A perspective imaging polarization three-dimensional reconstruction method

Abstract

The present application provides a kind of perspective imaging polarization three-dimensional reconstruction method.The process includes: A1, using polarization camera to shoot target polarization image;A2, intensity map, polarization degree map and phase angle map are solved using polarization image calculation;A3, using intensity map, rough depth map about object surface is solved by perspective projection dark shadow recovery shape method;A4, using polarization degree map and phase angle map to solve object surface normal vector;A5, using rough depth map to optimize object surface normal vector;A6, using the normal vector after optimization and integral algorithm to solve target high-precision depth map;A7, high-precision depth map is represented by point cloud, and final three-dimensional reconstruction is realized.The present application can be used to alleviate the polarization three-dimensional imaging error caused by the perspective effect of polarization camera, realize high-precision polarization three-dimensional imaging, and can be widely applied in target detection, automatic driving and other fields.

Description

Technical Field

[0001] This invention relates to a perspective imaging polarization 3D reconstruction method, which alleviates polarization 3D imaging errors caused by the perspective effect of polarization cameras, and belongs to the field of computational optics. Background Technology

[0002] As digitalization deepens, people's demand for multi-dimensional target information continues to increase. Compared to traditional two-dimensional imaging methods, three-dimensional imaging can provide comprehensive spatial data, including depth and height, breaking through the two-dimensional limitations of traditional imaging methods. For humans, it is closer to how we perceive the objective world, providing more intuitive information. For computers, higher-dimensional information can facilitate better reasoning and judgment. Today, three-dimensional imaging is used in fields such as underwater exploration, autonomous driving, and remote sensing terrain exploration. Compared to other three-dimensional imaging methods, polarization three-dimensional imaging has advantages such as strong resistance to environmental interference, long imaging distance, and suitability for low-texture targets. Polarization three-dimensional imaging has achieved high-precision three-dimensional reconstruction of single targets, achieving good results in both qualitative and quantitative evaluation.

[0003] Currently, polarization-based 3D imaging still has a long way to go before it can be applied in practical engineering. This is because most previous polarization-based 3D research relied on the imaging assumption of orthogonal projection. However, actual camera models are constructed based on pinhole imaging models, where the incident light rays and photosensitive pixels satisfy a perspective projection relationship. Perspective projection inevitably introduces errors into polarization-based 3D imaging. In previous studies, to reduce errors caused by perspective effects, the target surface to be reconstructed was usually positioned directly facing the lens and at the center of the image. It is often difficult for the lens and the target to maintain this relative positional relationship, leading to significant errors in polarization-based 3D reconstruction results in practical applications, especially when the target appears at the edge of the lens's image. This critical issue has received attention, but currently, there is no method that can truly solve the errors caused by perspective projection and bring polarization-based 3D imaging to practical applications.

[0004] This invention discloses a perspective imaging polarization-based 3D reconstruction method. This method utilizes a polarization camera to capture a polarization image of the target. An intensity map, polarization degree map, and phase angle map are then derived from the polarization image. A rough target surface depth map is obtained using the intensity map and the perspective shadow recovery shape method. Based on the characteristic that the rough depth map estimated by the perspective shadow recovery shape method is unaffected by camera perspective effects, this rough depth map is used to optimize the target surface normal vector obtained from the polarization degree map and phase angle map, thus mitigating the polarization-based 3D reconstruction error caused by perspective effects. The optimized normal vector is integrated to obtain a high-precision depth map, which is then represented as a point cloud to obtain the final 3D reconstruction result. Using this method, 3D reconstruction can achieve high-precision polarization-based 3D reconstruction without requiring the target to be directly facing the camera lens or centered in the image, facilitating its application in practical production. Summary of the Invention

[0005] The purpose of this invention is to provide a method that can alleviate perspective effect errors and facilitate the practical application of polarization three-dimensional imaging.

[0006] The objective of this invention is achieved through the following technical means:

[0007] A method for polarization-based three-dimensional reconstruction using perspective imaging, comprising:

[0008] A1, Use a polarization camera to capture a polarization image of the target;

[0009] A2 uses polarization images to calculate and solve for the intensity map, degree of polarization map, and phase angle map;

[0010] A3. Using the intensity map, the roughness depth map of the object's surface is obtained by restoring the shape through perspective projection and dark shadows.

[0011] A4, use the polarization degree diagram and phase angle diagram to solve for the surface normal vector of the object;

[0012] A5, using a rough depth map to optimize the surface normal vectors of an object;

[0013] A6 uses optimized normal vectors and integral algorithms to solve for high-precision depth maps of the target;

[0014] A7 uses point cloud representation to represent the depth map, thus achieving the final 3D reconstruction.

[0015] Furthermore, the intensity map, polarization degree map, and phase angle map in A2 can be obtained using the polarization image captured in A1, and they satisfy the following relationship:

[0016] (1)

[0017] in For polarization angle, Phase angle; This represents the maximum intensity that the light reflected from the target surface can achieve after passing through a polarizer at different polarization angles; conversely, This indicates the minimum strength that can be achieved.

[0018] Using polarization images from at least three different angles and the linear least squares method to solve formula (1), we obtain... , and Then, further solving yields the intensity map and polarization map. and With intensity map And polarization diagram The following relationship must be satisfied:

[0019] (2)

[0020] The intensity diagram, polarization diagram, and phase angle diagram obtained from formula (2) are as follows: Figure 1 , Figure 2 and Figure 3 As shown.

[0021] Furthermore, in A3, the focal length of the polarizing camera lens is set to... Then the points on the surface of the object Projection point on the camera imaging plane Satisfy Figure 4 The geometric relationships shown. Strength With the surface normal vector of the object The following relationship must be satisfied:

[0022] (3)

[0023] in, This indicates the direction of the light source; let's assume... ,but and express exist and Partial derivatives in direction.

[0024] In the perspective projection dark shadow shape recovery method, the rough depth map can be obtained by solving formula (3) using the fast travel method.

[0025] Furthermore, in A4, the polarization degree diagram and phase angle diagram are related to the object surface normal vector. The following relationship must be satisfied:

[0026] (4)

[0027] (5)

[0028] (6)

[0029] Furthermore, in A5, the normal vector map is solved using the finite difference method based on the roughness depth map. Optimize using energy cost function To obtain the accurate surface normal vector of the object. .

[0030] Furthermore, in A6, the optimized normal vector is used. The final depth map is obtained by solving the integral formula. The integral formula is shown below:

[0031] (7)

[0032] in, This represents the inverse discrete Fourier transform. Represents the Discrete Fourier Transform; and They represent the normal vectors respectively. exist and Components in direction; It represents frequency domain coordinates.

[0033] Furthermore, in A7, in the known depth map In this case, by calibrating the parameters of the polarization camera, the coordinates of points on the object's surface in three-dimensional space can be determined. The final 3D reconstruction is completed by using point clouds to represent all coordinate points on the surface of the object. Attached Figure Description

[0034] Figure 1 It is a target intensity map obtained by taking and calculating images using a polarization camera.

[0035] Figure 2 It is a target polarization map obtained by taking and calculating images using a polarization camera.

[0036] Figure 3 It is a target phase angle map obtained by taking pictures and calculating with a polarization camera.

[0037] Figure 4 A schematic diagram of the geometric relationship between points on the object's surface and the projection points on the camera's imaging plane.

[0038] Figure 5 This is a flowchart of a perspective imaging polarization three-dimensional reconstruction method according to the present invention. Detailed Implementation

[0039] The present invention will be further illustrated below with reference to specific embodiments.

[0040] like Figure 4 As shown, the present invention provides a perspective imaging polarization three-dimensional reconstruction method, comprising:

[0041] A1, Use a polarization camera to capture a polarization image of the target;

[0042] A2 uses polarization images to calculate and solve for the intensity map, degree of polarization map, and phase angle map;

[0043] A3. Using the intensity map, the roughness depth map of the object's surface is obtained by restoring the shape through perspective projection and dark shadows.

[0044] A4, use the polarization degree diagram and phase angle diagram to solve for the surface normal vector of the object;

[0045] A5, using a rough depth map to optimize the surface normal vectors of an object;

[0046] A6 uses optimized normal vectors and integral algorithms to solve for high-precision depth maps of the target;

[0047] A7 represents high-precision depth maps in three dimensions to achieve final three-dimensional reconstruction.

[0048] Specifically, the intensity map, linear polarization map, and phase angle map of step A2 can be obtained using MATLAB software. The polarization images (0°, 45°, 90°, and 135°) captured by a polarization camera are used as input to formula (1), and the phase angle map is obtained by solving the linear least squares method. and Furthermore, with and Using the input formula (2), the intensity map and polarization map can be obtained.

[0049] Specifically, in step A3, the intensity map is used as input. According to formula (3), the "seed points" of the algorithm are set from the intensity map. Starting from these points, the depth information is propagated outward through the fast traversal method to obtain a coarse depth map.

[0050] Specifically, in step A4, using the polarization degree diagram and phase angle diagram as input, the surface normal vector of the object can be obtained by calculating formulas (4), (5) and (6). .

[0051] Specifically, in step A5, the roughness depth map is used as input, and the normal vector map is obtained through the finite difference method. Iterative optimization is achieved through the energy cost function. This minimizes the energy cost function, ultimately yielding the accurate surface normal vector of the object. Among them, the data item optimized in the energy cost function is... and The difference between them, the smoothing term optimizes is Abnormal mutations in adjacent data.

[0052] Specifically, in step A6, the optimized object surface normal vector is used. As input, these values ​​are substituted into formula (7) to obtain the accurate depth map. .

[0053] Specifically, in step A7, the polarization camera used in the experiment is calibrated to obtain the camera's intrinsic and extrinsic parameters, thereby obtaining the depth map. The coordinates of each pixel in the image are given in three-dimensional space. Using these coordinates and the camera's intrinsic and extrinsic parameters, a point cloud algorithm is used to obtain the point cloud of points on the target surface, thus completing the final three-dimensional shape representation of the target.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to specific embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention. Technical, shape, and structural parts not described in detail in this invention are all well-known technologies.

Claims

1. A method for polarization-based three-dimensional reconstruction using perspective imaging, the specific process of which includes the following steps: A1, Use a polarization camera to capture a polarization image of the target; A2 uses polarization images to calculate and solve for the intensity map, degree of polarization map, and phase angle map; A3. Using the intensity map, the roughness depth map of the object's surface is obtained by restoring the shape through perspective projection and dark shadows. A4, use the polarization degree diagram and phase angle diagram to solve for the surface normal vector of the object; A5, using a rough depth map to optimize the surface normal vectors of an object; A6 uses optimized normal vectors and integral algorithms to solve for high-precision depth maps of the target; A7 uses point cloud representation to represent high-precision depth maps, achieving final 3D reconstruction.

2. The perspective imaging polarization three-dimensional reconstruction method according to claim 1, characterized in that: This paper integrates perspective projection shadow shape restoration with polarized 3D imaging. Leveraging the fact that perspective shadow shape restoration is not affected by the perspective effect of polarized cameras, a rough depth map of the target is estimated. The rough depth map and an energy optimization function are then used to optimize the surface normal vectors of the object obtained using polarization information. The optimized surface normal vectors and an integral algorithm are then used to obtain a high-precision depth map and 3D reconstruction results. This mitigates the polarized 3D imaging errors caused by the perspective effect of polarized cameras to a certain extent, enabling polarized 3D imaging to perform 3D reconstruction of scene targets and obtain high-precision 3D reconstruction results under different camera poses, thus improving its general applicability.

Images