Narrow space picture imaging method and device based on binocular fisheye, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of operation and maintenance inspection, and provides a binocular fisheye-based narrow space picture imaging method and device, electronic equipment and a storage medium, and the method comprises the steps: image collection, image correction, image expansion, key point matching, distance mean value calculation and image synthesis. According to the method, the image is expanded by using the ControlNet improved model, so that the picture information restoration effect and the visual angle unified switching effect are improved; through coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation and coordinate remapping, the visual field of the image is expanded, and the display effect of the synthesized image is optimized.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance inspection technology, and in particular to a method, device, electronic device and storage medium for imaging images in a narrow space based on binocular fisheye. Background Technology

[0002] Current mainstream circuit breaker status recognition solutions are primarily applicable to open environments, relying on conventional cameras to directly capture target images in open spaces. While such solutions can achieve effective recognition under ideal lighting and unobstructed conditions, they have significant limitations: they cannot adapt to the enclosed application environment of control cabinets. Because the control cabinet doors must be kept closed during routine maintenance, ordinary cameras cannot capture images of the circuit breakers inside the cabinet due to physical obstruction, rendering this type of technology completely ineffective. This fundamental flaw makes existing solutions unable to meet the actual needs of intelligent inspection of substation control cabinets. Furthermore, existing technologies do not employ ultra-wide-angle lenses to expand the field of view and lack hardware-level optimization, resulting in potentially incomplete images and affecting recognition performance. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a method, device, electronic device and storage medium for imaging images in a narrow space based on binocular fisheye, so as to solve the problems that conventional methods cannot be applied to closed control cabinets and the images are incomplete.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for imaging images in confined spaces based on binocular fisheye imaging includes:

[0006] The target narrow space is captured using a pre-set dual fisheye camera to obtain the original captured image;

[0007] The original acquired image is sequentially subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping to obtain the corrected image;

[0008] The corrected image is then subjected to conditional encoding, input construction, and viewpoint transformation using an improved ControlNet model to obtain an unfolded image.

[0009] Hamming distance is used to perform key point matching on the two unfolded images to obtain the nearest matching points of the target number. The average distances of the nearest matching points in the X-axis and Y-axis directions of the two original acquired images are calculated to obtain the average horizontal distance and the average vertical distance.

[0010] The two unfolded images are fused based on the mean horizontal distance and the mean vertical distance to obtain a composite image.

[0011] Preferably, a preset dual fisheye camera is used to acquire images of the target confined space to obtain the original acquired image, including:

[0012] Set an ideal incident angle; the range of the ideal incident angle is [0, 220°];

[0013] Define a projection function; the expression for the projection function is: ;in, The distortion radius; The incident angle; The projection function; For fisheye focal length parameters;

[0014] According to the projection function, the target narrow space is image acquired by a preset left fisheye camera and a right fisheye camera respectively, and the original acquired image is obtained; the original acquired image includes: the left fisheye acquired image and the right fisheye acquired image.

[0015] Preferably, the original acquired image is subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping in sequence to obtain a corrected image, including:

[0016] Construct a distortion model; the expression of the distortion model is: ;in, , , , These are the first radial distortion coefficient, the second radial distortion coefficient, the third radial distortion coefficient, and the fourth radial distortion coefficient, respectively.

[0017] The original acquired image is normalized using a pre-constructed normalization formula to obtain a normalized image; the normalization formula includes: , ;in, , These are the x-coordinate and y-coordinate of the normalized image, respectively; , These are the original coordinates in the X and Y directions, respectively; This represents the offset in the X direction; The focal length is in the X direction; This represents the offset along the Y-axis. The focal length is along the Y-axis.

[0018] The distortion radius is obtained by calculating the radius of the normalized image based on the distortion model.

[0019] Preferably, the original acquired image is subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping in sequence to obtain a corrected image, including:

[0020] The normalized image is calculated using a preset distortion coordinate formula based on the distortion radius to obtain horizontal and vertical distortion coordinate data; the distortion coordinate formula includes: , ;in, The lateral distortion coordinate data; The longitudinal distortion coordinate data;

[0021] The horizontal and vertical distortion coordinate data are calculated using a preset pixel coordinate formula to obtain the horizontal and vertical pixel coordinate data; the pixel coordinate formula includes: , ;in, The horizontal pixel coordinate data; The vertical pixel coordinate data;

[0022] The original acquired image is adjusted according to the horizontal and vertical pixel coordinate data to obtain the corrected image.

[0023] Preferably, the corrected image is sequentially subjected to conditional encoding, input construction, and viewpoint transformation using an improved ControlNet model to obtain an unfolded image, including:

[0024] The improved ControlNet model is obtained by replacing the bilinear interpolation layer in the original ControlNet model with a view consistency transformation layer; the expression of the view consistency transformation layer includes: , ;in, Input feature map; The distortion coefficient; It is a rotation matrix; The translation coefficient; Represents a multilayer perceptron; The output of the multilayer perceptron; This represents a vector concatenation operation; The coordinates of a two-dimensional point in a plane; This is the output of the view consistency transformation layer;

[0025] Construct ControlNet conditional input;

[0026] The expanded image is obtained by processing the ControlNet conditional input using the improved ControlNet model.

[0027] Preferably, the two unfolded images are fused according to the mean horizontal distance and the mean vertical distance to obtain a composite image, including:

[0028] The unfolded image corresponding to the left fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the left fisheye image. The unfolded image corresponding to the right fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the right fisheye image. The translated left fisheye image and right fisheye image are then merged to obtain the composite image.

[0029] Preferably, a narrow space image imaging device based on binocular fisheye includes: a card camera A side, a card camera B side, a first fisheye, a second fisheye, an expansion interface, a PoE separation module, a third fisheye, a switch, and a client.

[0030] The expansion interface and the PoE separation module are disposed on surface A of the card reader; the third fisheye lens is disposed on surface B of the card reader; the first fisheye lens and the second fisheye lens are located to the left and right of the center line of surface A of the card reader, respectively; the distance between the first fisheye lens and the second fisheye lens and the center line of surface A of the card reader is 5cm to 10cm; the lens of the first fisheye lens is offset to the left by 30°; the lens of the second fisheye lens is offset to the right by 45°; the switch is connected to the first fisheye lens, the second fisheye lens, and the third fisheye lens; the switch is connected to the client.

[0031] Preferably, both the first fisheye and the second fisheye are 220° ultra-wide-angle fisheye lenses with a focal length of 1.2mm.

[0032] Preferably, an electronic device includes: at least one processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned binocular fisheye-based narrow-space image imaging method.

[0033] Preferably, a non-transitory computer-readable storage medium stores computer instructions for causing a computer to execute the aforementioned binocular fisheye-based narrow-space image imaging method.

[0034] The present invention discloses the following technical effects:

[0035] This invention provides a method, device, electronic device, and storage medium for imaging images in confined spaces based on binocular fisheye. By utilizing an improved ControlNet model for image unfolding, it solves the problem of poor edge distortion unfolding effect in conventional methods, achieving high-fidelity image information restoration and unified viewpoint switching. Through coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping, it solves the problems that conventional methods cannot be applied to closed control cabinets and incomplete captured images, thereby expanding the image field of view. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A schematic diagram of the narrow space image imaging process based on binocular fisheye provided in an embodiment of the present invention;

[0038] Figure 2 This is a structural diagram of the A-side of a card-type camera provided in an embodiment of the present invention;

[0039] Figure 3 This is a structural diagram of the B-side of a card-type camera provided in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of a system provided in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram illustrating the synthesis effect provided in an embodiment of the present invention. Detailed Implementation

[0042] 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.

[0043] The purpose of this invention is to provide a method, device, electronic device, and storage medium for imaging images in confined spaces based on binocular fisheye imaging, thereby solving the problems of conventional methods being unsuitable for closed control cabinets and incomplete captured images.

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Figure 1 This is a schematic diagram of the imaging process for images in confined spaces based on binocular fisheye imaging, as provided in an embodiment of the present invention. Figure 1 As shown, this invention provides a method for imaging images in confined spaces based on binocular fisheye, comprising:

[0046] Step 100: Use a preset dual fisheye camera to acquire images of the target space to obtain the original acquired images;

[0047] Step 200: Perform coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping on the original acquired image in sequence to obtain the corrected image;

[0048] Step 300: Using the ControlNet improved model, conditional encoding, input construction, and viewpoint transformation are performed on the corrected image sequentially to obtain the unfolded image;

[0049] Step 400: Use Hamming distance to perform key point matching on the two unfolded images to obtain the target number of nearest matching points. Calculate the average distance between the nearest matching points of the two original acquired images in the X-axis and Y-axis directions to obtain the average horizontal distance and the average vertical distance.

[0050] Step 500: Fuse the two unfolded images according to the mean horizontal distance and the mean vertical distance to obtain a composite image.

[0051] Furthermore, using a pre-set dual fisheye camera, images are acquired from the narrow space of the target area to obtain the original acquired images, including:

[0052] Set an ideal incident angle; the range of the ideal incident angle is [0, 220°];

[0053] Define a projection function; the expression for the projection function is: ;in, The distortion radius; The incident angle; The projection function; For fisheye focal length parameters;

[0054] According to the projection function, the target narrow space is image acquired by a preset left fisheye camera and a right fisheye camera respectively, and the original acquired image is obtained; the original acquired image includes: the left fisheye acquired image and the right fisheye acquired image.

[0055] Specifically, the original acquired image is sequentially subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping to obtain a corrected image, including:

[0056] Construct a distortion model; the expression of the distortion model is: ;in, , , , These are the first radial distortion coefficient, the second radial distortion coefficient, the third radial distortion coefficient, and the fourth radial distortion coefficient, respectively.

[0057] The original acquired image is normalized using a pre-constructed normalization formula to obtain a normalized image; the normalization formula includes: , ;in, , These are the x-coordinate and y-coordinate of the normalized image, respectively; , These are the original coordinates in the X and Y directions, respectively; This represents the offset in the X direction; The focal length is in the X direction; This represents the offset along the Y-axis. The focal length is along the Y-axis.

[0058] The distortion radius is obtained by calculating the radius of the normalized image based on the distortion model.

[0059] Further, the original acquired image is sequentially subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping to obtain a corrected image, including:

[0060] The normalized image is calculated using a preset distortion coordinate formula based on the distortion radius to obtain horizontal and vertical distortion coordinate data; the distortion coordinate formula includes: , ;in, The lateral distortion coordinate data; The longitudinal distortion coordinate data;

[0061] The horizontal and vertical distortion coordinate data are calculated using a preset pixel coordinate formula to obtain the horizontal and vertical pixel coordinate data; the pixel coordinate formula includes: , ;in, The horizontal pixel coordinate data; The vertical pixel coordinate data;

[0062] The original acquired image is adjusted according to the horizontal and vertical pixel coordinate data to obtain the corrected image.

[0063] Specifically, the corrected image is conditionally encoded, input constructed, and viewpoint transformed sequentially using the ControlNet improved model to obtain the unfolded image, including:

[0064] The improved ControlNet model is obtained by replacing the bilinear interpolation layer in the original ControlNet model with a view consistency transformation layer; the expression of the view consistency transformation layer includes: , ;in, Input feature map; The distortion coefficient; It is a rotation matrix; The translation coefficient; Represents a multilayer perceptron; The output of the multilayer perceptron; This represents a vector concatenation operation; The coordinates of a two-dimensional point in a plane; This is the output of the view consistency transformation layer;

[0065] Construct ControlNet conditional input;

[0066] The expanded image is obtained by processing the ControlNet conditional input using the improved ControlNet model.

[0067] Further, the two unfolded images are fused based on the mean horizontal distance and the mean vertical distance to obtain a composite image, including:

[0068] The unfolded image corresponding to the left fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the left fisheye image. The unfolded image corresponding to the right fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the right fisheye image. The translated left fisheye image and right fisheye image are then merged to obtain the composite image.

[0069] refer to Figures 2 to 4 A narrow-space imaging device based on binocular fisheye lenses includes: a side A of a compact camera, a side B of a compact camera, a first fisheye lens, a second fisheye lens, an expansion interface, a PoE separation module, a third fisheye lens, a switch, and a client; the first fisheye lens and the second fisheye lens are both 220° ultra-wide-angle fisheye lenses with a focal length of 1.2mm.

[0070] The first fisheye lens, the second fisheye lens, the expansion interface, and the PoE separation module are disposed on surface A of the card camera; the third fisheye lens is disposed on surface B of the card camera; surface A and surface B of the card camera form a closed structure; the first fisheye lens and the second fisheye lens are located to the left and right of the center line of surface A of the card camera, respectively; the distance between the first fisheye lens and the second fisheye lens and the center line of surface A of the card camera ranges from 5cm to 10cm; the lens of the first fisheye lens is offset to the left by 30°; the lens of the second fisheye lens is offset to the right by 45°; the switch is connected to the first fisheye lens, the second fisheye lens, and the third fisheye lens; the switch is connected to the client.

[0071] The client is used to synthesize the images acquired from the first fisheye and the second fisheye.

[0072] Specifically, reference to the A-side of the card camera. Figure 2 The first fisheye lens is used to capture the image from the right side of the camera door; the second fisheye lens is used to capture the image from the left side of the camera door. Expansion interfaces include a network port, a power interface, and two programmable LED lights, allowing for brightness adjustment in low-light conditions. The PoE separation module separates the PoE cable into a network cable and an input power interface. All components are custom-assembled using a board similar to a Raspberry Pi. (See B-side of the compact camera for reference.) Figure 3 The third fisheye lens is used to capture images outside the equipment door. Two symmetrically arranged 220° ultra-wide-angle fisheye lenses (focal length f=1.2mm) are embedded in the upper inner side of the equipment door, allowing observation of equipment inside and outside the door. The installation position is offset from the center line of the equipment door by 5cm to 10cm, and the two lenses are asymmetrically distributed facing the cabinet (fisheye 1 is 30° to the left, fisheye 2 is 45° to the right). LED light strips are integrated on both sides, and the brightness of the light can be adjusted in low-light environments (color temperature 5000K, illuminance ≥800lux); Communication interface: PoE network port supporting IEEE802.3at standard.

[0073] Further, a summary of the process:

[0074] 1) Initiating image capture command via client: The client sends a command via TCP protocol specifying which fisheye image to capture and waits for the result.

[0075] 2) Imaging with dual fisheye cameras inside the door: The fisheye lens uses a negative meniscus perspective group, achieving an ultra-wide-angle incident angle range of [0, 220°] through extreme light bending; distortion radius ;in This is the projection function. The projection function can be adjusted according to the actual situation. By default, the projection function is... Distance is proportional to angle, maximizing the imaging field of view. Edges will be stretched, but since the devices are all in the center, this stretching is just enough to compensate for the disadvantage of edge stretching. Binocular fisheye simultaneous acquisition, distortion correction, and unfolding are used.

[0076] 3) Distortion correction and unfolding of multiple images: Distortion model: ,in For the ideal angle of incidence, arrive The radial distortion coefficient is... The radius of the distorted image.

[0077] 31) Distortion correction:

[0078] Normalized coordinates: camera intrinsic parameter matrix Points in a coordinate system with the camera center as the origin :

[0079]

[0080]

[0081] in For the camera intrinsic parameter matrix, Indicates the offset in the x-direction. This represents the focal length in the x-direction. Indicates the offset in the y-direction. This represents the focal length in the y-direction;

[0082] Find the normalized radius:

[0083]

[0084] and Normalized coordinates;

[0085] The normalized radius here Approximately the radius of the distorted image, according to 2) Substitute the distortion model from step 3) to obtain the incident angle of the distortion. ;

[0086] Find the distorted coordinates:

[0087]

[0088]

[0089] Find pixel coordinates:

[0090]

[0091]

[0092] point The corrected coordinates are By remapping coordinates, the corrected image is obtained. .

[0093] 32) Image expansion:

[0094] Conditional encoding and input construction: mapping from raw image coordinates to spherical coordinates:

[0095]

[0096] in To construct the azimuth angle, construct the ControlNet conditional input:

[0097]

[0098]

[0099]

[0100]

[0101] Add a transformation layer to maintain a consistent perspective:

[0102]

[0103]

[0104] in It is the input feature map (intermediate layer). The distortion coefficient is... (rotation matrix) and The translation coefficients are learnable spatial transformation matrices that replace traditional bilinear interpolation to achieve pixel-level viewpoint alignment.

[0105] Generate unfolded images: Input the preprocessed model input parameters into the ControlNet model (end-to-end learning, distortion adaptation, generating images from a unified perspective) to generate unfolded images. .

[0106] 4) Composite image:

[0107] Keypoint matching: Using binary descriptors (such as ORB, BRIEF, BRISK, etc.) and Hamming distance for matching, the fastest way to obtain matching keypoints. , This represents the first point in the first image, for example. => This represents the matching point between the first and second images, that is, the first image's... The coordinates of the second image are ; and Given a pair of matching points, n defaults to 5. The five closest matching points are selected, and the average distance between the two fisheye matching points along the X and Y axes is calculated. , , , :

[0108]

[0109]

[0110] in The width of the image is known from its physical structure. The error won't be significant; the two fisheyes are basically on the same horizontal line. Finally, here's the image of the unfolded first fisheye. Translation of X and Y coordinates , Distance and the second fisheye image unfolding Translation of X and Y coordinates , The distances are calculated and merged to obtain the composite image. Final rendering reference. Figure 5 .

[0111] As an optional implementation, this embodiment also provides a small-space image imaging system based on binocular fisheye, including:

[0112] The image acquisition module is used to acquire images of a target confined space using a preset dual fisheye camera to obtain the original acquired images; the original acquired images include: a left fisheye acquired image and a right fisheye acquired image;

[0113] The image correction module is used to perform coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation and coordinate remapping on the original acquired image in sequence to obtain the corrected image;

[0114] The image unfolding module is used to perform conditional encoding, input construction, and viewpoint transformation on the corrected image sequentially using the ControlNet improved model to obtain the unfolded image.

[0115] The key point matching module is used to perform key point matching on the two unfolded images using Hamming distance to obtain the nearest matching points of the target number;

[0116] The distance mean calculation module is used to calculate the average distance between the nearest matching point in the left fisheye image and the right fisheye image in the X-axis and Y-axis directions, respectively, to obtain the horizontal distance mean and the vertical distance mean;

[0117] The image synthesis module is used to translate the unfolded image corresponding to the left fisheye image based on the average horizontal distance and the average vertical distance corresponding to the left fisheye image, translate the unfolded image corresponding to the right fisheye image based on the average horizontal distance and the average vertical distance corresponding to the right fisheye image, and merge the translated left fisheye image and right fisheye image to obtain a synthesized image.

[0118] As an optional implementation, this embodiment also provides an electronic device, including: at least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned binocular fisheye-based narrow space image imaging method.

[0119] As an optional implementation, this embodiment also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the aforementioned binocular fisheye-based narrow-space image imaging method.

[0120] The beneficial effects of this invention are as follows:

[0121] This invention improves the image information restoration and viewpoint switching effects by utilizing the ControlNet improved model for image unfolding; it expands the image's field of view and optimizes the display effect of the synthesized image through coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping.

[0122] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for imaging images in confined spaces based on binocular fisheye, characterized in that, include: The target narrow space is captured using a pre-set dual fisheye camera to obtain the original captured image; The original acquired image is sequentially subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping to obtain the corrected image; The corrected image is then subjected to conditional encoding, input construction, and viewpoint transformation using an improved ControlNet model to obtain an unfolded image. Hamming distance is used to perform key point matching on the two unfolded images to obtain the nearest matching points of the target number. The average distances of the nearest matching points in the X-axis and Y-axis directions of the two original acquired images are calculated to obtain the average horizontal distance and the average vertical distance. The two unfolded images are fused based on the mean horizontal distance and the mean vertical distance to obtain a composite image.

2. The method for imaging images in a confined space based on binocular fisheye as described in claim 1, characterized in that, Using a pre-set dual fisheye camera, images are acquired from the narrow space of the target area to obtain the original acquired images, including: Set an ideal incident angle; the range of the ideal incident angle is [0, 220°]; Define a projection function; the expression for the projection function is: ;in, The distortion radius; The incident angle; The projection function; For fisheye focal length parameters; Based on the projection function, images are acquired from the target narrow space using a preset left fisheye camera and a right fisheye camera to obtain the original acquired images; the original acquired images include: the left fisheye acquired image and the right fisheye acquired image.

3. The method for imaging images in a confined space based on binocular fisheye as described in claim 2, characterized in that, The original acquired image is subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping in sequence to obtain a corrected image, including: Construct a distortion model; the expression of the distortion model is: ;in, , , , These are the first radial distortion coefficient, the second radial distortion coefficient, the third radial distortion coefficient, and the fourth radial distortion coefficient, respectively. The original acquired image is normalized using a pre-constructed normalization formula to obtain a normalized image; the normalization formula includes: , ;in, , These are the x-coordinate and y-coordinate of the normalized image, respectively; , These are the original coordinates in the X and Y directions, respectively; This represents the offset in the X direction; The focal length is in the X direction; This represents the offset along the Y-axis. The focal length is along the Y-axis. The distortion radius is obtained by calculating the radius of the normalized image based on the distortion model.

4. The method for imaging images in a confined space based on binocular fisheye as described in claim 3, characterized in that, The original acquired image is subjected to coordinate normalization, radius and angle calculation, distortion coordinate calculation, pixel coordinate calculation, and coordinate remapping in sequence to obtain a corrected image, including: The normalized image is calculated using a preset distortion coordinate formula based on the distortion radius to obtain horizontal and vertical distortion coordinate data; the distortion coordinate formula includes: , ;in, The lateral distortion coordinate data; The longitudinal distortion coordinate data; The horizontal and vertical distortion coordinate data are calculated using a preset pixel coordinate formula to obtain the horizontal and vertical pixel coordinate data; the pixel coordinate formula includes: , ;in, The horizontal pixel coordinate data; The vertical pixel coordinate data; The original acquired image is adjusted according to the horizontal and vertical pixel coordinate data to obtain the corrected image.

5. The method for imaging images in a confined space based on binocular fisheye as described in claim 4, characterized in that, The corrected image is then subjected to conditional encoding, input construction, and viewpoint transformation using an improved ControlNet model to obtain an unfolded image, including: The improved ControlNet model is obtained by replacing the bilinear interpolation layer in the original ControlNet model with a view consistency transformation layer; the expression of the view consistency transformation layer includes: , ;in, Input feature map; The distortion coefficient; It is a rotation matrix; The translation coefficient; This represents a multilayer perceptron; The output of the multilayer perceptron; This represents a vector concatenation operation; The coordinates of a two-dimensional point in a plane; This is the output of the view consistency transformation layer; Construct ControlNet conditional input; The expanded image is obtained by processing the ControlNet conditional input using the improved ControlNet model.

6. The method for imaging images in a confined space based on binocular fisheye as described in claim 5, characterized in that, The two unfolded images are fused based on the mean horizontal distance and the mean vertical distance to obtain a composite image, including: The unfolded image corresponding to the left fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the left fisheye image. The unfolded image corresponding to the right fisheye image is translated based on the average horizontal distance and the average vertical distance corresponding to the right fisheye image. The translated left fisheye image and right fisheye image are then merged to obtain the composite image.

7. A device for imaging images in a confined space based on binocular fisheye imaging, characterized in that, The device used to implement the narrow space image imaging method based on binocular fisheye as described in claim 1 includes: a side A of a compact camera, a side B of a compact camera, a first fisheye, a second fisheye, an expansion interface, a PoE separation module, a third fisheye, a switch, and a client. The expansion interface and the PoE separation module are disposed on surface A of the card reader; the third fisheye lens is disposed on surface B of the card reader; the first fisheye lens and the second fisheye lens are located to the left and right of the center line of surface A of the card reader, respectively; the distance between the first fisheye lens and the second fisheye lens and the center line of surface A of the card reader is 5cm to 10cm; the lens of the first fisheye lens is offset to the left by 30°; the lens of the second fisheye lens is offset to the right by 45°; the switch is connected to the first fisheye lens, the second fisheye lens, and the third fisheye lens; the switch is connected to the client.

8. The narrow-space image imaging device based on binocular fisheye as described in claim 7, characterized in that, Both the first fisheye and the second fisheye are 220° ultra-wide-angle fisheye lenses with a focal length of 1.2mm.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform a binocular fisheye-based narrow-space image imaging method according to any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute any one of claims 1 to 6, a method for imaging images in a confined space based on binocular fisheye.

Images