A VR 3D visualization simulation system for marine geology that integrates panoramic real-time footage.

Abstract

This invention relates to the field of marine geological 3D visualization technology, specifically disclosing a marine geological VR 3D visualization simulation system that integrates panoramic real-world footage. The system includes a visualization simulation platform, which is communicatively connected to the following modules: a coordinate mapping and alignment module, used to precisely align the spherical UV coordinates of the panoramic real-world footage with the terrain world coordinate system in Unity3D through inverse projection transformation using a unified virtual-real coordinate mapping algorithm, and to smooth the parallax jumps during head rotation using quaternion interpolation. This invention precisely aligns the spherical UV coordinates of the panoramic real-world image with the 3D terrain world coordinate system using a unified virtual-real coordinate mapping algorithm, and uses quaternion spherical linear interpolation to smoothly transition the parallax changes during head rotation, eliminating the visual disconnect between the panoramic background and the 3D terrain, and ensuring the continuity and stability of spatial perception when the user's perspective changes.

Description

Technical Field

[0001] This invention relates to the field of marine geological three-dimensional visualization technology, and in particular to a marine geological VR three-dimensional visualization simulation system that integrates panoramic real-time photography. Background Technology

[0002] With the rapid development of virtual reality and 3D visualization technologies, the field of marine geology is gradually introducing the integration of high-precision simulation and panoramic live-action photography to digitally represent complex marine environments. By combining on-site real-scene image acquisition with 3D modeling technology, it is possible to construct virtual environments with spatial realism of seabed topography, geological structures, and marine survey scenarios. This helps to improve the intuitive presentation of marine geological data, support the spatial structure analysis and process retrospection of marine geological phenomena, and provide immersive visual support for marine geological research, data display, and professional training.

[0003] In existing VR marine geological simulation technologies, the lack of spatial coordinate system fusion between panoramic real-world images and 3D reconstructed terrain leads to parallax jumps when the user's head rotates, disrupting the visual immersion of the VR 3D visualization simulation. Furthermore, without eliminating parallax, the lack of scale and orientation mapping of the 3D simulation equipment within the panoramic background further weakens the spatial interaction consistency of the VR 3D visualization simulation. Simultaneously, the lack of real-time coupling between the dynamic evolution of seabed geology and the user's observation path causes the dynamic process to become disconnected from the surrounding real-world environment, undermining the credibility of the geological process. Therefore, how to maintain a reliable linkage between the dynamic process simulation of the VR 3D visualization simulation of marine geology and the surrounding real-world environment has become a key issue restricting the overall credibility of the system. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a marine geological VR three-dimensional visualization simulation system that integrates panoramic real-time photography, which can effectively solve the problems involved in the prior art.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a marine geological VR 3D visualization simulation system that integrates panoramic real-time photography, including a visualization simulation platform, wherein the visualization simulation platform is communicatively connected to the following modules: The coordinate mapping and alignment module is used to precisely align the spherical UV coordinates of the panoramic real shot with the terrain world coordinate system in Unity3D through inverse projection transformation using a unified virtual and real coordinate mapping algorithm. It also uses quaternion interpolation to smooth the parallax jump when the head rotates, laying a spatial consistency foundation for marine geological visualization simulation, eliminating the visual separation between panoramic images and terrain, and ensuring the uniformity of spatial reference. The spatial anchor point calibration module is based on Unity3D to build a spatial anchor point calibrator. It uses C++ to extract feature points from panoramic images and calculates the scale and rotation offset of the embedded 3D geological model through the perspective projection inverse algorithm. In marine geological visualization simulation, it achieves physical consistency and viewpoint-independent positioning of interactive hotspots, eliminates the lack of scale and orientation mapping, and achieves physical consistency of model embedding. The observation path-driven state machine module is used to build a state machine in C++. It receives the user's viewpoint orientation and dwell time in Unity3D in real time as input, and dynamically switches the geological evolution stage state according to the preset observation path rules. This enables the user's behavior to accurately drive the model's evolution process and avoids the dynamic process from becoming disconnected from the real environment. The LOD deformation and texture fusion module is used to dynamically schedule the multi-level LOD deformation sequence of the 3D geological model according to the trigger signal output by the state machine, to simulate the continuous deformation of the seabed topography, and to fuse the deformation process with the panoramic real-time texture frame by frame to ensure that the model evolution is synchronized with the surrounding real-world feedback. The real-scene closed-loop feedback module is used to analyze the user's gaze and head rotation behavior in real time as trigger signals, driving the panoramic real-shot texture to dynamically update according to the evolution state of the three-dimensional geological model, realizing the synchronous evolution of the real scene and the model, forming a two-way closed-loop synchronization between user gaze drive and real-scene texture evolution.

[0006] Preferably, the coordinate mapping alignment module includes a spherical UV inverse projection transformation unit and a quaternion parallax smoothing interpolation unit; The spherical UV inverse projection transformation unit, based on the unified mapping algorithm of virtual and real coordinates, calculates the spherical UV coordinates of the panoramic real-shot image in real time through the inverse projection transformation matrix written in C++, and converts it into the Unity3D world coordinate system. This establishes a unified spatial mapping benchmark for marine geological visualization simulation, eliminates the original coordinate deviation between the image and the terrain, realizes the accurate mapping from the spherical image to the world coordinate system, and lays the foundation for spatial alignment. The quaternion parallax smoothing interpolation unit is used to smooth the parallax changes during the user's head rotation using the quaternion spherical linear interpolation algorithm, eliminating jumps and tearing sensations, and ensuring the continuity of perspective switching in marine geological visualization simulation.

[0007] Preferably, the spherical UV inverse projection transformation unit specifically includes: Acquire spherical panoramic images captured by panoramic shooting equipment, extract the spherical UV coordinates and corresponding timestamps of each frame image, establish a spherical coordinate sequence with temporal continuity, and provide a continuous data source for subsequent coordinate transformation; Based on the preset terrain world coordinate system in Unity3D, a 4×4 inverse projection homogeneous transformation matrix containing rotation matrix and translation vector is constructed in C++. The spherical UV coordinates are solved in real time into dense spatial point cloud in world coordinate system, and the panoramic image and 3D terrain are coarsely aligned. This achieves spatial alignment of panoramic image and 3D terrain in a unified coordinate system and eliminates the original coordinate deviation between the two. The calculated spatial point cloud coordinates are written to Unity3D's coordinate cache using spatial hash indexes. The UV mapping table of the original terrain map is replaced based on nearest neighbor interpolation to complete the accurate update of the terrain map UV mapping and ensure spatial consistency between the panoramic texture and the terrain mesh.

[0008] Preferably, the quaternion disparity smoothing interpolation unit specifically includes: The system collects quaternion pose data during the user's head rotation in real time, calculates the six-degree-of-freedom disparity offset vector between the current frame and the previous frame through Kalman filtering, identifies the critical frame interval of disparity jump based on the disparity gradient change rate as the target interval for interpolation processing, eliminates disparity abrupt changes during head rotation, and ensures visual continuity. Based on the unified coordinate system output by the spherical UV inverse projection transformation unit, the quaternion spherical linear interpolation algorithm is used to generate intermediate attitude sequences that satisfy the continuity of angular acceleration for the disparity offset in the target interval. This fills in the attitude gaps and kinematically unconstrained intervals between jump frames, fills in the attitude loss caused by jumps, and maintains the natural smoothness of the motion trajectory. The interpolated intermediate pose sequence is written frame by frame into the camera transformation matrix of Unity3D, replacing the original parallax switching logic. This completes the smooth transition of the six-degree-of-freedom parallax change and the consistent output of rigid body motion during head rotation, achieving consistent output of camera pose and rigid body motion, and enhancing the realism of spatial interaction.

[0009] Preferably, the spatial anchor point calibration module includes a panoramic feature point extraction unit and a perspective projection inverse solution positioning unit; The panoramic feature point extraction unit is used in marine geological visualization simulation to automatically extract stable and repeatable feature points from panoramic real-shot images using C++, based on a spatial anchor point calibrator built with Unity3D, and to establish the correspondence between panoramic images and three-dimensional terrain. The perspective projection inverse solution positioning unit, based on the correspondence between the panoramic image and the three-dimensional terrain, calculates the embedding ratio and rotation offset of the three-dimensional geological model in the panoramic background through the perspective projection inverse solution algorithm, so as to achieve physical consistency of model embedding, ensure that the interactive hotspots maintain stable mapping under any viewpoint, realize the accurate embedding of the model in the panoramic background, and eliminate scale orientation deviation.

[0010] Preferably, the panoramic feature point extraction unit specifically includes: The panoramic image frame sequence after processing by the spherical UV inverse projection transformation unit is obtained. The sequence image is loaded into the spatial anchor point calibrator built in Unity3D. The Harris corner point and FAST edge feature in the image are extracted by the multi-scale spatial extremum detection method. An initial candidate feature point set is constructed to improve the robustness and spatial coverage integrity of feature point detection. Multi-scale spatial extremum detection is performed on the candidate feature point set to eliminate pseudo feature points with low contrast and unstable edge response. Stable feature points that can be repeatedly detected are selected as spatial reference marks for anchor point calibration to ensure the repeatability of spatial reference marks under different conditions. A stable bidirectional mapping table of feature points between the panoramic image UV space and the Unity3D world coordinate system is established. After removing mismatched point pairs using the RANSAC algorithm, the correspondence of feature points is output, thus establishing a high-precision spatial mapping relationship and eliminating mismatch interference.

[0011] Preferably, the perspective projection inverse positioning unit specifically includes: The system receives the bidirectional mapping table of feature points output by the panoramic feature point extraction unit, constructs a nonlinear perspective projection model including radial distortion compensation terms, and extends the two-dimensional homogeneous coordinates of feature points in the panoramic image into three-dimensional spatial constraint overdetermined equations by adding depth degrees of freedom, thereby eliminating the interference of lens distortion on spatial mapping and improving the integrity of constraint equations. The Levenberg-Marquardt iterative optimization algorithm is used to solve the overdetermined equations by least squares. The six-degree-of-freedom embedding transformation parameters of the three-dimensional geological model are back-derived using the feature point spatial constraints as boundary conditions. These parameters include the rotation angles around the three coordinate axes, the translations along the three axes, and the anisotropic scaling factor. This achieves the global optimal estimation of the embedding parameters and ensures the spatial positioning accuracy of the model. The obtained embedded transformation parameters are normalized by quaternions and then written into the Unity3D model transformation matrix. At the same time, the transformation residual is output as the initial value for the next frame iteration correction, thereby realizing the inter-frame temporal consistency constraint of the model embedded parameters, suppressing the inter-frame jump of parameters, and maintaining the continuity and stability of the model transformation.

[0012] Preferably, the observation path-driven state machine module specifically includes: In C++, a finite state machine is constructed, defining the state set and state transition conditions for each geological evolution stage. The user's viewpoint orientation range and dwell time threshold are preset as external trigger parameters for state switching, effectively dividing the geological evolution stages and making the state switching conditions clear and controllable. It receives user head posture sensor data in Unity3D in real time, analyzes the current viewpoint orientation and the cumulative dwell time in that orientation, uses the analysis results as input signals for the state machine to determine whether the migration conditions are met, and senses the user's attention area in real time to ensure the timeliness and accuracy of migration condition determination. When the migration conditions are met, the state machine outputs a geological evolution stage switching instruction to the LOD deformation and texture fusion module. The execution of the terrain deformation sequence is triggered by the user's observation path, and the deformation sequence is driven by the user's observation path, so as to realize the dynamic correlation between model evolution and user behavior.

[0013] Preferably, the LOD deformation and texture fusion module specifically includes: Receive the geological evolution stage switching command and historical trajectory signature output by the observation path driven state machine module, and index the corresponding vertex displacement field interpolation key frame from the pre-built quadtree LOD deformation library based on the signature to establish the spatiotemporal correlation constraint between the deformation process and the observation path, and eliminate the temporal misalignment between deformation and the user's observation path. In Unity3D, vertex displacement fields are applied frame by frame to drive the three-dimensional geological model to perform continuous deformation simulation. At the same time, the deformation calculation accuracy is dynamically adjusted according to the LOD level to realize the gradual deformation process of the seabed terrain from coarse to fine, and maintain the continuous evolution of the terrain under a high sense of immersion. The terrain mesh and panoramic real-shot temporal texture of each frame during the deformation process are fused frame by frame according to pixel depth, the texture mapping relationship of the terrain surface is updated, and a 3D geological scene with synchronous deformation and texture updates is output, ensuring the visual consistency between model evolution and environmental texture.

[0014] Preferably, the real-scene closed-loop feedback module specifically includes: Real-time monitoring of user gaze lingering behavior and head rotation trajectory; marking areas where the lingering time exceeds a preset threshold as areas of interest; extracting the spatiotemporal index of the panoramic image corresponding to the areas of interest; achieving accurate local area marking driven by gaze point; reducing the computational load of texture retrieval. Based on the current geological model evolution state output by the LOD deformation and texture fusion module, a texture frame sequence matching the evolution state is retrieved from the panoramic real-shot temporal texture library, and the original texture content in the area of ​​interest is replaced. The texture sequence is accurately matched according to the evolution stage to eliminate the temporal misalignment between texture and geological state. The replaced texture frame sequence is fed back to the Unity3D rendering pipeline, driving the panoramic real-shot texture to dynamically update according to the deformation progress of the 3D geological model, forming a two-way closed-loop synchronization between the real-scene texture and the model evolution, and constructing a two-way closed-loop synchronization between texture and deformation to ensure that the retrieved texture is always up-to-date.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This is a VR 3D visualization simulation system for marine geology that integrates panoramic real-shot images. It uses a unified mapping algorithm of virtual and real coordinates to accurately align the spherical UV coordinates of the panoramic real-shot images with the 3D terrain world coordinate system. Combined with quaternion spherical linear interpolation, it smoothly transitions the parallax changes during head rotation, eliminates the visual disconnect between the panoramic background and the 3D terrain, and ensures the continuity and stability of spatial perception when the user's perspective changes.

[0016] 2. This is a VR 3D visualization simulation system for marine geology that integrates panoramic real-time shooting. It uses a spatial anchor point calibrator to extract stable feature points in the panoramic image and uses a perspective projection inverse solution algorithm to accurately calculate the embedding ratio and rotation offset of the 3D geological model. This ensures that the position, orientation and scale of the 3D geological model in the panoramic background remain physically consistent, and the positioning of interactive hotspots does not shift with changes in viewing angle, significantly improving the reliability of spatial interaction.

[0017] 3. This marine geological VR 3D visualization simulation system integrates panoramic real-time shooting. By constructing a finite state machine to analyze the user's viewpoint orientation and dwell time in real time, the user's observation path is used as the trigger condition for switching geological evolution stages. This enables the user's behavior to accurately drive the terrain deformation process, keeping the geological dynamic process synchronized with the area of ​​user interest, and avoiding the disconnect between dynamic simulation and the real environment.

[0018] 4. This marine geological VR 3D visualization simulation system integrates panoramic real-shot footage. It adopts a LOD deformation and texture fusion mechanism to perform pixel-deep fusion of the continuous deformation simulation of the 3D geological model with the time-series texture of the panoramic real-shot footage. This ensures that the surface texture of the terrain mesh in each frame during the deformation process accurately corresponds to the current deformation state, and outputs a 3D geological scene with synchronous updates of deformation and texture. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the workflow of a marine geological VR 3D visualization simulation system that integrates panoramic real-time photography, as described in this invention. Figure 2 This is a line graph comparing the parallax smoothing effect of head rotation in this invention. Figure 3 This is a scatter plot of the distribution of stable feature points in the panoramic image of the present invention. Figure 4 This is a diagram showing the relationship between LOD levels and computational performance in this invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0021] Example 1, please refer to Figures 1 to 4 This invention provides a technical solution: a marine geological VR 3D visualization simulation system integrating panoramic real-time photography, including a visualization simulation platform, which has the following communication connection modules: The coordinate mapping and alignment module is used to precisely align the spherical UV coordinates of the panoramic real-shot with the terrain world coordinate system in Unity3D through inverse projection transformation using a unified virtual-real coordinate mapping algorithm. It also uses quaternion interpolation to smooth the parallax jump when the head rotates, laying a spatial consistency foundation for marine geological visualization simulation, eliminating the visual separation between panoramic images and terrain, and ensuring the uniformity of spatial reference. The coordinate mapping and alignment module includes a spherical UV inverse projection transformation unit and a quaternion parallax smoothing interpolation unit. The spherical UV inverse projection transformation unit, based on a unified virtual-real coordinate mapping algorithm, calculates the spherical UV coordinates of panoramic real-world images in real time using an inverse projection transformation matrix written in C++, converting them into the Unity3D world coordinate system. This establishes a unified spatial mapping benchmark for marine geological visualization simulation, eliminates the original coordinate deviation between images and terrain, achieves accurate mapping from spherical images to the world coordinate system, and lays the foundation for spatial alignment. It acquires spherical panoramic images captured by panoramic shooting equipment, extracts the spherical UV coordinates and corresponding timestamps of each frame, and establishes a temporally continuous spherical coordinate sequence, providing a continuous data source for subsequent coordinate transformations. Based on Unity... The pre-defined terrain world coordinate system in ty3D is used to construct a 4×4 inverse projection homogeneous transformation matrix containing rotation matrix and translation vector in C++. The spherical UV coordinates are solved in real time into dense spatial point clouds in the world coordinate system, which completes the coarse alignment of panoramic image and 3D terrain. This achieves spatial alignment of panoramic image and 3D terrain in a unified coordinate system, eliminates the original coordinate deviation between the two, and writes the solved spatial point cloud coordinates into the coordinate cache of Unity3D according to the spatial hash index. The UV mapping table of the original terrain map is replaced based on nearest neighbor interpolation to complete the accurate update of terrain map UV mapping and ensure the spatial consistency between panoramic texture and terrain mesh. It should be noted that the acquisition of the spherical panoramic image sequence was carried out by a panoramic imaging device deployed on a marine geological survey vessel. This device employs a six-lens surround optical assembly, outputting isometric cylindrical projection images with a resolution of 7680×3840 pixels at a frame rate of 30 frames per second. For each input image frame, the `remap` function of the OpenCV library was called in the C++ environment to extract the spherical UV coordinates based on the isometric cylindrical projection model. The U coordinate ranges from 0 to 2π radians, and the V coordinate ranges from -π / 2 to π / 2 radians. Simultaneously, the timestamp corresponding to each frame was read from the image metadata, with a timestamp precision of 1 millisecond. A spherical coordinate sequence is constructed using timestamps as indices. This sequence is stored in a double-ended queue data structure with a fixed queue capacity of 300 frames, corresponding to 10 seconds of continuous image data. For each frame, its spherical UV coordinates are pushed into the tail of the queue in chronological order, and header frames exceeding the capacity limit are popped, maintaining a continuous spherical coordinate sequence within a sliding time window. The origin of the default terrain world coordinate system in the Unity3D engine is located at the intersection of sea level and 120°E, 36°N. The positive X-axis points due east, the positive Z-axis points due north, and the positive Y-axis points vertically upwards, with a unit length of 1 meter. A 4×4 inverse projection is constructed using C++. The homogeneous transformation matrix consists of a rotation submatrix and translation vectors. The rotation submatrix is ​​constructed using the ZYX Euler angles sequence, with a rotation angle of 12.5 degrees around the Z-axis, -3.2 degrees around the Y-axis, and 0.8 degrees around the X-axis. These angle values ​​are based on field calibration experiments. The translation vector components are offsets of -25.3 meters along the X-axis, -12.7 meters along the Y-axis, and 8.6 meters along the Z-axis. Substituting each set of UV coordinates from the first segment of the established spherical coordinate sequence into this homogeneous transformation matrix, matrix multiplication is used to solve for a dense spatial point cloud in the world coordinate system in real time. Here, the U coordinate is mapped to the XZ plane azimuth angle, and the V coordinate is mapped... The calculation process, up to the Y-axis height value, is executed cyclically every 0.033 seconds, meaning it processes 30 complete spherical panoramic images per second, outputting a spatial point set containing approximately 7.86 million discrete point clouds, thus achieving coarse alignment between the panoramic image and the 3D terrain. The dense spatial point cloud coordinates generated in the second stage of calculation are written into the Unity3D coordinate buffer using a spatial hash indexing strategy. The spatial hash function employs a 3D mesh partitioning method, dividing the world coordinate system space into cubic voxels with a side length of 0.5 meters. Each voxel is assigned a unique 32-bit unsigned integer hash key value. The hash key value is calculated by dividing the X, Y, and Z components of the point cloud coordinates by 0.Rounded down to the nearest integer after 5 meters, the three integer values ​​are combined using bitwise operations to form a single hash key. For multiple point clouds contained within each voxel, the average of their coordinates is taken as the representative spatial location of that voxel. The coordinate buffer adopts a circular buffer structure with a capacity of 200 voxels, and uses a least recently used eviction policy to manage cache entries. Based on the constructed spatial hash index, nearest neighbor interpolation replacement is performed on the UV mapping table of the original terrain map: traversing each vertex of the terrain mesh, calculating the voxel hash key based on its world coordinates, retrieving the corresponding spatial point cloud coordinates in the coordinate buffer, and replacing the UV coordinate value of that vertex in the original UV mapping table with the retrieved point cloud coordinates. During the replacement process, if a vertex does not have a corresponding point cloud in the buffer, its original UV value is retained, and the replacement is attempted again when the next frame of point cloud data is updated. The quaternion disparity smoothing interpolation unit utilizes a quaternion spherical linear interpolation algorithm to smooth disparity changes during user head rotation, eliminating jumps and tearing sensations, ensuring the continuity of viewpoint switching in marine geological visualization simulations, and improving the stability and realism of spatial perception. It collects quaternion attitude data in real time during user head rotation, calculates the six-degree-of-freedom disparity offset vector between the current frame and the previous frame using Kalman filtering, and identifies the critical frame interval for disparity jumps based on the disparity gradient change rate as the target interval for interpolation processing, eliminating abrupt disparity changes during head rotation and ensuring... To improve visual continuity, based on the unified coordinate system output by the spherical UV inverse projection transformation unit, a quaternion spherical linear interpolation algorithm is used to generate intermediate pose sequences that satisfy angular acceleration continuity for the disparity offset within the target area. This fills in the pose gaps and kinematically unconstrained intervals between jump frames, fills in the pose loss caused by jumps, maintains the natural smoothness of the motion trajectory, and writes the interpolated intermediate pose sequences frame by frame into the camera transformation matrix of Unity3D, replacing the original disparity switching logic of jumps. This completes the smooth transition of six-degree-of-freedom disparity changes and the consistent output of rigid body motion during head rotation, achieving consistent output of camera pose and rigid body motion, and enhancing the realism of spatial interaction. It should be noted that the sampling frequency of the quaternion pose data is set to 90 Hz to keep synchronized with the refresh rate of the Unity3D rendering main loop. The quaternion pose data collected in the current frame and the quaternion pose data in the previous frame are input into a Kalman filter. The state vector of this filter is a six-DOF disparity offset vector, which includes rotational offset angles around the three coordinate axes and translational offsets along the three coordinate axes. The process noise covariance matrix is ​​initialized to 0.001 times the identity matrix, and the observation noise covariance matrix is ​​initialized to 0.01 times the identity matrix. The filter predicts the prior state estimate of the current frame based on the optimal state estimate of the previous frame, and updates the state by combining it with the observation data of the current frame, outputting the filtered six-DOF disparity offset vector. The disparity gradient change rate is calculated using the disparity offset vector, which is the difference between the disparity offset of the current frame and the disparity offset of the previous frame divided by the time interval between the two frames, 0.011 seconds. A disparity jump threshold is set at a rotation angle change greater than 0.5 degrees or a translation change greater than 0.02 meters per frame. When the disparity gradient change rate of three consecutive frames exceeds this threshold, these three frames are marked as the critical frame interval for disparity jump, and this entire interval is used as the target interval for interpolation. The unified coordinate system is defined using a right-handed Cartesian coordinate system, with the origin located at the intersection of sea level and 120 degrees east longitude and 36 degrees north latitude. The X-axis points due east, the Z-axis points due north, and the Y-axis points vertically upward. Quaternion spherical linear interpolation is performed on the disparity offset within the target interval to extract the starting frame of the target interval. The initial quaternion pose and the final quaternion pose of the final frame are used to calculate the cosine of the angle between the initial and final quaternions. If the cosine is less than 0.999, it indicates an interpolable angular difference between the two poses. An intermediate pose sequence is generated between the initial and final quaternions using a spherical linear interpolation formula, with an interpolation step size of 0.0056 seconds. One intermediate pose is inserted between each frame to ensure that the rate of change of angular acceleration between two consecutive frames is controlled within 2.5 degrees per square second. During interpolation, the quaternion modulus is kept at a unit length. Each interpolated intermediate quaternion corresponds to an intermediate pose that satisfies the continuity of angular acceleration. This pose also constrains the rate of change of rotational angular velocity around each coordinate axis, filling the pose gaps between transition frames. The system lacks kinematic constraints and addresses the camera transformation matrix as a 4x4 homogeneous matrix. A 3x3 submatrix in the top left corner stores rotation, and the first three rows of the fourth column store translation. For each intermediate pose, its quaternion is converted into an equivalent 3x3 rotation matrix. The rotation matrix and translation vector are then combined to form the complete camera transformation matrix. This directly replaces the camera component's local transformation matrix via the Unity3D underlying rendering interface. The original parallax switching logic is completely replaced. The new camera transformation matrix writing mechanism ensures kinematic continuity between the camera pose of each frame and the previous frame. The six-degree-of-freedom parallax change during head rotation is smoothly transitioned. The rate of change of rotational angular velocity is constrained to within 0.8 degrees per frame, and the rate of change of translational velocity is constrained to 0 degrees per frame.Within 0.3 meters, the camera attitude sequence is output, which conforms to the rigid body kinematic consistency constraints. The spatial anchor point calibration module is based on Unity3D to build a spatial anchor point calibrator. It uses C++ to extract feature points from panoramic images and calculates the scale and rotation offset of the embedded 3D geological model through the perspective projection inverse algorithm. In marine geological visualization simulation, it achieves physical consistency and viewpoint-independent positioning of interactive hotspots, eliminates the lack of scale and orientation mapping, and achieves physical consistency of model embedding. The spatial anchor point calibration module includes a panoramic feature point extraction unit and a perspective projection inverse positioning unit. The panoramic feature point extraction unit is used in marine geological visualization simulation. Based on the spatial anchor point calibrator built in Unity3D, it automatically extracts stable and repeatable feature points from panoramic real-world images using C++, establishes the correspondence between panoramic images and 3D terrain, obtains the panoramic image frame sequence processed by the spherical UV inverse projection transformation unit, loads the sequence image into the spatial anchor point calibrator built in Unity3D, and uses a multi-scale spatial extremum detection method to extract Harris corner points and FAST edge features in the image, constructs an initial candidate feature point set, improves the robustness and spatial coverage integrity of feature point detection, performs multi-scale spatial extremum detection on the candidate feature point set, removes pseudo feature points with low contrast and unstable edge response, and selects stable feature points that can be repeatedly detected as spatial reference marks for anchor point calibration, ensuring the repeatability of spatial reference marks under different conditions. It establishes a bidirectional mapping table of stable feature points between the panoramic image UV space and the Unity3D world coordinate system, uses the RANSAC algorithm to remove mismatched point pairs, outputs the feature point correspondence, establishes a high-precision spatial mapping relationship, and eliminates mismatch interference. It should be noted that the panoramic image frame sequence has established a coarse alignment between the spherical UV coordinates and the Unity3D world coordinate system. In the spatial anchor point calibrator built in Unity3D, the sequence images are loaded as a two-dimensional texture array. Each frame image has a resolution of 7680×3840 pixels and is stored using an equidistant cylindrical projection format. Multi-scale spatial extremum detection is performed on the loaded images: an eight-layer Gaussian pyramid is constructed, with a scale factor of 1.2 for each group. Harris corner response values ​​and FAST edge feature response values ​​are extracted respectively. The Harris corner detection window size is set to 7×7 pixels, and the FAST edge detection brightness difference threshold is 25. The responses of the two feature points are then compared. Weighted fusion was performed with weights of 0.6 for the Harris contribution factor and 0.4 for the FAST contribution factor. Feature points with response values ​​higher than 120 were retained, while those lower than this threshold were discarded, thus constructing an initial candidate feature point set. Each candidate point recorded its pixel position and corresponding spherical UV coordinates in the image coordinate system. A second multi-scale spatial extremum detection was performed on the initial candidate feature point set to remove false feature points. The candidate feature points were mapped to different layers of a Gaussian pyramid, and the contrast value of each feature point was calculated. Contrast was defined as the sum of the absolute values ​​of the grayscale differences between the feature point and its eight neighboring pixels. A contrast threshold of 35 was set; feature points with contrast values ​​lower than this threshold were marked as low-contrast. For the retained feature points, their edge response stability coefficients are calculated. This coefficient is the ratio of the square of the trace of the Hessian matrix at the feature point to its determinant. The effective range for the stability coefficient is set to 2.5 to 4.0. Feature points outside this range are considered unstable and are discarded. After screening, the remaining feature points are considered stable feature points that can be repeatedly detected and serve as spatial reference markers for anchor point calibration. These feature points can maintain stable detection under different lighting and viewing angle conditions. On average, about 320 stable feature points can be extracted per frame. The uniformity of distribution is characterized by the spatial variation coefficient, which is controlled below 0.35. For each stable feature point, its UV space is recorded. The system records the U and V coordinates of the feature point and the X, Y, and Z coordinates of the feature point in the Unity3D world coordinate system, forming a one-to-one mapping entry. The RANSAC algorithm is used to remove mismatched point pairs: three pairs of mapping points are randomly selected as the minimum sample set, and the initial transformation model is fitted. The threshold for determining the interior point is set to 0.05 pixels, and the maximum number of iterations is 500. After each iteration, the model with the most interior points that match the model is taken as the optimal model. The interior points are retained as the correct matching point pairs, and the exterior points are removed. After RANSAC filtering, the final feature point correspondence is output. The projection residual of each pair of mapping points in this correspondence is controlled within 0.03 pixels. The perspective projection inverse localization unit, based on the correspondence between panoramic images and 3D terrain, calculates the embedding ratio and rotation offset of the 3D geological model in the panoramic background using a perspective projection inverse algorithm. This achieves physical consistency in model embedding, ensuring stable mapping of interactive hotspots from any viewpoint, and enabling precise model embedding in the panoramic background. It eliminates scale and orientation biases, receives a bidirectional feature point mapping table output by the panoramic feature point extraction unit, and constructs a nonlinear perspective projection model including radial distortion compensation terms. The unit extends the 2D homogeneous coordinates of feature points in the panoramic image into 3D spatially constrained overdetermined equations by adding depth degrees of freedom, eliminating the interference of lens distortion on spatial mapping and improving the integrity of the constraint equations. The Levenberg-Marquardt iterative optimization algorithm solves the overdetermined equations using least squares. It uses the spatial constraints of feature points as boundary conditions to inversely deduce the six-degree-of-freedom embedding transformation parameters of the 3D geological model, including the rotation angles around the three coordinate axes, the translations along the three axes, and the anisotropic scaling factor. This achieves the global optimal estimation of the embedding parameters, ensuring the spatial positioning accuracy of the model. The obtained embedding transformation parameters are then normalized by quaternions and written into the model transformation matrix of Unity3D. At the same time, the transformation residuals are output as the initial values ​​for the next frame's iterative correction, thus achieving inter-frame temporal consistency constraints on the model embedding parameters, suppressing inter-frame parameter jumps, and maintaining the continuity and stability of the model transformation. It should be noted that the bidirectional mapping table of feature points records the correspondence between stable feature points in the spherical UV space and the Unity3D world coordinate system. To address the radial distortion of the lens during panoramic image acquisition, a nonlinear perspective projection model with a radial distortion compensation term is constructed. The distortion compensation term uses a third-order radial distortion polynomial with three radial distortion coefficients. The two-dimensional homogeneous coordinates (u, v, 1) of the feature points in the panoramic image are extended into three-dimensional spatial constraint equations by introducing a depth degree-of-freedom parameter λ. Each feature point provides two independent constraint equations, and all feature points form an overdetermined system of equations, with the number of equations being twice the number of feature points. Unknown parameters include six-degree-of-freedom embedded transformation parameters and depth degree-of-freedom parameters. The system of equations possesses redundant constraint characteristics. The Levenberg-Marquardt iterative optimization algorithm adaptively switches between gradient descent and Gauss-Newton methods, controlling the iteration step size by dynamically adjusting the damping factor. The spatial constraints of the feature points are substituted as boundary conditions into the optimization objective function. The term "embedded transformation parameters" refers to the weighted squared error between the predicted projection position and the actual feature point position after embedding the 3D geological model. Through iterative solving, the six-degree-of-freedom (DOF) embedding transformation parameters of the 3D geological model are derived. The parameter set includes rotation angles around the X, Y, and Z axes, translations along the X, Y, and Z axes, and anisotropic scaling factors independently set in the X, Y, and Z directions. The obtained six-DOF embedding transformation parameters are then quaternion-normalized, converting rotation angles to unit-modulus quaternions to eliminate rotation matrix orthogonalization errors, while maintaining the original numerical accuracy of translations and scaling factors. The processed parameters are written into the Unity3D model transformation matrix, replacing the original model's transformation attributes, achieving precise spatial alignment between the 3D geological model and the panoramic image. Simultaneously, the transformation residual, i.e., the objective function value of the optimized final value, is output as the initial damping factor and parameter update direction reference for the next frame's iterative correction. This achieves inter-frame temporal consistency constraints on the model embedding parameters, ensuring the continuity and stability of the embedding transformation parameters between adjacent frames and eliminating inter-frame parameter jumps. The observation path-driven state machine module is used to build a state machine in C++. It receives the user's viewpoint orientation and dwell time in Unity3D in real time as input, and dynamically switches the geological evolution stage state according to the preset observation path rules. This enables the user's behavior to accurately drive the model's evolution process and avoids the dynamic process from becoming disconnected from the real environment. The LOD deformation and texture fusion module is used to dynamically schedule the multi-level LOD deformation sequence of the 3D geological model according to the trigger signal output by the state machine, to simulate the continuous deformation of the seabed topography, and to fuse the deformation process with the panoramic real-time texture frame by frame to ensure that the model evolution is synchronized with the surrounding real-world feedback. The real-scene closed-loop feedback module is used to analyze the user's gaze and head rotation behavior in real time as trigger signals, driving the panoramic real-shot texture to dynamically update according to the evolution state of the three-dimensional geological model, realizing the synchronous evolution of the real scene and the model, forming a two-way closed-loop synchronization between user gaze drive and real-scene texture evolution.

[0022] Example 2, as Figures 1 to 4 As shown, based on Embodiment 1, the present invention provides a technical solution: the observation path driven state machine module specifically includes: constructing a finite state machine in C++, defining the state set and state transition conditions of the geological evolution stage, preset the user viewpoint orientation interval and dwell time threshold as external trigger parameters for state switching, effectively dividing the geological evolution stage, realizing clear and controllable state switching conditions, receiving user head posture sensor data in Unity3D in real time, analyzing the current viewpoint orientation and the cumulative dwell time in that orientation, using the analysis result as the input signal of the state machine, judging whether the transition conditions are met, sensing the user's attention area in real time, ensuring the timeliness and accuracy of the transition condition judgment, when the transition conditions are met, the state machine outputs the geological evolution stage switching instruction to the LOD deformation and texture fusion module, triggering the execution of the terrain deformation sequence with the user observation path as the driving source, driving the deformation sequence with the user observation path, and realizing the dynamic correlation between model evolution and user behavior; It should be noted that the finite state machine built in the C++ environment first initializes and defines the state set of geological evolution stages. This state set is divided into five discrete states based on the marine geological tectonic sequence: initial continental shelf state, continental slope development state, trench subduction state, ocean basin spreading state, and mid-ocean ridge eruption state. The binary criterion for state transition is based on the user's viewpoint azimuth interval and dwell time threshold. The viewpoint azimuth interval is expressed as a spherical angle in the world coordinate system, and the azimuth angle interval is divided into five intervals: 0° to 72°, 72° to 144°, 144° to 216°, 216° to 288°, and 288° to 360°. Each interval establishes a one-to-one correspondence with the five geological states mentioned above. The dwell time threshold... The value is uniformly set to 2.5 seconds. During the initialization process of the state machine, a preset state transition rule table is loaded. This rule table defines a unidirectional transition path from any current state to the target state, prohibiting cross-state jumps to ensure the natural progression of the geological evolution sequence. After the state machine enters the running phase, it receives real-time user head pose sensor data streams from the Unity3D rendering main thread. The data sampling frequency is synchronized with the main rendering loop and fixed at 90 Hz. The quaternion of the head pose input for each frame is parsed and converted into a viewpoint direction vector in the world coordinate system. Then, the horizontal azimuth angle between this direction vector and the reference plane is calculated. At the same time, an independent time accumulator is maintained for each azimuth angle interval. When the user viewpoint is continuous... When the viewpoint continues to fall within the same interval, the time accumulator corresponding to that interval continuously accumulates in increments of 0.011 seconds per frame. Once the viewpoint crosses out of the current interval, the time accumulator of the original interval is immediately reset to zero, and the accumulator of the new interval begins counting. The state machine's decision logic is executed once per rendering frame cycle. When the cumulative dwell time corresponding to a certain azimuth interval reaches a threshold of 2.5 seconds, the state machine determines that the transition condition is met and generates a geological evolution stage switching instruction. After determining that the transition condition is met, the state machine immediately executes the state switching and instruction output operations. First, the state machine updates the current state to the target state and records the system timestamp of the state switching moment, with an accuracy of 1 millisecond. Subsequently, the state machine packages and outputs the switching instruction data packet. The data package contains three fields: target geological stage identifier (an integer from 1 to 5, corresponding to five geological states), timestamp of the switching trigger time, and user historical trajectory signature. The historical trajectory signature is generated by hashing the viewpoint azimuth sequence of the most recent 10 frames. It is a 32-bit unsigned integer and is used by the LOD deformation and texture fusion module to establish spatiotemporal correlation constraints between the deformation process and the observation path. The switching command is sent to the LOD deformation and texture fusion module through the inter-process communication channel, which triggers the module to index the corresponding vertex displacement field interpolation keyframe from the quadtree LOD deformation library. The execution of the terrain deformation sequence is started with the user's observation path as the driving source. The state machine then enters the next monitoring loop, waiting for the triggering of the next migration condition. The LOD deformation and texture fusion module specifically includes: receiving geological evolution stage switching instructions and historical trajectory signatures output by the observation path driven state machine module; indexing the corresponding vertex displacement field interpolation keyframes from the pre-built quadtree LOD deformation library based on the signatures; establishing spatiotemporal correlation constraints between the deformation process and the observation path; eliminating temporal misalignment between deformation and the user's observation path; applying vertex displacement fields frame by frame in Unity3D to drive the 3D geological model to perform continuous deformation simulation; dynamically adjusting the deformation calculation accuracy according to the LOD level to realize the gradual deformation process of the seabed terrain from coarse-grained to fine-grained, maintaining the continuous evolution of terrain under high immersion; fusing the terrain mesh and panoramic real-shot temporal texture of each frame during the deformation process frame by frame according to pixel depth; updating the texture mapping relationship of the terrain surface; and outputting a 3D geological scene with synchronous updates of deformation and texture to ensure the visual consistency between model evolution and environmental texture. It should be noted that after receiving the geological evolution stage switching command and historical trajectory signature output by the observation path-driven state machine module, the corresponding vertex displacement field interpolation keyframes are indexed from the pre-built quadtree LOD deformation library based on the 32-bit unsigned integer hash signature. The root node of the quadtree LOD deformation library corresponds to the entire seabed topography region with a side length of 512 meters. It is recursively divided into four child nodes, each with a side length halved. The recursion depth is set to 5 levels, with the deepest node corresponding to a local topographic block with a side length of 16 meters. Each leaf node stores a set of vertex displacement field interpolation keyframes. The interpolation interval between keyframes is fixed at 0.1 seconds. Each keyframe contains a set of topographic vertex offset vectors at that time point. The component precision of the offset vectors is 0.001 meters. The signature is generated by inputting the viewpoint azimuth sequence of the most recent 10 frames into a hash function, with values ​​evenly distributed between 0 and 2. 32Between -1, the target leaf node position is determined by the remainder of the total number of leaf nodes of the quadrilateral tree based on the signature value modulo the target leaf node. The corresponding vertex displacement field interpolation keyframe sequence is loaded from this node to establish the spatiotemporal correlation constraint between the deformation process and the user's observation path. In the Unity3D rendering main loop, the vertex displacement field is applied frame-by-frame with a fixed time step of 0.033 seconds per frame to drive the 3D geological model for continuous deformation simulation. The total number of vertices in the terrain mesh corresponds to a regular mesh resolution of 1280×1280. The displacement of each vertex is determined by the linear interpolation result of the current frame timestamp in the interpolation keyframe sequence. The LOD level dynamic adjustment mechanism sets a switching threshold based on the viewpoint distance: when the distance to the viewpoint is less than 16 meters, depth level 5 fine deformation calculation is applied, with each vertex participating in 128 iterations; when the distance to the viewpoint is between 16 and 64 meters, depth level 3 deformation is applied, with each vertex participating in 32 iterations; when the distance to the viewpoint is greater than 64 meters, depth level 1 deformation is applied, with each vertex participating in 8 iterations. During the deformation simulation, the vertex displacement field is updated... The process is executed in batches according to the LOD level. In each frame, the vertex data of high-precision areas is updated first, and the vertex data of low-precision areas is gradually refreshed in subsequent frames, realizing the gradual deformation process of the seabed terrain from coarse-grained to fine-grained. The UV mapping table of the terrain mesh has been updated in the spherical UV inverse projection transformation unit. The texture pixel coordinates associated with each vertex are 32-bit floating-point precision, the texture resolution is 7680×3840 pixels, and it is stored in the equidistant cylindrical projection format. The depth fusion operation is performed in the pixel shader stage. For each pixel point covered by the terrain mesh surface, the depth value of the corresponding depth buffer of the pixel is read, and the actual mapping of the seabed depth range is determined. When the depth value is within the effective range, the terrain mesh color at that pixel is output with the panoramic real-shot texture color in a weighted ratio of 1.0:0.0, that is, the terrain mesh color is completely covered in the effective area of ​​the panoramic texture. The fused terrain surface texture mapping relationship is updated once per frame, and the update delay is controlled within 0.033 seconds, outputting a 3D geological scene with deformation and texture updated synchronously. The real-scene closed-loop feedback module specifically includes: real-time monitoring of user gaze lingering behavior and head rotation trajectory; marking areas where the lingering time exceeds a preset threshold as areas of interest; extracting the spatiotemporal index of the panoramic image corresponding to the areas of interest; achieving accurate marking of local areas driven by gaze point; reducing the computational load of texture retrieval; retrieving a texture frame sequence matching the evolution state from the panoramic real-shot temporal texture library based on the current geological model evolution state output by the LOD deformation and texture fusion module; replacing the original texture content in the areas of interest; accurately matching the texture sequence according to the evolution stage; eliminating the temporal misalignment between texture and geological state; feeding back the replaced texture frame sequence to the Unity3D rendering pipeline; driving the panoramic real-shot texture to dynamically update according to the deformation progress of the 3D geological model; forming a two-way closed-loop synchronization between real-scene texture and model evolution; constructing a two-way closed-loop synchronization between texture and deformation; and ensuring that the retrieved texture is always up-to-date. It should be noted that the user's head pose sensor data stream is monitored in real time at a sampling frequency of 90 Hz. The quaternion pose data of each frame is parsed in a C++ environment, converted into a viewpoint direction vector in the world coordinate system, and the intersection coordinates of this direction vector and the terrain mesh are calculated to determine the user's current gaze point. An independent time accumulator is maintained for each terrain patch. When the user's continuous dwell time on a terrain patch reaches a threshold of 2.5 seconds, that terrain patch is marked as a region of interest, and the spatial index coordinates of the region of interest are recorded. These coordinates are jointly determined by the hierarchical encoding of the quadtree LOD and the node number. Simultaneously, the corresponding spatiotemporal index of the region of interest in the panoramic image sequence is extracted. The spatiotemporal index includes the image frame number and pixel block. The system receives two fields: frame number (timestamp with 1 millisecond precision) and pixel block coordinates (boundary values ​​of a rectangular region at 7680×3840 resolution). After marking, the spatiotemporal index of the region of interest is stored in a circular buffer with a capacity of 120 entries, employing a first-in, first-out (FIFO) elimination strategy to ensure that subsequent texture retrieval operations are performed only on the region of interest, avoiding computational redundancy caused by full-frame processing. It also receives the current geological model evolution state parameters output by the LOD deformation and texture fusion module. These parameters include a state identifier, an integer ranging from 1 to 5, corresponding to five evolutionary stages: initial continental shelf, continental slope development, trench subduction, ocean basin expansion, and mid-ocean ridge eruption. Based on this identifier and the region of interest... The spatiotemporal index of the region is retrieved using a two-key search operation from a panoramic real-time temporal texture library. This library employs a B+ tree index structure, with the primary key being the evolutionary stage identifier and the secondary key being the timestamp range. The retrieved texture frame sequence is 30 frames long, corresponding to 1 second of continuous image data. Each frame has a resolution of 7680×3840 pixels and is stored using an equidistant cylindrical projection format. The system performs pixel-level alignment between the retrieved texture frame sequence and the spatial index of the region of interest, with an alignment accuracy controlled within 0.5 pixels. Subsequently, a texture replacement operation is performed, replacing the existing texture content within the region of interest frame by frame with the retrieved new texture sequence. The replacement process is executed at a cycle of 0.033 seconds per frame. After the replacement is complete, the texture map is updated. The update flag of the texture table; the replaced texture frame sequence is encapsulated into a GPU-recognizable texture data packet. Each data packet contains a memory address pointer of 30 frames of texture images and the corresponding update region mask. The mask resolution is 7680×3840 pixels. A value of 1 indicates that the texture of the pixel has been updated, and a value of 0 indicates that the original texture is retained. The texture data packet is submitted to the pixel shader through the Unity3D underlying rendering interface. The shader reads the deformation progress parameter of the current geological model in each rendering cycle. This parameter is a floating-point number between 0 and 1, determined by the interpolation position of the current frame timestamp in the keyframe sequence. The shader dynamically selects the corresponding frame in the texture frame sequence based on the deformation progress parameter, with each frame being 0.The system updates the panoramic real-world texture content within the area of ​​interest every 0.33 seconds, simultaneously writing the replaced texture frame sequence back to the panoramic real-world temporal texture library and updating the historical texture data in that library. This forms a two-way closed-loop synchronization between real-world textures and model evolution, ensuring that subsequent retrieval operations can obtain the latest texture state.

[0023] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A VR 3D visualization simulation system for marine geology integrating panoramic live-action footage, comprising a visualization simulation platform, characterized in that, The visualization simulation platform has the following communication connection modules: The coordinate mapping and alignment module is used to precisely align the spherical UV coordinates of the panoramic real shot with the terrain world coordinate system in Unity3D through inverse projection transformation using a unified virtual and real coordinate mapping algorithm, and to smooth the parallax jump when the head rotates by using quaternion interpolation. The spatial anchor point calibration module is based on Unity3D to build a spatial anchor point calibrator. It uses C++ to extract feature points from panoramic images and calculates the scale and rotation offset of the embedded 3D geological model through the perspective projection inverse solution algorithm to eliminate the lack of scale and orientation mapping. The observation path-driven state machine module is used to build a state machine in C++. It receives the user's viewpoint orientation and dwell time in Unity3D in real time as input, and dynamically switches the geological evolution stage state according to the preset observation path rules, so as to realize the precise driving of the model evolution process by user behavior. The LOD deformation and texture fusion module is used to dynamically schedule the multi-level LOD deformation sequence of the 3D geological model according to the trigger signal output by the state machine, to simulate the continuous deformation of the seabed topography, and to fuse the deformation process with the panoramic real-shot temporal texture frame by frame. The real-scene closed-loop feedback module is used to analyze the user's gaze and head rotation behavior in real time as trigger signals, driving the panoramic real-shot texture to be dynamically updated according to the evolution of the three-dimensional geological model.

2. The marine geology VR three-dimensional visualization simulation system of claim 1, wherein: The coordinate mapping alignment module includes a spherical UV inverse projection transformation unit and a quaternion parallax smoothing interpolation unit. The spherical UV inverse projection transformation unit, based on the unified mapping algorithm of virtual and real coordinates, calculates the spherical UV coordinates of the panoramic real-shot image in real time through the inverse projection transformation matrix written in C++, and converts it into the Unity3D world coordinate system, thus establishing a unified spatial mapping benchmark for marine geological visualization simulation. The quaternion parallax smoothing interpolation unit is used to smooth the parallax changes during the user's head rotation using a quaternion spherical linear interpolation algorithm, eliminating abrupt changes and tearing sensations.

3. The marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 2, characterized in that: The spherical UV inverse projection transformation unit specifically includes: Acquire spherical panoramic images captured by panoramic shooting equipment, extract the spherical UV coordinates and corresponding timestamps of each frame image, and establish a spherical coordinate sequence with temporal continuity. Based on the preset terrain world coordinate system in Unity3D, a 4×4 inverse projection homogeneous transformation matrix containing rotation matrix and translation vector is constructed in C++. The spherical UV coordinates are solved in real time into dense spatial point clouds in the world coordinate system, and the panoramic image and the 3D terrain are coarsely aligned. The calculated spatial point cloud coordinates are written to the Unity3D coordinate cache using spatial hash indexes, and the UV mapping table of the original terrain map is replaced based on nearest neighbor interpolation.

4. The marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 2, characterized in that: The quaternion disparity smoothing interpolation unit specifically includes: Quaternion pose data during the user's head rotation is collected in real time. The six-degree-of-freedom disparity offset vector between the current frame and the previous frame is calculated by Kalman filtering. The critical frame interval of disparity jump is identified based on the disparity gradient change rate as the target interval for interpolation processing. Based on the unified coordinate system output by the spherical UV inverse projection transformation unit, the quaternion spherical linear interpolation algorithm is used to generate intermediate attitude sequences that satisfy the continuity of angular acceleration for the disparity offset in the target interval, filling the attitude gap and kinematically unconstrained interval between jump frames. The interpolated intermediate pose sequence is written frame by frame into the camera transformation matrix of Unity3D, replacing the original parallax switching logic, thus completing the smooth transition of the six-degree-of-freedom parallax change and the output of rigid body motion consistency during the head rotation process.

5. The marine geology VR three-dimensional visualization simulation system of fusing panoramic real-shooting according to claim 2, characterized in that: The spatial anchor point calibration module includes a panoramic feature point extraction unit and a perspective projection inverse solution positioning unit; The panoramic feature point extraction unit is used in marine geological visualization simulation to automatically extract feature points from panoramic real-shot images using C++, based on a spatial anchor point calibrator built with Unity3D, and to establish the correspondence between panoramic images and three-dimensional terrain. The perspective projection inverse solution positioning unit, based on the correspondence between the panoramic image and the three-dimensional terrain, calculates the embedding ratio and rotation offset of the three-dimensional geological model in the panoramic background through the perspective projection inverse solution algorithm, ensuring that the interactive hotspots maintain stable mapping under any viewing angle.

6. The marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 5, characterized in that: The panoramic feature point extraction unit specifically includes: Obtain the panoramic image frame sequence after processing by the spherical UV inverse projection transformation unit, load the sequence image in the spatial anchor point calibrator built in Unity3D, and use the multi-scale spatial extremum detection method to extract Harris corner points and FAST edge features in the image to construct an initial candidate feature point set. Multi-scale spatial extremum detection is performed on the candidate feature point set to eliminate pseudo feature points with low contrast and unstable edge response, and stable feature points that can be repeatedly detected are selected as spatial reference marks for anchor point calibration. A bidirectional mapping table of stable feature points between the panoramic image UV space and the Unity3D world coordinate system is established. After removing mismatched point pairs using the RANSAC algorithm, the correspondence of feature points is output.

7. The VR three-dimensional visualization simulation system for fusing panoramic real shot of marine geology according to claim 5, characterized in that: The perspective projection inverse positioning unit specifically includes: The system receives the bidirectional mapping table of feature points output by the panoramic feature point extraction unit, constructs a nonlinear perspective projection model including radial distortion compensation terms, and extends the two-dimensional homogeneous coordinates of feature points in the panoramic image into three-dimensional spatially constrained overdetermined equations by adding depth degrees of freedom. The Levenberg-Marquardt iterative optimization algorithm is used to solve the overdetermined equations by least squares. The six-degree-of-freedom embedding transformation parameters of the three-dimensional geological model are back-derived using the feature point spatial constraints as boundary conditions. These parameters include the rotation angles around the three coordinate axes, the translations along the three axes, and the anisotropic scaling factor. The obtained embedded transformation parameters are normalized by quaternions and then written into the Unity3D model transformation matrix. At the same time, the transformation residual is output as the initial value for the next frame iteration correction, thus realizing the inter-frame temporal consistency constraint of the model embedded parameters.

8. The marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 5, characterized in that: The observation path-driven state machine module specifically includes: In C++, a finite state machine is constructed, defining the set of states for geological evolution stages and the conditions for state transitions. The user's viewpoint orientation range and dwell time threshold are preset as external trigger parameters for state switching. It receives user head posture sensor data in Unity3D in real time, analyzes the current viewpoint orientation and the cumulative dwell time in that orientation, and uses the analysis result as the input signal of the state machine to determine whether the transition conditions are met. When the migration conditions are met, the state machine outputs a geological evolution stage switching instruction to the LOD deformation and texture fusion module, which triggers the execution of the terrain deformation sequence with the user's observation path as the driving source.

9. A marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 8, characterized in that: The LOD deformation and texture blending module specifically includes: Receive the geological evolution stage switching instructions and historical trajectory signatures output by the observation path driven state machine module, and index the corresponding vertex displacement field interpolation keyframes from the pre-built quadtree LOD deformation library based on the signatures to establish the spatiotemporal correlation constraints between the deformation process and the observation path. In Unity3D, vertex displacement fields are applied frame by frame to drive the three-dimensional geological model to perform continuous deformation simulation, while the deformation calculation accuracy is dynamically adjusted according to the LOD level. The terrain mesh and panoramic real-shot temporal texture of each frame during the deformation process are fused frame by frame according to pixel depth, the texture mapping relationship of the terrain surface is updated, and a three-dimensional geological scene with synchronous deformation and texture updates is output.

10. A marine geological VR three-dimensional visualization simulation system integrating panoramic real-time shooting as described in claim 9, characterized in that: The real-scene closed-loop feedback module specifically includes: Real-time monitoring of user gaze lingering behavior and head rotation trajectory; marking areas where the lingering time exceeds a preset threshold as areas of interest; extracting the spatiotemporal index of the panoramic image corresponding to the areas of interest. Based on the current geological model evolution state output by the LOD deformation and texture fusion module, retrieve the texture frame sequence that matches the evolution state from the panoramic real-shot temporal texture library and replace the original texture content in the area of ​​interest. The replaced texture frame sequence is fed back to the Unity3D rendering pipeline, driving the panoramic real-shot texture to be dynamically updated according to the deformation progress of the 3D geological model, forming a two-way closed-loop synchronization between the real-scene texture and the model evolution.

Images