A method and system for depth-gradient consistency depth information completion based on sub-pixel edge guidance

Abstract

This invention discloses a method and system for depth-gradient consistency depth information completion based on subpixel edge guidance, comprising: acquiring color image information and depth image information of reactor internal components; performing subpixel-level edge detection of reactor internal components using Zernike moment edge detection on the color image; constructing a classification rule for depth value missing regions based on edge distance field and direction perception mechanism; performing initial completion on Class A depth value missing regions and optimizing depth-gradient consistency by constructing an energy function; performing completion on Class B / C depth value missing regions using a bilateral filtering method; obtaining a geometrically consistent depth map based on local enhancement of edge regions and Poisson reconstruction; this invention achieves depth-gradient consistency depth information completion based on subpixel edge guidance, and combined with the principle of binocular vision, realizes unmanned remote operation and high-precision guidance and positioning of reactor internal components.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method and system for depth-gradient consistency depth information completion based on sub-pixel edge guidance. Background Technology

[0002] The lower reactor internals are circular objects with a diameter of 6m. When positioning and installing the internals, the binocular system-assisted positioning method is more efficient than the traditional manual positioning method and can effectively reduce the safety hazards of radiation to workers.

[0003] However, when measuring the components inside the stack using a binocular system, it often happens that some points on the edge of the object are visible in one image but not in another due to the difference in the viewing angles of the left and right cameras. This phenomenon will seriously affect the stereo matching effect and cause the depth map to have missing depth information at the edge of the object. Summary of the Invention

[0004] In view of this, the present invention provides a depth-gradient consistency depth information completion method and system based on sub-pixel edge guidance to solve some of the technical problems mentioned in the background.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The depth-gradient consistency depth information completion method based on sub-pixel edge guidance includes the following steps:

[0007] S1. Acquire color and depth image information of the internal components of the reactor;

[0008] S2. Use Zernike moment edge detection on the color image to perform subpixel-level edge detection of internal reactor components and obtain subpixel coordinates;

[0009] S3. Construct a classification rule for depth value missing regions based on edge distance field and direction perception mechanism to obtain Class A and Class B / C depth value missing regions;

[0010] S4. Perform initial completion on the missing depth value regions of type A, and optimize the depth-gradient consistency of the depth map after initial completion by constructing an energy function; perform completion on the missing depth value regions of types B / C using a bilateral filtering method.

[0011] S5. Based on local enhancement of the edge region, and by returning the locally reconstructed gradient field to the depth map through Poisson reconstruction, a geometrically consistent depth map is obtained.

[0012] Preferably, the specific content of step S2 is as follows:

[0013] The four parameters required to determine the edge are:

[0014]

[0015] Where g is the background grayscale value, p is the step height, and l is the vertical distance from the center of the disk to the edge. Z is the angle between the edge normal and the x-axis. nm For the nth order m-th Zernike moment of a continuous image f(x,y), let z nm Rotation Get Z' nm ;

[0016]

[0017] Where n and m are integers, and n≥0, n-|m| is even, n>|m|, ρ is the edge position, θ is the angle between vector ρ and the x-axis, and V nm It is an integral function. is the complex conjugate of the integral function;

[0018] Sub-pixel coordinates (x) obtained from sub-pixel edge detection based on Zernike moments s ,y s )for:

[0019]

[0020] Where N is the template size used when calculating Zernike moments on a discrete image.

[0021] Preferably, in step S3, the classification rule for regions with missing depth values ​​based on the edge distance field and orientation perception mechanism is as follows:

[0022]

[0023] Where, θ l The angle between the line connecting the point with missing depth values ​​to the nearest edge point and the x-axis. The angle between the edge normal and the x-axis is denoted by , and Dis is the distance between the point where the depth value is missing and the nearest edge point.

[0024]

[0025] Where (x,y) are the coordinates of the points where depth information is missing, (x s ,y s ) represents the sub-pixel edge coordinates, and ε represents the sub-pixel edge point set.

[0026] Preferably, the specific content of step S4 is as follows:

[0027] S41. Perform initial completion on regions with missing depth values ​​in category A:

[0028]

[0029] Where (x,y) are the pixel coordinates of the missing depth information points in class A, and D valid (x k ,y k ) represents the depth value of the effective depth point within the neighborhood. Δx is the angle between the normal of the edge point closest to the missing depth value and the x-axis. k with Δy k Indicates coordinate offset, Δx k =x k -x, Δy k =y k -y, w k Direction-aware weights are used to guide the repair of vacant areas;

[0030] S42. Establish gradient magnitude field M target Used to control the trend of depth value changes in edge regions:

[0031]

[0032] S43. Establish the gradient direction field Φ target This ensures that the depth information gradient near the edge is consistent with the edge normal gradient:

[0033]

[0034] in, This represents the direction of the normal to the edge point closest to point (x, y). This represents the gradient vector at a certain depth point after bilateral filtering.

[0035] S44. Establish the target gradient field, combining the magnitude and direction to form the target gradient field:

[0036]

[0037] S45. Establish an energy function to achieve depth-gradient consistency optimization:

[0038]

[0039] Where D represents the depth information to be optimized, D k This is the depth information after bilateral filtering. To optimize the gradient of the depth map, valid represents the region with valid depth values, and Ω represents the region with missing depth information.

[0040] Preferably, step S4, which involves using a bilateral filtering method to complete the missing depth values ​​in the B / C class regions, is as follows:

[0041]

[0042]

[0043] Where S is the filtering window, p is the pixel at the current position, q is the pixel in the neighborhood, and σ s Let σ be the standard deviation parameter of the spatial domain. r I is the standard deviation parameter for the range. p I is the pixel value / grayscale value at the current position p. q W represents the pixel value / grayscale value at neighborhood location q. q This represents the weighted sum of each pixel value within the filtering window.

[0044] Preferably, step S5, based on the method of local enhancement of the edge region, is as follows:

[0045]

[0046] in, M is the gradient of the depth map after the energy function optimization. target (x,y) represents the target gradient magnitude. The direction of the edge normal;

[0047] The method for returning the locally reconstructed gradient field to the depth map using Poisson reconstruction is as follows:

[0048]

[0049] Preferably, the depth-gradient consistency depth information completion method based on sub-pixel edge guidance further includes: S6. using a target detection algorithm and an HSV threshold segmentation algorithm to obtain the pixel coordinates of internal reactor components, using the Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera, and combining the optimized depth map to complete the three-dimensional reconstruction based on the binocular vision principle, as well as the visualization of the localization results.

[0050] The depth-gradient consistency depth information completion system based on subpixel edge guidance, based on the aforementioned depth-gradient consistency depth information completion method based on subpixel edge guidance, includes an image information acquisition module, an image information processing module, and an image information visualization module.

[0051] The image information acquisition module is used to scan and photograph the internal components of the reactor using a binocular structured light camera to acquire color image information and depth image information of the internal components of the reactor.

[0052] The image information acquisition module is used to perform sub-pixel level edge detection of reactor internal components using Zernike moment edge detection on color images to obtain sub-pixel coordinates. It then uses a pre-constructed depth value missing region classification rule based on edge distance field and orientation awareness mechanism to obtain class A and class B / C depth value missing regions. For class A depth value missing regions, initial completion is performed, and depth-gradient consistency optimization is performed on the initially completed depth map using an energy function. For class B / C depth value missing regions, bilateral filtering is used for completion. Based on local enhancement of edge regions, the locally reconstructed gradient field is returned to the depth map through Poisson reconstruction to obtain a geometrically consistent depth map. Finally, the module uses object detection algorithms and HSV threshold segmentation algorithms to obtain the pixel coordinates of reactor internal components, and Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera. Combined with the optimized depth map, a 3D reconstruction based on binocular vision principles is completed.

[0053] The image information visualization module is used to display the 3D reconstruction results and positioning results in real time.

[0054] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned depth-gradient consistency depth information completion method based on subpixel edge guidance.

[0055] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the aforementioned depth-gradient consistency depth information completion method based on sub-pixel edge guidance.

[0056] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for depth-gradient consistency depth information completion based on subpixel edge guidance, which realizes depth-gradient consistency depth information completion guided by subpixel edge guidance. Combined with the principle of binocular vision, it realizes unmanned remote operation and high-precision guidance and positioning of reactor internal components, thereby improving the efficiency and automation level of reactor internal component positioning and installation.

[0057] Specifically, Zernike moments are used to extract sub-pixel precision edge positions and normal directions. Based on this, a depth value missing region classification mechanism based on the direction perception mechanism of the edge distance field is established. Different depth information completion algorithms are applied to different categories of depth information missing regions. A target gradient field strictly aligned with the edges of the color image is constructed to guide the restoration of depth information. Depth-gradient consistency is achieved by gradient-constrained energy functions. At the same time, the coordinated work of local gradient enhancement and Poisson reconstruction realizes the global optimization of the depth map. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0059] Figure 1 This is a schematic diagram of the depth-gradient consistency depth information completion method based on sub-pixel edge guidance provided by the present invention.

[0060] Figure 2 A schematic diagram illustrating the principle of the depth-gradient consistency depth information completion method based on sub-pixel edge guidance provided by the present invention;

[0061] Figure 3 This is a schematic diagram of the training process of the object detection model based on YOLOv5 provided by the present invention;

[0062] Figure 4 A schematic diagram of the ROI region calculation process for a stereo system based on target detection provided by this invention;

[0063] Figure 5 This is a schematic diagram of the depth-gradient consistency depth information completion system based on sub-pixel edge guidance provided by the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] This invention discloses a depth-gradient consistency depth information completion method based on sub-pixel edge guidance, such as... Figure 1 and Figure 2 This includes the following steps:

[0066] S1. Acquire color and depth image information of the internal components of the reactor;

[0067] S2. Use Zernike moment edge detection on the color image to perform subpixel-level edge detection of internal reactor components and obtain subpixel coordinates;

[0068] S3. Construct a classification rule for depth value missing regions based on edge distance field and direction perception mechanism to obtain Class A and Class B / C depth value missing regions;

[0069] S4. Perform initial completion on the missing depth value regions of type A, and optimize the depth-gradient consistency of the depth map after initial completion by constructing an energy function; perform completion on the missing depth value regions of types B / C using a bilateral filtering method.

[0070] S5. Based on local enhancement of the edge region, and by returning the locally reconstructed gradient field to the depth map through Poisson reconstruction, a geometrically consistent depth map is obtained.

[0071] To further implement the above technical solution, the specific content of step S2 is as follows:

[0072] The four parameters required to determine the edge are:

[0073]

[0074] Where g is the background grayscale value, p is the step height, and l is the vertical distance from the center of the disk to the edge. Z is the angle between the edge normal and the x-axis. nm For the nth order m-th Zernike moment of a continuous image f(x,Y), let Z... nm Rotation Get Z' nm ;

[0075]

[0076] Where n and m are integers, and n≥0, n-|m| is even, n>|m|, ρ is the edge position, θ is the angle between vector ρ and the x-axis, and V nm It is an integral function. is the complex conjugate of the integral function;

[0077] Sub-pixel coordinates (x) obtained from sub-pixel edge detection based on Zernike moments s ,y s )for:

[0078]

[0079] Where N is the template size used when calculating Zernike moments on a discrete image.

[0080] To further implement the above technical solution, in step S3, the classification rule for regions with missing depth values ​​based on the edge distance field and orientation sensing mechanism is as follows:

[0081]

[0082] Where, θ l The angle between the line connecting the point with missing depth values ​​to the nearest edge point and the x-axis. The angle between the edge normal and the x-axis is denoted by , and Dis is the distance between the point where the depth value is missing and the nearest edge point.

[0083]

[0084] Where (x,y) are the coordinates of the points where depth information is missing, (x s ,y s ) represents the sub-pixel edge coordinates, and ε represents the sub-pixel edge point set.

[0085] In this embodiment, the rule for determining whether depth information is missing is as follows:

[0086]

[0087] When fl is 1, it means that the depth value of the pixel is missing; when fl is 0, it means that the depth value of the pixel is not missing.

[0088] To further implement the above technical solution, the specific content of step S4 is as follows:

[0089] S41. Perform initial completion on regions with missing depth values ​​in category A:

[0090]

[0091] Where (x,y) are the pixel coordinates of the missing depth information points in class A, and D valid (x k ,y k ) represents the depth value of the effective depth point within the neighborhood. Δx is the angle between the normal of the edge point closest to the missing depth value and the x-axis. k with Δy k Indicates coordinate offset, Δx k =x k -x, Δy k =y k -y, w k Direction-aware weights are used to guide the repair of vacant areas;

[0092] S42. Establish gradient magnitude field M target Used to control the trend of depth value changes in edge regions:

[0093]

[0094] Near the edge, the depth value changes more dramatically, while further away from the edge, the depth value changes less dramatically.

[0095] S43. Establish the gradient direction field Φ target This ensures that the depth information gradient near the edge is consistent with the edge normal gradient:

[0096]

[0097] in, This represents the direction of the normal to the edge point closest to point (x, y). This represents the gradient vector at a certain depth point after bilateral filtering.

[0098] S44. Establish the target gradient field, combining the magnitude and direction to form the target gradient field:

[0099]

[0100] S45. Establish an energy function to achieve depth-gradient consistency optimization:

[0101]

[0102] Where D represents the depth information to be optimized, D k This is the depth information after bilateral filtering. To optimize the gradient of the depth map, valid represents the region with valid depth values, and Ω represents the region with missing depth information.

[0103] To further implement the above technical solution, step S4, which involves using a bilateral filtering method to fill in the missing depth values ​​in the B / C class regions, is as follows:

[0104]

[0105] Where S is the filtering window, p is the pixel at the current position, q is the pixel in the neighborhood, and σ s Let σ be the standard deviation parameter of the spatial domain. r I is the standard deviation parameter for the range. p I is the pixel value / grayscale value at the current position p. q W represents the pixel value / grayscale value at neighborhood location q. q This represents the weighted sum of each pixel value within the filtering window.

[0106] To further implement the above technical solution, step S5, based on the method of local enhancement of the edge region, is as follows:

[0107]

[0108] in, M is the gradient of the depth map after the energy function optimization. target (x,y) represents the target gradient magnitude. The direction of the edge normal;

[0109] The method for returning the locally reconstructed gradient field to the depth map using Poisson reconstruction is as follows:

[0110]

[0111] To further implement the above technical solution, the depth-gradient consistency depth information completion method based on sub-pixel edge guidance also includes: S6. Using a target detection algorithm and an HSV threshold segmentation algorithm to obtain the pixel coordinates of internal reactor components, using the Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera, and combining the optimized depth map to complete the 3D reconstruction based on the binocular vision principle, as well as the visualization of the localization results.

[0112] In this embodiment, the object detection algorithm is a pre-trained YOLOv5-based object detection model. A dataset of color images of the target object is collected to train the YOLOv5-based object detection model, such as... Figure 3 The specific training process is as follows: within the target detection result range, pixel coordinates are obtained through the HSV threshold segmentation algorithm. Since the internal components and the hoisting mechanism are rigidly connected, color blocks of different colors are set on the surface of the hoisting mechanism of the internal components. The positioning coordinates of the internal components are obtained through the relative spatial relationship between the internal components and the hoisting mechanism.

[0113] In practical applications, the intrinsic parameter matrix of a binocular structured light camera is: The translation matrix is ​​[t] x ,t y ,t z ] T , where f x f y These represent the focal lengths on the x-axis and y-axis, respectively, and u0 and v0 are the optical centers of the image.

[0114] A depth-gradient consistency depth information completion system based on sub-pixel edge guidance, such as Figure 5 The method for depth-gradient consistency depth information completion based on subpixel edge guidance includes an image information acquisition module, an image information processing module, and an image information visualization module.

[0115] The image information acquisition module is used to scan and photograph the internal components of the reactor using a binocular structured light camera to acquire color image information and depth image information of the internal components of the reactor.

[0116] The image information acquisition module is used to perform sub-pixel level edge detection of reactor internal components using Zernike moment edge detection on color images to obtain sub-pixel coordinates. It then uses a pre-constructed depth value missing region classification rule based on edge distance field and orientation awareness mechanism to obtain class A and class B / C depth value missing regions. For class A depth value missing regions, initial completion is performed, and depth-gradient consistency optimization is performed on the initially completed depth map using an energy function. For class B / C depth value missing regions, bilateral filtering is used for completion. Based on local enhancement of edge regions, the locally reconstructed gradient field is returned to the depth map through Poisson reconstruction to obtain a geometrically consistent depth map. Finally, the module uses object detection algorithms and HSV threshold segmentation algorithms to obtain the pixel coordinates of reactor internal components, and Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera. Combined with the optimized depth map, a 3D reconstruction based on binocular vision principles is completed.

[0117] The image information visualization module is used to display the 3D reconstruction results and positioning results in real time.

[0118] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements a depth-gradient consistency depth information completion method based on subpixel edge guidance.

[0119] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements a depth-gradient consistency depth information completion method based on subpixel edge guidance.

[0120] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0121] 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 depth-gradient consistency depth information completion method based on sub-pixel edge guidance, characterized in that, Includes the following steps: S1. Acquire color and depth image information of the internal components of the reactor; S2. Use Zernike moment edge detection on the color image to perform subpixel-level edge detection of internal reactor components and obtain subpixel coordinates; S3. Construct a classification rule for depth value missing regions based on edge distance field and direction perception mechanism to obtain Class A and Class B / C depth value missing regions; S4. Perform initial completion on the missing depth value regions of type A, and optimize the depth-gradient consistency of the depth map after initial completion by constructing an energy function; perform completion on the missing depth value regions of types B / C using a bilateral filtering method. S5. Based on local enhancement of the edge region, and by returning the locally reconstructed gradient field to the depth map through Poisson reconstruction, a geometrically consistent depth map is obtained. The specific content of step S4 is as follows: S41. Perform initial completion on regions with missing depth values ​​in category A: in, The pixel coordinates of the points where depth information is missing in Class A. The depth value of the effective depth point in the neighborhood. The normal of the edge point closest to the point where the depth value is missing is The angle between the axes, and Indicates coordinate offset. , , Direction-aware weights are used to guide the repair of vacant areas; S42. Establish the gradient magnitude field Used to control the trend of depth value changes in edge regions: S43. Establish gradient direction field This ensures that the depth information gradient near the edge is consistent with the edge normal gradient: in, Indicates distance point The normal direction of the nearest edge point This represents the gradient vector at a certain depth point after bilateral filtering. S44. Establish the target gradient field, combining the magnitude and direction to form the target gradient field: S45. Establish an energy function to achieve depth-gradient consistency optimization: in, For depth information to be optimized, This is the depth information after bilateral filtering. To optimize the gradient of the depth map, Ω represents the valid depth value region, and Ω represents the region where depth information is missing.

2. The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to claim 1, characterized in that, The specific content of step S2 is as follows: The four parameters required to determine the edge are: in, The background grayscale value. The step height, The vertical distance from the center of the disk to its edge. The angle between the edge normal and the x-axis. For continuous images of Step The second Zernike moment, Rotation get ; in, It is an integer, and , Even number, , For edge positions, For vectors and The included angle of the axis, It is an integral function. is the complex conjugate of the integral function; Sub-pixel coordinates obtained from sub-pixel edge detection based on Zernike moments for: in, The template size used when calculating Zernike moments on a discrete image.

3. The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to claim 1, characterized in that, Step S3, the classification rule for regions with missing depth values ​​based on edge distance field and orientation awareness mechanism is as follows: in, The angle between the line connecting the point with missing depth values ​​to the nearest edge point and the x-axis. The angle between the edge normal and the x-axis. The distance between the point with missing depth values ​​and the nearest edge point; in, The coordinates of the points where depth information is missing. Subpixel edge coordinates It is a sub-pixel edge point set.

4. The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to claim 1, characterized in that, Step S4, specifically the use of bilateral filtering to complete the missing depth values ​​in classes B / C, is as follows: in, For filtering window, The pixel at the current position. For pixels in the neighborhood, For the spatial domain standard deviation parameter, The parameter is the standard deviation of the range. Current location pixel value / grayscale value, for neighborhood location Pixel value / grayscale value at that location This represents the weighted sum of each pixel value within the filtering window.

5. The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to claim 1, characterized in that, Step S5, based on the method of local enhancement of the edge region, is as follows: in, The gradient of the depth map after energy function optimization. For the target gradient magnitude, The direction of the edge normal; The method for returning the locally reconstructed gradient field to the depth map using Poisson reconstruction is as follows: 。 6. The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to claim 1, characterized in that, Also includes: S6. Use the target detection algorithm and HSV threshold segmentation algorithm to obtain the pixel coordinates of the internal components of the reactor, use the Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera, and combine the optimized depth map to complete the 3D reconstruction based on the binocular vision principle and visualize the localization results.

7. A depth-gradient consistency depth information completion system based on sub-pixel edge guidance, characterized in that, The depth-gradient consistency depth information completion method based on sub-pixel edge guidance according to any one of claims 1-6 includes an image information acquisition module, an image information processing module, and an image information visualization module. The image information acquisition module is used to scan and photograph the internal components of the reactor using a binocular structured light camera to acquire color image information and depth image information of the internal components of the reactor. The image information acquisition module is used to perform sub-pixel-level edge detection of reactor internal components using Zernike moment edge detection on color images to obtain sub-pixel coordinates. It then uses a pre-constructed depth value missing region classification rule based on edge distance field and orientation awareness mechanism to obtain Class A and Class B / C depth value missing regions. Initial completion is performed on Class A depth value missing regions, and depth-gradient consistency optimization is applied to the initially completed depth map using an energy function. For Class B / C depth value missing regions, bilateral filtering is used for completion. Based on local enhancement of edge regions, the locally reconstructed gradient field is returned to the depth map through Poisson reconstruction to obtain a geometrically consistent depth map. Finally, the module uses object detection algorithms and HSV threshold segmentation algorithms to obtain the pixel coordinates of reactor internal components, and Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and translation matrix of the binocular structured light camera. Combined with the optimized depth map, a 3D reconstruction based on binocular vision principles is completed. The image information visualization module is used to display the 3D reconstruction results and positioning results in real time.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the depth-gradient consistency depth information completion method based on sub-pixel edge guidance as described in any one of claims 1-6.

9. A processing terminal, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the depth-gradient consistency depth information completion method based on sub-pixel edge guidance as described in any one of claims 1-6.

Images