A method and system for improving the accuracy of visual field simulation

Abstract

This invention belongs to the field of video image processing technology, specifically relating to a method and system for improving the accuracy of visual field simulation. The method includes the following steps: acquiring basic geographic information data to construct a macroscopic 3D scene; acquiring on-site image data and camera parameters, preprocessing the image data to provide standardized data input for geometric inversion; generating a microscopic 3D proxy scene representing on-site occlusion details through 3D geometric inversion; performing coordinate calibration and fusion on the macroscopic 3D scene and the microscopic 3D proxy scene to obtain an enhanced 3D scene model; performing visual field collision detection simulation calculations based on the enhanced 3D scene model to generate basic visual field simulation data and outputting simulation results. This invention's simulation accuracy improvement method achieves an order-of-magnitude improvement in simulation accuracy and realism, while significantly reducing implementation costs and greatly improving efficiency.

Description

Technical Field

[0001] This invention belongs to the field of video image processing technology, specifically relating to a method and system for improving the accuracy of visual field simulation. Background Technology

[0002] Video surveillance systems are core infrastructure in fields such as safe cities, smart security, traffic control, and emergency management. The scientific planning of camera locations and the accurate assessment of visual coverage directly determine the surveillance system's prevention and control capabilities and construction effectiveness. Visual area simulation technology, through 3D scene modeling and line-of-sight collision detection, simulates the actual monitoring coverage of cameras. It is a core technical means for realizing camera location planning, blind spot identification, and solution optimization, and has become a core foundational technology in the security industry. The core logic of current visual area simulation technology is based on 3D scene data, simulating the camera's field of view through virtual camera modeling and visibility assessment.

[0003] Patent application CN107368656A provides a method for BIM-based security monitoring simulation and blind spot analysis, capable of simulating and analyzing the coverage of security monitoring cameras deployed within a target building. The method includes: establishing a BIM model of the target building using BIM technology, wherein the established BIM model reflects the complete target building and records its three-dimensional coordinates; deploying security monitoring cameras based on the established BIM model; generating the monitoring range of each security monitoring camera according to the deployed cameras; calculating the obstruction of the monitoring range by surrounding objects, performing blind spot analysis and overlapping area analysis; and marking the effective monitoring area, blind spot range, and overlapping area range within the target building based on the analysis results. Patent application CN117376705A discloses a genetic optimization method and system for the spatial layout of urban surveillance cameras. This method includes acquiring and processing oblique photogrammetry three-dimensional model data, selecting alternative camera locations based on the model, calculating the optimal camera position and angle, calculating the true visible range of the cameras, and optimizing using a genetic algorithm.

[0004] Existing technical solutions all revolve around the acquisition and processing of 3D scene data, but none have broken through the technical paradigm of "relying on pre-collected geographic information / survey data." They can only incorporate macroscopic terrain and building outlines, failing to reproduce microscopic obstructions such as trees, temporary structures, and building facade details. The simulation results deviate greatly from the actual situation, easily leading to the problem of full simulation coverage but numerous blind spots on site. To improve simulation accuracy, existing technologies rely on laser scanning and manual on-site surveys to acquire detailed 3D data. This results in high equipment purchase costs, long field survey cycles, and difficult data processing. Furthermore, they cannot quickly adapt to dynamic changes in the on-site environment (such as the addition of temporary fencing or changes in tree growth), hindering large-scale application and failing to simultaneously resolve the industry's core contradiction of "simulation accuracy" and "cost efficiency." Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method for improving the accuracy of visual field simulation. By utilizing video image data from existing cameras on-site, geometric inversion is used to achieve 3D scene enhancement and improve the accuracy of visual field simulation.

[0006] In a first aspect, the present invention provides a method for improving the accuracy of visual field simulation, comprising the following steps: Acquire basic geographic information data and construct a macroscopic 3D scene; Acquire on-site image data and camera parameters, preprocess image data, and provide standardized data input for geometric inversion; A microscopic 3D proxy scene representing the occlusion details of the scene is generated through 3D geometric inversion; The coordinates of the macroscopic 3D scene and the microscopic 3D proxy scene are calibrated and fused to obtain an enhanced 3D scene model; Based on the enhanced 3D scene model, a viewpoint collision detection simulation is performed to generate viewpoint simulation base data and output the simulation results.

[0007] Furthermore, the acquisition of basic geographic information data and the construction of a macroscopic three-dimensional scene include the following steps: Obtain digital elevation model data and building outline vector data of the target area, and combine them with regional coordinate system parameters to construct a macroscopic basic 3D scene that includes terrain elevation, building outline, and overall spatial layout.

[0008] Furthermore, the step of generating a microscopic 3D proxy scene representing on-site occlusion details through 3D geometric inversion includes the following steps: The preprocessed image keyframe sequence is reconstructed in three dimensions to generate a sparse three-dimensional point cloud; based on the sparse three-dimensional point cloud, dense depth estimation and point cloud optimization are performed to generate a dense three-dimensional point cloud; the dense three-dimensional point cloud is reconstructed to generate a three-dimensional mesh model of key occlusions; the mesh model is lightweighted to generate a microscopic three-dimensional proxy scene representing the details of occlusion on site.

[0009] Furthermore, the step of generating a microscopic 3D proxy scene representing on-site occlusion details through 3D geometric inversion also includes the following steps: Semantic segmentation and attribute annotation are performed on the models in the microscopic 3D proxy scene.

[0010] Furthermore, the coordinate calibration and fusion of the macroscopic 3D scene and the microscopic 3D proxy scene includes the following steps: A unified world coordinate system is constructed based on the engineering coordinate system of the basic geographic information data; the micro-3D proxy scene is registered to the unified world coordinate system based on the camera extrinsic coordinates or scene-named control points; the registered micro-3D proxy scene is superimposed and fused with the macro-3D scene, and the repeated areas are replaced with the micro-3D proxy scene as the highest priority, finally generating an enhanced 3D scene model.

[0011] Further, the step of performing view-domain collision detection simulation calculations based on the enhanced 3D scene model includes the following steps: Configure virtual camera parameters in the enhanced 3D scene model; generate a view ray with the optical center of the virtual camera as the origin; perform two-level collision detection on the view ray, namely macroscopic collision detection and microscopic fine collision detection; statistically analyze the collision detection results of the view ray, mark the visible area and blind area, and quantify and classify the effective monitoring distance of the visible area based on the distance of the collision point and the type of occlusion, thereby generating basic data for the visual field simulation.

[0012] Furthermore, the simulation results include a simulation view with realistic on-site texture mapping, a coverage heatmap, vector data of the visible field coverage, and a simulation analysis report.

[0013] Secondly, this application provides a system for improving the accuracy of visual field simulation to implement the above method, comprising: The data preprocessing module is used for the acquisition, parsing, and preprocessing of basic geographic information data, on-site video image data, and camera parameter data; The image geometry inversion module is used to complete 3D sparse reconstruction of the scene, dense depth estimation, mesh reconstruction and lightweighting, semantic annotation, and generate a lightweight 3D proxy scene with attribute annotations. The scene fusion modeling module is used to construct a unified coordinate system, complete the accurate registration, global optimization and overlay fusion of two-layer scenes, and generate an enhanced 3D scene model; The high-precision field of view calculation module is used to configure virtual camera parameters, perform macro-micro hierarchical collision detection, complete high-precision field of view connectivity calculation, and output basic field of view data. The results output and visualization module is used to generate simulation views with realistic on-site texture maps, overlay heatmaps, vector data of the visible field of view, and simulation analysis reports.

[0014] Furthermore, the output of the data preprocessing module is communicatively connected to the inputs of the image geometric inversion module and the scene fusion modeling module, respectively; the input of the image geometric inversion module is communicatively connected to the data preprocessing module, and its output is communicatively connected to the input of the scene fusion modeling module; the input of the scene fusion modeling module is communicatively connected to both the data preprocessing module and the image geometric inversion module, and its output is communicatively connected to the input of the high-precision visual field calculation module; the input of the high-precision visual field calculation module is communicatively connected to the scene fusion modeling module, and its output is communicatively connected to the input of the result output and visualization module; the input of the result output and visualization module is communicatively connected to the high-precision visual field calculation module, and its output is connected to the front-end business application platform.

[0015] Furthermore, the data preprocessing module has a built-in camera calibration unit, which can calculate the intrinsic and extrinsic parameters of cameras with unknown parameters, and complete preprocessing operations such as image distortion removal, noise reduction, and key frame extraction.

[0016] Furthermore, the image geometric inversion module incorporates a sparse reconstruction unit, a dense depth estimation unit, a mesh reconstruction and lightweighting unit, and a semantic annotation unit.

[0017] Furthermore, the scene fusion modeling module has a built-in coordinate registration unit, a global optimization unit, and a scene fusion unit.

[0018] Furthermore, the high-precision field-of-view calculation module has a built-in virtual camera construction unit, a hierarchical collision detection unit, and a field-of-view calculation unit.

[0019] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the above-described method for improving the accuracy of visual field simulation.

[0020] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method for improving the accuracy of visual field simulation.

[0021] Fifthly, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned method for improving the accuracy of visual field simulation.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Significant improvement in simulation accuracy and realism: Compared with traditional simulation schemes based on DEM / DSM, this invention incorporates microscopic occlusion details such as trees, temporary buildings, and fine building facades through video image inversion, thereby improving the overlap between the simulation results in the visible field and the actual scene, the accuracy of blind spot identification, and the first-pass rate of point planning.

[0023] 2. Significantly reduced implementation costs and greatly improved efficiency: Compared with high-precision solutions based on laser point clouds, this invention does not require the purchase of expensive 3D laser scanning equipment or manual field surveys. It can directly utilize video image data from existing cameras on site to complete detailed scene reconstruction. The data acquisition and processing cycle is shortened from several days to several hours, reducing the implementation cost per site and making it suitable for large-scale promotion and application.

[0024] 3. Excellent timeliness and environmental adaptability: This invention can quickly respond to dynamic changes in the on-site environment. When there are new temporary fences, changes in tree growth, or adjustments to structures, the scene can be updated and the simulation recalculated simply by acquiring the latest on-site video images. This solves the problem of fixed scene data and inability to adapt to dynamic changes in the environment in traditional solutions.

[0025] 4. Significantly enhanced business support capabilities: This invention can accurately identify monitoring blind spots caused by microscopic occlusion, and output simulated views with realistic textures, quantified coverage heat maps and blind spot data. It can provide accurate and reliable data support for security camera location planning, coverage assessment, optimization and adjustment, and deployment in key areas, greatly improving the scientific nature and effectiveness of video surveillance system construction.

[0026] 5. High compatibility and low implementation threshold: This invention is compatible with existing mainstream basic geographic information data formats and can be seamlessly connected to existing security monitoring platforms and geographic information systems (GIS). It does not require large-scale transformation of existing infrastructure and can be quickly implemented. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating the overall process of improving the accuracy of visual field simulation. Figure 2 This is a schematic diagram of a two-layer fusion scenario; Figure 3 This is a schematic diagram of graded collision detection; Figure 4 System structure block diagram; Figure 5 This is a block diagram of the internal structure of the image geometric inversion module. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0031] The core innovation of this invention is to construct a two-layer fusion simulation model of "geographic information-image geometry". On the basis of the traditional macro-layer of basic geographic information, a micro-layer of image geometry inversion based on on-site video is added. Through the precise calibration and fusion of the two layers of data, a high-precision visual field simulation that takes into account both macro-terrain and micro-occlusion details is achieved.

[0032] This embodiment provides a method for improving the accuracy of field-of-view simulation based on geometric inversion and fusion of on-site video images. The overall process is detailed in the appendix. Figure 1 The flowchart is shown below, with the specific steps as follows: Step S1: Acquire basic geographic information data and construct a macroscopic 3D scene. Basic geographic information data acquisition and scene construction: acquire digital elevation model (DEM) data, building outline vector data, and regional coordinate system parameters of the target simulation area, and construct a macro-level basic 3D scene. This scene includes the terrain elevation, building outline, and overall spatial layout of the target area, providing a unified geographic coordinate benchmark for subsequent simulations.

[0033] Step S2: Acquire on-site image data and camera parameters, and preprocess the image data. On-site video image and camera parameter acquisition: Acquire on-site video streams or still image data collected by one or more existing cameras in the target area; simultaneously acquire the intrinsic parameters (focal length, pixel size, distortion coefficient) and extrinsic parameters (installation position coordinates, installation height, pitch angle, azimuth angle, coordinate system transformation parameters) of the corresponding cameras; if the camera parameters are unknown, calculate the camera intrinsic and extrinsic parameters from the video image sequence using the Zhang Zhengyou calibration method or self-calibration algorithm.

[0034] The acquired video image data is preprocessed by distortion correction, Gaussian denoising, keyframe extraction, and ORB feature point matching. Invalid, blurry, and duplicate frames are removed, and high-quality keyframe sequences containing rich scene textures and occlusion features are selected to provide standardized data input for subsequent geometric inversion.

[0035] Step S3: Generate a microscopic 3D proxy scene through 3D geometric inversion. This step is the core innovation of this invention. It corresponds to the image geometry inversion layer that is missing in the existing technology. It can solve the three-dimensional geometric information of the real occlusion in the scene from the two-dimensional image without laser scanning. The specific implementation is as follows: 1) Scene 3D sparse reconstruction: Based on the preprocessed keyframe sequence, the structure of motion recovery (SfM) algorithm is used to match and track the feature points in the keyframes, calculate the camera pose corresponding to each keyframe, and generate a sparse 3D point cloud of the scene to initially restore the spatial position and relative geometric relationship of key occluders in the scene.

[0036] 2) Dense Depth Estimation and Point Cloud Optimization: For sparse reconstruction results, the Multi-View Stereo Matching (MVS) algorithm (for multi-camera / multi-frame scenes) or monocular depth estimation algorithm (for single-camera single-frame scenes) is used to perform dense depth estimation on key frame sequences to generate dense 3D point clouds of the scene; at the same time, through statistical filtering, outlier removal, and point cloud registration optimization, noise points and redundant data are removed, and dense point cloud data of key occlusion objects such as trees, walls, billboards, balconies, eaves, and temporary structures in the scene are accurately preserved.

[0037] In some embodiments, for scenarios with a single camera and no continuous video sequence, Neural Radiation Field (NeRF) and Instant NGP algorithms can be used to reconstruct the 3D geometry and texture of the scene using a single frame or a small number of frames, generating a 3D proxy scene. For edge scenarios with limited computing power, a lightweight monocular depth estimation algorithm based on a pre-trained deep learning model can be used to directly solve the depth map from a single frame image, generating a 3D point cloud of the scene and reducing computing power consumption.

[0038] 3) 3D Mesh Reconstruction and Lightweight Processing: The optimized dense point cloud is reconstructed, and the Poisson reconstruction algorithm is used to generate a 3D mesh model of the key occlusions. At the same time, the mesh model is lightweighted by simplifying the mesh through edge folding and topology optimization. While fully preserving the geometric contours and spatial boundaries of the occlusions, the number of model faces and the amount of data are reduced, generating a lightweight 3D proxy scene that represents the real occlusion details on site.

[0039] 4) Semantic segmentation and attribute labeling of occluded objects: A pre-trained semantic segmentation neural network is used to perform semantic classification on the models in the 3D proxy scene, and to label different types of occluded objects such as trees, building attachments, temporary structures, walls, billboards, etc. At the same time, each occluded object model is bound with corresponding spatial coordinates, geometric dimensions, and contour boundary attributes to provide semantic support for subsequent fusion and collision detection.

[0040] Step S4: Perform coordinate calibration and fusion of macroscopic and microscopic scenes. This step achieves precise fusion of the macroscopic layer of geographic information and the microscopic layer of image geometric inversion, constructing an enhanced 3D scene model. (See attached document.) Figure 2 The schematic diagram is shown below, and the specific implementation is as follows: 1) Construction of a unified coordinate system: Based on the engineering coordinate system of the basic geographic information data obtained in step S1, a unified world coordinate system is constructed to ensure that the spatial reference of the two scenes is consistent.

[0041] 2) Proxy Scene Coordinate Registration and Calibration: Based on the camera's extrinsic coordinates and corresponding control points in the scene (such as building corners, road signs, and fixed landmarks with known coordinates), the Iterative Closest Point (ICP) algorithm is used to register the lightweight 3D proxy scene generated in step S3 to the unified world coordinate system. At the same time, the registration result is globally optimized by the bundle adjustment (BA) method to eliminate coordinate deviations and ensure that the spatial position error between the proxy scene and the basic geographic information scene is controlled within the centimeter level.

[0042] In some embodiments, for scenarios without clearly defined control points of the same name, a feature-matching-based global registration algorithm can be used instead. This algorithm extracts line features and surface features (such as building outlines and road boundary lines) from the basic geographic information scene and performs feature matching to complete the automatic registration of the scene. For large-scale scene data from multiple cameras, a hierarchical registration strategy can be used instead. This strategy first completes the local registration of the single-camera proxy scene and then performs global stitching and optimization of multiple scenes to improve the registration efficiency of large-scale scenes.

[0043] 3) Dual-layer scene overlay and fusion: The calibrated lightweight 3D proxy scene is overlaid and fused with the macroscopic 3D scene constructed from basic geographic information; priority processing is performed on the model of the repeated area, with the proxy scene model retrieved from the image having the highest priority, replacing the corresponding coarse-grained building outlines and terrain data in the basic scene, and finally generating an enhanced 3D scene model that includes both macroscopic terrain and building body, as well as microscopic on-site occlusion details.

[0044] Step S5: Perform viewpoint collision detection simulation calculations based on the enhanced 3D scene model. This step achieves accurate visibility judgment based on enhanced 3D scenes. The core is macro-micro hierarchical collision detection, balancing computational efficiency and simulation accuracy. The specific implementation is as follows: 1) Virtual camera parameter configuration: In the enhanced 3D scene model, configure the parameters of the virtual camera to be simulated, including installation position coordinates, installation height, field of view, focal length, pitch angle, azimuth angle, distortion coefficient, and restore the imaging model of the real camera at a 1:1 scale.

[0045] 2) View Ray Generation and Hierarchical Collision Detection: Using the optical center of the virtual camera as the origin, uniformly distributed view rays are generated within the field of view according to a preset angular resolution; for each view ray, two levels of collision detection are performed (see attached). Figure 3 Schematic diagram of the principle: Level 1: Macro collision detection, quickly determines whether the view ray collides with the terrain and building outlines in the basic geographic information scene, filters out rays that are completely blocked by macro terrain / buildings, and improves computational efficiency; Level 2: Microscopic fine collision detection. For the view rays that pass the first level of detection, it is further determined whether they collide with fine occlusion models such as trees, building facade details, and temporary structures obtained from the image inversion in the fused scene, so as to accurately identify the interruption of visibility caused by microscopic occlusion.

[0046] In some embodiments, for large-scale scene multi-camera batch simulation scenarios, a GPU-accelerated spatial partitioning collision detection algorithm can be used instead. The enhanced 3D scene is spatially partitioned using octrees and KD-Trees, and the collision detection between the view rays and the scene model is accelerated in parallel. While ensuring detection accuracy, the simulation computation efficiency of large-scale scenes is greatly improved.

[0047] 3) Visual area result calculation: The collision detection results of all visible rays are statistically analyzed. The area covered by rays that do not have any collisions is marked as the visible area, and the area covered by rays that have collisions is marked as the blind zone. At the same time, based on the distance of the collision point and the type of obstruction, the effective monitoring distance of the visible area is quantified and classified to generate the basic data for visual area simulation.

[0048] Step S6: Output simulation results 1) Based on the enhanced 3D scene model, a simulation view with realistic texture maps of the scene is generated. The simulation view and the actual shooting screen of the scene camera achieve a precise correspondence between viewpoint and content. 2) Based on the visual field calculation results, a coverage heat map is generated, and different color levels are used to accurately mark the monitoring blind spots, half-coverage areas and full-coverage areas caused by microscopic occlusion details on site. At the same time, quantitative data such as the location, area and type of occlusion of the blind spots are output. 3) Output vector data and simulation analysis reports of the visible field of view coverage, providing accurate data basis for the planning, evaluation, and optimization of camera locations.

[0049] In some embodiments, an edge-cloud collaborative deployment architecture can be used to replace the centralized processing system. Video image preprocessing and key frame extraction are completed at the edge camera, while 3D geometric inversion, scene fusion, and field-of-view simulation calculations are completed in the cloud, reducing data transmission bandwidth consumption and improving system response speed.

[0050] This application also provides a system for improving the accuracy of visual field simulation based on geometric inversion and fusion of on-site video images. This system is a dedicated system for implementing the above method. The overall architecture is divided into 5 core modules, which are connected in sequence. Refer to the appendix. Figure 4 The architecture diagram, including its components, functions, and connections, is as follows: 1) Data preprocessing module Its functions include acquiring, parsing, and preprocessing basic geographic information data, on-site video image data, and camera parameter data; it has a built-in camera calibration unit that can calculate the intrinsic and extrinsic parameters of cameras with unknown parameters; and it also performs image distortion correction, noise reduction, and keyframe extraction preprocessing operations to provide standardized data input for subsequent modules.

[0051] 2) Image geometric inversion module, with built-in sparse reconstruction unit, dense depth estimation unit, mesh reconstruction and lightweight unit, and semantic annotation unit, such as... Figure 5 As shown.

[0052] Its function is to receive keyframe sequences and camera parameters output by the data preprocessing module, complete 3D sparse reconstruction of the scene, dense depth estimation, mesh reconstruction and lightweighting, semantic annotation, and generate a lightweight 3D proxy scene with attribute annotations.

[0053] 3) Scene fusion modeling module, with built-in coordinate registration unit, global optimization unit, and scene fusion unit.

[0054] Function: Receives basic geographic information scene and 3D proxy scene, constructs a unified coordinate system, completes accurate registration, global optimization and overlay fusion of the two scenes, and generates an enhanced 3D scene model.

[0055] 4) High-precision field of view calculation module, with built-in virtual camera construction unit, hierarchical collision detection unit, and field of view calculation unit.

[0056] Its functions include receiving enhanced 3D scene models, configuring virtual camera parameters, performing macro-micro hierarchical collision detection, completing high-precision visibility calculation, and outputting basic visibility data.

[0057] 5) Results Output and Visualization Module Its function is to receive basic data of the visible field, generate simulation views with realistic textures and overlay heatmaps, and output quantized data of the visible field, simulation reports and vector data, providing data and visualization support for front-end business applications.

[0058] The output of the data preprocessing module is communicatively connected to the inputs of the image geometric inversion module and the scene fusion modeling module, respectively. The input of the image geometric inversion module is communicatively connected to the data preprocessing module, and its output is communicatively connected to the input of the scene fusion modeling module. The input of the scene fusion modeling module is communicatively connected to both the data preprocessing module and the image geometric inversion module, and its output is communicatively connected to the input of the high-precision visual field calculation module. The input of the high-precision visual field calculation module is communicatively connected to the scene fusion modeling module, and its output is communicatively connected to the input of the result output and visualization module. The input of the result output and visualization module is communicatively connected to the high-precision visual field calculation module, and its output is connected to the front-end business application platform.

[0059] On the other hand, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the above-described method for improving the accuracy of visual field simulation.

[0060] On the other hand, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for improving the accuracy of visual field simulation.

[0061] On the other hand, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-mentioned method for improving the accuracy of visual field simulation.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A visual field simulation accuracy enhancement method, characterized in that, Including the following steps: Acquire basic geographic information data and construct a macroscopic 3D scene; Acquire on-site image data and camera parameters, preprocess image data, and provide standardized data input for geometric inversion; A microscopic 3D proxy scene representing the occlusion details of the scene is generated through 3D geometric inversion; The coordinates of the macroscopic 3D scene and the microscopic 3D proxy scene are calibrated and fused to obtain an enhanced 3D scene model; Based on the enhanced 3D scene model, a viewpoint collision detection simulation calculation is performed to generate viewpoint simulation basic data and output the simulation results. The step of generating a microscopic 3D proxy scene representing on-site occlusion details through 3D geometric inversion includes the following steps: Three-dimensional sparse reconstruction is performed on the preprocessed image keyframe sequence to generate a sparse three-dimensional point cloud. Based on the sparse 3D point cloud, dense depth estimation and point cloud optimization are performed to generate a dense 3D point cloud. The dense 3D point cloud is surface reconstructed, and a 3D mesh model of the key occlusion is generated. The mesh model is then lightweighted to generate a microscopic 3D proxy scene representing the details of the occlusion on site.

2. The method of claim 1, wherein, The process of acquiring basic geographic information data and constructing a macroscopic 3D scene includes the following steps: Obtain digital elevation model data and building outline vector data of the target area, and combine them with regional coordinate system parameters to construct a macroscopic basic 3D scene that includes terrain elevation, building outline, and overall spatial layout.

3. The method of claim 1, wherein, The method of generating a microscopic 3D proxy scene representing on-site occlusion details through 3D geometric inversion also includes the following steps: Semantic segmentation and attribute annotation are performed on the models in the microscopic 3D proxy scene.

4. The method according to claim 1 or 2, characterized in that, The coordinate calibration and fusion of the macroscopic 3D scene and the microscopic 3D proxy scene includes the following steps: A unified world coordinate system is constructed based on the engineering coordinate system of basic geographic information data; Based on the camera extrinsic coordinates or scene-named control points, the microscopic 3D proxy scene is registered to the unified world coordinate system; The registered microscopic 3D proxy scene is overlaid and merged with the macroscopic 3D scene. Repeated areas are replaced with the microscopic 3D proxy scene as the reference, and finally an enhanced 3D scene model is generated.

5. The method according to claim 1 or 2, characterized in that, The step of performing view-domain collision detection simulation calculations based on the enhanced 3D scene model includes the following steps: Configure virtual camera parameters in the enhanced 3D scene model; A view ray is generated with the optical center of the virtual camera as the origin, and two-level collision detection, namely macroscopic collision detection and microscopic fine collision detection, is performed on the view ray. The collision detection results of the visual beam are statistically analyzed, the visible area and blind area are marked, and the effective monitoring distance of the visible area is quantified and graded based on the distance of the collision point and the type of occlusion, generating basic data for visual field simulation.

6. The method of claim 1 or 2, wherein, The simulation results include a simulation view with realistic on-site texture mapping, a coverage heatmap, vector data of the visible field of view, and a simulation analysis report.

7. A visual field simulation accuracy enhancement system for implementing the method of any one of claims 1-6, characterized by include: The data preprocessing module is used for the acquisition, parsing, and preprocessing of basic geographic information data, on-site video image data, and camera parameter data; The image geometry inversion module is used to complete 3D sparse reconstruction of the scene, dense depth estimation, mesh reconstruction and lightweighting, semantic annotation, and generate a lightweight 3D proxy scene with attribute annotations. The scene fusion modeling module is used to construct a unified coordinate system, complete the accurate registration, global optimization and overlay fusion of two-layer scenes, and generate an enhanced 3D scene model; The high-precision field of view calculation module is used to configure virtual camera parameters, perform macro-micro hierarchical collision detection, complete high-precision field of view connectivity calculation, and output basic field of view data. The results output and visualization module is used to generate simulation views with realistic on-site texture maps, overlay heatmaps, vector data of the visible field of view, and simulation analysis reports.

8. The system of claim 7, wherein, The output of the data preprocessing module is communicatively connected to the inputs of the image geometric inversion module and the scene fusion modeling module, respectively. The input of the image geometric inversion module is communicatively connected to the data preprocessing module, and its output is communicatively connected to the input of the scene fusion modeling module. The input of the scene fusion modeling module is communicatively connected to both the data preprocessing module and the image geometric inversion module, and its output is communicatively connected to the input of the high-precision visual field calculation module. The input of the high-precision visual field calculation module is communicatively connected to the scene fusion modeling module, and its output is communicatively connected to the input of the result output and visualization module. The input of the result output and visualization module is communicatively connected to the high-precision visual field calculation module, and its output is connected to the front-end business application platform.

9. The system of claim 7 or 8, wherein, The data preprocessing module has a built-in camera calibration unit that calculates intrinsic and extrinsic parameters for cameras with unknown parameters, and completes preprocessing operations such as image distortion removal, noise reduction, and key frame extraction.

10. The system of claim 7 or 8, wherein, The image geometric inversion module includes a sparse reconstruction unit, a dense depth estimation unit, a mesh reconstruction and lightweighting unit, and a semantic annotation unit.

11. The system of claim 7 or 8, wherein, The scene fusion modeling module has a built-in coordinate registration unit, a global optimization unit, and a scene fusion unit.

12. The system according to claim 7 or 8, characterized in that, The high-precision field-of-view calculation module has a built-in virtual camera construction unit, a hierarchical collision detection unit, and a field-of-view calculation unit.

13. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the visual field simulation accuracy improvement method as described in any one of claims 1-6.

14. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the visual field simulation accuracy improvement method as described in any one of claims 1-6.

15. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the visual field simulation accuracy improvement method as described in any one of claims 1-6.

Images