A water surface three-dimensional reconstruction method based on strabismus geometric constraint and attitude blind correction

Abstract

The application discloses a water surface three-dimensional reconstruction method based on oblique view geometric constraint and attitude blind correction, and belongs to the technical field of precision engineering measurement, computer vision and physical oceanography, and comprises the following steps: acquiring point clouds by oblique scanning of a 905nm laser radar near Brewster angle, removing noises through straight-through filtering and statistical outlier removal, realizing IMU-free attitude leveling based on SVD and Rodrigues formula, taking a preset low quantile of a Z axis as a dynamic reference to perform height normalization, and finally generating a continuous three-dimensional wave surface model through two-dimensional gridding, morphological closing operation and Gaussian smoothing. By using the above method, the low-cost near-infrared laser radar can be used to realize high-precision three-dimensional reconstruction of micro-scale waves without external inertial measurement hardware and physical targets, so that the hardware cost is greatly reduced, and the convenience and precision of non-contact measurement are improved.

Description

Technical Field

[0001] This invention relates to the fields of precision engineering measurement, computer vision, and physical oceanography, and in particular to a method for three-dimensional reconstruction of the water surface based on oblique geometric constraints and attitude blind correction. Background Technology

[0002] With the development of marine engineering monitoring and digital twin technology, non-contact, high spatiotemporal resolution 3D reconstruction of dynamic water surfaces has become an important requirement. Traditional contact-based measurements such as wave buoys have limitations such as deployment difficulties, while non-contact sensing methods, although the future trend, still face technical bottlenecks in balancing low-cost hardware and high-precision reconstruction.

[0003] In terms of optical detection, conventional 905nm near-infrared lidar is susceptible to strong absorption by water and specular reflection from still water surfaces, resulting in low echo rates and difficulty in obtaining effective three-dimensional data. Some existing technologies (such as CN109490906A, published on 2019.03.19) indirectly invert waves by manually throwing physical targets, but their operability in large water areas is poor. Other direct scanning technologies (such as CN114755675A, published on 2022.07.15) are prone to signal distortion and voids in the model if the coupling effect between the laser grazing angle and water surface ripple is not effectively handled.

[0004] In terms of spatial reconstruction and attitude calculation, mainstream measurement technologies (such as CN113970756B, published on 2024.07.09) usually rely on high-precision inertial navigation units (IMU) or GNSS systems to eliminate point cloud tilt. However, this solution based on high-end hardware is expensive and prone to sensor drift in complex environments. Once calibration fails, it will cause layered distortion and reduce reconstruction accuracy.

[0005] Therefore, there is an urgent need for a method that can eliminate the dependence on physical targets and expensive inertial navigation hardware and achieve high-precision wave reconstruction using only low-cost near-infrared radar. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional reconstruction method for water surface based on oblique geometric constraints and attitude blind correction, which aims to solve the problems of signal sparsity, noise interference, and difficulty in accurately calibrating sensor pose caused by water absorption and specular reflection when near-infrared lidar detects water surface.

[0007] To achieve the above objectives, this invention provides a method for 3D reconstruction of the water surface based on oblique gaze geometric constraints and blind attitude correction, comprising the following steps: S1. Physical high-pass filtering data acquisition: A non-repetitive scanning lidar with a working wavelength of 905nm is set up above the water body to be measured. The lidar is controlled to perform oblique scanning within the oblique elevation angle range set near the Brewster angle of the water body. Based on the Fresnel reflection attenuation characteristics, the specular reflection beam of the still water surface is deviated from the receiving field of view, and background noise is suppressed from the physical end to obtain the original point cloud with wave signals as the main component. S2. Statistical denoising and region of interest extraction: The original point cloud is filtered to remove environmental noise outside the set detection range, and isolated flying point noise is removed by using a statistical outlier removal algorithm based on Gaussian distribution to obtain a denoised point cloud. The denoised point cloud retains the original tilted posture. S3. Adaptive attitude leveling based on SVD: Calculate the geometric centroid of the denoised point cloud and perform decentralization processing, construct the coordinate covariance matrix and perform singular value decomposition, take the right singular vector corresponding to the minimum singular value as the instantaneous water surface normal vector, combine the Rodrigues rotation formula to generate a rotation matrix, and correct the denoised point cloud to the horizontal reference plane to obtain the horizontally corrected point cloud. S4. Dynamic water level locking and height normalization: The preset low quantile of the Z-axis coordinate of the point cloud after statistical level correction is used as the dynamic zero water level reference. The height of each point is subtracted from the dynamic zero water level reference to achieve height normalization and obtain a normalized point cloud that directly expresses the wave height. S5. Spatial Continuity Reconstruction and Geometric Morphology Restoration: The normalized point cloud is resampled in two dimensions using a rasterization method. Sampling holes are filled using morphological closing operations, and Gaussian kernel convolution is applied for smoothing to finally generate a spatially continuous three-dimensional wavefront model.

[0008] Preferably, in step S1, the set oblique tilt angle range is 50° to 65°.

[0009] Preferably, in step S1, the geometric configuration of the lidar installation satisfies the following conditions: The vertical height of the radar optical center from the water surface is 1.6m, and the horizontal distance from the center of the water area is 2.5m.

[0010] Preferably, in step S1, the integration sampling time of a single frame point cloud data of the lidar is controlled to be 200ms.

[0011] Preferably, in step S2, the number of local spatial neighborhood points of the Statistical Outlier Removal (SOR) algorithm based on Gaussian distribution is 20, and the standard deviation multiplier threshold is 1.0.

[0012] Preferably, in step S3, the adaptive attitude leveling method extracts the normal vector based solely on the three-dimensional coordinate covariance matrix of the denoised point cloud itself, thereby achieving adaptive compensation for radar installation angle deviation without external inertial measurement data.

[0013] Preferably, in step S4, the lower percentile is preset to the 5th percentile.

[0014] Preferably, in step S5, the grid resolution of the two-dimensional raster resampling is 0.015m.

[0015] Preferably, in step S5, during two-dimensional rasterization resampling, only the point with the largest Z value is retained in each grid cell as the representative point of the wavefront to overcome the depth ambiguity caused by underwater body scattering.

[0016] Preferably, in step S5, the morphological closing operation uses a disk structuring element, and the standard deviation of the Gaussian kernel convolution is 1.0.

[0017] Therefore, the above-mentioned method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction has the following beneficial effects: (1) Significantly reduced hardware costs. Only a conventional 905nm near-infrared lidar is needed to complete the operation, without the need for an external inertial measurement unit (IMU) or differential positioning module.

[0018] (2) Improved the convenience of non-contact measurement. It eliminates the cumbersome steps of traditional wave height meters and throwing floating targets, and realizes direct, non-contact dynamic wave surface reconstruction.

[0019] (3) Improved measurement accuracy and stability. By adaptively overcoming the reference drift caused by sensor vibration or installation deviation, high-precision three-dimensional reconstruction of microscale waves was achieved, with the relative error stably controlled within 5%.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a three-dimensional water surface reconstruction method based on oblique gaze geometric constraints and attitude blind correction according to the present invention; Figure 2 This is a schematic diagram of the layout of the lidar oblique scanning geometric architecture in an embodiment of a water surface three-dimensional reconstruction method based on oblique gaze geometric constraints and attitude blind correction of the present invention. Figure 3 This is a schematic diagram illustrating the principle of extracting instantaneous water surface normal vectors and performing geometric correction based on singular value decomposition (SVD) in an embodiment of the water surface three-dimensional reconstruction method based on oblique gaze geometric constraints and attitude blind correction of the present invention. Figure 4 This is a comparison diagram of the sea surface reconstruction effect of the method of the present invention under different wave conditions. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0024] Example 1 This embodiment uses a Livox Tele-15 solid-state lidar (operating wavelength 905nm, non-repetitive scanning mode) to perform three-dimensional reconstruction of the water surface in an indoor wave tank. The tank diameter is 1.2m and the water depth is 10.2cm. The wave-generating device can generate micro-scale waves of different heights and frequencies.

[0025] like Figure 1 As shown, a method for 3D reconstruction of water surface based on oblique gaze geometric constraints and blind attitude correction includes the following steps: S1: Physical high-pass filter data acquisition The lidar is mounted above the water tank, with the specific geometric configuration as follows: Figure 2 As shown: The vertical height H from the radar optical center to the water surface is 1.6m, and the horizontal distance D from the center of the water tank to the water surface is 2.5m. The resulting pitch angle (the angle between the radar optical axis and the vertical direction) is approximately 57°, which falls near the Brewster angle of the water body (the range of 50° to 65°).

[0026] After activating the wave-generating device, the lidar is controlled to continuously acquire point clouds with a single-frame integration time of 200ms. Based on the Fresnel reflection attenuation characteristics, the specular reflection beam from a smooth, still water surface deviates from the receiving field of view due to the oblique viewing angle, thus physically suppressing still water background noise and primarily acquiring diffuse reflection signals from the wave surface. This step outputs the raw point cloud, with each point containing three-dimensional coordinate information.

[0027] S2: Statistical Denoising and Region of Interest Extraction First, apply a pass-through filter to the original point cloud: set a spatial threshold (e.g., the boundary range of a pool) to remove environmental noise outside the detection range and obtain the point cloud within the region of interest.

[0028] Then, a statistical outlier removal algorithm based on Gaussian distribution is used: for each point, its 20 nearest neighbors are searched, and the average distance from that point to all neighboring points is calculated. Assuming that the global average distance follows a Gaussian distribution, if the average distance of a point exceeds the global mean plus 1.0 standard deviation, the point is determined to be an isolated flying point (mainly caused by the absorption effect of 905nm laser by water) and is removed. After denoising, a clean point cloud is obtained, which still retains the tilted posture of the original oblique scan.

[0029] S3: SVD-based adaptive attitude leveling The core of this step is to extract the instantaneous water surface normal vector from the spatial distribution of the point cloud itself using Singular Value Decomposition (SVD), thereby achieving attitude correction without an external inertial measurement unit. The specific process is as follows: First, assume that the denoised point cloud contains The point, the first The three-dimensional coordinate vector of each point is Calculate the geometric centroid of the denoised point cloud. ,pass Decentralization is achieved. Then, the point cloud 3D coordinate covariance matrix is ​​constructed. And perform singular value decomposition on it. ,in, and Both are 3×3 orthogonal matrices (referred to as the left singular vector matrix and the right singular vector matrix, respectively). It is a diagonal matrix, and It is a singular value.

[0030] Since the water surface is approximately horizontal under the influence of gravity, the direction with the minimum variance in the point cloud distribution corresponds to the minimum singular value. Therefore, take The third column (i.e., the right singular vector corresponding to the minimum singular value) is taken as the instantaneous water surface normal vector, denoted as... .

[0031] Next, the rotation matrix is ​​calculated using Rodriguez's rotation formula. , normal vector Correct to vertical direction ,in The rotation angle is... For antisymmetric matrices, This is a 3×3 identity matrix. Finally, all points in the original denoised point cloud are left-multiplied by this rotation matrix to obtain the horizontally corrected point cloud. After correction, the principal plane of the point cloud is parallel to the horizontal plane, and the Z-axis coordinate directly reflects the height of each point relative to an unknown reference. The geometric principle of this step is as follows: Figure 3 As shown.

[0032] S4: Dynamic water level locking and height normalization To eliminate reference drift caused by water level fluctuations (such as changes in water volume caused by evaporation and wave generation), a dynamic zero-water-level locking method based on quantiles is adopted.

[0033] First, the coordinate distribution of the point cloud in the Z-axis direction after horizontal correction is statistically analyzed, and its 5th percentile is calculated as the dynamic zero water level reference. ,in It is a quantile function. This is the set of Z-axis coordinates of all points in the point cloud after horizontal correction. This represents the total number of point clouds. This statistical value can stably lock the instantaneous static reference plane. Subsequently, height normalization is achieved by calculating the relative deviation between the Z-axis coordinates of each sampling point and this reference plane. The calculation formula is as follows: Normalized values This represents the true physical wave height of the sampling point relative to the instantaneous reference surface.

[0034] S5: Spatial Continuity Reconstruction and Geometric Morphology Restoration This step converts the discrete point cloud into a continuous 3D wavefront model and repairs local data holes caused by laser absorption in the water. The specific operations are as follows: 1. Two-dimensional rasterization (digital elevation model generation) The normalized point cloud is projected onto a horizontal plane and divided into a regular grid S with a resolution of 0.015m. Within each grid cell, only the point with the largest Z value (wave height) is retained as the representative value of that grid. This is to overcome the depth ambiguity caused by underwater body scattering and ensure that the reconstructed point is the highest point of the wavefront (i.e., the actual water surface).

[0035] 2. Morphological closing operation Using a disk-shaped structuring element B (with a radius of 3 pixels), a morphological closing operation is performed on the mesh image S, the mathematical expression of which is: That is, first perform the expansion operation. (Connect the broken wave crests), and then perform the etching operation. (Restore the original outline) to effectively fill the sampling void caused by the absorption of 905nm laser by water, while maintaining the overall shape of the wave surface.

[0036] 3. Gaussian smoothing Construct the standard deviation as Two-dimensional Gaussian kernel function Its mathematical expression is: , For pixel coordinates, the mesh data after morphological closing operation, Perform a two-dimensional convolution operation with this Gaussian kernel: , This is the final output of a 3D wave digital elevation model. This convolutional calculation eliminates the jagged edges of digital quantization, making the final output 3D wave surface closer to the natural form of water surface tension.

[0037] Figure 4 These are comparative images showing the sea surface reconstruction effects of the method of this invention under different wave conditions. Each set of figures, from left to right, consists of: the original three-dimensional discrete point cloud of the water surface after noise removal and attitude error correction; a continuous three-dimensional transient wave topography model generated based on interpolation fitting of the corrected point cloud; and a verification image of the fitting accuracy between the measured ground truth discrete points and the reconstructed wave surface. The maximum wave height for the first sample group is 6.1 cm, for the second sample group it is 13.0 cm, and for the third sample group it is 7.6 cm. In the figures, the blue-cyan scattered points represent the measured point cloud of the sea surface after correction, the blue-green gradient surface represents the wave surface reconstructed by the algorithm of this invention, and the red discrete points represent the measured ground truth sampling points of the sea surface.

[0038] Experimental verification results By comparing with wave height meter reference data, this method can transform discrete, tilted point clouds into spatially continuous three-dimensional wave digital elevation models without the aid of an inertial measurement unit. It can reconstruct micro-scale waves of varying intensities (maximum wave height 13.0 cm) with high precision, maintaining a relative error within 5%, and ensuring continuous, unbroken wave surfaces. This significantly reduces hardware costs and improves the convenience and accuracy of non-contact measurement.

[0039] Therefore, this invention employs the aforementioned method for 3D water surface reconstruction based on oblique-view geometric constraints and attitude blind correction. It suppresses still water background noise from a physical perspective through Brewster angle oblique scanning, achieves point cloud horizontal alignment without an IMU by combining SVD adaptive attitude correction, and fills data gaps using dynamic quantile benchmark locking and morphological repair. Experimental results show that this method, without inertial navigation assistance, can transform discrete, tilted point clouds into spatially continuous 3D wave digital elevation models, enabling high-precision reconstruction of microscale waves of varying intensities (maximum wave height 13.0 cm). The relative error is stably controlled within 5%, and the wave surface is continuous and unbroken, significantly reducing hardware costs and improving the convenience and accuracy of non-contact measurement.

[0040] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for 3D reconstruction of water surface based on oblique gaze geometric constraints and blind attitude correction, characterized in that, Includes the following steps: S1. Physical high-pass filtering data acquisition: A non-repetitive scanning lidar with a working wavelength of 905nm is set up above the water body to be measured. The lidar is controlled to perform oblique scanning within the oblique elevation angle range set near the Brewster angle of the water body. Based on the Fresnel reflection attenuation characteristics, the specular reflection beam of the still water surface is deviated from the receiving field of view, and the background noise is suppressed from the physical end to obtain the original point cloud with wave signals as the main component. S2. Statistical denoising and region of interest extraction: The original point cloud is filtered to remove environmental noise outside the set detection range, and isolated flying point noise is removed by using a statistical outlier removal algorithm based on Gaussian distribution to obtain a denoised point cloud. The denoised point cloud retains the original tilted posture. S3. Adaptive attitude leveling based on SVD: Calculate the geometric centroid of the denoised point cloud and perform decentralization processing, construct the coordinate covariance matrix and perform singular value decomposition, take the right singular vector corresponding to the minimum singular value as the instantaneous water surface normal vector, combine the Rodrigues rotation formula to generate a rotation matrix, and correct the denoised point cloud to the horizontal reference plane to obtain the horizontally corrected point cloud. S4. Dynamic water level locking and height normalization: The preset low quantile of the Z-axis coordinate of the point cloud after statistical level correction is used as the dynamic zero water level reference. The height of each point is subtracted from the dynamic zero water level reference to achieve height normalization and obtain a normalized point cloud that directly expresses the wave height. S5. Spatial Continuity Reconstruction and Geometric Morphology Restoration: The normalized point cloud is resampled in two dimensions using a rasterization method. Sampling holes are filled using morphological closing operations, and Gaussian kernel convolution is applied for smoothing to finally generate a spatially continuous three-dimensional wavefront model.

2. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S1, the set oblique viewing pitch angle range is 50° to 65°.

3. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S1, the geometric configuration of the lidar installation satisfies the following conditions: The vertical height of the radar optical center from the water surface is 1.6m, and the horizontal distance from the center of the water area is 2.5m.

4. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S1, the integration sampling time of a single frame point cloud data of the lidar is controlled to be 200ms.

5. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S2, the number of local spatial neighborhood points in the Gaussian distribution-based statistical outlier removal algorithm is 20, and the standard deviation multiplier threshold is 1.

0.

6. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S3, the adaptive attitude leveling method extracts the normal vector based solely on the three-dimensional coordinate covariance matrix of the denoised point cloud itself, thereby achieving adaptive compensation for radar installation angle deviation without external inertial measurement data.

7. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S4, the lower percentile is preset to the 5th percentile.

8. The method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S5, the grid resolution of the two-dimensional raster resampling is 0.015m.

9. A method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 8, characterized in that, In step S5, during two-dimensional rasterization resampling, only the point with the largest Z value is retained in each grid cell as the representative point of the wavefront to overcome the depth ambiguity caused by underwater body scattering.

10. A method for 3D reconstruction of water surface based on oblique gaze geometric constraints and attitude blind correction according to claim 1, characterized in that, In step S5, the morphological closing operation uses a disk structuring element, and the standard deviation of the Gaussian kernel convolution is 1.0.

Images