An underwater binocular camera calibration method and system fusing polarization information

Abstract

The application discloses a kind of underwater binocular camera calibration methods and systems of fusion polarization information, belong to marine environment monitoring technical field.The application constructs the isomerous binocular system under active polarization illumination, left eye camera is equipped with rotatable linear polarizer, right eye camera is full transparent;By collecting parallel and orthogonal polarization images, clear polarization feature image is solved using orthogonal polarization difference denoising model, left eye polarization feature point and right eye gray feature point are extracted respectively;Further construct joint error optimization objective function, water body real-time refractive index and polarization plate installation error are taken as state variable and included in objective function, and nonlinear optimization algorithm global iteration optimization is used.The application can effectively suppress water body backscattering noise, eliminate dynamic refractive index and mechanical error influence, realize the high-precision in-situ calibration of binocular camera under turbid water, and hardware change is small, engineering adaptability is strong.

Description

Technical Field

[0001] This invention relates to the field of marine environmental monitoring technology, and in particular to a method and system for calibrating an underwater binocular camera that integrates polarization information. Background Technology

[0002] Underwater stereo vision technology has wide applications in marine engineering, seabed exploration, and underwater robot navigation. The accuracy of its three-dimensional measurements relies heavily on the high-precision parameter calibration of binocular camera systems. However, in real marine operating environments, underwater optical conditions face severe physical challenges, and traditional calibration methods struggle to meet the stability and accuracy requirements in turbid waters.

[0003] On the one hand, the large amount of suspended particulate matter in turbid water causes strong optical backscattering, leading to severe degradation of visual image features. Traditional calibration boards based on gray-level gradients (such as checkerboard patterns) exhibit a low-pass filtering effect underwater, blurring edges and resulting in extremely low success rates for conventional corner extraction algorithms, even producing significant sub-pixel-level positioning errors. This feature degradation directly affects the accuracy of binocular stereo matching and 3D reconstruction.

[0004] On the other hand, the refractive index of seawater dynamically changes with salinity, temperature, and depth, while water pressure causes micron-level deformation of the camera's sealed housing, introducing complex nonlinear refractive errors. Traditional methods typically perform calibration in a laboratory clean water environment. When the system is deployed to real ocean conditions, the calibration parameters experience non-negligible system drift, significantly reducing measurement reliability.

[0005] To address these issues, existing underwater polarization vision technologies often employ a "cascaded strategy," which involves first using deep learning or physical prior models to restore underwater defogging and descattering images, and then extracting corner features from the restored images. However, these black-box image restoration algorithms are prone to introducing local artifacts and pixel-level geometric distortions, fundamentally undermining the stringent geometric constraints required for camera calibration. Furthermore, some studies utilize projector-based active structured light polarization calibration systems, which are not only costly in terms of hardware and difficult to seal and dissipate underwater, but their refraction models are often based on "small-angle approximations," leading to exponential amplification of refraction errors at the edges of wide-angle lens fields of view.

[0006] More importantly, conventional passive polarization defogging methods are essentially ineffective in low-light or extremely turbid environments due to their low signal-to-noise ratio; while most existing active polarization schemes lack a closed-loop design that integrates "bottom-level physical polarization denoising" with "multi-layer refraction geometric nonlinearity optimization" at the system level. Therefore, there is an urgent need for an in-situ calibration method that requires minimal hardware modifications, has strong anti-scattering capabilities, and can eliminate underwater nonlinear refraction and mechanical errors from a mathematical perspective, in order to meet the high-precision measurement requirements under harsh water conditions. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method and system for calibrating underwater binocular cameras by fusing polarization information. This method overcomes the bottleneck in feature extraction under high scattering conditions in turbid water, eliminates dynamic refractive index changes and mechanical installation errors of polarizers from a physical perspective, and achieves high-precision, high-robust in-situ calibration of underwater binocular cameras under real harsh conditions.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows:

[0009] A method for calibrating an underwater binocular camera by fusing polarization information includes the following steps: Step S1: Construct a heterogeneous binocular system and a passive calibration board based on active polarization illumination. The heterogeneous binocular system includes an illumination source, a left eye camera, and a fully transparent right eye camera. A first linear polarizer is fixedly installed in front of the illumination source to output active polarized light. A motor-rotatable second linear polarizer is added to the left eye camera. The passive calibration board adopts a diffuse black and white checkerboard pattern. Step S2: Synchronously trigger acquisition in turbid water, control the rotation of the first linear polarizer, and acquire parallel polarization image and orthogonal polarization image respectively; utilize the polarization preservation characteristics of backscattering of suspended matter in water and the depolarization characteristics of diffuse reflection of calibration plate to construct orthogonal polarization difference denoising mathematical model, and obtain bi-angle difference image, that is, solve the clear polarization feature image without scattering. Step S3: Perform sub-pixel level detection on the calculated clear polarization feature image to obtain the left eye polarization feature points; at the same time, perform adaptive contrast stretching preprocessing on the turbid grayscale image acquired by the right eye camera to extract the right eye grayscale feature points. Step S4: Using the obtained left-eye polarization feature points and right-eye grayscale feature points as observation values, and the binocular camera's intrinsic and extrinsic parameters, the real-time refractive index of the water body, and the mechanical installation error of the second linear polarizer as unknown state variables, construct a joint error optimization objective function, and use a nonlinear optimization algorithm to perform global iterative optimization of the objective function to complete the calibration.

[0010] In the above scheme, in step S1, the left eye camera is specifically configured as follows: its lens front end is rigidly connected to a miniature electric rotary stage through a threaded interface. The rotary stage includes a stator frame and a hollow rotor, and a second linear polarizer is embedded inside the hollow rotor. An absolute zero physical marker is provided between the stator frame and the hollow rotor to establish the mechanical initial reference position of the polarization axis. When the system is running, the edge computing main control unit outputs a control pulse signal to the motor drive amplification module, drives the hollow rotor to rotate the second linear polarizer to the target phase, and sends a hardware trigger synchronization signal to the left eye camera to ensure that the camera shutter opening time is aligned with the phase locking time of the second linear polarizer.

[0011] In the above scheme, the orthogonal polarization difference denoising mathematical model in step S2 is specifically as follows:

[0012] First, calculate the polarization coefficient of the water backscattering. : ; in, This is the background area of ​​pure water in the non-calibrated plate. Points of orthogonal polarization image , Points in a parallel polarization image Then, based on the regularization truncation of the damping factor, a clear polarization feature image is constructed. The formula: ; in, It is a preset minimum constant. It is an orthogonal polarization image. This is a parallel polarization image.

[0013] In the above scheme, the objective function in step S4 is specifically: ; in, For underwater forward projection function that includes the nonlinear refraction effect of the camera window; and These are independent indices of the effective feature points successfully extracted from the right eye grayscale channel and the left eye polarization channel, respectively. The first one extracted for the right eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; The first one extracted for the left eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; and These are the intrinsic and extrinsic parameter matrices for the right and left eyes, respectively; For bimodal confidence weights, This is the subpixel optical translation compensation vector caused by the mechanical installation error of the second linear polarizer of the left eye camera. This represents the real-time dynamic refractive index of the water body.

[0014] In a further technical solution, the dual-modal confidence weight A nonlinear smoothing function is used, based on the global polarization degree. Dynamic mapping: ; in, As a smoothing adjustment factor, This is the modal equilibrium threshold; when the water body is relatively clear, hour, Optimize the dominant bias towards grayscale reprojection error; when the water is extremely turbid, hour, Optimize the smooth transition of dominance to the left eye's polarization reprojection error.

[0015] In the above scheme, the nonlinear optimization algorithm in step S4 adopts the Levenberg-Marquardt algorithm and introduces a physical prior initial value setting strategy: the initial values ​​of the camera's intrinsic and extrinsic parameters are pre-calibrated parameters in a pure air environment; the real-time refractive index of the water body... The initial value is set to 1.333; the sub-pixel compensation vector caused by the installation error of the left eye polarizer. The initial value is set to the zero vector.

[0016] An underwater binocular camera calibration system that fuses polarization information, for performing the above-described method, includes: An active polarization heterogeneous binocular vision unit includes an illumination source, a left eye camera, and a fully transparent right eye camera; a first linear polarizer is fixedly installed in front of the illumination source, and a motor-rotatable second linear polarizer is added to the left eye camera; The passive calibration board uses a diffuse black and white checkerboard pattern. The synchronous acquisition and control unit is used to synchronously trigger the left and right eye cameras in turbid water and control the rotation of the second polarizer of the left eye camera to acquire parallel polarized images and orthogonal polarized images respectively. The image processing unit is used to construct an orthogonal polarization difference denoising mathematical model to obtain a clear polarization feature image, and to extract the polarization feature points of the left eye and the grayscale feature points of the right eye respectively. The global optimization calibration unit is used to construct the objective function of the polarization-grayscale dual-mode joint reprojection error optimization model. The intrinsic and extrinsic parameters of the binocular camera, the real-time refractive index of the water body, and the mechanical installation error of the second linear polarizer of the left eye camera are used as unknown state variables. The nonlinear optimization algorithm is used to perform global iterative optimization and output the calibration parameters.

[0017] In the above scheme, the left eye camera specifically includes: a miniature electric rotary stage rigidly connected to the front end of the lens, the rotary stage including a stator frame, a hollow rotor with an absolute zero physical mark, and a second linear polarizer embedded inside the hollow rotor; and an edge computing main control unit, used to output control pulse signals to the motor drive amplification module and send hardware trigger synchronization signals to the left eye camera.

[0018] In the above scheme, the lighting source and the binocular camera are encapsulated together in a pressure-resistant sealed chamber, and the transparent window of the sealed chamber is made of optical glass with low stress birefringence.

[0019] The above scheme also includes an alternative acquisition mode: for static high-precision calibration scenarios, the main control unit controls the left eye camera to sequentially acquire images in four polarization directions: 0°, 45°, 90°, and 135°. The Stokes parameter method is used to analyze the linear polarization degree image and polarization angle image of the entire optical field, in order to replace the double-angle difference image calculated from the parallel and orthogonal polarization images.

[0020] The underwater binocular camera calibration method and system that fuses polarization information provided by the above technical solution have the following beneficial effects: 1. Significantly improves the robustness and accuracy of feature extraction in turbid water bodies. This invention employs active polarization illumination combined with an orthogonal polarization difference denoising mathematical model. Utilizing the polarization-preserving properties of backscattered light from suspended particles in water and the depolarization properties of diffuse reflected light from the calibration plate, it effectively separates and suppresses backscattered noise, directly calculating a clear polarization feature image with a high signal-to-noise ratio. This overcomes the shortcomings of traditional calibration methods, such as excessive reliance on grayscale gradients and the tendency of cascaded dehazing strategies to introduce local artifacts and geometric shifts. In turbid water bodies with low visibility and strong scattering, it significantly improves the success rate and sub-pixel positioning accuracy of extracting corner points from ordinary diffuse reflectance checkerboard patterns.

[0021] 2. Eliminate nonlinear refraction and mechanical installation errors from the physical level to achieve high-precision in-situ calibration. This invention directly incorporates the dynamic refractive index of the water under complex seabed conditions (which varies with salinity, temperature, and depth) and the sub-pixel-level optical offset error of the linear polarizer of the left eye camera caused by mechanical installation and rotation into a precise underwater refraction geometry model based on Snell's law for joint nonlinear optimization. This avoids the exponential amplification of edge field-of-view refraction errors caused by the "small-angle approximation" in traditional calibration. Simultaneously, it explicitly models the polarizer mechanical installation error compensation vector in the joint objective function. This eliminates the microscopic disturbances to the imaging optical path caused by hardware assembly deviations, enabling the calibration parameters to truly reflect the underwater in-situ state and significantly reducing systematic measurement drift.

[0022] 3. It possesses an adaptive dual-modal confidence weighting mechanism to ensure the stability of optimization convergence under harsh water conditions. This invention constructs a joint reprojection error optimization model for polarization-grayscale dual modes and introduces a nonlinear smoothing function based on global polarization degree (DoP) dynamic mapping to adjust the dual-mode confidence weights. When the water is relatively clear, the optimization bias favors the right-eye fully transparent grayscale channel; when the water is extremely turbid and the grayscale characteristics are severely degraded, the bias smoothly and without oscillations transitions to the left-eye polarization channel, which has stronger anti-scattering capabilities. This adaptive mechanism ensures that the nonlinear optimization solver can always converge to the global optimum along a continuous and uninterrupted gradient direction under different visibility conditions.

[0023] 4. Minimal hardware modifications, strong engineering adaptability, and good feasibility for in-situ calibration. This invention eliminates the need for customized, complex active-emitting polarizing panels or underwater projector equipment. It only requires adding a first linear polarizer to the front of a conventional binocular light source and a miniature, electrically driven, rotatable linear polarizer to the left eye camera lens to achieve anti-scattering calibration. The hardware modification cost is low, the sealing and heat dissipation risks are minimal, and it is compatible with the pressure-resistant sealed chamber structure of existing marine monitoring equipment, making it particularly suitable for in-situ rapid calibration needs in harsh marine environments. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0025] Figure 1 This is a schematic diagram of the spatial physical configuration and optical path topology of the active polarization heterogeneous binocular vision system provided in an embodiment of the present invention; Figure 2 This is a magnified front view of the mechanical structure of the left eye camera; Figure 3 This is a block diagram of the hardware synchronization triggering and underlying control logic of the left eye polarization sensing component provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the overall process of an underwater binocular camera calibration method that fuses polarization information, provided in an embodiment of the present invention. Figure 5 The underwater precise refraction geometry model and forward projection diagram of the optical path are provided for embodiments of the present invention.

[0026] In the diagram: 1. Left eye camera; 2. Right eye camera; 3. Illumination source; 4. First linear polarizer; 5. Second linear polarizer; 6. Passive calibration plate; 7. Pressure-resistant sealed chamber; 8. Transparent window; 9. Electric rotary table; 10. Stator frame; 11. Hollow rotor; 12. Absolute zero point physical marker; 13. Threaded interface. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0028] This invention provides a method and system for calibrating underwater binocular cameras by fusing polarization information, which is suitable for high-precision in-situ calibration of binocular vision systems in turbid water environments.

[0029] I. System Hardware Configuration like Figure 1 As shown, the active polarization heterogeneous binocular vision system constructed in this embodiment includes: a left eye camera 1 (polarization channel), a right eye camera 2 (grayscale channel), an illumination source 3, a passive calibration board 6, and an edge computing main control unit.

[0030] like Figure 2 As shown, the left eye camera 1 is specifically configured as follows: its lens front end is rigidly connected to a miniature electric rotary stage 9 through a threaded interface 13. The rotary stage includes a stator frame 10 and a hollow rotor 11. A second linear polarizer 5 is embedded inside the hollow rotor 11. An absolute zero physical marker 12 is provided between the stator frame 10 and the hollow rotor 11 to establish the mechanical initial reference position of the polarization axis.

[0031] The front of the lens of the right eye camera 2 is completely transparent and does not have any optical attenuation devices. The illumination source 3 is fixedly installed at the midpoint of the baseline of the binocular camera, and a first linear polarizer 4 with a transmission axis of 0° is attached to its light outlet to output actively linearly polarized light.

[0032] The binocular camera and light source are encapsulated together in a pressure-resistant sealed chamber 7. The transparent window 8 of the sealed chamber is made of optical glass with a low stress birefringence coefficient to prevent pressure deformation from destroying the physical polarization state of the light.

[0033] The passive calibration plate 6 is made of alumina ceramic material with strong diffuse reflection characteristics, and its surface is etched with a standard black and white checkerboard pattern. When the 0° active polarized light emitted by the light source penetrates the water and hits the surface of the calibration plate, the rough surface produces strong multiple scattering, which induces a significant depolarization effect, converting the reflected light into unpolarized light, allowing it to partially penetrate the orthogonal polarizer of the left eye camera 1, and become a stable geometric feature beacon.

[0034] Figure 3 The hardware synchronization trigger and underlying control logic of the left eye polarization sensing component are shown. During system operation, the edge computing main control unit outputs control pulse signals to the motor drive amplification module, which drives the hollow rotor 11 to rotate the second linear polarizer 5 to the target phase and sends a hardware trigger synchronization signal to the left eye camera 1 to ensure that the camera shutter opening moment is strictly aligned with the phase locking moment of the second linear polarizer 5.

[0035] II. Specific Procedures for Calibration like Figure 4 As shown, the calibration method in this embodiment includes the following steps.

[0036] Step S1: System Setup and Calibration Board Deployment Complete the system setup using the hardware configuration described above. Fix the passive calibration board within the underwater field of view to be tested, ensuring that both binocular cameras can clearly capture the complete checkerboard image.

[0037] Step S2: Polarization image acquisition and physical separation of scattering noise Before calibration begins in turbid water, a mechanical zero point is first established using a physical marker of absolute zero. The second linear polarizer of the left eye camera is controlled to perform micro-step scanning near the expected zero point, and the average light intensity in the central region of the image is calculated in real time. Based on Malus's law, when the transmitted light intensity reaches the highest peak of the quadratic fitting curve, this position is calibrated as the absolute "parallel polarization phase" under the current environment, and a parallel polarization image is acquired. Subsequently, the micro stepper motor is controlled to precisely step 90° from this peak position to obtain an "orthogonal polarization phase" without phase alignment error, and an orthogonal polarization image is acquired. .

[0038] As an alternative embodiment, for static high-precision calibration scenarios, the left eye camera can be controlled to sequentially acquire images of four polarization directions: 0°, 45°, 90°, and 135°. The Stokes parameter method is then used to analyze the linear polarization degree image and polarization angle image of the entire optical field, replacing the dual-angle difference image.

[0039] The target signal is further separated by utilizing the physical properties of high polarization preservation of backscattered light from suspended particles in water and depolarization of diffuse reflected light from the calibration plate. The polarization coefficient of water backscattering is defined. The ratio of the pixel mean of the orthogonal and parallel polarized images of the background region: ; in, This refers to the background region of the uncalibrated plate pure water body extracted through image thresholding. Points of orthogonal polarization image , Points in a parallel polarization image ; To prevent under extreme conditions The approach to 1 causes the denominator of the difference model to approach zero and high-frequency noise to amplify exponentially. Therefore, a regularization truncation based on the damping factor is introduced to construct an orthogonal polarization difference denoising model, which solves for a clear polarization feature image without scattering. : ; in, A minimum constant preset for the project (which can be taken as 0.01~0.05). When At that time, the denominator is forced to clamp at While preserving the underlying geometric features of the image, the algorithm's stability under harsh water conditions is ensured.

[0040] Step S3: Feature Point Extraction Left eye polarization feature extraction: Extracting the high signal-to-noise ratio image from step S2. A gradient-based sub-pixel corner detection algorithm is applied to accurately extract the coordinates of checkerboard corner points, which are denoted as polarization feature points. .

[0041] Right Eye Grayscale Feature Extraction: For the original murky grayscale image captured by the right eye's full-transmittance camera, Limit Contrast Adaptive Histogram Equalization (CLAHE) is used to highlight local high-frequency features. Then, the same sub-pixel corner detection algorithm as the left eye is applied to accurately extract the checkerboard corner points, which are denoted as grayscale feature points. As an alternative, when the right eye image cannot stably extract feature points in an extremely murky environment, a rotatable linear polarizer can be added to the right eye camera and the same polarization difference processing can be used.

[0042] Step S4: Joint optimization and calibration of polarization-grayscale dual modes Step S41: Establish an accurate underwater refraction geometry model. For example... Figure 5 As shown, a forward projection function of light penetrating the camera's waterproof casing interface is established based on Snell's law. : ; in, Predict the coordinates of a model projecting a point in three-dimensional space onto the two-dimensional image plane of the camera; The underwater forward projection function includes the nonlinear refraction effect of the camera window. This function simulates the process by which the physical light path passes through refraction points I1 and I2, then precisely passes through the camera optical center C and is projected onto the sensor plane. ideal This is the ideal projection point where light does not refract; the underwater refractive projection model includes the known air refractive index. Refractive index of glass window and system constants such as the physical thickness d of the window; The coordinates of the three-dimensional corner points on the calibration board; This is the camera intrinsic parameter matrix; and These are the rotation matrix and translation vector of the camera coordinate system relative to the world coordinate system, respectively. This represents the real-time dynamic refractive index of the water body.

[0043] The above projection function The internal optical path geometry mapping is governed by the following Snell's law and the window thickness constraint equation: ; ; in, , , , These represent the angle of incidence and angle of refraction at the interfaces between the water, glass, and air, respectively, and the physical thickness of the viewing window. It is a known constant.

[0044] By combining the aforementioned physical constraints with the camera's intrinsic and extrinsic parameters, the model's predicted coordinates can be solved using a ray tracing model. .

[0045] Step S42: Construct the joint objective function. To compensate for the minute deflection of the imaging optical path caused by the mechanical mounting and rotation of the left eye polarizer, the sub-pixel-level optical offset error is addressed. Perform explicit modeling. Let the rotation angle of the second linear polarizer of the left eye camera be... In the image pixel coordinate system, this optical offset model is defined as a two-dimensional compensation vector containing rotation and translation components: ; in, The angle of rotation with the polarizer Dynamically changing subpixel optical offset vector; and These offsets are located in the image's physical coordinate system. Axis (horizontal direction) and Projected components on the axis (vertical direction).

[0046] In fixed parallel and orthogonal differential modes, this error degenerates into a constant-order subpixel translation vector. Finally, the aforementioned compensation term is applied to the observed feature points and aligned with the right eye grayscale reprojection error, constructing the following polarization-grayscale dual-mode joint optimization objective function. : ; in, and These are independent indices of the effective feature points successfully extracted from the right eye grayscale channel and the left eye polarization channel, respectively. The first one extracted for the right eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; The first one extracted for the left eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; and These are the intrinsic and extrinsic parameter matrices for the right and left eyes, respectively; For bimodal confidence weights.

[0047] Since multiple scattering of suspended particles in turbid water leads to a significant increase in the global degree of polarization (DoP), the system uses a nonlinear smoothing function to dynamically map DoP to confidence weights. Its adaptive adjustment rule is as follows: ; in, This is the preset curve smoothing adjustment factor. The modal balance threshold set for the system. The specific weight allocation logic is as follows: when... < At this time, it indicates that the water is relatively clear, and the signal-to-noise ratio of the right eye's fully transparent grayscale channel is high. The calculation at this time yields... <0.5, the joint objective function is biased towards the right eye's grayscale reprojection error; when > At this time, it indicates that the water is extremely turbid and the grayscale characteristics of the right eye have severely degraded. The calculations at this point show... >0.5 and gradually approaches 1 as turbidity increases, the dominance of the objective function is determined by... The continuous differentiability of the function allows for a smooth, non-oscillating transition to the left-eye polarization reprojection error, which is more resistant to scattering. This adaptive rule ensures that the nonlinear optimization solver always converges along a continuous, uninterrupted gradient direction in water conditions with varying visibility.

[0048] Step S43: Global nonlinear iterative solution and initial value setting.

[0049] To minimize the joint optimization objective function L constructed in step S42, and to decouple and output the globally optimal system state variables, this step employs the Levenberg-Marquardt (LM) algorithm for nonlinear iterative optimization. Since this objective function contains multi-dimensional nonlinear coupling parameters, to ensure the convergence direction of the solution domain and avoid getting trapped in local minima, this invention introduces an initial value setting strategy based on strict physical priors before the algorithm begins.

[0050] The specific rules for setting initial values ​​during iterations are as follows: (1) Initial values ​​of camera intrinsic and extrinsic parameters: The camera intrinsic parameter matrix, distortion coefficients and binocular extrinsic parameters obtained by the system in a pure air environment before being launched into the water are used as the iterative benchmark; (2) Initial value of water refractive index: The real-time refractive index of the water body The initial value for the iteration is set to the standard optical refractive index constant of pure water, which is 1.333. (3) Initial value of offset error: Assuming that the initial mechanical assembly is in an ideal state, the polarizer sub-pixel optical translation compensation vector is... The initial value is set to the zero vector. .

[0051] As the algorithm converges, it finally decouples and outputs the globally optimal binocular camera intrinsic and extrinsic parameters and the compensated polarizer optical offset error. and the real-time dynamic refractive index of the current turbid water body The calibration is now complete.

[0052] III. Experimental Verification and Results This embodiment conducted a comparative experiment on dual-target calibration in a turbid water experimental platform. Experimental observations show that as the turbidity of the water increases, the traditional dual-target calibration method based on gray-level gradients is affected by backscattering noise, resulting in a decrease in the success rate of corner detection and an increase in the frequency of feature loss or mismatch, which to some extent limits the effectiveness of calibration.

[0053] The polarization information fusion calibration method described in this embodiment uses an orthogonal polarization difference model to separate some scattering noise. Test data shows that, under the same turbidity conditions, this method improves the completeness of checkerboard corner feature extraction, enhances the stability of sub-pixel localization, and thus effectively suppresses reprojection errors.

[0054] Further underwater mounting tests showed that the binocular system optimized by this method exhibited superior anti-interference capabilities and three-dimensional measurement accuracy compared to traditional calibration methods in turbid water environments, objectively verifying the robustness of this invention under complex water conditions.

[0055] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for calibrating an underwater binocular camera by fusing polarization information, characterized in that, Includes the following steps: Step S1: Construct a heterogeneous binocular system and a passive calibration board based on active polarization illumination. The heterogeneous binocular system includes an illumination source, a left eye camera, and a fully transparent right eye camera. A first linear polarizer is fixedly installed in front of the illumination source to output active polarized light. A motor-rotatable second linear polarizer is added to the left eye camera. The passive calibration board adopts a diffuse black and white checkerboard pattern. Step S2: Synchronously trigger acquisition in turbid water, control the rotation of the first linear polarizer, and acquire parallel polarization image and orthogonal polarization image respectively; utilize the polarization preservation characteristics of backscattering of suspended matter in water and the depolarization characteristics of diffuse reflection of calibration plate to construct orthogonal polarization difference denoising mathematical model, and obtain bi-angle difference image, that is, solve the clear polarization feature image without scattering. Step S3: Perform sub-pixel level detection on the calculated clear polarization feature image to obtain the left eye polarization feature points; at the same time, perform adaptive contrast stretching preprocessing on the turbid grayscale image acquired by the right eye camera to extract the right eye grayscale feature points. Step S4: Using the obtained left-eye polarization feature points and right-eye grayscale feature points as observation values, and the binocular camera's intrinsic and extrinsic parameters, the real-time refractive index of the water body, and the mechanical installation error of the second linear polarizer as unknown state variables, construct a joint error optimization objective function, and use a nonlinear optimization algorithm to perform global iterative optimization of the objective function to complete the calibration.

2. The method according to claim 1, characterized in that, In step S1, the left eye camera is specifically configured as follows: its lens front end is rigidly connected to a miniature electric rotary stage via a threaded interface. The rotary stage includes a stator frame and a hollow rotor, with a second linear polarizer embedded inside the hollow rotor. An absolute zero physical marker is provided between the stator frame and the hollow rotor to establish the mechanical initial reference position of the polarization axis. When the system is running, the edge computing main control unit outputs control pulse signals to the motor drive amplification module, driving the hollow rotor to rotate the second linear polarizer to the target phase, and sending a hardware trigger synchronization signal to the left eye camera to ensure that the camera shutter opening time is aligned with the phase locking time of the second linear polarizer.

3. The method according to claim 1, characterized in that, In step S2, the orthogonal polarization difference denoising mathematical model is specifically as follows: First, calculate the polarization coefficient of the water backscattering. : ； in, This is the background area of ​​pure water in the non-calibrated plate. Points of orthogonal polarization image , Points in a parallel polarization image Then, based on the regularization truncation of the damping factor, a clear polarization feature image is constructed. The formula: ； in, It is a preset minimum constant. It is an orthogonal polarization image. This is a parallel polarization image.

4. The method according to claim 1, characterized in that, In step S4, the objective function is specifically: ； in, For underwater forward projection function that includes the nonlinear refraction effect of the camera window; and These are independent indices of the effective feature points successfully extracted from the right eye grayscale channel and the left eye polarization channel, respectively. The first one extracted for the right eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; The first one extracted for the left eye Coordinates of two-dimensional observation feature points These are its corresponding real three-dimensional space corner points; and These are the intrinsic and extrinsic parameter matrices for the right and left eyes, respectively; For bimodal confidence weights, This is the subpixel optical translation compensation vector caused by the mechanical installation error of the second linear polarizer of the left eye camera. This represents the real-time dynamic refractive index of the water body.

5. The method according to claim 4, characterized in that, The bimodal confidence weight A nonlinear smoothing function is used, based on the global polarization degree. Dynamic mapping: ； in, As a smoothing adjustment factor, This is the modal equilibrium threshold; when the water body is relatively clear, hour, Optimize the dominant bias towards grayscale reprojection error; when the water is extremely turbid, hour, Optimize the smooth transition of dominance to the left eye's polarization reprojection error.

6. The method according to claim 1, characterized in that, The nonlinear optimization algorithm described in step S4 employs the Levenberg-Marquardt algorithm and introduces a physical prior initial value setting strategy: the initial values ​​of the camera's intrinsic and extrinsic parameters are pre-calibrated parameters in a pure air environment; the real-time refractive index of the water body... The initial value is set to 1.333; the sub-pixel compensation vector caused by the installation error of the left eye polarizer. The initial value is set to the zero vector.

7. An underwater binocular camera calibration system that fuses polarization information, used to perform the method according to any one of claims 1 to 6, characterized in that, include: An active polarization heterogeneous binocular vision unit includes an illumination source, a left eye camera, and a fully transparent right eye camera; a first linear polarizer is fixedly installed in front of the illumination source, and a motor-rotatable second linear polarizer is added to the left eye camera; The passive calibration board uses a diffuse black and white checkerboard pattern. The synchronous acquisition and control unit is used to synchronously trigger the left and right eye cameras in turbid water and control the rotation of the second polarizer of the left eye camera to acquire parallel polarized images and orthogonal polarized images respectively. The image processing unit is used to construct an orthogonal polarization difference denoising mathematical model to obtain a clear polarization feature image, and to extract the polarization feature points of the left eye and the grayscale feature points of the right eye respectively. The global optimization calibration unit is used to construct the objective function of the polarization-grayscale dual-mode joint reprojection error optimization model. The intrinsic and extrinsic parameters of the binocular camera, the real-time refractive index of the water body, and the mechanical installation error of the second linear polarizer of the left eye camera are used as unknown state variables. The nonlinear optimization algorithm is used to perform global iterative optimization and output the calibration parameters.

8. The system according to claim 7, characterized in that, The left eye camera specifically includes: a miniature electric rotary stage rigidly connected to the front end of the lens, the rotary stage including a stator frame, a hollow rotor with an absolute zero physical mark, and a second linear polarizer embedded inside the hollow rotor; and an edge computing main control unit, used to output control pulse signals to the motor drive amplification module and send hardware trigger synchronization signals to the left eye camera.

9. The system according to claim 7, characterized in that, The illumination source and the binocular camera are encapsulated together in a pressure-resistant sealed chamber, and the transparent window of the sealed chamber is made of optical glass with a low stress birefringence coefficient.

10. The system according to claim 7, characterized in that, It also includes an alternative acquisition mode: for static high-precision calibration scenarios, the main control unit controls the left eye camera to sequentially acquire images in four polarization directions: 0°, 45°, 90°, and 135°. The Stokes parameter method is used to analyze the linear polarization degree image and polarization angle image of the entire optical field, in order to replace the double-angle difference image calculated from the parallel and orthogonal polarization images.

Images