A depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging

Abstract

The application provides a depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging. Image acquisition is performed on two kinds of complementary three-dimensional imaging technologies by using a single two-dimensional camera, so that a complex image registration problem in the image fusion process is avoided. Further, an azimuth angle and a zenith angle are calculated based on depth images obtained by the polarized three-dimensional shape and the polarization modulation ranging respectively, and an azimuth angle fusion model and a zenith angle fusion model are constructed to obtain fused azimuth angle and zenith angle, so that the respective advantages of the two kinds of three-dimensional data can be maximally transferred to the fusion result. Integration and linear fitting are performed on the fused azimuth angle and the fused zenith angle to obtain a final depth reconstruction result. The depth reconstruction result has high resolution and contains accurate texture detail information and accurate low-frequency absolute depth information.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of three-dimensional depth reconstruction of object surface, and relates to a depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging, which is suitable for various three-dimensional reconstruction application scenarios with high quality requirements. BACKGROUND

[0002] Compared with two-dimensional images, three-dimensional data of object surface can provide more comprehensive feature information and spatial position information, and three-dimensional reconstruction technology has been applied in many fields such as automatic driving, biological imaging, robot system and cultural relic protection.

[0003] Polarized three-dimensional shape is a target surface feature calculated by using the polarization state information of light reflected by the object surface, which can reflect the information such as material, roughness and distribution of the target object surface. The biggest advantage is that it can accurately reconstruct the texture detail information of the target object surface. However, there are two difficult problems in the calculation process of polarized three-dimensional shape: (1) the solution result of the azimuth component of the surface normal vector has binary nature, and the difference between the two solution results is π radian. This leads to the concave-convex ambiguity problem of the reconstructed polarized three-dimensional shape; (2) the calculation of the zenith component of the surface normal vector needs the refractive index of the target object surface to be known, but in actual application scenarios, the refractive index of the target object surface cannot be obtained.

[0004] The existing three-dimensional reconstruction methods for solving the above difficult problems can be mainly divided into two categories. The first category of methods is to combine light and shadow information to constrain the polarized three-dimensional shape. For example, multiple light sources with known positions are set to illuminate the target, and the more accurate polarized three-dimensional shape is obtained by combining the light source direction and the shadow information of the target surface. However, this kind of method often needs a relatively complex lighting system, which limits the practical application ability of this kind of method. Moreover, this kind of method can only correct the azimuth ambiguity problem of the polarized three-dimensional shape, and cannot solve the zenith solving problem when the reflectivity of the target surface is unknown, so the error between the reconstructed polarized three-dimensional shape and the real three-dimensional surface of the target is still large. The second category of methods is to fuse the polarized three-dimensional shape and the depth data obtained by the depth sensor. The depth sensor based on structured light, multi-view stereo vision and laser radar can provide rough target three-dimensional depth data. The stable low-frequency information of the depth data and the high-frequency texture details of the polarized three-dimensional shape can be fused by using model-based methods or deep learning methods, so as to obtain more accurate target three-dimensional depth information. The reconstruction quality of the second category of methods is obviously improved compared with the first category of methods. However, the second category of methods often needs to accurately register the polarized image and the depth image of the depth sensor, which is very difficult in many application scenarios.

[0005] Polarization modulation ranging is a novel 3D imaging method. By introducing a polarization modulation module in front of a common 2D image sensor, the polarization modulation images of the target can be captured, and the depth images of the target can be further calculated. The depth images obtained by the polarization modulation ranging method have accurate low-frequency information and can be effectively fused with the polarization 3D shape. More importantly, the polarization images and the polarization modulation images required to reconstruct the polarization 3D shape can be captured by a common 2D camera. How to design an imaging system to capture the two kinds of images by a single camera to avoid the complex image registration problem and establish an effective image fusion method to reconstruct high-quality 3D depth is a key problem of fusion 3D reconstruction. SUMMARY

[0006] The present application is directed to the deficiencies of the prior art, and a depth reconstruction algorithm for fusing the polarization 3D shape and the polarization modulation ranging is proposed by using the similarities and differences between the polarization 3D reconstruction and the polarization modulation ranging. The depth reconstruction result with real depth and accurate texture detail information is obtained by solving the concave-convex ambiguity and the depth unreality problem of the polarization 3D shape through the depth images of the polarization modulation ranging.

[0007] The technical scheme adopted by the present application is: a depth reconstruction method for fusing the polarization 3D shape and the polarization modulation ranging, first, the initial azimuth component and the initial zenith component of the target surface normal vector are calculated from the polarization image and the depth image generated based on the polarization modulation ranging respectively. Then, by analyzing the complementary features of the different initial azimuth and initial zenith obtained by the two methods, the azimuth fusion model and the zenith fusion model are established respectively, and the alternating direction multiplier method is used to iteratively solve the fusion model to generate the accurate fusion azimuth and fusion zenith. Finally, the initial 3D shape is generated from the fusion azimuth and the fusion zenith by using the normal vector integral algorithm, and the 3D reconstruction result with absolute depth is generated by using linear fitting. The method includes the following steps:

[0008] Step 1: obtain the polarization image by a 2D camera, and obtain the azimuth component and the zenith component of the target surface normal vector by using the polarization 3D reconstruction method, which are represented as polarization azimuth Φ p and polarization zenith Θ p respectively.

[0009] Step 2: based on the polarization modulation ranging technology, obtain the polarization modulation image by using the 2D camera, and further calculate the depth image, then estimate the azimuth component and the zenith component of the target surface normal vector, which are represented as ranging azimuth Φ r and ranging zenith Θ r respectively.

[0010] Step 3: for the polarization azimuth Φ pPre-correction is performed to address the ambiguity issue, and the polarization azimuth angle Φ is selected. p Center and ranging azimuth angle Φ r The difference is greater than The pixels are calculated, and the azimuth angles of these pixels are increased by π radians to obtain the corrected polarization azimuth angle Φ′. p ;

[0011] Step 4: By adjusting the corrected polarization azimuth angle Φ′ p and ranging azimuth Φ r Azimuth fusion model is constructed by separately building data fidelity terms and constraint terms;

[0012] Step 5: By adjusting the polarization zenith angle Θ p and the zenith angle Θ r A zenith angle fusion model is constructed by separately building data fidelity terms and constraint terms;

[0013] Step 6: Iteratively solve the azimuth fusion model and the zenith angle fusion model using the alternating direction multiplier method to obtain the fused azimuth angle Φ. f and the fusion zenith angle Θ f ;

[0014] Step 7: Utilize the fused azimuth angle Φ f and the fusion zenith angle Θ f Surface normal vectors are generated, and the three-dimensional shape is reconstructed based on an integral algorithm. Finally, the reconstructed three-dimensional shape is linearly fitted with the depth measured by polarization modulation ranging to obtain the depth reconstruction result.

[0015] Furthermore, the specific implementation of step 1 includes the following sub-steps:

[0016] Step 1.1: Place a linear polarizer in front of the 2D camera, and set the polarization angle at the specified angles. Polarization images of the target object's surface were acquired at 0°, 45°, 90°, and 135°. According to Fresnel's formula, the change in light intensity at each pixel of the polarization image is related to the polarization angle. The relationship is:

[0017]

[0018] The light intensity change at each point is fitted with a cosine curve, and the maximum light intensity I at each point is obtained from the fitted curve. max Minimum light intensity I min and phase angle φ;

[0019] Step 1.2, polarization azimuth angle of a single pixel It can be obtained from the following formula:

[0020] or

[0021] Step 1.3, the polarization degree p of the object surface of a single pixel point can be obtained by formula (3):

[0022]

[0023] Divide the target surface into a specular reflection region and a diffuse reflection region, for the specular reflection region, the polarization zenith angle θ of a single pixel point p can be obtained by formula (4):

[0024]

[0025] In the formula, n is the reflectivity of the target surface.

[0026] For the diffuse reflection region, the polarization zenith angle θ of a single pixel point p can be obtained by formula (5):

[0027]

[0028] Finally, for an i x j pixel point image, the polarization azimuth angle image Φ p is a matrix composed of all pixel points , and the polarization zenith angle image Θ p is a matrix composed of all pixel points θ p .

[0029] Further, the specific implementation of step 2 includes the following sub-steps:

[0030] Step 2.1, the laser of the pulse laser is expanded and collimated and then vertically irradiated to the surface of the detection target, the pulse laser is triggered externally by a signal source, and the pulse period of the laser is set to T r . Adjust the receiving system to capture the echo signal reflected by the target surface, so that the echo signal passes through the polarization modulation module composed of a polarizer, an electro-optic phase modulator and a polarizer in turn, and then is collected by a two-dimensional camera. Use the signal source to apply a linearly increasing periodic voltage signal to the electro-optic phase modulator, and the period T G of the voltage signal is the same as and in phase with the pulse period T r of the laser emitting pulsed laser. Set the polarization direction of the polarizer to the vertical direction, and then collect the polarization modulation images I -45° and I 45° when the included angle between the polarization angle of the polarizer and the polarization direction of the polarizer is -45° and 45° clockwise, respectively. The relationship between the depth image Z r and the polarization modulation images I -45° and I 45° is

[0031]

[0032] where c is the speed of light.

[0033] Step 2.2, for a single pixel point q = (x, y), the surface normal vector can be calculated by the surface gradient:

[0034]

[0035] where (x0, y0) is the principal point of the camera, and f is the focal length of the camera.

[0036] Ranging zenith angle θ of a single pixel r which can be obtained from equation (8)

[0037]

[0038] Ranging azimuth angle of a single pixel which can be obtained from equation (9)

[0039]

[0040] where θ r ∈ [0, π / 2), For a depth image of i x j pixels, the ranging azimuth angle image Φ r is a matrix composed of θ of all pixel points, the polarization zenith angle image Θ r is a matrix composed of θ r of all pixel points.

[0041] Further, the specific implementation of step 4 includes the following sub-steps:

[0042] Step 4.1, considering the corrected polarization azimuth angle Φ′ p still has certain angular deviation and uncorrected blur error, and its relationship with the fused azimuth angle Φ f can be approximated as linear, while the sparsity of the translation component in the gradient in the linear relationship is described by the total variation constraint (TV constraint). Φ′ p and Φ f can be described by the P1(Φ f ) term:

[0043]

[0044] where denotes the Frobenius norm, S1 is a scaling matrix, T1 is a translation matrix, ⊙ denotes Hadamard product, and λ and γ are balance parameters for different terms.

[0045] Step 4.2, further utilize the R1(Φ f ) term to correct the ranging azimuth angle Φr The azimuth fusion model can be written as f The low-frequency part and the ranging noise:

[0046]

[0047] wherein, Ψ represents a Gaussian low-frequency filter operator, E1 is a noise matrix of Φ r , ||·||1 represents an L1 norm describing the sparsity of the ranging noise, and ρ represents a balance parameter between different terms.

[0048] Based on the formulas (10) and (11), the azimuth fusion model can be written as

[0049]

[0050] Further, the specific implementation of step 5 includes the following sub-steps:

[0051] Step 5.1, when the target surface reflectivity n is unknown or inaccurate, the polarization zenith angle Θ p and the fusion zenith angle Θ f can be simply regarded as a linear relationship, and can be described by using a data fitting term P2(Θ f ):

[0052]

[0053] wherein, S1 is a scaling matrix, and T2 is a translation matrix.

[0054] Step 5.2, further using the R2(Φ f ) term, the ranging zenith angle Θ r can be described as a low-frequency part of the fusion zenith angle Θ f and the ranging noise:

[0055]

[0056] wherein, α represents a balance parameter between different terms, and E2 is a noise matrix of Θ r .

[0057] Based on the formulas (13) and (14), the zenith angle fusion model can be written as

[0058]

[0059] wherein, β represents a balance parameter between different terms.

[0060] Compared with the prior art, the application has the following advantages and beneficial effects:

[0061] The application provides a depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging. Image acquisition is performed on two three-dimensional imaging technologies with complementarity by using a single two-dimensional camera, so that the complex image registration problem in the image fusion process is avoided. Further, the azimuth and zenith angles are calculated based on the depth images obtained by the polarized three-dimensional shape and the polarization modulation ranging respectively, and the azimuth fusion model and the zenith fusion model are constructed to obtain the fused azimuth and zenith angles, so that the respective advantages of the two three-dimensional data can be maximally transmitted to the fusion result. The fused azimuth and the fused zenith are integrated and linearly fitted to obtain the final depth reconstruction result. The depth reconstruction result has high resolution and contains accurate texture detail information and accurate low-frequency absolute depth information. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a schematic diagram of an imaging system device.

[0063] Figure 2 is a grayscale image of the target object of the embodiment.

[0064] Figure 3 is the polarized azimuth, the polarized zenith and the polarized three-dimensional shape of the target of the embodiment, (a) is the polarized azimuth, (b) is the polarized zenith, and (c) is the polarized three-dimensional shape.

[0065] Figure 4 is the ranging azimuth, the ranging zenith and the depth image of the polarization modulation ranging of the target of the embodiment, (a) is the ranging azimuth, (b) is the ranging zenith, and (c) is the depth image of the polarization modulation ranging.

[0066] Figure 5 is the fused azimuth and the fused zenith of the surface of the target object of the embodiment, (a) is the fused azimuth, and (b) is the fused zenith.

[0067] Figure 6 is the depth reconstruction result of the target of the embodiment.

[0068] Figure 7 is a comparison diagram of the depth reconstruction results of different methods of the target of the embodiment.

[0069] Figure 8 is a comparison diagram of the quantitative experimental results of the depth reconstruction results of different methods. DETAILED DESCRIPTION

[0070] In order to facilitate those skilled in the art to understand and implement the application, the application is further described in detail below in combination with the drawings and embodiments. It should be understood that the embodiments described herein are only used to explain the application and do not limit the application.

[0071] The present application is mainly aimed at the application requirement of surface three-dimensional depth reconstruction with fine details. According to the characteristics of polarized three-dimensional shape and polarization modulation ranging, a depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging is proposed. Polarization images and polarization modulation images are collected by a two-dimensional camera, which directly avoids the complex image registration problem in fusion reconstruction. By constructing an azimuth angle fusion model and a zenith angle fusion model, the accurate texture details of polarized three-dimensional shape and the accurate low-frequency information of depth image of polarization modulation ranging are transferred to the fused depth to obtain a high-quality depth fusion result.

[0072] Figure 1 The imaging system device schematic diagram of the proposed depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging is shown in the figure, and the resolution of the two-dimensional CCD camera (Sang Nond SN1200W) used is 4000*3000. Figure 2 is a gray-scale image of the target object. The embodiment provides a depth reconstruction method fusing polarized three-dimensional shape and polarization modulation ranging to reconstruct the three-dimensional depth of the surface of the target object, which specifically comprises the following steps:

[0073] Step 1: obtain the polarization image by the two-dimensional camera, and obtain the azimuth angle component and the zenith angle component of the surface normal vector of the target by the polarized three-dimensional reconstruction method, which are represented as polarization azimuth angle p and polarization zenith angle p respectively. The specific implementation comprises the following sub-steps:

[0074] Step 1.1: place a linear polarizer in front of the CCD camera, and obtain the polarization images of the target object surface at polarization angles of 0°, 45°, 90° and 135° respectively. According to the Fresnel formula, the relationship between the light intensity change of each pixel point of the polarization image and the polarization angle is as follows:

[0075]

[0076] Fit the light intensity change of each point with a cosine curve, and obtain the maximum light intensity I max , the minimum light intensity I min and the phase angle φ of each point according to the fitting curve;

[0077] Step 1.2: the polarization azimuth angle of a single pixel point can be obtained as follows:

[0078] or

[0079] In the calculation, take as the result, and take this value for all targets.

[0080] Step 1.3: The degree of polarization ρ of the object surface at a single pixel can be calculated:

[0081]

[0082] The target surface is divided into specular reflection and diffuse reflection regions. For the specular reflection region, the polarization zenith angle θ of a single pixel is... p It can be found that:

[0083]

[0084] In the formula, n is the reflectivity of the target surface, which is estimated to be 1.5 in the calculation.

[0085] For the diffuse reflection region, the polarization zenith angle θ of a single pixel p It can be found that:

[0086]

[0087] Finally, polarization azimuth image For all pixels The matrix formed, polarization zenith angle image For θ of all pixels p The matrix formed. The polarization azimuth angle, polarization zenith angle, and three-dimensional polarization shape of the target in the embodiment are as follows: Figure 3 As shown.

[0088] Step 2: Based on polarization modulation ranging technology, a polarization modulation image is obtained using a CCD camera, and the depth image is further calculated. Then, the azimuth and zenith angle components of the target surface normal vector are estimated, which are expressed as the ranging azimuth angle Φ. r and the zenith angle Θ r The specific implementation includes the following sub-steps:

[0089] Step 2.1: After expanding and collimating the laser beam from the pulsed laser, illuminate the surface of the target perpendicularly. Trigger the pulsed laser externally from the signal source, and set the laser's pulse period to T. r =100. The receiving system is adjusted to capture the echo signal reflected from the target surface. The echo signal passes sequentially through a polarization modulation module consisting of a polarizer, an electro-optic phase modulator, and an analyzer, and is then acquired by the CCD camera. A linearly increasing periodic voltage signal is applied to the electro-optic phase modulator using a signal source. The period T of the voltage signal is... G The pulse period T of the laser emitted by the laser r Same polarization, consistent phase. The polarization direction of the polarizer is set to vertical, and then polarization modulation images I are acquired when the polarization angle of the analyzer is -45° and 45° clockwise relative to the polarization direction of the polarizer. -45° and I45° Then, the depth image Z r is obtained.

[0090]

[0091] where c = 3 x 10 8 is the speed of light.

[0092] Step 2.2, for a single pixel point q = (x, y), the surface normal vector can be calculated by the surface gradient:

[0093]

[0094] The ranging zenith angle θ r of a single pixel

[0095]

[0096] The ranging azimuth angle of a single pixel

[0097]

[0098] where θ r ∈ [0, π / 2), The ranging azimuth angle image is a matrix composed of θ The polarization zenith angle image is a matrix composed of θ r of all pixel points. The ranging azimuth angle, ranging zenith angle and polarization modulation ranging depth image of the example target are shown in FIG. 6. Figure 4

[0099] Step 3: Pre-correction is performed on the ambiguity problem of the polarization azimuth angle Φ p , and pixel points whose polarization azimuth angle Φ p differs from the ranging azimuth angle Φ r by more than are screened out, and the azimuth angle of these pixel points is added by π radians to obtain the corrected polarization azimuth angle Φ' p ;

[0100] Step 4: A azimuth angle fusion model is constructed by constructing a data fidelity term and a constraint term from the corrected polarization azimuth angle Φ' p and the ranging azimuth angle Φ r , and specifically includes the following sub-steps:

[0101] Step 4.1, considering that the corrected polarization azimuth angle Φ' p still has certain angle deviation and uncorrected ambiguity error, it is fused with the fusion azimuth angle Φ f ​​The relationship can be approximated as linear, and the total variation constraint (TV constraint) is used to describe the sparsity of the translation component on the gradient in the linear relationship. Φ′ p With Φ f The relation can be defined by the relation descriptor P1(Φ) f Item description:

[0102]

[0103] In the formula, Let S1 denote the Frobenius norm, S1 be the scaling matrix, T1 be the translation matrix, ⊙ denote the Hadamard product, and λ and γ be the balancing parameters for different terms.

[0104] Step 4.2, further utilize the relation description term R1(Φ) f The item will measure the azimuth angle Φ. r It can be described as a fused azimuth angle Φ f Low-frequency components and ranging noise:

[0105]

[0106] In the formula, Ψ represents the Gaussian low-frequency filter operator, and E1 is Φ r The noise matrix is ​​denoted by ||·||1, which represents the L1 norm describing the sparsity of the ranging noise, and ρ represents the balance parameter between different terms.

[0107] Based on formulas (10) and (11), the azimuth fusion model can be written as follows:

[0108]

[0109] Step 5: By adjusting the polarization zenith angle Θ p and the zenith angle Θ r To construct the zenith angle fusion model, data fidelity terms and constraint terms are constructed separately, which includes the following sub-steps:

[0110] Step 5.1, when the target surface reflectivity n is unknown or inaccurate, the polarization zenith angle Θ p With the fusion zenith angle Θ f It can be simplified as a linear relationship, and the data fitting term P2(Θ) can be used. f To describe:

[0111]

[0112] In the formula, S1 is the scaling matrix and T2 is the translation matrix.

[0113] Step 5.2, further utilize the relation descriptor R2(Φ) f The item will measure the zenith angle Θ. rThe low-frequency part can be described as fusion zenith angle Θ f and ranging noise:

[0114]

[0115] wherein a represents the balance parameter between different terms.

[0116] Based on formula (13) and (14), the zenith angle fusion model can be written as

[0117]

[0118] wherein β represents the balance parameter between different terms, and E2 is the noise matrix of Θ r .

[0119] Step 6: Based on the alternating direction multiplier method, the azimuth angle fusion model and the zenith angle fusion model are iteratively solved respectively to obtain the fused azimuth angle Φ f and the fused zenith angle Θ f . The fused azimuth angle and the fused zenith angle of the example target are shown in Figure 5 .

[0120] Step 7: The surface normal vector is generated using the fused azimuth angle Φ f and the fused zenith angle Θ f , and the three-dimensional shape is reconstructed based on the integral algorithm. Finally, the reconstructed three-dimensional shape is linearly fitted with the depth of the polarization modulation ranging to obtain the depth reconstruction result. The depth reconstruction result of the example target is shown in Figure 6 .

[0121] Based on the depth reconstruction result of the target object surface obtained in the above steps, in order to compare with other methods, we use the SFPEI [1] , SFP+BSV [2] method to compare with our method, and the results are shown in the accompanying Figure 7 .

[0122] In order to quantitatively evaluate the three-dimensional reconstruction result, we select another target object: a standard sphere, and use the SFPEI, SFP+BSV method to compare with our method, and introduce the mean absolute error (MSE) as the evaluation index, and the results are shown in the accompanying Figure 8 , and the quantitative comparison results are as follows:

[0123] Table 1 Quantitative analysis of different reconstruction methods

[0124]

[0125] It can be seen that the proposed method combines the object surface polarization three-dimensional shape and polarization modulation ranging, better solves the concave-convex ambiguity problem of polarization three-dimensional shape, and the depth reconstruction result has more accurate surface shape and depth information, and has strong reconstruction ability for surface texture characteristics.

[0126] It should be understood that parts not elaborated in the specification are all prior art.

[0127] It should be understood that the above description of the embodiments is more detailed, and therefore should not be considered as limiting the scope of patent protection of the present application. Those skilled in the art can make substitutions or modifications without departing from the scope of protection of the present application, and all fall within the scope of protection of the present application. The scope of protection of the present application shall be subject to the appended claims.

[0128] References

[0129] [1] Smith W A P, Ramamoorthi R, Tozza S. Height-from-polarisation with unknown lighting or albedo [J]. IEEE transactions on pattern analysis and machine intelligence, 2018, 41(12): 2875-2888.

[0130] [2] Tian X, Liu R, Wang Z, et al. High quality 3D reconstruction based on fusion of polarization imaging and binocular stereo vision [J]. Information Fusion, 2022, 77: 19-28.

Claims

1. A depth reconstruction method integrating polarization three-dimensional shape and polarization modulation ranging, characterized in that, Includes the following steps: Step 1: Obtain a polarized image using a 2D camera, and use a polarization 3D reconstruction method to obtain the azimuth and zenith angle components of the target surface normal vector, which are represented as polarization azimuth angles, respectively. and polarization zenith angle ; Step 2: Based on polarization modulation ranging technology, a polarization modulation image is obtained using a two-dimensional camera, and the depth image is further calculated. Then, the azimuth and zenith components of the target surface normal vector are estimated, and expressed as the ranging azimuth angle. and ranging zenith angle ; Step 3: Targeting the polarization azimuth angle Pre-correction is performed to address the ambiguity issue, and the polarization azimuth angle is selected. Center and ranging azimuth angle The difference is greater than The pixels, and the azimuth angle of these pixels plus Radius, to obtain the corrected polarization azimuth angle ; The specific implementation method of obtaining polarization modulation images using a CCD camera in step 2 is as follows; After beam expansion and collimation, the pulsed laser beam is perpendicularly projected onto the surface of the target. The pulsed laser is triggered externally by a signal source, and the pulse period of the laser is set to [value missing]. The receiving system is adjusted to capture the echo signal reflected from the target surface. The echo signal is then passed sequentially through a polarization modulation module consisting of a polarizer, an electro-optic phase modulator, and an analyzer, and finally acquired by a two-dimensional camera. A linearly increasing periodic voltage signal is applied to the electro-optic phase modulator using a signal source, the period of which is... The pulse period of the laser emitted by the laser. With identical polarization and phase, the polarizer's polarization direction is set to vertical. Then, the angle between the analyzer's polarization angle and the polarizer's polarization direction (clockwise) is measured. and Polarization modulation image at time and ; Step 4: By adjusting the corrected polarization azimuth angle and ranging azimuth Azimuth fusion model is constructed by separately building data fidelity terms and constraint terms; Step 5: By adjusting the polarization zenith angle and ranging zenith angle A zenith angle fusion model is constructed by separately building data fidelity terms and constraint terms; Step 6: Iteratively solve the azimuth fusion model and the zenith angle fusion model using the alternating direction multiplier method to obtain the fused azimuth angle. and fusion zenith angle ; Step 7: Utilize the fused azimuth angle and fusion zenith angle Surface normal vectors are generated, and the three-dimensional shape is reconstructed based on an integral algorithm. Finally, the reconstructed three-dimensional shape is linearly fitted with the depth measured by polarization modulation ranging to obtain the depth reconstruction result.

2. The depth reconstruction method that integrates polarization three-dimensional shape and polarization modulation ranging as described in claim 1, characterized in that: Step 1 includes the following sub-steps: Step 1.1: Place a linear polarizer in front of the 2D camera, and set the polarization angle at the specified angles. for , , and At the same time, a polarization image of the target object's surface is acquired. According to Fresnel's formula, the change in light intensity at each pixel of the polarization image is related to the polarization angle. The relationship is: (1) The light intensity change at each point is fitted with a cosine curve, and the maximum light intensity at each point is obtained from the fitted curve. Minimum light intensity and phase angle ; Step 1.2, polarization azimuth angle of a single pixel It can be obtained from the following formula: (2) Step 1.3, Polarization degree of the object surface of a single pixel. It can be obtained from equation (3): (3) The target surface is divided into specular reflection and diffuse reflection regions. For the specular reflection region, the polarization zenith angle of a single pixel is... It can be obtained from equation (4): (4) In the formula, The reflectivity of the target surface; For the diffuse reflection region, the polarization zenith angle of a single pixel It can be obtained from equation (5): (5) Finally, for An image of 1 pixel, its polarization azimuth image For all pixels The matrix formed, polarization zenith angle image For all pixels The matrix formed.

3. The depth reconstruction method that integrates polarization three-dimensional shape and polarization modulation ranging as described in claim 1, characterized in that: Depth image in step 2 With polarization modulation image and The relationship is: (6) In the formula, The speed of light; Step 2.2, for a single pixel The surface normal vector can be calculated from the surface gradient: (7) In the formula, It is the main point of the camera. It is the camera's focal length; Zenith angle of a single pixel It can be obtained from equation (8) (8) Azimuth of a single pixel It can be obtained from equation (9) (9) In the formula, , , ;for A depth image of 1 pixel, and its ranging azimuth image For all pixels The matrix formed, polarization zenith angle image For all pixels The matrix formed.

4. The depth reconstruction method that integrates polarization three-dimensional shape and polarization modulation ranging as described in claim 1, characterized in that: The azimuth fusion model is constructed in step 4 as follows; Step 4.1, taking into account the corrected polarization azimuth angle It still has a certain angular deviation and uncorrected fuzziness error, which, when compared with the fused azimuth angle... The relationship can be approximated as linear, and the sparsity of the translation component on the gradient in the linear relationship can be described using total variation constraints. and Relationships can be defined by relation descriptors Item Description: (10) In the formula, Describing the Frobenius norm, It is a scaling matrix. It is a translation matrix. Represents the Hadama product. and These are the balance parameters for different terms; Step 4.2, further utilize relational description items The item will measure the azimuth angle. It can be described as a fused azimuth angle Low-frequency components and ranging noise: (11) In the formula, This represents the Gaussian low-frequency filtering operator. for The noise matrix, The L1 norm represents the sparsity of ranging noise. This represents the balance parameter between different terms; Based on formulas (10) and (11), the azimuth fusion model can be written as: (12)。 5. The depth reconstruction method that integrates polarization three-dimensional shape and polarization modulation ranging as described in claim 1, characterized in that: The zenith angle fusion model is constructed in step 5 as follows; Step 5.1, when the target surface reflectivity When unknown or inaccurate, the polarization zenith angle With fusion zenith angle It can be simplified as a linear relationship, and data fitting terms can be used. To describe: (13) In the formula, It is a scaling matrix. It is a translation matrix; Step 5.2, further utilize relational description items The project will measure the zenith angle. It can be described as a fused zenith angle Low-frequency components and ranging noise: (14) In the formula, This represents the balance parameter between different terms; Based on formulas (13) and (14), the zenith angle fusion model can be written as follows: (15) In the formula, This represents the balance parameter between different terms. for The noise matrix.

6. The depth reconstruction method that integrates polarization three-dimensional shape and polarization modulation ranging as described in claim 1, characterized in that: It also includes step 8, which uses the SFPEI, SFP+BSV method to obtain the depth reconstruction results and compares them with the depth reconstruction results obtained in step 7, and introduces the mean absolute error as an evaluation index to quantitatively evaluate the 3D reconstruction results.

Images