Underwater three-dimensional reconstruction method based on spatial signal double physical field adaptive constraint

Abstract

This invention provides an underwater 3D reconstruction method based on adaptive constraints of dual-physics fields for spatial signals, comprising: acquiring multi-view RGB image data using underwater image acquisition equipment; establishing 3D Gaussian particles for coordinate points on an initial point cloud; and performing parameterization processing on the 3D Gaussian particles in Cartesian coordinate space to obtain parameter information such as their geometric center, 3D covariance matrix, normal vector, opacity, and color; establishing a variable-space Gaussian particle field to represent and reconstruct the scene based on the variable-space Gaussian particles at infinity in the far-viewing distance space; establishing a Gaussian frequency signal filtering field for near-viewing distance, performing frequency modeling and filtering constraints on Gaussian particles within the near-viewing distance to eliminate 3D Gaussian particle aliasing artifacts and achieve clear reconstruction of near-viewing distance targets in the underwater environment; and establishing an adaptive Gaussian particle opacity optimization model to eliminate redundant distribution of Gaussian particles in the near-viewing distance scattering region and compensate for sparse distribution of Gaussian particles at the far-viewing distance by observing the distance constraints of 3D Gaussian particles through variable-space parameters.

Description

Technical Field

[0001] This invention relates to the field of 3D reconstruction and rendering technology, and more particularly to an underwater 3D reconstruction method based on adaptive constraints of spatial signal dual physics fields. Background Technology

[0002] With the continuous development of fields such as marine exploration, underwater robotics, underwater archaeology, seabed mapping, and underwater engineering inspection, 3D reconstruction of underwater scenes using optical imaging has become an important technical approach for obtaining target geometric structure information and scene environment information. Existing 3D reconstruction methods are mostly based on multi-view images, point cloud modeling, or neural radiation fields to reconstruct target scenes, achieving good reconstruction results in conventional air environments. However, underwater environments are generally plagued by complex imaging interferences such as light absorption, scattering, refraction, and non-uniform illumination, easily leading to problems such as decreased contrast, blurred textures, weakened edges, and color distortion in acquired images. On the one hand, traditional methods are usually based on Cartesian coordinate space for modeling, which makes it difficult to achieve effective coordination between particle parameters and viewing distance under conditions of wide field of view and coexistence of near and far targets, resulting in insufficient representation of distant targets and areas at infinity. On the other hand, near-field regions are prone to aliasing artifacts during imaging sampling, affecting the clear recovery of target edges and texture details. Simultaneously, the Gaussian particle distribution in underwater scattering environments is prone to redundancy, increasing invalid computations and affecting overall rendering efficiency, making it difficult to balance reconstruction fidelity and real-time performance. Therefore, there is an urgent need to propose a new underwater 3D reconstruction technology to solve the problems of insufficient modeling of distant scenes, obvious aliasing in close-range areas, and low rendering efficiency in existing technologies. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an underwater 3D reconstruction method based on adaptive constraints of dual physical fields using spatial signals. The method first initializes the scene using multi-view underwater RGB images and a motion recovery structure algorithm, generating point clouds and establishing a 3D Gaussian particle model. Based on this, a variable-space Gaussian particle field is constructed. By introducing a fourth-dimensional variable-space parameter, a multi-parameter mapping of the 3D Gaussian particles from Cartesian coordinate space to homogeneous coordinate space is achieved, allowing particle position and scale parameters to adaptively change with viewing distance, thereby enhancing the expressive power of wide-field scenes. Simultaneously, a Gaussian frequency signal filter field is constructed for near-viewing distance regions. Frequency constraints are applied to the Gaussian particles based on the sampling characteristics of the imaging system to suppress aliasing artifacts and improve the detail reconstruction effect of near-range targets. Combined with an adaptive opacity optimization model, distance constraints and pruning are applied to the Gaussian particles to reduce redundant particles at close range and improve the sparsity problem of particles at long distances. Finally, by combining spherical harmonic function color mapping and an Alpha-Blending fusion rendering mechanism, high-fidelity 3D reconstruction of complex underwater scenes and high-quality new perspective synthesis are achieved.

[0004] The technical means employed in this invention are as follows:

[0005] An underwater 3D reconstruction method based on adaptive constraints of dual physics fields of spatial signals includes: S1. Use underwater image acquisition equipment to acquire multi-view RGB image data, and use motion recovery structure algorithm to acquire point cloud data generated by initialization. Establish three-dimensional Gaussian particles for the coordinate points on the initial point cloud, and perform parameterization processing on the three-dimensional Gaussian particles in Cartesian coordinate space to obtain parameter information such as geometric center, three-dimensional covariance matrix, normal vector, opacity and color. S2, establish a variable space Gaussian particle field, introduce a fourth-dimensional variable space parameter for three-dimensional Gaussian particles, and use the variable space parameter to perform multi-parameter mapping of three-dimensional Gaussian particles from Cartesian coordinate space to homogeneous coordinate space, and perform scene expression and reconstruction based on variable space Gaussian particles at infinite distance in the far-viewing space. S3. Establish a Gaussian frequency signal filtering field for near-field range, perform frequency modeling and filtering constraints on Gaussian particles within the near-field range, eliminate 3D Gaussian particle aliasing artifacts, and perform clear reconstruction of near-field reconstruction targets in the underwater environment. S4. An adaptive Gaussian particle opacity optimization model is established. By observing the distance constraint of three-dimensional Gaussian particles through variable spatial parameters, the redundant distribution of Gaussian particles in the near-viewing distance scattering region is eliminated, and the sparse distribution of Gaussian particles in the far-viewing distance is compensated. S5 establishes a Gaussian multi-parameter rendering algorithm, which uses spherical harmonic functions to map RGB values ​​in multi-view image data onto Gaussian particles. Combined with parameters such as color and opacity, it performs fusion rendering based on the Alpha-Blending algorithm to finally obtain a new perspective synthetic image.

[0006] Furthermore, the parameter initialization involves obtaining initialized point cloud data from RGB image data acquired by underwater image acquisition equipment combined with a motion structure recovery algorithm. Based on the acquired point cloud data, parameters such as the geometric center of a three-dimensional Gaussian particle and the three-dimensional covariance matrix are defined. The initialized parameter information is then combined with a Gaussian mathematical distribution to form a three-dimensional Gaussian particle.

[0007] Furthermore, the three-dimensional Gaussian particle It internally stores four data types: the position of the particle's center point. That is, the geometric center of the three-dimensional Gaussian particle in Cartesian coordinate space; covariance matrix It contains the scaling and rotation matrices that determine the shape and orientation of the three-dimensional Gaussian particle; the opacity parameter Used for subsequent 3D Gaussian particle rendering; spherical harmonic functions are used to fit the viewpoint-dependent color appearance. 3D Gaussian particles follow a distribution in space: .

[0008] Furthermore, the fourth-dimensional variable space parameter As a switching value for evaluating and characterizing points within the normal myopic distance and points within the hyperopic distance, when When, homogeneous coordinates correspond to the point at infinity, when At that time, homogeneous coordinates correspond to three-dimensional Gaussian particle points within the normal myopic distance, through this variable distance parameter Perform the transformation between the Cartesian coordinate space and the homogeneous coordinate space of a three-dimensional Gaussian particle; Furthermore, the variable-space Gaussian particle field model integrates parameterized information such as the Gaussian geometric center of the three-dimensional Gaussian particles and the three-dimensional covariance matrix with the fourth-dimensional variable-space parameters. This combination allows the position, scaling matrix, and other parameters of the three-dimensional Gaussian particles in the underwater space to be integrated with the variable space parameters. Proportional correlation.

[0009] Furthermore, the Nyquist sampling theorem requires that the sampling frequency must be greater than or equal to twice the highest frequency of the original signal in order to recover the original continuous signal without distortion from the discretely sampled signal. The lowest sampling frequency that satisfies this condition is called the Nyquist frequency.

[0010] Furthermore, the Gaussian frequency signal filtering field for near-field imaging, based on the sampling characteristics of the imaging system, performs Gaussian spatial frequency correlation with the underwater scene depth. The frequency constraint threshold of the Gaussian particles is determined based on the Nyquist sampling theorem, and by constructing a geometric similarity relationship between the two-dimensional imaging space and the three-dimensional world space, the underwater near-field region is... Frequency modeling is performed on the three-dimensional Gaussian particles to determine an adaptive frequency threshold that is compatible with the environment. Based on this threshold, a low-pass filter constraint is applied to the three-dimensional Gaussian particles to eliminate aliasing.

[0011] Furthermore, the adaptive Gaussian particle opacity optimization model establishes adaptive pruning of 3D Gaussian particles in the underwater scene based on the opacity of the 3D Gaussian particles and the observation distance, defining a density dynamic optimization function: when the opacity of the 3D Gaussian particles... At that time, the Gaussian primitive is pruned. Adaptive pruning follows an underwater scene adaptability strategy. The closer the observation distance, the larger the pruning threshold. This rule can effectively eliminate redundant Gaussian primitives in near-field scattering regions, while retaining enough 3D Gaussian primitives in far-field unbounded regions to ensure the modeling fidelity of the entire underwater scene.

[0012] Furthermore, the Gaussian multi-parameter rendering algorithm uses spherical harmonic functions to map the RGB values ​​in multi-view image data onto Gaussian variable space particles, and combines parameters such as color and opacity to perform fusion rendering based on the Alpha-Blending algorithm, ultimately obtaining a new perspective synthetic image.

[0013] Compared with the prior art, the present invention has the following advantages: 1. The underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraint provided by this invention realizes the multi-parameter mapping of 3D Gaussian particles from Cartesian coordinate space to homogeneous coordinate space by constructing a variable spatial Gaussian particle field and introducing a fourth-dimensional variable spatial parameter. This enables the position and scale parameters of the particles to adapt to the viewing distance, thereby effectively improving the modeling ability of near-viewing distance area and far-viewing field scene, and enhancing the overall accuracy and stability of underwater scene reconstruction.

[0014] 2. The underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraint provided by the present invention establishes a Gaussian frequency signal filter field for the near-field region, and adaptively controls Gaussian particles according to the sampling characteristics of the imaging system and the frequency constraint mechanism, which effectively suppresses aliasing artifacts in the near-field imaging process and improves the reconstruction clarity of underwater target edge contours, texture details and local structures.

[0015] 3. The underwater 3D reconstruction method based on spatial signal dual physics field adaptive constraints provided by this invention achieves effective pruning of redundant particles at near-view distance and reasonable compensation for sparse regions at far-view distance by establishing an adaptive Gaussian particle opacity optimization model. Under the premise of ensuring the fidelity of underwater scene modeling and the quality of new perspective synthesis, it reduces invalid computational overhead and improves overall rendering efficiency.

[0016] In summary, the technical solution of this invention improves upon existing 3D reconstruction and rendering technologies in practical applications, addressing issues such as poor scene geometric fidelity, appearance modeling distortion, and detail loss during wide-field 3D reconstruction. It achieves high-precision fitting of complex geometry and edge structures on object surfaces, optimizes the scaling limitations of wide-field scenes for near and far targets, and significantly improves rendering quality and efficiency.

[0017] Based on the above reasons, this invention can be widely applied in the field of underwater long-range 3D reconstruction and rendering technology. Attached Figure Description

[0018] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This invention relates to an underwater 3D reconstruction method based on adaptive constraints of dual physical fields of spatial signals.

[0020] Figure 2 This is a flowchart illustrating the underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraints according to the present invention.

[0021] Figure 3 This is a schematic diagram of the variable spatial Gaussian particle field in the underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraints of the present invention.

[0022] Figure 4 This is a schematic diagram of the Gaussian frequency signal filtering field in the underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraints of the present invention.

[0023] Figure 5 This is a schematic diagram comparing the rendered images in underwater space using the underwater 3D reconstruction method based on spatial signal dual-physics field adaptive constraints according to the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] like Figure 1 As shown, this invention provides an underwater 3D reconstruction method based on adaptive constraints of dual physics fields of spatial signals. A flowchart of this method is shown below. Figure 2 As shown, it includes: Underwater image acquisition equipment is used to acquire multi-view RGB image data. Based on the structure-of-motion regression algorithm, point cloud data is obtained to initialize the data. Three-dimensional Gaussian particles are established for the coordinate points on the initialized point cloud. The three-dimensional Gaussian particles are parameterized in Cartesian coordinate space to obtain their geometric center, three-dimensional covariance matrix, normal vector, opacity and color parameters. In a specific implementation, as a preferred embodiment of the present invention, the parameter initialization is achieved by combining the RGB image data obtained by the underwater image acquisition equipment with the motion structure recovery algorithm to obtain initialized point cloud data. Based on the obtained point cloud data, parameters such as the geometric center of the three-dimensional Gaussian particle and the three-dimensional covariance matrix are defined, and the initialized parameter information is combined with the Gaussian mathematical distribution to form a three-dimensional Gaussian particle.

[0027] The three-dimensional Gaussian particles It internally stores four data types: the position of the particle's center point. That is, the geometric center of the three-dimensional Gaussian particle in Cartesian coordinate space; covariance matrix It contains the scaling and rotation matrices that determine the shape and orientation of the three-dimensional Gaussian particle; the opacity parameter Used for subsequent 3D Gaussian particle rendering; spherical harmonic functions are used to fit the viewpoint-dependent color appearance. 3D Gaussian particles follow a distribution in space: .

[0028] S2, as Figure 3 As shown, a variable space Gaussian particle field is established, and a fourth-dimensional variable space parameter is introduced for the three-dimensional Gaussian particles. The variable space parameter is used to perform multi-parameter mapping of the three-dimensional Gaussian particles from Cartesian coordinate space to homogeneous coordinate space, and to express and reconstruct the scene based on the variable space Gaussian particles at infinite distance in the far-viewing space. In a specific implementation, as a preferred embodiment of the present invention, the fourth-dimensional variable spatial parameter serves as a switching value for evaluating and characterizing points within conventional myopic distances and points within hyperopic distances. When, homogeneous coordinates correspond to the point at infinity, when At that time, homogeneous coordinates correspond to three-dimensional Gaussian particle points within the normal myopic distance, through this variable distance parameter Perform the transformation between the Cartesian coordinate space and the homogeneous coordinate space of a three-dimensional Gaussian particle.

[0029] The variable-space Gaussian particle field model integrates parameterized information such as the Gaussian geometric center of the three-dimensional Gaussian particles and the three-dimensional covariance matrix with the fourth-dimensional variable-space parameters. This combination allows the position, scaling matrix, and other parameters of the three-dimensional Gaussian particles in the underwater space to be integrated with the variable space parameters. Proportional correlation.

[0030] S3, as Figure 4 As shown, a Gaussian frequency signal filtering field for near-field observation is established. Frequency modeling and filtering constraints are performed on Gaussian particles within the near-field observation range to eliminate the aliasing artifacts of three-dimensional Gaussian particles and achieve clear reconstruction of near-field reconstructed targets in the underwater environment. In a specific implementation, as a preferred embodiment of the present invention, the Nyquist sampling theorem requires that the sampling frequency must be greater than or equal to twice the highest frequency of the original signal when recovering the original continuous signal without distortion from the discretely sampled signal. The lowest sampling frequency that satisfies this condition is called the Nyquist frequency.

[0031] The Gaussian frequency signal filtering field for near-field imaging correlates the underwater scene depth with Gaussian spatial frequency based on the sampling characteristics of the imaging system. The frequency constraint threshold for Gaussian particles is determined based on the Nyquist sampling theorem. By constructing a geometric similarity relationship between the two-dimensional imaging space and the three-dimensional world space, the near-field underwater region is analyzed. Frequency modeling is performed on the three-dimensional Gaussian particles to determine an adaptive frequency threshold that is compatible with the environment. Based on this threshold, a low-pass filter constraint is applied to the three-dimensional Gaussian particles to eliminate aliasing.

[0032] S4. An adaptive Gaussian particle opacity optimization model is established. By observing the distance constraint of three-dimensional Gaussian particles through variable spatial parameters, the redundant distribution of Gaussian particles in the near-viewing distance scattering region is eliminated, and the sparse distribution of Gaussian particles in the far-viewing distance is compensated. In a specific implementation, as a preferred embodiment of the present invention, the adaptive Gaussian particle opacity optimization model establishes adaptive pruning of the three-dimensional Gaussian particles in the underwater scene based on the opacity of the three-dimensional Gaussian particles and the observation distance, and defines an opacity dynamic optimization function: when the opacity of the three-dimensional Gaussian particles... At that time, the Gaussian primitive is pruned. Adaptive pruning follows an underwater scene adaptability strategy. The closer the observation distance, the larger the pruning threshold. This rule can effectively eliminate redundant Gaussian primitives in near-field scattering regions, while retaining enough 3D Gaussian primitives in far-field unbounded regions to ensure the modeling fidelity of the entire underwater scene.

[0033] S5. Establish a Gaussian multi-parameter rendering algorithm, using spherical harmonic functions to map the RGB values ​​in multi-view image data to Gaussian particles, and combine parameters such as color and opacity to perform fusion rendering based on the Alpha-Blending algorithm to finally obtain a new perspective synthetic image.

[0034] The Gaussian multi-parameter rendering algorithm utilizes spherical harmonic functions to map RGB values ​​from multi-view image data onto Gaussian variable-space particles. Combined with parameters such as color and opacity, it performs fusion rendering based on the Alpha-Blending algorithm, ultimately obtaining a composite image from a new perspective. The composite image result is as follows: Figure 5 As shown.

[0035] Example First, three-dimensional Gaussian particles are used as the reconstructed particles for the underwater scene. A set of three-dimensional Gaussian particles is defined. Each three-dimensional Gaussian particle Opacity Geometric parameterization of the 3D Gaussian is performed, and the geometric center of the 3D Gaussian is defined. Scaling matrix Rotation matrix 3D covariance matrix Parameterization Three-dimensional Gaussian particles follow a distribution in space:

[0036] Extending the Cartesian coordinate space containing the Gaussian particle to a homogeneous coordinate space, the point in the Cartesian coordinate space... In homogeneous coordinate space, it is represented as ,Right now

[0037] in As a variable spatial parameter, it serves as a switch value for evaluating and characterizing points within the normal viewing distance and points at infinity. It is used to represent translation and projection through matrix and vector multiplication. When, homogeneous coordinates correspond to the point at infinity, when At this time, homogeneous coordinates correspond to a three-dimensional Gaussian particle point within a normal near-vision distance, and the homogeneous coordinates are simplified to:

[0038] Parameter mapping is performed on the positions of three-dimensional Gaussian particles, and the three-dimensional covariance matrix is... scaling matrix in The transformation from Cartesian coordinate space to homogeneous coordinate space is represented as follows: ,Right now

[0039] When filtering Gaussian frequency signals at near-field distances, a geometric similarity relationship is established by projecting from the two-dimensional imaging space to the three-dimensional world space. In the two-dimensional imaging space, the focal length of the underwater serialized image is defined as... The sampling interval is 1 (pixel unit). In three-dimensional world space, depth The sampling interval corresponding to the point at that location Represented as the physical size of a pixel in three-dimensional space, and thus geometrically similar after projection: ,Right now Establish sampling frequency in three-dimensional world space .

[0040] According to the Nyquist sampling theorem, the maximum frequency that can be accurately sampled in the absence of aliasing is half the sampling frequency. Therefore, for each 3D Gaussian element in 3D world space... Its frequency threshold is: .

[0041] For each three-dimensional Gaussian element Define Gaussian frequency: ,in The total number of underwater sequence images. For the first Three-dimensional Gaussian elements of a camera's view frustum sampling frequency, Let be the visibility evaluation function of the three-dimensional Gaussian primitive, when the... Gaussian geometric center Located in the The value is 1 (visible) when the camera is within its field of view cone, and 0 (invisible) otherwise.

[0042] To address the sampling frequency fluctuations caused by underwater light attenuation and image noise, a frequency threshold was set. Set to the maximum value among all visible viewpoints to ensure that at least one camera is in 3D high-order primitives. Achieve accurate sampling.

[0043] Establishing Gaussian frequencies in the underwater environment With threshold Then, a low-pass filter was introduced. For three-dimensional Gaussian elements Regularization and constraints are applied to ensure Gaussian frequency. Not exceeding the threshold .

[0044] Increase filter covariance To limit the highest frequency of the three-dimensional Gaussian element to no more than This is to reduce 3D Gaussian aliasing caused by continuous fluctuations in sensor observation and imaging distances in underwater environments. To control the hyperparameters of the filter scale, It is a three-dimensional identity matrix, ensuring the isotropic characteristics of the low-pass filter.

[0045] For three-dimensional Gaussian elements Apply low-pass filtering:

[0046] in The covariance of the frequency-constrained 3D Gaussian elements in space can be defined as follows: .

[0047] Therefore, the frequency-constrained three-dimensional Gaussian element is... The spatial distribution characteristics are expressed as:

[0048] in As a normalization term, it ensures that the Gaussian integral value remains 1 after convolution. To address the redundancy and sparse distribution of Gaussian elements caused by underwater medium scattering, an adaptive Gaussian opacity optimization strategy incorporating distance awareness is introduced.

[0049] Opacity based on 3D Gaussian primitives and observation distance An adaptive pruning mechanism for 3D Gaussian primitives in underwater scenes is established, and a density dynamic optimization function is defined. :

[0050] in The base pruning threshold for the original 3DGS is 0.005. The feature distance for threshold enhancement (10m). These are the weighting coefficients.

[0051] When the opacity of the three-dimensional Gaussian primitive At that time, the Gaussian primitive is pruned. Adaptive pruning follows an underwater scene adaptability strategy. The closer the observation distance, the higher the pruning threshold. This rule can effectively eliminate redundant Gaussian primitives in near-field scattering regions, while retaining enough 3D Gaussian primitives in far-field unbounded regions to ensure the modeling fidelity of the entire underwater scene.

[0052] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0053] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0055] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0056] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0057] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An underwater 3D reconstruction method based on adaptive constraints of dual physics fields of spatial signals, characterized in that, include: Underwater image acquisition equipment is used to acquire multi-view RGB image data. Based on the structure-of-motion regression algorithm, point cloud data is obtained to initialize the data. Three-dimensional Gaussian particles are established for the coordinate points on the initialized point cloud. The three-dimensional Gaussian particles are parameterized in Cartesian coordinate space to obtain their geometric center, three-dimensional covariance matrix, normal vector, opacity and color parameter information. A variable-space Gaussian particle field is established, and a fourth-dimensional variable-space parameter is introduced for the three-dimensional Gaussian particles. The variable-space parameter is used to perform multi-parameter mapping of the three-dimensional Gaussian particles from Cartesian coordinate space to homogeneous coordinate space. Scene representation and reconstruction based on variable-space Gaussian particles are performed at infinite distance in the far-viewing space. A Gaussian frequency signal filtering field for near-field observation is established. Frequency modeling and filtering constraints are performed on Gaussian particles within the near-field observation range to eliminate three-dimensional Gaussian particle aliasing artifacts and achieve clear reconstruction of near-field reconstructed targets in the underwater environment. An adaptive Gaussian particle opacity optimization model is established. By observing the distance constraint of three-dimensional Gaussian particles through variable spatial parameters, the redundant distribution of Gaussian particles in the near-view distance scattering region is eliminated, and the sparse distribution of Gaussian particles in the far-view distance is compensated. A Gaussian multi-parameter rendering algorithm is established, which uses spherical harmonic functions to map the RGB values ​​in multi-view image data onto Gaussian particles. Combined with color and opacity parameters, the algorithm is used for fusion rendering to finally obtain a new perspective synthetic image.

2. The underwater 3D reconstruction method based on spatial signal dual-physics adaptive constraints according to claim 1, characterized in that: The parameter initialization is achieved by combining RGB image data acquired by underwater image acquisition equipment with motion structure recovery algorithm to obtain initialized point cloud data. Based on the acquired point cloud data, parameters such as the geometric center of three-dimensional Gaussian particles and three-dimensional covariance matrix are defined. The initialized parameter information is combined with Gaussian mathematical distribution to form three-dimensional Gaussian particles. The three-dimensional Gaussian particles It internally stores four data types: the position of the particle's center point. , representing the geometric center of the corresponding three-dimensional Gaussian particle in Cartesian coordinate space; covariance matrix It includes scaling and rotation matrices that determine the shape and orientation of the 3D Gaussian particle; opacity parameter. Used for subsequent 3D Gaussian particle rendering; spherical harmonic function, used to fit the viewpoint-dependent color appearance, the 3D Gaussian particles follow a distribution in space.

3. The underwater 3D reconstruction method based on spatial signal dual-physics adaptive constraints according to claim 1, characterized in that: The fourth-dimensional variable space parameter As a switching value for evaluating and characterizing points within the normal myopic distance and points within the hyperopic distance, when When, homogeneous coordinates correspond to the point at infinity, when At that time, homogeneous coordinates correspond to three-dimensional Gaussian particle points within the normal near-vision distance, through this variable space parameter Perform the transformation between the Cartesian coordinate space and the homogeneous coordinate space of a three-dimensional Gaussian particle; The variable-space Gaussian particle field model works by parametrically combining the Gaussian geometric center of the three-dimensional Gaussian particles with the parameterized information of the three-dimensional covariance matrix and the fourth-dimensional spatial parameters. This combination allows for the determination of the position, scaling matrix parameters, and spatial parameters of three-dimensional Gaussian particles within the underwater space. Proportional correlation.

4. The underwater 3D reconstruction method based on spatial signal dual-physics adaptive constraints according to claim 1, characterized in that: Based on the Nyquist sampling theorem, when recovering the original continuous signal without distortion from a discretely sampled signal, the sampling frequency must be greater than or equal to twice the highest frequency of the original signal. The lowest sampling frequency that satisfies this condition is called the Nyquist frequency. The Gaussian frequency signal filtering field for near-field imaging, based on the sampling characteristics of the imaging system, performs Gaussian spatial frequency correlation with underwater scene depth, determines the frequency constraint threshold of Gaussian particles based on the Nyquist sampling theorem, and constructs a geometric similarity relationship between the two-dimensional imaging space and the three-dimensional world space to filter the underwater near-field region. Frequency modeling is performed on the three-dimensional Gaussian particles to determine an adaptive frequency threshold that is compatible with the environment. Based on this threshold, a low-pass filter constraint is applied to the three-dimensional Gaussian particles to eliminate aliasing.

5. The underwater 3D reconstruction method based on spatial signal dual-physics adaptive constraints according to claim 1, characterized in that: The adaptive Gaussian particle opacity optimization model works by establishing adaptive pruning of 3D Gaussian particles within the underwater scene based on their opacity and observation distance, and defining a density dynamic optimization function: when the opacity of the 3D Gaussian particles... At that time, the Gaussian primitive is pruned. The adaptive pruning follows the underwater scene adaptability strategy. The closer the observation distance, the larger the pruning threshold. This rule removes redundant Gaussian primitives in the near-view distance scattering region, while retaining enough three-dimensional Gaussian primitives in the far-view distance unbounded region to ensure the modeling fidelity of the entire underwater scene.

6. The underwater 3D reconstruction method based on spatial signal dual-physics adaptive constraints according to claim 1, characterized in that: By using spherical harmonic functions to map the RGB values ​​in multi-view image data onto Gaussian variable space particles, and combining color and opacity parameters, the Alpha-Blending algorithm is used for fusion rendering to finally obtain a new perspective synthetic image.

Images