Virtual reality-based engineering project construction safety education interaction method and system

Abstract

The invention relates to the technical field of engineering safety education, in particular to an engineering project construction safety education interaction method and system based on virtual reality. Performing dynamic diffuse reflection voxel tracking of rendering light on the three-dimensional model of the construction scene of the port engineering in combination with the construction progress information and the real-time environment parameters, and performing spatial time sequence fusion with actual scene scanning data to construct a virtual construction scene model of the port engineering; an optical positioning base station and an inertial measurement unit are adopted to capture the limb movement of the trainee in real time, and the joint position of the skeleton is subjected to forward and reverse iterative adjustment, so that the tail end posture of the virtual character approaches to the real-time operation posture feature for mapping; and carrying out registration operation on the real-time operation attitude characteristics and standard operation attitudes in the port construction safety operation attitude database, and judging whether the trainee violates the rule or not through the registration sensitivity of characteristic transformation. According to the invention, a vivid virtual reality scene can be constructed for students to carry out safety education interaction of port engineering construction, and the efficiency and the intelligent level of construction safety training are improved.

Description

Technical Field

[0001] This invention relates to the field of engineering safety education technology, and in particular to an interactive method and system for construction safety education in engineering projects based on virtual reality. Background Technology

[0002] Port construction projects are characterized by complex environments and a concentration of high-risk operations, involving multiple high-risk aspects such as high-altitude operations, heavy machinery operation, dangerous goods loading and unloading, and ship berthing and unberthing coordination. Construction safety issues cannot be ignored.

[0003] Traditional port construction safety education methods have many limitations: First, practical training is costly, requiring significant financial investment in areas such as building physical simulation sites and purchasing training equipment, and equipment maintenance costs are also high. Second, high-risk scenarios are difficult to simulate realistically; for safety reasons, trainees cannot experience extremely dangerous scenarios such as crane overload and breakage, hazardous material leaks and explosions, and falls from heights. Third, training effectiveness is poor; traditional classroom lectures and picture displays are rather abstract, resulting in low trainee participation and difficulty in developing a deep safety awareness and operational memory, which can easily lead to safety accidents due to improper operation in actual construction. Summary of the Invention

[0004] This invention overcomes the shortcomings of existing technologies and provides an interactive method and system for construction safety education in engineering projects based on virtual reality.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The first aspect of this invention provides an interactive method for construction safety education in engineering projects based on virtual reality, comprising the following steps:

[0007] S102: Construct a 3D model of the port project construction scene using BIM modeling software, combine construction progress information and real-time environmental parameters to render the 3D model of the construction scene, perform dynamic diffuse reflection voxel tracking of light, and integrate it with the spatial and temporal data of the actual scene scan to generate a virtual construction scene model of the port project.

[0008] S104: The optical positioning base station and inertial measurement unit are used to capture the trainee's limb movements in real time, obtain the real-time operation posture characteristics, and adjust the joint position of the skeleton through forward and reverse iterations so that the end posture of the virtual character approaches the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character.

[0009] S106: Establish a port construction safety operation posture database, and perform a maximum information gain transformation registration operation between the real-time operation posture characteristics and each standard operation posture in the port construction safety operation posture database. Determine whether the trainee has violated the rules by using the registration sensitivity of the feature transformation. If a violation is found, the corresponding safety warning will be triggered in a timely manner.

[0010] S108: Based on the power energy required by the current operating object in the virtual port construction scene model and the actual operating force, a corresponding force feedback control law is designed. The force feedback and tactile feedback of the force feedback handle are dynamically adjusted according to the force feedback control law.

[0011] S110: Collects real-time eye-tracking orientation datasets of trainees through virtual reality headsets, assesses the trainees' attention distribution in the virtual scene based on the real-time eye-tracking orientation datasets, and generates attention assessment results.

[0012] More specifically, step S102 includes the following steps:

[0013] Obtain construction scene information for port engineering projects, perform scene analysis on the construction scene information using BIM modeling software, and construct a three-dimensional model of the construction scene for port engineering projects.

[0014] By acquiring and measuring the light wavefront gradient at different environmental response levels in the BIM model software based on the construction progress information and real-time environmental parameters of the port project, the propagation of the light wavefront gradient is superimposed using geometric optics theory to form light wave projection matrices at different environmental response levels in the three-dimensional model of the construction scene.

[0015] The 3D model of the voxelized construction scene is a sparse voxel mesh. The rendering light of the voxels is calculated by radiating and injecting the light wave projection matrix into the sparse voxel mesh, and the rendering light voxel field is constructed.

[0016] An octree model is introduced, and its anisotropy is used to perform multi-resolution asymptotic texture construction on the rendered light voxel field, generating several asymptotic texture layers.

[0017] Extract the geometric scalar field of the 3D model of the construction scene from the BIM model software, as well as the viewpoint transformation pixels under the progress perspective where the construction progress information of the geometric scalar field is located. Then, send one or more tracking normal directions to the rendering light voxel field, which are aligned with the scalar guides in the geometric scalar field that record the changes of the construction scene as the construction progress information changes.

[0018] Starting from the viewpoint transformation pixel, the color value and directionality of each rendered light voxel are blurred and sampled in each asymptotic texture layer along the tracking normal direction. This simulates the diffuse reflection propagation mileage of the rendered light on the 3D model of the construction scene, and generates a multi-frame dynamic radiation cache texture where the rendered light is in a continuous temporal sequence when the construction progress information of the 3D model of the construction scene is generated in real time.

[0019] By acquiring actual scene scan data of the port construction scenario through 3D scanning technology, multi-frame dynamic radiation cache textures are superimposed and fused with the actual scene scan data in spatial frequency bands in a temporal sequence to finally generate a virtual construction scene model of the port project.

[0020] More specifically, the process of acquiring actual scene scan data of a port construction scenario using 3D scanning technology, and then temporally overlaying and fusing multi-frame dynamic radiosity buffer maps with the actual scene scan data in spatial frequency bands to ultimately generate a virtual construction scene model of the port project includes the following steps:

[0021] The actual port construction scene was scanned using 3D scanning technology to obtain actual scene scanning data, and at the same time, multiple dynamic radiation cache textures in the time sequence of construction progress information were obtained.

[0022] The Hilbert transform algorithm is introduced to calculate the spatial frequency band of the actual scene scan data located in the 3D model of the construction scene. Each dynamic radiosity buffer texture is offset and modulated to the corresponding spatial frequency band, and a spatial carrier frequency is assigned to the dynamic radiosity buffer texture.

[0023] Based on the spatial carrier frequency, all modulated dynamic radiation buffer maps are time-series superimposed to form a composite hologram. By encoding and modeling the composite hologram, a virtual construction scene model of the port project is finally generated.

[0024] More specifically, in step S104, the joint positions of the skeleton are adjusted iteratively in both forward and reverse directions to make the end-effector posture of the virtual character approach the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character. This specifically includes the following steps:

[0025] Human body knowledge graphs are introduced to perform topology analysis on the real-time limb movements and postures, and a skeletal topology map of the real-time operation posture features is obtained. The position of the intention of the movement is identified through the skeletal topology map to extract the target joint landing point of the student's real-time operation posture features.

[0026] Obtain the position of the virtual character's end joint, calculate the extension distance from the virtual character's end joint position to the target joint landing point, define it as the initial extension ecological distance, and at the same time obtain the total chain length of all bones summed up in the real-time operation posture feature;

[0027] If the initial stretching distance is greater than the total chain length, the skeletal links in the chain are straightened and directed toward the end joints of the virtual character; if the stretching distance is less than the total chain length, the original location of the root joint is saved and marked.

[0028] For a given joint, starting from the original position of the root joint, move each given joint forward in sequence, so that the extension distance between the previous given joint and the given joint phase remains equal to the total chain length. Then update the extension of the previous given joint to obtain a new type of extension position.

[0029] The end joint position is directly placed at the target joint landing point in advance. For a given joint, the real-time operation posture feature joint chain is moved in the reverse direction from the previous given joint to the given joint, so that the extension distance between the given joint and the previous given joint is equal to the total chain length. At this time, the extension of the given joint is updated to obtain the second type of new extension position.

[0030] The extension distance during the update and generation process of the first-class and second-class newborn extension positions is obtained and defined as the current extension ecological distance. If the initial extension ecological distance is less than the total chain length, the real-time operation posture characteristics are mapped to the virtual character according to the first-class and second-class newborn extension positions.

[0031] More specifically, S106 includes the following steps:

[0032] The standard operating postures for different port construction operations are obtained through big data networks, and a database of safe operating postures for port construction is established based on the various standard operating postures.

[0033] Machine learning algorithms are introduced to identify and analyze the movement step length of trainees' real-time limb operation postures in a virtual port construction scenario model and the capture timing of the real-time limb operation postures. Based on the movement step length and capture timing, multiple sets of virtual geometric posture operation motion control points are preset.

[0034] The geometric position of the real-time limb operation posture is determined using the operation motion control points described in each group, and the geometric transformation estimation parameters of the operation motion control points are obtained.

[0035] Based on the geometric feature contour of the standard operating posture characteristics, a planning constraint boundary is preset, and the limit information gain of each standard operating posture characteristic is planned in the port construction safety operating posture database until the planning constraint boundary is reached, forming the maximum directional information gain of the standard operating posture characteristics.

[0036] The real-time limb operation posture features are imported into the port construction safety operation posture database. According to the geometric transformation estimation parameters, they are projected and registered one by one to the corresponding standard operation posture features towards the maximum information gain. The projection registration transformation of each group of operation motion control points is observed, and the Fisher registration information matrix on the operation motion control points is output.

[0037] Based on the operation motion control points, the real-time limb operation posture is divided into N auxiliary registration sub-regions. The registration sensitivity of the geometric transformation estimation parameters of each auxiliary registration sub-region to the standard operation posture features is obtained by Fisher information matrix.

[0038] If the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a high-sensitivity registration region; if the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a low-sensitivity registration region.

[0039] The ratio of high-sensitivity registration areas to low-sensitivity registration areas is considered. If the ratio is lower than a preset threshold, the student's real-time operation is deemed to be non-compliant with safety regulations and constitutes a violation.

[0040] More specifically, S108 includes the following steps:

[0041] The system obtains the current operating object of the trainee in the virtual port construction scene model, and collects the real-time operating force of the trainee when interacting with the current operating object through the force feedback handle. Based on the real-time operating force, the system calculates the end-effector dynamic energy of the trainee's operation and obtains the actual Hamiltonian dynamic function.

[0042] By acquiring the dynamic laws of the current operation object under the conditions of port construction scenario through big data network, and simulating the dynamic state of the virtual port construction scenario model based on the actual Hamiltonian dynamic function in the dynamic laws, a dynamic model of the current operation object of student interaction is constructed and defined as the original dynamic energy structure.

[0043] Obtain the safety education needs for port engineering construction, and extract the prescribed force feedback gradient for different real-time operating forces to achieve different levels of danger based on the safety education needs;

[0044] Based on the predefined force feedback gradient, the expected Hamiltonian dynamic function of the current operating object is preset. According to the expected Hamiltonian dynamic function, the dynamic model of closed-loop feedback with different real-time operating forces is constructed using the laws of dynamics, which is defined as the expected dynamic energy structure.

[0045] By using big data networks, we can obtain a force-feedback mapping table of the applied force that follows the laws of dynamics of the current operating object. We can then query the real-time operating force through the force-feedback mapping table to obtain a set of force feedback parameters that match the real-time operating force.

[0046] A set of force feedback parameters is introduced into the original dynamic energy structure to change the energy shape. During the introduction process, the deviation of the original dynamic energy structure from the desired dynamic energy structure is calculated to obtain the energy structure drift amplitude.

[0047] If the energy structure drift amplitude is greater than the preset amplitude threshold, the adjustment force feedback energy is enhanced; if the energy structure drift amplitude is greater than the preset amplitude threshold, the force feedback energy is weakened, thus forming an energy matching partial differential equation.

[0048] Solve the energy matching partial differential equation to obtain the force feedback control parameters, establish the force feedback control law based on the force feedback control parameters, and upload the force feedback control law to the terminal of the force feedback handle.

[0049] More specifically, S110 includes the following steps:

[0050] The eye-tracking sensor integrated into the virtual reality headset collects the eye movement orientation of the trainees in real time, and obtains a real-time eye movement orientation dataset.

[0051] In the two-dimensional interpolation domain of constructing a virtual construction scene model, the Mahalanobis distance between each real-time eye-tracking azimuth data and its adjacent real-time eye-tracking azimuth data in the real-time eye-tracking azimuth dataset is calculated one by one to obtain several Mahalanobis distances.

[0052] An inverse distance weighting algorithm is introduced. Based on Mahalanobis distance, the spatial influence of each real-time eye-tracking orientation data point pair is defined in the inverse distance weighting algorithm to obtain the inverse distance weighting function. The real-time eye-tracking orientation dataset is then interpolated into a two-dimensional interpolation domain according to the inverse distance weighting function to obtain the real-time eye-tracking orientation heatmap.

[0053] Obtain the distribution pattern of critical and non-critical safety information, and divide the virtual construction scene model into M sub-virtual blocks based on the distribution pattern. Extract the number of hotspots in each sub-virtual block through real-time eye-tracking orientation hotspot map.

[0054] An evaluation radar framework is constructed with key security items and non-key security items as the guide for reading. The number of hotspots is statistically fitted to the evaluation radar framework to obtain the evaluation radar panel with key attention, which is defined as the first evaluation radar panel, and the evaluation radar panel with non-key attention is defined as the second evaluation radar panel.

[0055] Calculate the area of ​​the non-overlapping panel region between the first evaluation radar panel and the second evaluation radar panel to obtain the key attention tendency area value and the non-key attention tendency area value.

[0056] If the area of ​​critical attention tendency is greater than the area of ​​non-critical attention tendency, the student is marked as a student with low safety awareness; otherwise, the student is marked as a student with high safety awareness, and an attention assessment result is generated.

[0057] A second aspect of this invention provides a virtual reality-based interactive system for construction safety education in engineering projects, used to implement any of the virtual reality-based interactive methods for construction safety education in engineering projects. The system specifically includes:

[0058] Virtual Reality Module: The virtual reality module is responsible for constructing a dynamic and realistic virtual construction scene model of the port project by combining construction progress information, real-time environmental parameters, and actual scene scanning data of the port construction scene.

[0059] Optical positioning base station module: The optical positioning base station module is used to continuously emit infrared beams to illuminate the trainee and capture the reflected light, analyze the angle and time difference of the reflected light and other information to accurately calculate the real-time posture of the trainee's limb movements in the virtual construction scene model;

[0060] IMU (Inertial Measurement Unit): The IMU is used to measure the three-axis attitude angular rate and acceleration of the trainee in the virtual construction scene model, and infer the trainee's real-time attitude in conjunction with the optical positioning base station module.

[0061] Force feedback handle: The force feedback handle provides corresponding force feedback according to different port construction operations in the virtual construction scenario model;

[0062] Virtual Reality Headset: The virtual reality headset integrates an eye-tracking sensor for real-time collection of the trainee's eye movement data;

[0063] Intelligent Analysis Module: The intelligent analysis module is used to analyze the relationship between the registration sensitivity and the preset registration sensitivity to determine whether the student's real-time operation has violated any rules;

[0064] Violation warning module: The violation warning module is responsible for triggering corresponding safety warnings in the virtual construction scene model in a timely manner and simulating the possible consequences of violations.

[0065] This invention addresses the technical deficiencies in the prior art, and its beneficial technical effects are as follows:

[0066] This invention achieves highly realistic reproduction of port construction scenes by introducing BIM modeling, voxel tracking of construction progress, and dynamic rendering of lighting conditions under real-time environmental parameters, significantly enhancing the realism and immersion of the virtual scene. High-precision posture capture is achieved through a combination of optical positioning and inertial measurement, and a forward and reverse iterative skeletal mapping algorithm accurately recreates trainees' real-time operational actions, effectively improving the human-computer interaction realism in port safety education. A posture registration method based on maximum information gain transformation enables intelligent recognition and violation judgment of trainees' operational behaviors, enhancing the system's safety education and early warning capabilities. By adaptively generating force feedback control laws based on actual operational load and power requirements, real-time force-tactile feedback linked to trainees' operational behaviors is achieved, improving the educational experience and training effect. Simultaneously, an eye-tracking mechanism is introduced to assess attention distribution, providing a scientific basis for trainee behavior analysis and teaching quality evaluation. Through highly realistic virtual scenes, posture interaction, force feedback, and tactile interaction technologies, this invention creates an immersive safety education environment for trainees, enabling them to fully engage in training and improving their learning enthusiasm and initiative. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0068] Figure 1 A flowchart of the first method of an interactive approach to construction safety education for engineering projects based on virtual reality is shown.

[0069] Figure 2 A flowchart of the second method for interactive construction safety education in engineering projects based on virtual reality is shown.

[0070] Figure 3 A system framework diagram of an interactive system for construction safety education based on virtual reality is shown. Detailed Implementation

[0071] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0072] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0073] The first aspect of this invention provides an interactive method for construction safety education in engineering projects based on virtual reality, such as... Figure 1 As shown, it includes the following steps:

[0074] S102: Construct a 3D model of the port project construction scene using BIM modeling software, combine construction progress information and real-time environmental parameters to render the 3D model of the construction scene, perform dynamic diffuse reflection voxel tracking of light, and integrate it with the spatial and temporal data of the actual scene scan to generate a virtual construction scene model of the port project.

[0075] S104: The optical positioning base station and inertial measurement unit are used to capture the trainee's limb movements in real time, obtain the real-time operation posture characteristics, and adjust the joint position of the skeleton through forward and reverse iterations so that the end posture of the virtual character approaches the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character.

[0076] S106: Establish a port construction safety operation posture database, and perform a maximum information gain transformation registration operation between the real-time operation posture characteristics and each standard operation posture in the port construction safety operation posture database. Determine whether the trainee has violated the rules by using the registration sensitivity of the feature transformation. If a violation is found, the corresponding safety warning will be triggered in a timely manner.

[0077] S108: Based on the power energy required by the current operating object in the virtual port construction scene model and the actual operating force, a corresponding force feedback control law is designed. The force feedback and tactile feedback of the force feedback handle are dynamically adjusted according to the force feedback control law.

[0078] S110: Collects real-time eye-tracking orientation datasets of trainees through virtual reality headsets, assesses the trainees' attention distribution in the virtual scene based on the real-time eye-tracking orientation datasets, and generates attention assessment results.

[0079] It should be noted that the 3D model of the port engineering construction scene covers the wharf shoreline, storage yard, lifting machinery, transport vehicles, various buildings, safety signs, and pipelines. Environmental parameters include weather, tides, and sunlight.

[0080] More specifically, step S102 includes the following steps:

[0081] Obtain construction scene information for port engineering projects, perform scene analysis on the construction scene information using BIM modeling software, and construct a three-dimensional model of the construction scene for port engineering projects.

[0082] By acquiring and measuring the light wavefront gradient at different environmental response levels in the BIM model software based on the construction progress information and real-time environmental parameters of the port project, the propagation of the light wavefront gradient is superimposed using geometric optics theory to form light wave projection matrices at different environmental response levels in the three-dimensional model of the construction scene.

[0083] The 3D model of the voxelized construction scene is a sparse voxel mesh. The rendering light of the voxels is calculated by radiating and injecting the light wave projection matrix into the sparse voxel mesh, and the rendering light voxel field is constructed.

[0084] An octree model is introduced, and its anisotropy is used to perform multi-resolution asymptotic texture construction on the rendered light voxel field, generating several asymptotic texture layers.

[0085] Extract the geometric scalar field of the 3D model of the construction scene from the BIM model software, as well as the viewpoint transformation pixels under the progress perspective where the construction progress information of the geometric scalar field is located. Then, send one or more tracking normal directions to the rendering light voxel field, which are aligned with the scalar guides in the geometric scalar field that record the changes of the construction scene as the construction progress information changes.

[0086] Starting from the viewpoint transformation pixel, the color value and directionality of each rendered light voxel are blurred and sampled in each asymptotic texture layer along the tracking normal direction. This simulates the diffuse reflection propagation mileage of the rendered light on the 3D model of the construction scene, and generates a multi-frame dynamic radiation cache texture where the rendered light is in a continuous temporal sequence when the construction progress information of the 3D model of the construction scene is generated in real time.

[0087] By acquiring actual scene scan data of the port construction scenario through 3D scanning technology, multi-frame dynamic radiation cache textures are superimposed and fused with the actual scene scan data in spatial frequency bands in a temporal sequence to finally generate a virtual construction scene model of the port project.

[0088] It should be noted that due to the complexity of port construction projects, traditional virtual reality (VR) models for port construction struggle to present realistic and immersive scenes. Furthermore, the dynamic changes and disturbances in construction progress and the surrounding real-time environment can lead to localized misalignments and rendering errors in the port construction scene during safety education interactions, potentially obscuring safety information. The distortion inherent in traditional VR scene models significantly reduces the quality of training. To address this, this method uses rendering measurements to determine the gradient projection of light waves propagating layer by layer onto the 3D model of the construction scene as construction progress information and real-time environmental parameters dynamically change. This allows for the detailed decomposition of the overall disturbances in construction progress information and real-time environmental parameters into disturbance trends at different height levels, thus characterizing the dynamic phase changes of light waves jointly exerted on the 3D model of the construction scene by both. This provides a reliable basis for subsequent scene lighting rendering positioning and interference offset elimination in the VR model. Next, the 3D model of the port construction scene is converted into a sparse voxel mesh, containing information such as color, normals, and material properties. The phase radiation of the light wave gradient propagation reflected by the light wave projection matrix is ​​injected into the sparse voxel mesh, enabling the voxel rendering of the 3D model to include static scene lighting information and dynamic contributions from construction progress and real-time environmental parameters. Subsequently, a multi-resolution structure creation of the construction interaction voxel field, namely the asymptotic texture layer, is performed. This asymptotic texture layer is used for blurring and accelerating long-distance sampling, improving sampling efficiency and accelerating lighting calculations, while controlling the virtual reality blurriness of dynamic light wave propagation and disturbances. Starting from the viewpoint transformation pixel of the known geometric scalar field of the construction scene, tracing guide normals are emitted along multiple directions. Blurred sampling and rendering of light voxels along these tracing normals on the asymptotic texture layer simulates the indirect light bounce effect between the construction progress and the real-time environment on the port scene, achieving dynamic and realistic global rendering light perception. This effectively reduces the misalignment and rendering errors caused by external dynamic factors in the virtual reality interaction of the port construction scene, more accurately restoring the realism of the displayed port construction scene. Among them, the rendering light propagation simulation that follows the normal direction can replace a large number of random rays in traditional path tracing, providing a controllable and accelerated real-time rendering method, and improving the stability and occlusion redundancy of light interaction in virtual reality dynamic scenes.

[0089] It should be noted that, regarding the construction steps of the light wave projection matrix, this method obtains the construction progress information and real-time environmental parameters of the port project, sets the sampling perspective based on the construction progress information, and pre-sets different environmental response levels of the 3D model of the construction scene based on the real-time environmental parameters in the documentary monitoring distribution of the sensor array, and assigns a height critical threshold for each environmental response level. This environmental response level establishes the basic framework for the transmission of light waves disturbed by real-time environmental parameters, defines the structure of the understanding space, and the hierarchical division of environmental responses can more accurately simulate actual wavefront disturbances, improving the inversion accuracy of the rendered light. According to the sampling perspective and sampling step size, the phase distribution of each environmental response level under the construction progress information condition reaches the height critical threshold when the equiphase surface after passing through the real-time environmental parameters reaches the height critical threshold in the light wave rendering unit of the BIM model software. This obtains the light wavefront gradient displayed in the 3D model of the construction scene when the real-time environmental parameters interfere with the construction progress information. This light wavefront gradient is the projection information between adjacent environmental response levels of the disturbed light wavefront of the 3D model of the construction scene in the BIM model software. By introducing geometrical optics theory, the wavefront gradient contributions of each layer are superimposed from the wavefront perturbations of adjacent environmental response layers to the termination measurement plane, forming the wavefront projection matrix between different environmental response layers in the 3D model of the construction scene. Geometrical optics theory is used to establish the contribution relationship of the wavefront to the measurement results between each environmental response layer, establishing a potential link between the construction progress and real-time environmental perturbations at each layer and the wavefront observation data. This significantly improves the realism and interactive accuracy of the virtual reality scene model under construction progress and real-time environmental perturbations.

[0090] More specifically, the process of acquiring actual scene scan data of a port construction scenario using 3D scanning technology, and then temporally overlaying and fusing multi-frame dynamic radiosity buffer maps with the actual scene scan data in spatial frequency bands to ultimately generate a virtual construction scene model of the port project includes the following steps:

[0091] The actual port construction scene was scanned using 3D scanning technology to obtain actual scene scanning data, and at the same time, multiple dynamic radiation cache textures in the time sequence of construction progress information were obtained.

[0092] The Hilbert transform algorithm is introduced to calculate the spatial frequency band of the actual scene scan data located in the 3D model of the construction scene. Each dynamic radiosity buffer texture is offset and modulated to the corresponding spatial frequency band, and a spatial carrier frequency is assigned to the dynamic radiosity buffer texture.

[0093] Based on the spatial carrier frequency, all modulated dynamic radiation buffer maps are time-series superimposed to form a composite hologram. By encoding and modeling the composite hologram, a virtual construction scene model of the port project is finally generated.

[0094] It should be noted that traditional interactive methods for virtual reality scene models typically rely on historical data for construction. This limits learners' understanding of port construction simulations to specific, old-fashioned scenarios, making it difficult to adapt to the diverse changes encountered in actual construction. Therefore, integrating real-world scenarios into port construction safety education is crucial. To address this, this method analyzes the spatial changes in the scanned data of the actual scene within the 3D model of the construction scene, i.e., the spatial frequency band. This frequency band reflects the distribution and frequency step size of the actual scene at different directions and locations. By modulating the dynamic radiometric cache texture offset at each spatial temporal time to its corresponding spatial frequency band, the rendering voxels between the dynamic radiometric cache textures make decisive frequency offsets according to the direction and position changes of the actual scene, forming a spatial carrier frequency. This allows the dynamic radiometric cache textures on a holographic map to coexist spatially and temporally without interference, based on the changing trajectory of the actual scene. Finally, by superimposing dynamic radiometric cache textures onto the spatial carrier frequency time sequence, spatial multiplexing encoding is completed, compressing multiple different contents into a single hologram. This modulates the different directional and positional information carried by the actual scene scanning data onto different spatial time sequences displayed in the dynamic radiometric cache texture, achieving holographic fusion of the actual scene scanning data and the 3D model of the construction scene. This constructs a virtual reality scene model with actual scene changes for student interaction, further enhancing the realism of the virtual scene. Through the fusion of actual scene data and scene model using this method, precise matching between the virtual scene and the actual construction environment is achieved, ensuring high-precision interaction of the virtual reality model.

[0095] More specifically, in step S104, the joint positions of the skeleton are adjusted iteratively in both forward and reverse directions to make the end-effector posture of the virtual character approach the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character. This specifically includes the following steps:

[0096] Human body knowledge graph is introduced to perform topology analysis on the real-time operation posture features to obtain the skeletal topology map of the real-time operation posture features. The position of the action intention is identified through the skeletal topology map to extract the target joint landing point of the student performing the real-time operation posture features.

[0097] Obtain the position of the virtual character's end joint, calculate the extension distance from the virtual character's end joint position to the target joint landing point, define it as the initial extension ecological distance, and at the same time obtain the total chain length of the sum of the lengths of all bones of the virtual character;

[0098] If the initial stretching distance is greater than the total chain length, the skeletal links in the chain are straightened and directed toward the target joint landing point; if the stretching distance is less than the total chain length, the original location of the root joint is preserved and marked.

[0099] For a given joint, starting from the original position of the root joint, move each given joint forward sequentially, so that the extension distance between the previous given joint and this given joint remains equal to the total chain length. Then update the extension of the previous given joint to obtain a new type of extension position.

[0100] The end joint position is directly placed at the target joint landing point in advance. For a given joint, the real-time operation posture feature joint chain is moved in the reverse direction from the previous given joint to the given joint, so that the extension distance between the given joint and the previous given joint is equal to the total chain length. At this time, the extension of the given joint is updated to obtain the second type of new extension position.

[0101] The extension distance during the update and generation process of the first-class and second-class newborn extension positions is obtained and defined as the current extension ecological distance. If the initial extension ecological distance is less than the total chain length, the real-time operation posture characteristics are mapped to the virtual character according to the first-class and second-class newborn extension positions.

[0102] It should be noted that, to ensure accuracy and immersion, the physical movements of trainees interacting with the virtual character in a virtual reality (VR) scenario must maintain low latency and real-time responsiveness. However, traditional VR models generally suffer from high latency or discontinuous action responses, and the physical movements do not accurately reflect reality, significantly reducing the reliability of VR safety training. To address this, this method first uses a standardized anthropometric knowledge graph to topologically identify the target joint points where the trainee's real-time operational posture features control the force feedback handle to reach the end of a specified scene or process during actual interaction. These target joint points reflect the target positions that the virtual character's movements require to reach. The end-joint position represents the current resting position of the virtual character within the VR construction scenario model. Next, this method determines whether the initial stretch ecological distance from the current joint position of the virtual character to the target joint landing point is within the maximum stretch range of the real-time joint chain. If the initial stretch ecological distance is greater than the total chain length, the current joint position is too far from the target landing point, and the entire real-time joint chain cannot reach the target. Therefore, the method chooses to directly straighten the joint chain towards the target landing point, ensuring that the length of all joint segments remains unchanged, and constructing a standard state that is closest to the target. This makes the joint skeleton chain of the virtual character as close as possible to the real-time posture target, effectively avoiding unnecessary posture mapping iteration calculations and maintaining relative consistency.

[0103] It should be noted that if the initial extension distance is less than the total chain length, it indicates that the distance to the target landing point is not very far. In this case, this method reconstructs the position of the entire chain from the direction of the root joint point, thereby maintaining the length of each joint segment while making the entire joint chain as close to the target as possible while returning to the fixed root node. Since the position of the root joint point may shift, forward joint position adjustment can fix the root joint point and readjust the entire joint chain of the virtual character according to the real-time operation posture characteristics. This makes the interactive actions of the virtual character in the virtual reality scene model more standardized and closer to the real-time target, improving the accuracy of mapping real-time operation posture characteristics to the virtual character. In addition, this method also updates the extension position of each joint point sequentially from the end joint position of the virtual character towards the root joint point in the chain, so that the extension distance between the previous predetermined joint and the current predetermined joint remains equal to the total chain length. This effectively keeps the joint chain of the virtual character close to the actual posture target while maintaining the length of each joint segment. This allows the end joint of the virtual character to be accurately placed on the target local area, realizing the retrospective adjustment of real-time operation posture mapping to the virtual character's action joints. This method can accurately map the joint positions required for the ends of the skeletal joints of a virtual character to reach the target position, thereby accurately mapping real-time operational posture characteristics onto the virtual character in the virtual scene. This ensures that the virtual character's movements are consistent with the trainee's movements, enabling real-time interaction between the trainee and the virtual scene, reducing interaction delays and inconsistencies in body movements, and improving the immersive experience of trainees' safety education and learning.

[0104] More specifically, S106, as Figure 2 As shown, the specific steps include:

[0105] The standard operating postures for different port construction operations are obtained through big data networks, and a database of safe operating postures for port construction is established based on the various standard operating postures.

[0106] Machine learning algorithms are introduced to identify and analyze the movement step length of trainees' real-time limb operation postures in a virtual port construction scenario model and the capture timing of the real-time limb operation postures. Based on the movement step length and capture timing, multiple sets of virtual geometric posture operation motion control points are preset.

[0107] The geometric position of the real-time limb operation posture is determined using the operation motion control points described in each group, and the geometric transformation estimation parameters of the operation motion control points are obtained.

[0108] Based on the geometric feature contour of the standard operating posture characteristics, a planning constraint boundary is preset, and the limit information gain of each standard operating posture characteristic is planned in the port construction safety operating posture database until the planning constraint boundary is reached, forming the maximum directional information gain of the standard operating posture characteristics.

[0109] The real-time limb operation posture features are imported into the port construction safety operation posture database. According to the geometric transformation estimation parameters, they are projected and registered one by one to the corresponding standard operation posture features towards the maximum information gain. The projection registration transformation of each group of operation motion control points is observed, and the Fisher registration information matrix on the operation motion control points is output.

[0110] Based on the operation motion control points, the real-time limb operation posture is divided into N auxiliary registration sub-regions. The registration sensitivity of the geometric transformation estimation parameters of each auxiliary registration sub-region to the standard operation posture features is obtained by Fisher information matrix.

[0111] If the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a high-sensitivity registration region; if the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a low-sensitivity registration region.

[0112] The ratio of high-sensitivity registration areas to low-sensitivity registration areas is considered. If the ratio is lower than a preset threshold, the student's real-time operation is deemed to be non-compliant with safety regulations and constitutes a violation.

[0113] It should be noted that violations by trainees in the virtual construction scenario model need to be analyzed and captured in real time. Therefore, this method constructs a database of safe operating postures for port construction, storing a large number of standard operating postures for different port construction operations, to compare and judge the real-time operating posture characteristics of trainees. By identifying the step length of the trainee's output real-time operating posture characteristics and the capture sequence of real-time limb operating postures, multiple sets of virtual geometric posture operation motion control points are further preset. The introduction of operation motion control points can virtually linearly combine the geometric positions of real-time operating posture characteristics, so that the parameters that originally needed to be solved for each posture landing point can be realized by inferring the virtual positions of a small number of operation motion control points, which greatly simplifies the degrees of freedom of posture features and the amount of comparison calculation. Traditional feature comparison methods usually calculate the degree of overlap or similarity between actual features and standard features to determine the degree of matching between the two. However, this comparison method is relatively simple and has a large deviation, making it difficult to maximize the accuracy of comparison using standard features as a characterization template. To address this, our method uses the geometric feature contour of the standard operating posture as a boundary constraint to plan a maximum information gain that conforms to the standard operating posture feature, namely the maximum information gain. Under the constraint of this maximum information gain, the real-time operating posture feature comparison can be aligned towards the direction containing the maximum feature information of the standard operating posture feature. Compared with traditional methods, this method can significantly improve the local fineness and global registration performance of feature comparison, ensure the credibility and accuracy of feature comparison results, and improve the reliability of subsequent analysis to determine the student's violation of operation based on the feature comparison results.

[0114] It should be noted that the Fisher registration information matrix at the operation motion control points, also known as the sensitive registration information matrix, measures the sensitivity of the projection registration transformation of the operation motion control points to the standard operation posture features during real-time operation posture feature registration. If the registration sensitivity is greater than the preset registration sensitivity, it means that the registration rejection between the real-time operation posture features and the standard operation posture features in the subordinate registration sub-region is low, and the two features have a large fit, indicating a high local registration rate. Therefore, this subordinate registration region is labeled as a high-sensitivity registration region. Conversely, it means that the feature registration rate of the subordinate registration sub-region is low, and therefore, this subordinate registration region is labeled as a low-sensitivity registration region. Thus, if the ratio of high-sensitivity registration regions to low-sensitivity registration regions is lower than a preset threshold, it means that the trainee's actual operation posture does not match the standard posture in the database, and the trainee's real-time operation does not comply with safety regulations, thus constituting a violation. If any violation occurs, the corresponding safety warning will be triggered in the virtual construction scenario model in a timely manner, such as popping up a prompt box and issuing a warning sound, and simulating the possible consequences of the violation, such as the virtual character falling or the equipment being damaged, to warn, educate, and correct the trainee's wrong operation.

[0115] More specifically, S108 includes the following steps:

[0116] The system obtains the current operating object of the trainee in the virtual port construction scene model, and collects the real-time operating force of the trainee when interacting with the current operating object through the force feedback handle. Based on the real-time operating force, the system calculates the end-effector dynamic energy of the trainee's operation and obtains the actual Hamiltonian dynamic function.

[0117] By acquiring the dynamic laws of the current operation object under the conditions of port construction scenario through big data network, and simulating the dynamic state of the virtual port construction scenario model based on the actual Hamiltonian dynamic function in the dynamic laws, a dynamic model of the current operation object of student interaction is constructed and defined as the original dynamic energy structure.

[0118] Obtain the safety education needs for port engineering construction, and extract the prescribed force feedback gradient for different real-time operating forces to achieve different levels of danger based on the safety education needs;

[0119] Based on the predefined force feedback gradient, the expected Hamiltonian dynamic function of the current operating object is preset. According to the expected Hamiltonian dynamic function, the dynamic model of closed-loop feedback with different real-time operating forces is constructed using the laws of dynamics, which is defined as the expected dynamic energy structure.

[0120] By using big data networks, we can obtain a force-feedback mapping table of the applied force that follows the laws of dynamics of the current operating object. We can then query the real-time operating force through the force-feedback mapping table to obtain a set of force feedback parameters that match the real-time operating force.

[0121] A set of force feedback parameters is introduced into the original dynamic energy structure to change the energy shape. During the introduction process, the deviation of the original dynamic energy structure from the desired dynamic energy structure is calculated to obtain the energy structure drift amplitude.

[0122] If the energy structure drift amplitude is greater than the preset amplitude threshold, the adjustment force feedback energy is enhanced; if the energy structure drift amplitude is greater than the preset amplitude threshold, the force feedback energy is weakened, thus forming an energy matching partial differential equation.

[0123] Solve the energy matching partial differential equation to obtain the force feedback control parameters, establish the force feedback control law based on the force feedback control parameters, and upload the force feedback control law to the terminal of the force feedback handle.

[0124] It should be noted that the force feedback handle integrates force feedback and tactile interaction technologies, enabling learners to provide corresponding force feedback to the objects they interact with in the virtual construction scenario model (such as grasping goods or operating valves), or to provide different operating resistances based on the equipment's operating status (such as lifting weight or travel resistance). However, existing interaction methods struggle to dynamically control and adjust the force feedback handle to apply appropriate force based on the interactive operation content, thus failing to achieve precise force feedback for port construction learning. To address this, this method infers the endpoint energy value of the power output when the learner interacts with the current object in the virtual construction scenario model through real-time operating force, forming a real-time energy feedback loop and providing a basis for subsequent modulation of the power energy structure. Subsequently, based on the actual Hamiltonian dynamic function, the original power energy structure of the learner's interaction with the current object is constructed within the laws of dynamics, revealing the actual internal structure of the power transmission state when the learner operates the force feedback handle. The defined force feedback gradient is a standard force feedback range specification. For example, when a trainee moves goods in a virtual port construction scenario model by manipulating a force feedback handle, the goods will generate a certain amount of mutual gravitational feedback to the trainee, and the movement will create a certain degree of gravitational oscillation. This forms a constantly changing, explorable benchmark gradient for force feedback, namely the defined force feedback gradient. Since this defined force feedback gradient clarifies the expected state of the applied force that the force feedback handle needs to make the trainee perceive, this method constructs the expected dynamic energy structure of closed-loop feedback for different real-time operating forces based on its preset expected Hamiltonian dynamic function.

[0125] It should be noted that introducing the force feedback parameter set into the native kinetic energy structure allows the native system of the force feedback handle to accept external input for planned control, making the interactive energy malleable and realizing the force feedback modulation mechanism required for interaction with the learner, thus providing an adaptive interface for control law design. During the process, the energy structure drift amplitude of the native kinetic energy structure compared to the desired kinetic energy structure is calculated. This energy structure drift amplitude reflects the compliance of the closed-loop system in approaching the desired Hamiltonian dynamic function. Therefore, this energy structure drift amplitude should be minimized to ensure that the interactive closed-loop system of the force feedback handle can reach the desired force feedback state. Thus, if the energy structure drift amplitude is greater than the preset amplitude threshold, it indicates that the kinetic energy of the force feedback parameter set is too small to reach the desired state, and the force feedback energy injection of the interactive closed-loop system can be appropriately increased; conversely, it indicates that the kinetic energy of the force feedback parameter set is too large, resulting in inappropriate force feedback control and deviation from the desired state, requiring a corresponding reduction in the force feedback energy injection. This makes the final constructed force feedback control law more accurate and reliable. This method allows for the analysis and design of control laws based on the trainee's actual operational force in a virtual scenario and the expected force feedback from the current object being manipulated. This enables dynamic adjustment of the force feedback intensity, damping sensitivity, and tactile feedback mode of the force feedback handle. Furthermore, for certain high-risk scenarios, the feedback intensity can be appropriately increased, allowing trainees to more clearly perceive the force and danger of their actions.

[0126] More specifically, S110 includes the following steps:

[0127] The eye-tracking sensor integrated into the virtual reality headset collects the eye movement orientation of the trainees in real time, and obtains a real-time eye movement orientation dataset.

[0128] In the two-dimensional interpolation domain of constructing a virtual construction scene model, the Mahalanobis distance between each real-time eye-tracking azimuth data and its adjacent real-time eye-tracking azimuth data in the real-time eye-tracking azimuth dataset is calculated one by one to obtain several Mahalanobis distances.

[0129] An inverse distance weighting algorithm is introduced. Based on Mahalanobis distance, the spatial influence of each real-time eye-tracking orientation data point pair is defined in the inverse distance weighting algorithm to obtain the inverse distance weighting function. The real-time eye-tracking orientation dataset is then interpolated into a two-dimensional interpolation domain according to the inverse distance weighting function to obtain the real-time eye-tracking orientation heatmap.

[0130] Obtain the distribution pattern of critical and non-critical safety information, and divide the virtual construction scene model into M sub-virtual blocks based on the distribution pattern. Extract the number of hotspots in each sub-virtual block through real-time eye-tracking orientation hotspot map.

[0131] An evaluation radar framework is constructed with key security items and non-key security items as the guide for reading. The number of hotspots is statistically fitted to the evaluation radar framework to obtain the evaluation radar panel with key attention, which is defined as the first evaluation radar panel, and the evaluation radar panel with non-key attention is defined as the second evaluation radar panel.

[0132] Calculate the area of ​​the non-overlapping panel region between the first evaluation radar panel and the second evaluation radar panel to obtain the key attention tendency area value and the non-key attention tendency area value.

[0133] If the area of ​​critical attention tendency is greater than the area of ​​non-critical attention tendency, the student is marked as a student with low safety awareness; otherwise, the student is marked as a student with high safety awareness, and an attention assessment result is generated.

[0134] It should be noted that this method collects real-time eye-tracking datasets of trainees using eye-tracking sensors integrated into the VR headset. These datasets are then interpolated into the two-dimensional space of the virtual construction scene model (i.e., the two-dimensional interpolation domain) using inverse distance weighting calculations, generating a heatmap of the trainees' real-time eye-tracking positions. This heatmap provides a clearer and more intuitive understanding of the distribution of trainees' eye attention, offering a reliable basis for subsequent attention assessment. If the area of ​​focus on critical safety information is greater than that on non-critical safety information, it indicates that the trainee's focus on safety information is lower than their focus on other areas. Critical safety information includes safety signs, equipment status indicators, and hazardous areas, suggesting a weak safety awareness among the trainee. Furthermore, based on the attention assessment results, corresponding attention guidance measures are implemented in the virtual scene. For critical safety information that trainees have not yet noticed, highlighting or flashing prompts are used to attract their attention, improving their sensitivity to safety information and their concentration.

[0135] A second aspect of this invention provides an interactive system for construction safety education in engineering projects based on virtual reality, such as... Figure 3 As shown, the system, applied to implement any of the virtual reality-based interactive methods for construction safety education in engineering projects, specifically includes:

[0136] Virtual Reality Module: The virtual reality module is responsible for constructing a dynamic and realistic virtual construction scene model of the port project by combining construction progress information, real-time environmental parameters, and actual scene scanning data of the port construction scene.

[0137] Optical positioning base station module: The optical positioning base station module is used to continuously emit infrared beams to illuminate the trainee and capture the reflected light, analyze the angle and time difference of the reflected light and other information to accurately calculate the real-time posture of the trainee's limb movements in the virtual construction scene model;

[0138] IMU (Inertial Measurement Unit): The IMU is used to measure the three-axis attitude angular rate and acceleration of the trainee in the virtual construction scene model, and infer the trainee's real-time attitude in conjunction with the optical positioning base station module.

[0139] Force feedback handle: The force feedback handle provides corresponding force feedback according to different port construction operations in the virtual construction scenario model;

[0140] Virtual Reality Headset: The virtual reality headset integrates an eye-tracking sensor for real-time collection of the trainee's eye movement data;

[0141] Intelligent Analysis Module: The intelligent analysis module is used to analyze the relationship between the registration sensitivity and the preset registration sensitivity to determine whether the student's real-time operation has violated any rules;

[0142] Violation warning module: The violation warning module is responsible for triggering corresponding safety warnings in the virtual construction scene model in a timely manner and simulating the possible consequences of violations.

[0143] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An interactive method for construction safety education in engineering projects based on virtual reality, characterized in that, Includes the following steps: S102: Construct a 3D model of the port project construction scene using BIM modeling software, combine construction progress information and real-time environmental parameters to render the 3D model of the construction scene, perform dynamic diffuse reflection voxel tracking of light, and integrate it with the spatial and temporal data of the actual scene scan to generate a virtual construction scene model of the port project. S104: The optical positioning base station and inertial measurement unit are used to capture the trainee's limb movements in real time, obtain the real-time operation posture characteristics, and adjust the joint position of the skeleton through forward and reverse iterations so that the end posture of the virtual character approaches the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character. S106: Establish a port construction safety operation posture database, and perform a maximum information gain transformation registration operation between the real-time operation posture characteristics and each standard operation posture in the port construction safety operation posture database. Determine whether the trainee has violated the rules by using the registration sensitivity of the feature transformation. If a violation is found, the corresponding safety warning will be triggered in a timely manner. S108: Based on the power energy required by the current operating object in the virtual port construction scene model and the actual operating force, a corresponding force feedback control law is designed. The force feedback and tactile feedback of the force feedback handle are dynamically adjusted according to the force feedback control law. S110: Collects real-time eye-tracking orientation datasets of trainees through virtual reality headsets, assesses the trainees' attention distribution in the virtual scene based on the real-time eye-tracking orientation datasets, and generates attention assessment results.

2. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 1, characterized in that, S102 specifically includes the following steps: Obtain construction scene information for port engineering projects, perform scene analysis on the construction scene information using BIM modeling software, and construct a three-dimensional model of the construction scene for port engineering projects. By acquiring and measuring the light wavefront gradient at different environmental response levels in the BIM model software based on the construction progress information and real-time environmental parameters of the port project, the propagation of the light wavefront gradient is superimposed using geometric optics theory to form light wave projection matrices at different environmental response levels in the three-dimensional model of the construction scene. The 3D model of the voxelized construction scene is a sparse voxel mesh. The rendering light of the voxels is calculated by radiating and injecting the light wave projection matrix into the sparse voxel mesh, and the rendering light voxel field is constructed. An octree model is introduced, and its anisotropy is used to perform multi-resolution asymptotic texture construction on the rendered light voxel field, generating several asymptotic texture layers. Extract the geometric scalar field of the 3D model of the construction scene from the BIM model software, as well as the viewpoint transformation pixels under the progress perspective where the construction progress information of the geometric scalar field is located. Then, send one or more tracking normal directions to the rendering light voxel field, which are aligned with the scalar guides in the geometric scalar field that record the changes of the construction scene as the construction progress information changes. Starting from the viewpoint transformation pixel, the color value and directionality of each rendered light voxel are blurred and sampled in each asymptotic texture layer along the tracking normal direction. This simulates the diffuse reflection propagation mileage of the rendered light on the 3D model of the construction scene, and generates a multi-frame dynamic radiation cache texture where the rendered light is in a continuous temporal sequence when the construction progress information of the 3D model of the construction scene is generated in real time. By acquiring actual scene scan data of the port construction scenario through 3D scanning technology, multi-frame dynamic radiation cache textures are superimposed and fused with the actual scene scan data in spatial frequency bands in a temporal sequence to finally generate a virtual construction scene model of the port project.

3. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 2, characterized in that, The process involves acquiring actual scene scan data of a port construction scenario using 3D scanning technology, then temporally overlaying and fusing multi-frame dynamic radiosity buffer maps with the actual scene scan data in spatial frequency bands to ultimately generate a virtual construction scene model for the port project. This process specifically includes the following steps: The actual port construction scene was scanned using 3D scanning technology to obtain actual scene scanning data, and at the same time, multiple dynamic radiation cache textures in the time sequence of construction progress information were obtained. The Hilbert transform algorithm is introduced to calculate the spatial frequency band of the actual scene scan data located in the 3D model of the construction scene. Each dynamic radiosity buffer texture is offset and modulated to the corresponding spatial frequency band, and a spatial carrier frequency is assigned to the dynamic radiosity buffer texture. Based on the spatial carrier frequency, all modulated dynamic radiation buffer maps are time-series superimposed to form a composite hologram. By encoding and modeling the composite hologram, a virtual construction scene model of the port project is finally generated.

4. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 1, characterized in that, In step S104, the joint positions of the skeleton are adjusted iteratively in both forward and reverse directions to make the end-effector posture of the virtual character approach the real-time operation posture characteristics, thereby mapping the trainee's real-time posture to the virtual character. This specifically includes the following steps: Human body knowledge graphs are introduced to perform topology analysis on the real-time limb movements and postures, and a skeletal topology map of the real-time operation posture features is obtained. The position of the intention of the movement is identified through the skeletal topology map to extract the target joint landing point of the student's real-time operation posture features. Obtain the position of the virtual character's end joint, calculate the extension distance from the virtual character's end joint position to the target joint landing point, define it as the initial extension ecological distance, and at the same time obtain the total chain length of all bones summed up in the real-time operation posture feature; If the initial stretching distance is greater than the total chain length, the skeletal links in the chain are straightened and directed toward the end joints of the virtual character; if the stretching distance is less than the total chain length, the original location of the root joint is saved and marked. For a given joint, starting from the original position of the root joint, move each given joint forward in sequence, so that the extension distance between the previous given joint and the given joint phase remains equal to the total chain length. Then update the extension of the previous given joint to obtain a new type of extension position. The end joint position is directly placed at the target joint landing point in advance. For a given joint, the real-time operation posture feature joint chain is moved in the reverse direction from the previous given joint to the given joint, so that the extension distance between the given joint and the previous given joint is equal to the total chain length. At this time, the extension of the given joint is updated to obtain the second type of new extension position. The extension distance during the update and generation process of the first-class and second-class newborn extension positions is obtained and defined as the current extension ecological distance. If the initial extension ecological distance is less than the total chain length, the real-time operation posture characteristics are mapped to the virtual character based on the first-class and second-class newborn extension positions.

5. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 1, characterized in that, S106 specifically includes the following steps: The standard operating postures for different port construction operations are obtained through big data networks, and a database of safe operating postures for port construction is established based on these standard operating postures. Machine learning algorithms are introduced to identify and analyze the movement step length of trainees' real-time limb operation postures in a virtual port construction scenario model and the capture timing of the real-time limb operation postures. Based on the movement step length and capture timing, multiple sets of virtual geometric posture operation motion control points are preset. The geometric position of the real-time limb operation posture is determined using the operation motion control points described in each group, and the geometric transformation estimation parameters of the operation motion control points are obtained. Based on the geometric feature contour of the standard operating posture characteristics, a planning constraint boundary is preset, and the limit information gain of each standard operating posture characteristic is planned in the port construction safety operating posture database until the planning constraint boundary is reached, forming the maximum directional information gain of the standard operating posture characteristics. The real-time limb operation posture features are imported into the port construction safety operation posture database. According to the geometric transformation estimation parameters, they are projected and registered one by one to the corresponding standard operation posture features towards the maximum information gain. The projection registration transformation of each group of operation motion control points is observed, and the Fisher registration information matrix on the operation motion control points is output. Based on the operation motion control points, the real-time limb operation posture is divided into N auxiliary registration sub-regions. The registration sensitivity of the geometric transformation estimation parameters of each auxiliary registration sub-region to the standard operation posture features is obtained by Fisher information matrix. If the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a high-sensitivity registration region; if the registration sensitivity is greater than the preset registration sensitivity, the auxiliary registration region is marked as a low-sensitivity registration region. The ratio of high-sensitivity registration areas to low-sensitivity registration areas is considered. If the ratio is lower than a preset threshold, the student's real-time operation is deemed to be non-compliant with safety regulations and constitutes a violation.

6. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 1, characterized in that, S108 specifically includes the following steps: The system obtains the current operating object of the trainee in the virtual port construction scene model, and collects the real-time operating force of the trainee when interacting with the current operating object through the force feedback handle. Based on the real-time operating force, the end power energy of the trainee's operation is calculated to obtain the actual Hamiltonian dynamic function. By acquiring the dynamic laws of the current operation object under the conditions of port construction scenario through big data network, and simulating the dynamic state of the virtual port construction scenario model based on the actual Hamiltonian dynamic function in the dynamic laws, a dynamic model of the current operation object of student interaction is constructed and defined as the original dynamic energy structure. Obtain the safety education needs for port engineering construction, and extract the prescribed force feedback gradient for different real-time operating forces to achieve different levels of danger based on the safety education needs; Based on the predefined force feedback gradient, the expected Hamiltonian dynamic function of the current operating object is preset. According to the expected Hamiltonian dynamic function, the dynamic model of closed-loop feedback with different real-time operating forces is constructed using the laws of dynamics, which is defined as the expected dynamic energy structure. By using big data networks, we can obtain a force-feedback mapping table of the applied force and the dynamic laws of the current operating object. We can then query the real-time operating force through the force-feedback mapping table to obtain a set of force feedback parameters that match the real-time operating force. A set of force feedback parameters is introduced into the original dynamic energy structure to change the energy shape. During the introduction process, the deviation of the original dynamic energy structure from the desired dynamic energy structure is calculated to obtain the energy structure drift amplitude. If the energy structure drift amplitude is greater than the preset amplitude threshold, the adjustment force feedback energy is enhanced; if the energy structure drift amplitude is greater than the preset amplitude threshold, the force feedback energy is weakened, thus forming an energy matching partial differential equation. Solve the energy matching partial differential equation to obtain the force feedback control parameters, establish the force feedback control law based on the force feedback control parameters, and upload the force feedback control law to the terminal of the force feedback handle.

7. The interactive method for construction safety education in engineering projects based on virtual reality according to claim 1, characterized in that, S110 specifically includes the following steps: The eye-tracking sensor integrated into the virtual reality headset collects the eye movement orientation of the trainees in real time, and obtains a real-time eye movement orientation dataset. In the two-dimensional interpolation domain of constructing a virtual construction scene model, the Mahalanobis distance between each real-time eye-tracking azimuth data and its adjacent real-time eye-tracking azimuth data in the real-time eye-tracking azimuth dataset is calculated one by one to obtain several Mahalanobis distances. An inverse distance weighting algorithm is introduced. Based on Mahalanobis distance, the spatial influence of each real-time eye-tracking orientation data point pair is defined in the inverse distance weighting algorithm to obtain the inverse distance weighting function. The real-time eye-tracking orientation dataset is then interpolated into a two-dimensional interpolation domain according to the inverse distance weighting function to obtain the real-time eye-tracking orientation heatmap. Obtain the distribution pattern of critical and non-critical safety information, and divide the virtual construction scene model into M sub-virtual blocks based on the distribution pattern. Extract the number of hotspots in each sub-virtual block through real-time eye-tracking orientation hotspot map. An evaluation radar framework is constructed with key security items and non-key security items as the guide for reading. The number of hotspots is statistically fitted to the evaluation radar framework to obtain the evaluation radar panel with key attention, which is defined as the first evaluation radar panel, and the evaluation radar panel with non-key attention is defined as the second evaluation radar panel. Calculate the area of ​​the non-overlapping panel region between the first evaluation radar panel and the second evaluation radar panel to obtain the key attention tendency area value and the non-key attention tendency area value. If the area of ​​critical attention tendency is greater than the area of ​​non-critical attention tendency, the student is marked as a student with low safety awareness; otherwise, the student is marked as a student with high safety awareness, and an attention assessment result is generated.

8. A virtual reality-based interactive system for construction safety education in engineering projects, characterized in that, The system, applied to the implementation of the virtual reality-based interactive method for construction safety education in engineering projects as described in any one of claims 1-7, specifically includes: Virtual Reality Module: The virtual reality module is responsible for constructing a dynamic and realistic virtual construction scene model of the port project by combining construction progress information, real-time environmental parameters, and actual scene scanning data of the port construction scene. Optical positioning base station module: The optical positioning base station module is used to continuously emit infrared beams to illuminate the trainee and capture the reflected light, analyze the angle and time difference of the reflected light and other information to accurately calculate the real-time posture of the trainee's limb movements in the virtual construction scene model; IMU (Inertial Measurement Unit): The IMU is used to measure the three-axis attitude angular rate and acceleration of the trainee in the virtual construction scene model, and infer the trainee's real-time attitude in conjunction with the optical positioning base station module. Force feedback handle: The force feedback handle provides corresponding force feedback according to different port construction operations in the virtual construction scenario model; Virtual Reality Headset: The virtual reality headset integrates an eye-tracking sensor for real-time collection of the trainee's eye movement data; Intelligent Analysis Module: The intelligent analysis module is used to analyze the relationship between the registration sensitivity and the preset registration sensitivity to determine whether the student's real-time operation has violated any rules; Violation warning module: The violation warning module is responsible for triggering corresponding safety warnings in the virtual construction scene model in a timely manner and simulating the possible consequences of violations.

Images