for the scene to be risk surveyed.

2. The method according to claim 1, characterized in that The method for obtaining the adaptive weighted result based on data quality includes: By using an adaptive fusion algorithm, the data collected by each data collection device is weighted according to quality indicators, and the scene image is weighted based on the weight distribution result; the quality indicators include noise level, resolution and viewing angle coverage.

3. The method according to claim 1, characterized in that The method for obtaining the spatiotemporal adaptive weighting result based on scene feature points includes: The weights of static objects and dynamic objects in the scene image are allocated through a spatiotemporal adaptive weight allocation mechanism based on scene feature points, and the weights allocated to the dynamic objects are adjusted through time-synchronous filtering, and the scene image is weighted based on the adjusted weight allocation results.

4. The method according to claim 1, characterized in that: The method for obtaining the multi-resolution-based adaptive weighted result includes: The Laplace pyramid multi-scale representation method is used to decompose the scene image into different scales according to different resolutions, and weights are adaptively allocated to the scene images of different scales; an adaptive resolution enhancement algorithm based on regional complexity adjusts the weight allocation according to the complexity of the scene region, and weights the scene image based on the adjusted weight allocation result.

5. The method according to any one of claims 1 to 4, characterized in that The method of reconstructing the scene image after preprocessing and adaptive fusion includes: The preprocessed and adaptively fused scene image is input into a Gaussian mixture model or a neural rendering module for reconstruction.

6. The method according to claim 5, characterized in that Inputting the preprocessed and adaptively fused scene image into a Gaussian mixture model for reconstruction, including: The Gaussian mixture model adaptively reconstructs the three-dimensional structure of the scene to be risk surveyed based on the preprocessed and adaptively fused scene images, including adaptive adjustment of the number of Gaussian distributions, adaptive optimization of the Gaussian mixture model parameters, and adaptive enhancement of the robustness of the Gaussian mixture model.

7. The method according to claim 6, characterized in that The adaptive adjustment of the Gaussian distribution quantity includes: The geometric complexity of the scene is monitored based on an adaptive complexity estimator, and the number of required Gaussian distributions is adaptively adjusted according to the geometric complexity of the scene through an adaptive clustering algorithm.

8. The method according to claim 7, characterized in that The adaptive optimization of the Gaussian mixture model parameters comprises: The parameters of the Gaussian mixture model are adaptively optimized according to the characteristics of the area covered by each Gaussian distribution by using an adaptively extended expectation-maximization algorithm; a dynamic update mechanism is introduced to dynamically adjust the weight of each Gaussian distribution according to the local characteristics of the data distribution during each iteration of expectation-maximization; the parameters include mean and covariance matrix parameters; the local characteristics include gradient changes and texture complexity.

9. The method according to claim 7, characterized in that: The adaptive enhancement of the robustness of the Gaussian mixture model includes: A robust Gaussian filtering mechanism is introduced to automatically adjust the accuracy of the Gaussian distribution according to the noise level in the scene image, and combined with an adaptive noise detection mechanism to automatically detect the noise level of the scene image and perform adaptive adjustment.

10. The method according to claim 5, characterized in that Inputting the preprocessed and adaptively fused scene image into a neural rendering module for reconstruction, including: The neural rendering module predicts a scene depth map according to the preprocessed and adaptively fused scene image, generates a three-dimensional point cloud of the scene based on the predicted scene depth map, and converts the three-dimensional point cloud into a mesh model to form a complete three-dimensional structure.

11. A risk survey device based on digital twins, characterized in that: include: A scene image acquisition module, used to acquire scene images through a variety of data acquisition devices; the data acquisition devices include drones, cameras, and mobile phones; the scene images include photos or video data of the scene to be risk surveyed; A scene image adaptive fusion module, used for preprocessing and adaptively fusing the scene image; the adaptive fusion at least includes fusing two of the adaptive weighted results based on data quality, the spatiotemporal adaptive weighted results based on scene feature points, or the adaptive weighted results based on multiple resolutions; A scene three-dimensional structure reconstruction module is used to reconstruct the three-dimensional structure of the scene to be risk surveyed based on the digital twin technology using the scene images after preprocessing and adaptive fusion; The scenario risk assessment module is used to input the three-dimensional structure into a risk assessment model, and the risk assessment model performs risk assessment on the three-dimensional structure according to the risk assessment rules of the target industry for the scenario, and generates a risk survey report for the scenario to be risk surveyed.

Images