Unmanned aerial vehicle battery interface identification method and system for unmanned aerial vehicle battery replacement platform

Abstract

The invention relates to the technical field of image processing, in particular to an unmanned aerial vehicle battery interface recognition method and system for an unmanned aerial vehicle battery replacement platform, and the method comprises the steps: collecting images at an unmanned aerial vehicle battery interface, and screening out a blurred image from the images at the unmanned aerial vehicle battery interface; obtaining a strong expression area of the blurred image according to edge pixel points in the blurred image; according to the frequency domain information of the strong expression area of the blurred image, the blurring direction and the blurring length of the strong expression area of the blurred image are obtained, then the dense blurring field of the blurred image is obtained, the blurring kernel of each pixel point in the blurred image is obtained, and then the blurred image is enhanced to accurately identify the unmanned aerial vehicle battery interface. According to the method, the frequency domain information of the image at the blurred unmanned aerial vehicle battery interface is analyzed, the optimal blurred kernel of each pixel point in the blurred image is adaptively obtained, and the blurred image is enhanced so as to accurately identify the unmanned aerial vehicle battery interface.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method and system for identifying the battery interface of a drone for a drone battery swapping platform. Background Technology

[0002] An automated battery swapping platform for drones is a core infrastructure for enabling uninterrupted drone operation. Its task is to automatically grab, unload, and install new batteries on landing drones using a robotic arm, without human intervention. Drone battery interfaces typically consist of multi-pin electrical connectors and sophisticated mechanical locking mechanisms. For the battery swapping platform to accurately and quickly replace batteries, it needs to accurately identify the drone's battery interface. However, when a drone is parked on the platform, residual vibrations within the drone's internal machinery can cause motion blurring at the battery interface location, making accurate identification difficult.

[0003] To accurately and quickly identify the drone battery interface, the image of the blurred drone battery interface needs to be enhanced. However, the traditional method of enhancing blurred images is to use a globally uniform blur kernel to enhance the blurred image. But different regions of the captured image of the drone battery interface may correspond to completely different blur characteristics, resulting in poor enhancement effect of the traditional method of enhancing the blurred image of the drone battery interface with a globally uniform blur kernel. Summary of the Invention

[0004] This invention provides a method and system for identifying drone battery interfaces for drone battery swapping platforms, in order to solve the existing problem: the traditional method of enhancing images at the blurred drone battery interface using a globally uniform fuzzy kernel has poor enhancement effect.

[0005] The UAV battery interface identification method and system for UAV battery swapping platforms of the present invention adopts the following technical solution: One embodiment of the present invention provides a method for identifying the battery interface of a drone for a drone battery swapping platform, the method comprising the following steps: Acquire images of the drone's battery interface and filter out blurry images from those images. Based on the edge pixels and gradient characteristics in the blurred image, obtain several closed edges in the blurred image and the reference factor of each closed edge; based on the reference factor and the contour of the closed edges in the blurred image, obtain the reference degree of the closed edges in the blurred image; based on the reference degree of the closed edges in the blurred image, filter out the strong performance regions from all closed edges in the blurred image. Based on the frequency domain information of the strong representation region of the blurred image, the blur direction and blur length of the strong representation region of the blurred image are obtained; based on the blur direction of the strong representation region of the blurred image, combined with the gradient value and gradient direction of all pixels in the blurred image, the blur confidence weight of each pixel in the blurred image is obtained; combined with the blur direction and blur length of the strong representation region of the blurred image, the dense blur field of the blurred image is obtained, and thus the blur kernel of each pixel in the blurred image is obtained. The blurred image is enhanced based on the blur kernel of each pixel in the blurred image, thereby accurately identifying the drone battery interface.

[0006] Preferably, the specific method for acquiring images from the drone battery interface and filtering out blurry images from those images includes: The camera mounted on the drone battery swapping platform captures images of the drone's battery interface at each moment. For an image of the drone battery interface at any given time, the Laplacian operator is used to perform a convolution operation on the image of the drone battery interface at that time. The result of the convolution operation is recorded as the Laplacian image of the drone battery interface at that time. The SSV method is used to obtain the energy of the Laplacian image of the drone battery interface at that time. Preset an energy threshold If the energy of the Laplacian image at the UAV battery interface at the stated time is less than The image of the drone's battery interface at the specified time is used as a blurred image.

[0007] Preferably, the specific method for obtaining several closed edges in the blurred image and a reference factor for each closed edge based on the edge pixels and gradient characteristics in the blurred image is as follows: For any blurred image, the Canny operator is used to obtain all edge pixels in the blurred image, and a contour search algorithm is used to obtain several closed edges in the blurred image. For any closed edge in the blurred image, the gradient value of each pixel on the closed edge is obtained using the Sobel operator. The difference between the maximum and minimum gradient values ​​on the closed edge is taken as the gradient interval of the closed edge in the blurred image. The product of the ratio of the average gradient value of all pixels on the closed edge to the gradient interval and the number of pixels on the closed edge is taken as the reference factor of the closed edge in the blurred image.

[0008] Preferably, the specific method for obtaining the reference degree of the closed edges in the blurred image based on the reference factor and the contour of the closed edges in the blurred image includes: For any closed edge in any blurred image, the Euclidean distance between the center of the closed edge and the center of the blurred image is taken as the eccentricity distance of the closed edge in the blurred image. The convex hull detection result of the closed edge in the blurred image is obtained by using a convex hull detection algorithm. The absolute value of the difference between the area of ​​the closed edge in the blurred image and the area of ​​the convex hull detection result is multiplied by the eccentricity distance of the closed edge in the blurred image, and the product is taken as the degree of unevenness of the closed edge in the blurred image. The ratio of the reference factor of the closed edge in the blurred image to its degree of unevenness is used as the reference degree of the closed edge in the blurred image.

[0009] Preferably, the method for selecting strong representation regions from all closed edges in the blurred image based on the reference degree of the closed edges in the blurred image includes: For any blurred image, all closed edges in the blurred image are sorted in ascending order according to their reference degree to obtain a sequence of closed edges in the blurred image; for any two adjacent closed edges in the sequence of closed edges in the blurred image, the absolute value of the difference in reference degree between the two adjacent closed edges in the sequence of closed edges in the blurred image is used as the reference difference between the two adjacent closed edges in the sequence of closed edges in the blurred image. The reference difference of all adjacent closed edges in the closed edge sequence of the blurred image is obtained. The adjacent closed edge corresponding to the largest reference difference is taken as the critical adjacent closed edge. The closed edge in the blurred image whose reference degree is greater than or equal to the maximum value of the reference degree corresponding to the critical adjacent closed edge is taken as the reference edge. Each reference edge in the blurred image and the region inside the reference edge are taken as the strong representation region of the blurred image.

[0010] Preferably, the specific method for obtaining the blur direction and blur length of the strong representation region of the blurred image based on the frequency domain information of the strong representation region of the blurred image is as follows: For any strong representation region of any blurred image, perform a two-dimensional Fourier transform on the strong representation region of the blurred image to obtain the spectrum of the strong representation region of the blurred image; perform a Radon transform on the spectrum of the strong representation region of the blurred image to obtain a projection sequence of several angles, and record the angle corresponding to the projection sequence with the largest projection integral as the projection direction, and take the direction perpendicular to the projection direction as the blur angle of the strong representation region in the blurred image. The average interval between all adjacent peaks in the projection sequence corresponding to the projection direction is taken as the spatial period of the fluctuation in the projection sequence corresponding to the projection direction, and the reciprocal of the spatial period of the fluctuation in the projection sequence corresponding to the projection direction is taken as the blur length of the strong representation region in the blurred image.

[0011] Preferably, the specific method for obtaining the blur confidence weight of each pixel in the blurred image based on the blur direction of the strong representation region of the blurred image, combined with the gradient values ​​and gradient directions of all pixels in the blurred image, includes: For any blurred image, the gradient value and gradient direction of each pixel in the blurred image are obtained using the Sobel operator. The sum of the gradient values ​​of all pixels in the eight neighborhoods of each pixel in the blurred image is recorded as the local texture feature of the corresponding pixel in the blurred image. Let any pixel in the blurred image be designated as the target point; let the region with the strongest representation closest to the target point in the blurred image be designated as the reference region of the target point; the formula for calculating the fuzzy confidence weight of the target point is as follows: In the formula, The fuzzy confidence weights represent the target points; Indicates the gradient direction at the target point; The fuzzy direction representing the reference area of ​​the target point; Represents the local texture features of the target point; This represents the local texture features of all pixels in the blurred image; This represents the function that takes the absolute value. This represents the cosine function.

[0012] Preferably, the specific method for obtaining the dense blur field of the blurred image is as follows: For any blurred image, the blur angle and blur length of each strong representation region in the blurred image are used as the label of each strong representation region in the blurred image, and the blur confidence weight of each pixel in the blurred image is used as a smoothing term to construct a Markov random field model of the blurred image. The Markov random field model of the blurred image is optimized and solved using the belief propagation algorithm to obtain the dense blurred field of the blurred image.

[0013] Preferably, the method for enhancing the blurred image based on the blur kernel of each pixel in the blurred image includes: For any blurred image, the Wiener filtering algorithm is used to enhance the blurred image based on the blur kernel of each pixel in the blurred image.

[0014] Another embodiment of the present invention provides a drone battery interface identification system for a drone battery swapping platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any of the above-described drone battery interface identification methods for a drone battery swapping platform.

[0015] The beneficial effects of the technical solution of the present invention are as follows: This application acquires images of the UAV battery interface and filters out blurred images from the images of the UAV battery interface; analyzes the morphological features of the edges after being affected by blurring, and obtains the reference degree of the closed edges in the blurred image. The greater the reference degree of the closed edges, the more reliable the blurred features extracted from the corresponding regions of the closed edges. Therefore, based on this, strong performance regions can be selected from all closed edges in the blurred image. Then, using the modal features of the regions that can extract blurred features, the blurred features of all pixels in the blurred image can be obtained, providing a reliable theoretical basis for the subsequent adaptive generation of the best blur kernel for each pixel in the blurred image. It should be further explained that, since reliable blur features can be extracted from strong regions, the blur confidence weight of each pixel in the blurred image can be obtained based on the blur features of the strong regions. This weight is then used to construct a dense blur field for the blurred image, resulting in a blur kernel for each pixel. By using known precise blur parameters as fixed constraint points and establishing correlation weights between adjacent pixels based on pixel matching, an optimization model covering the entire image is constructed. This model is calculated using mathematical optimization algorithms, allowing blur parameters to adaptively propagate from the measured strong regions along the image edge structure and texture features, ultimately generating a dense blur parameter field that maintains spatial coherence. This field then adaptively generates an optimal blur kernel for each pixel in the blurred image to enhance it. Subsequently, the blurred image can be enhanced based on the blur kernel of each pixel. This application analyzes the frequency domain information of the blurred image at the drone battery interface to adaptively obtain the optimal blur kernel for each pixel in the blurred image, enhancing the blurred image to accurately identify the drone battery interface. Attached Figure Description

[0016] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the steps of the drone battery interface identification method for drone battery swapping platforms according to the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the drone battery interface identification method and system for drone battery swapping platforms proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific solution of the drone battery interface identification method and system for drone battery swapping platforms provided by this invention.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a method for identifying the battery interface of a drone for a drone battery swapping platform, according to an embodiment of the present invention. The method includes the following steps: Step S001: Acquire an image of the drone's battery interface and filter out blurry images from the images of the drone's battery interface.

[0022] It should be noted that this embodiment, as a method for identifying the battery interface of a drone on a drone battery swapping platform, specifically uses a camera mounted on the drone battery swapping platform to detect and analyze the battery interface of the drone parked on the platform, accurately obtaining the spatial position of the drone battery interface. This allows the drone battery swapping platform to quickly and accurately connect the battery pins to the drone charging interface via a robotic arm. To achieve this, the camera on the drone battery swapping platform first needs to capture images of the drone battery interface. However, when the drone is parked on the platform, residual vibrations exist within the drone's internal machinery. These vibrations, superimposed on the robotic arm's movement, can cause motion blur in the images of the drone battery interface, making accurate identification difficult. To accurately and quickly identify the drone battery interface, it is necessary to capture images of the drone battery interface and filter out the blurred images. Subsequently, the blurred images are enhanced to accurately identify the drone battery interface.

[0023] Specifically, the camera mounted on the drone battery swapping platform captures images of the drone battery interface at each moment, which are described in this embodiment as 0.05 seconds per moment. For an image of the drone battery interface at any given time, the Laplacian operator is used to perform a convolution operation on the image of the drone battery interface at that time. The result of the convolution operation is recorded as the Laplacian image of the drone battery interface at that time. The SSV (Sum of Squared Values) method is used to obtain the energy of the Laplacian image of the drone battery interface at that time. Furthermore, a preset energy threshold is established. The The specific value can be set according to the actual situation. This embodiment does not make a hard requirement. In this embodiment, it is used as... For example, if the energy of the Laplacian image at the UAV battery interface at the stated time is greater than or equal to... If the image at the drone battery interface at that moment is clear, then the image at that moment is clear; if the energy of the Laplacian image at the drone battery interface at that moment is less than... The image of the drone's battery interface at the specified time is used as a blurred image.

[0024] It should be noted that since the Laplacian operator and the SSV method are both well-known existing technologies, they will not be described in detail in this embodiment. The Laplacian image at the drone battery interface represents the second-order derivative image of the image at the drone battery interface, highlighting the high-frequency information (edge ​​details) in the image at the drone battery interface. The more blurred the image, the weaker its high-frequency information. Therefore, the smaller the energy of the Laplacian image at the drone battery interface, the more blurred the image at the drone battery interface. Therefore, this can be used as a basis to obtain a blurred image, which can then be added to the blurred image to accurately identify the drone battery interface.

[0025] At this point, several blurry images were obtained.

[0026] Step S002: Based on the edge pixels and gradient characteristics in the blurred image, obtain several closed edges in the blurred image and the reference factor of each closed edge; based on the reference factor and contour of the closed edges in the blurred image, obtain the reference degree of the closed edges in the blurred image; based on the reference degree of the closed edges in the blurred image, select strong performance regions from all closed edges in the blurred image.

[0027] It should be noted that this embodiment, as a method for identifying drone battery interfaces for drone battery swapping platforms, specifically enhances the image of the blurred drone battery interface to accurately identify it. Traditional methods for enhancing blurred images use a globally uniform blur kernel, but different regions of the acquired image at the drone battery interface may correspond to completely different blur characteristics. A globally uniform blur kernel cannot accurately model this non-uniform blur, resulting in poor enhancement effects from traditional methods. Therefore, this embodiment identifies edge regions with high contrast and sensitivity to blur in the blurred image. By analyzing the morphological characteristics of these edges after being affected by blur, it obtains the reference level of closed edges in the blurred image. Then, it selects strong-performing regions from all closed edges in the blurred image, using these as a reference for subsequently adaptively generating the optimal blur kernel for each pixel in the blurred image.

[0028] Preferably, in a specific embodiment of the present invention, for any blurred image, the Canny operator is used to obtain all edge pixels in the blurred image, and a contour search algorithm is combined to obtain several closed edges in the blurred image. Since the Canny operator and the contour search algorithm are both well-known prior art, they will not be described in detail in this embodiment. Furthermore, for any closed edge in the blurred image, the gradient value of each pixel on the closed edge in the blurred image is obtained using the Sobel operator. The difference between the maximum and minimum gradient values ​​on the closed edge in the blurred image is taken as the gradient interval of the closed edge in the blurred image. The product obtained by multiplying the ratio of the average gradient value of all pixels on the closed edge in the blurred image to the gradient interval by the number of pixels on the closed edge in the blurred image is taken as the reference factor of the closed edge in the blurred image. Since the Sobel operator is a well-known prior art, it will not be described in detail in this embodiment.

[0029] As an example, the specific formula for calculating the reference factor for obtaining the closed edge in the blurred image is as follows: In the formula, A reference factor representing the closed edge in the blurred image; This represents the average gradient value of all pixels on the closed edge in the blurred image; This represents the maximum gradient value on the closed edge in the blurred image; This represents the minimum gradient value on the closed edge in the blurred image; This represents the number of all pixels on the closed edge in the blurred image.

[0030] It should be noted that the larger the gradient value of the pixels on the closed edge in the blurred image and the smaller the gradient range of the closed edge, the clearer the closed edge in the blurred image. The better it is to use the closed edge in the blurred image as a reference for the optimal blur kernel of each pixel in the subsequent adaptive generation of the blurred image. On the other hand, the more pixels there are on the closed edge in the blurred image, the easier it is to identify the blur trailing feature produced by the closed edge in the blurred image. The better it is to use it as a reference for the optimal blur kernel of each pixel in the subsequent adaptive generation of the blurred image.

[0031] It should be further explained that the center position of the image at the blurred drone battery interface is least affected by the blur. Therefore, the closer the closed edge in the blurred image is to the center of the blurred image, the better it is to track the blur trailing features produced by the closed edge in the blurred image. Furthermore, since areas with regular contours will form simple and pure linear trailing when blurred, they can more clearly and reliably represent their blur features. Therefore, the reference degree of the closed edge in the blurred image can be obtained by combining the contour of the closed edge in the blurred image and its distance from the image center with the reference factor of the closed edge in the blurred image. This provides the optimal blur kernel for each pixel in the subsequent adaptive generation of the blurred image.

[0032] Preferably, in a specific embodiment of the present invention, for any closed edge in any blurred image, the Euclidean distance between the center of the closed edge in the blurred image and the center of the blurred image is taken as the eccentricity distance of the closed edge in the blurred image; the convex hull detection result of the closed edge in the blurred image is obtained by using a convex hull detection algorithm, and the product obtained by multiplying the absolute value of the difference between the area of ​​the closed edge in the blurred image and the area of ​​the convex hull detection result by the eccentricity distance of the closed edge in the blurred image is taken as the degree of unevenness of the closed edge in the blurred image; since the convex hull detection algorithm is a well-known prior art, it will not be described in detail in this embodiment; Furthermore, the ratio of the reference factor of the closed edge in the blurred image to its degree of unevenness is used as the reference degree of the closed edge in the blurred image.

[0033] As an example, the specific formula for calculating the reference level of the closed edge in the blurred image is as follows: In the formula, This indicates the degree of reference for the closed edges in the blurred image; A reference factor representing the closed edge in the blurred image; This represents the eccentricity distance of the closed edge in the blurred image; The area of ​​the closed edge in the blurred image; The area represents the convex hull detection result of the closed edge in the blurred image; This represents the function that takes the absolute value.

[0034] It should be noted that the eccentricity of the closed edge in a blurred image represents the Euclidean distance between the eccentricity of the closed edge and the center of the blurred image. The closer the closed edge is to the center of the blurred image, the better it is for tracking the blur trailing features produced by the closed edge. It is also more effective in using the closed edge as a reference for the subsequent adaptive generation of the optimal blur kernel for each pixel in the blurred image. The smaller the absolute value of the difference between the area of ​​the closed edge in the blurred image and the area of ​​its convex hull detection result, the more regular the outline of the closed edge in the blurred image. The resulting blur trailing effect is a simpler and purer linear blur trailing effect, which can more clearly and reliably represent its blurring features. It is also more effective in using the closed edge as a reference for the subsequent adaptive generation of the optimal blur kernel for each pixel in the blurred image.

[0035] Preferably, in a specific embodiment of the present invention, for any blurred image, all closed edges in the blurred image are sorted in ascending order according to the reference degree of all closed edges in the blurred image to obtain a sequence of closed edges in the blurred image; for any two adjacent closed edges in the sequence of closed edges in the blurred image, the absolute value of the difference in reference degree between the two adjacent closed edges in the sequence of closed edges in the blurred image is used as the reference difference between the two adjacent closed edges in the sequence of closed edges in the blurred image. Furthermore, the reference difference of all adjacent closed edges in the closed edge sequence of the blurred image is obtained. The adjacent closed edge corresponding to the largest reference difference is taken as the critical adjacent closed edge. The closed edge in the blurred image whose reference degree is less than or equal to the minimum value of the reference degree corresponding to the critical adjacent closed edge is taken as the non-reference edge. The closed edge in the blurred image whose reference degree is greater than or equal to the maximum value of the reference degree corresponding to the critical adjacent closed edge is taken as the reference edge. Each reference edge in the blurred image and the region inside the reference edge are taken as the strong representation region of the blurred image.

[0036] At this point, several regions with strong representation of the blurred image have been obtained.

[0037] Step S003: Based on the frequency domain information of the strong representation region of the blurred image, obtain the blur direction and blur length of the strong representation region of the blurred image; based on the blur direction of the strong representation region of the blurred image, combined with the gradient value and gradient direction of all pixels in the blurred image, obtain the blur confidence weight of each pixel in the blurred image; combined with the blur direction and blur length of the strong representation region of the blurred image, obtain the dense blur field of the blurred image, and then obtain the blur kernel of each pixel in the blurred image.

[0038] It should be noted that after obtaining several strong representation regions of the blurred image through step S002, the blur features of each strong representation region can be obtained based on the frequency domain information of each strong representation region. Since reliable blur features can be extracted from the strong representation regions, the blur confidence weight of each pixel in the blurred image can be obtained based on the blur features of the strong representation regions. Based on this, a dense blur field of the blurred image is constructed to obtain the blur kernel of each pixel in the blurred image. Then, the blurred image can be enhanced according to the blur kernel of each pixel in the blurred image to accurately identify the drone battery interface.

[0039] Preferably, in a specific embodiment of the present invention, for any strong representation region of any blurred image, a two-dimensional Fourier transform is performed on the strong representation region of the blurred image to obtain the spectrum of the strong representation region of the blurred image; a Radon transform is performed on the spectrum of the strong representation region of the blurred image to obtain a projection sequence of several angles, and the angle corresponding to the projection sequence with the largest projection integral is recorded as the projection direction, and the direction perpendicular to the projection direction is taken as the blur angle of the strong representation region in the blurred image. Furthermore, the average interval between all adjacent peaks in the projection sequence corresponding to the projection direction is taken as the spatial period of the fluctuation in the projection sequence corresponding to the projection direction, and the reciprocal of the spatial period of the fluctuation in the projection sequence corresponding to the projection direction is taken as the blur length of the strongly represented region in the blurred image; since Radon transform is a well-known prior art, it will not be described in detail in this embodiment.

[0040] It should be noted that vibration blur in the frequency domain manifests as a series of parallel zero-value fringes on the spectrogram. The direction of these fringes is perpendicular to the motion blur direction, and the fringe spacing is inversely proportional to the motion blur length. After transforming the spectrogram from a Cartesian coordinate system to a projection domain with angle and distance as parameters using Radon transform, the zero-value fringe feature in the projection domain manifests as a projection sequence at a specific angle exhibiting obvious periodic fluctuations. The projection direction with the largest projection integral is the normal direction of the zero-value fringe, and its perpendicular direction is the motion blur direction. The fluctuation period of this projection sequence directly reflects the interval frequency of the zero-value fringes in the spectrogram. According to the frequency domain theoretical model of motion blur, this interval frequency has a definite mathematical inverse relationship with the motion blur length. Therefore, by calculating the spatial period of the fluctuation of the projection sequence and taking its reciprocal, the corresponding motion blur length can be accurately calculated. Furthermore, since reliable blur features can be extracted from strong representation regions, the blur confidence weight of each pixel in the blurred image can be obtained based on the blur features of the strong representation regions. Based on this, a dense blur field of the blurred image can be constructed to obtain the blur kernel of each pixel in the blurred image.

[0041] Preferably, in a specific embodiment of the present invention, for any blurred image, the gradient value and gradient direction of each pixel in the blurred image are obtained by using the Sobel operator, and the sum of the gradient values ​​of all pixels in the eight neighborhoods of each pixel in the blurred image is recorded as the local texture feature of the corresponding pixel in the blurred image. Let any pixel in the blurred image be designated as the target point; let the region with the strongest representation closest to the target point in the blurred image be designated as the reference region of the target point; based on the local texture features of the target point and the local texture features of all pixels in the blurred image, combined with the gradient direction of the target point and the blur direction of the reference region of the target point, obtain the blur confidence weight of the target point, the specific calculation formula of which is as follows: In the formula, The fuzzy confidence weights represent the target points; Indicates the gradient direction at the target point; The fuzzy direction representing the reference area of ​​the target point; Represents the local texture features of the target point; This represents the local texture features of all pixels in the blurred image; This represents the function that takes the absolute value. This represents the decosine function.

[0042] It should be noted that the blur confidence weight of a pixel represents the probability that the pixel is part of the motion blur caused by vibration. It represents the angular difference between the gradient direction of the target point and the blur direction of the reference area of ​​the target point. The smaller the value, the more the gradient direction of the pixel matches the local blur trail direction, and the more similar its blur feature is to the blur feature of the nearest strong performance area. This represents the local texture richness of a pixel. Since areas with rich texture exhibit more pronounced blurring and trailing characteristics when subjected to vibration, the blur information they provide is more reliable. Therefore, a calculation weight is assigned to each pixel based on its local texture richness, thus obtaining the blur confidence weight for each pixel in the blurred image. Furthermore, based on the blur confidence weight of each pixel in the blurred image, combined with the blur angle and blur length of each strongly represented region in the blurred image, the blur kernel for each pixel in the blurred image can be obtained.

[0043] Preferably, in a specific embodiment of the present invention, for any blurred image, the blur angle and blur length of each strong representation region in the blurred image are used as the label of each strong representation region in the blurred image, and the blur confidence weight of each pixel in the blurred image is used as a smoothing term to construct a Markov random field model of the blurred image. Furthermore, the Markov random field model of the blurred image is optimized using the belief propagation algorithm to obtain the dense blur field of the blurred image. Since both the Markov random field model and the belief propagation algorithm are well-known existing technologies, they will not be described in detail in this embodiment. For any pixel in the blurred image, the blur direction and blur length corresponding to the pixel in the dense blur field of the blurred image are obtained as the blur kernel of the pixel in the blurred image.

[0044] It should be noted that by using known precise blur parameters as fixed constraint points and establishing correlation weights between adjacent pixels based on pixel matching degree, an optimization model covering the entire image is constructed. This model is calculated using mathematical optimization algorithms, allowing the blur parameters to adaptively propagate from the measured strong representation areas along the image edge structure and texture features, ultimately generating a dense blur parameter field that maintains spatial coherence. This field then adaptively generates an optimal blur kernel for each pixel in the blurred image to enhance the image and accurately identify the drone battery interface.

[0045] At this point, the blur kernel for each pixel in the blurred image is obtained.

[0046] Step S004: Enhance the blurred image based on the blur kernel of each pixel in the blurred image, thereby accurately identifying the drone battery interface.

[0047] It should be noted that after obtaining the blur kernel of each pixel in the blurred image through step S003, the blurred image can be enhanced based on the blur kernel of each pixel in the blurred image. Then, the enhanced blurred image is used to enable the drone battery swapping platform to accurately identify the drone battery interface.

[0048] Specifically, for any blurred image, the Wiener filtering algorithm is used to enhance the blurred image based on the blur kernel of each pixel in the blurred image, and the enhanced image is obtained. Since the Wiener filtering algorithm is a well-known prior art, it will not be described in detail in this embodiment. Furthermore, the enhanced results of all blurred images are obtained, and the enhanced results of all blurred images, along with the clear images, are input into the UAV battery interface recognition model, thereby enabling the UAV battery swapping platform to accurately identify the UAV battery interface. The UAV battery interface recognition model extracts key features by fusing deep learning and traditional geometric features. The deep learning network is responsible for recognizing the overall structure and orientation of the interface, while the traditional algorithm precisely locates subtle features such as pins and slots. Each recognition result undergoes multi-level verification: first, feature matching degree verification is performed to ensure that the similarity threshold of the main features reaches above 99%; second, geometric consistency verification is performed to ensure that the detected features conform to the physical constraints of the interface; finally, spatial pose rationality verification is performed.

[0049] Another embodiment of the present invention provides a drone battery interface identification system for a drone battery swapping platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the drone battery interface identification method for a drone battery swapping platform in steps S001 to S004.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying the battery interface of a drone for a drone battery swapping platform, characterized in that, The method includes the following steps: Acquire images of the drone's battery interface and filter out blurry images from those images. Based on the edge pixels and gradient characteristics in the blurred image, obtain several closed edges in the blurred image and the reference factor of each closed edge; based on the reference factor and the contour of the closed edges in the blurred image, obtain the reference degree of the closed edges in the blurred image; based on the reference degree of the closed edges in the blurred image, filter out the strong performance regions from all closed edges in the blurred image. Based on the frequency domain information of the strong representation region of the blurred image, the blur direction and blur length of the strong representation region of the blurred image are obtained; based on the blur direction of the strong representation region of the blurred image, combined with the gradient value and gradient direction of all pixels in the blurred image, the blur confidence weight of each pixel in the blurred image is obtained; combined with the blur direction and blur length of the strong representation region of the blurred image, the dense blur field of the blurred image is obtained, and thus the blur kernel of each pixel in the blurred image is obtained. The blurred image is enhanced based on the blur kernel of each pixel in the blurred image, thereby accurately identifying the drone battery interface.

2. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The specific method for acquiring images from the drone's battery interface and filtering out blurry images from those images includes: The camera mounted on the drone battery swapping platform captures images of the drone's battery interface at each moment. For an image of the drone battery interface at any given time, the Laplacian operator is used to perform a convolution operation on the image of the drone battery interface at that time. The result of the convolution operation is recorded as the Laplacian image of the drone battery interface at that time. The SSV method is used to obtain the energy of the Laplacian image of the drone battery interface at that time. Preset an energy threshold If the energy of the Laplacian image at the UAV battery interface at the stated time is less than The image of the drone's battery interface at the specified time is used as a blurred image.

3. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The specific method for obtaining several closed edges in the blurred image and the reference factor for each closed edge based on the edge pixels and gradient characteristics in the blurred image includes: For any blurred image, the Canny operator is used to obtain all edge pixels in the blurred image, and a contour search algorithm is used to obtain several closed edges in the blurred image. For any closed edge in the blurred image, the gradient value of each pixel on the closed edge is obtained using the Sobel operator. The difference between the maximum and minimum gradient values ​​on the closed edge is taken as the gradient interval of the closed edge in the blurred image. The product of the ratio of the average gradient value of all pixels on the closed edge to the gradient interval and the number of pixels on the closed edge is taken as the reference factor of the closed edge in the blurred image.

4. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The specific method for obtaining the reference degree of closed edges in a blurred image based on the reference factor and contour of the closed edges in the blurred image includes: For any closed edge in any blurred image, the Euclidean distance between the center of the closed edge and the center of the blurred image is taken as the eccentricity distance of the closed edge in the blurred image. The convex hull detection result of the closed edge in the blurred image is obtained by using a convex hull detection algorithm. The absolute value of the difference between the area of ​​the closed edge in the blurred image and the area of ​​the convex hull detection result is multiplied by the eccentricity distance of the closed edge in the blurred image, and the product is taken as the degree of unevenness of the closed edge in the blurred image. The ratio of the reference factor of the closed edge in the blurred image to its degree of unevenness is used as the reference degree of the closed edge in the blurred image.

5. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The method for selecting strong representation regions from all closed edges in a blurred image based on the reference degree of closed edges in the blurred image includes the following specific methods: For any blurred image, all closed edges in the blurred image are sorted in ascending order according to their reference degree to obtain a sequence of closed edges in the blurred image; for any two adjacent closed edges in the sequence of closed edges in the blurred image, the absolute value of the difference in reference degree between the two adjacent closed edges in the sequence of closed edges in the blurred image is used as the reference difference between the two adjacent closed edges in the sequence of closed edges in the blurred image. The reference difference of all adjacent closed edges in the closed edge sequence of the blurred image is obtained. The adjacent closed edge corresponding to the largest reference difference is taken as the critical adjacent closed edge. The closed edge in the blurred image whose reference degree is greater than or equal to the maximum value of the reference degree corresponding to the critical adjacent closed edge is taken as the reference edge. Each reference edge in the blurred image and the region inside the reference edge are taken as the strong representation region of the blurred image.

6. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The specific method for obtaining the blur direction and blur length of the strong representation region of the blurred image based on the frequency domain information of the strong representation region of the blurred image includes: For any strong representation region of any blurred image, perform a two-dimensional Fourier transform on the strong representation region of the blurred image to obtain the spectrum of the strong representation region of the blurred image; perform a Radon transform on the spectrum of the strong representation region of the blurred image to obtain a projection sequence of several angles, and record the angle corresponding to the projection sequence with the largest projection integral as the projection direction, and take the direction perpendicular to the projection direction as the blur angle of the strong representation region in the blurred image. The average interval between all adjacent peaks in the projection sequence corresponding to the projection direction is taken as the spatial period of the fluctuation in the projection sequence corresponding to the projection direction, and the reciprocal of the spatial period of the fluctuation in the projection sequence corresponding to the projection direction is taken as the blur length of the strong representation region in the blurred image.

7. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The method for obtaining the blur confidence weight of each pixel in the blurred image based on the blur direction of the strong representation region of the blurred image, combined with the gradient values ​​and gradient directions of all pixels in the blurred image, includes the following specific methods: For any blurred image, the gradient value and gradient direction of each pixel in the blurred image are obtained using the Sobel operator. The sum of the gradient values ​​of all pixels in the eight neighborhoods of each pixel in the blurred image is recorded as the local texture feature of the corresponding pixel in the blurred image. Let any pixel in the blurred image be designated as the target point; let the region with the strongest representation closest to the target point in the blurred image be designated as the reference region of the target point; the formula for calculating the fuzzy confidence weight of the target point is as follows: In the formula, The fuzzy confidence weights represent the target points; Indicates the gradient direction at the target point; The fuzzy direction representing the reference area of ​​the target point; Represents the local texture features of the target point; This represents the local texture features of all pixels in the blurred image; This represents the function that takes the absolute value. This represents the cosine function.

8. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The specific method for obtaining the dense blur field of the blurred image is as follows: For any blurred image, the blur angle and blur length of each strong representation region in the blurred image are used as the label of each strong representation region in the blurred image, and the blur confidence weight of each pixel in the blurred image is used as a smoothing term to construct a Markov random field model of the blurred image. The Markov random field model of the blurred image is optimized and solved using the belief propagation algorithm to obtain the dense blurred field of the blurred image.

9. The method for identifying the battery interface of a drone for a drone battery swapping platform according to claim 1, characterized in that, The method for enhancing the blurred image based on the blur kernel of each pixel in the blurred image includes the following specific methods: For any blurred image, the Wiener filtering algorithm is used to enhance the blurred image based on the blur kernel of each pixel in the blurred image.

10. A drone battery interface identification system for drone battery swapping platforms, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the drone battery interface identification method for drone battery swapping platforms as described in any one of claims 1-9.

Images