Pose information determination method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a pose information determination method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a plurality of inspection images of a power transmission line acquired by inspection equipment, and performing image enhancement processing on each inspection image to obtain each inspection image after image enhancement processing; performing feature extraction on each inspection image after image enhancement processing by using a pre-configured feature extraction network to obtain a plurality of feature points; determining a plurality of inter-frame feature matching pairs based on each feature point, and determining pose information of the inspection equipment according to the plurality of inter-frame feature matching pairs; each inter-frame feature matching pair comprises two feature points, and the inspection images to which the two feature points belong are different. By adopting the method, the pose information of the inspection equipment, which is adaptive to a complex scene of a power transmission line and is relatively high in accuracy, can be determined.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, computer device, computer-readable storage medium, and computer program product for determining pose information. Background Technology

[0002] With the continuous expansion of my country's power grid, the total length of transmission lines has exceeded one million kilometers. Traditional manual inspection methods have problems such as low efficiency, high risk, and limited coverage. Intelligent inspection equipment has become the core development direction of intelligent operation and maintenance of power grids.

[0003] Specifically, inspection images of transmission lines can be collected by inspection equipment, and a three-dimensional environmental map of the transmission lines can be constructed based on the inspection images, so as to realize the inspection of transmission lines based on the three-dimensional environmental map.

[0004] However, the accurate construction of 3D environmental maps of power transmission lines heavily relies on the real-time pose information of inspection equipment. If the pose information is inaccurate, problems such as map stitching misalignment and map distortion can easily occur. Therefore, there is an urgent need for a method to determine the pose information of inspection equipment that is adaptable to the complex scenarios of power transmission lines and has high accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for determining the position and orientation information of inspection equipment that can adapt to complex scenarios of power transmission lines and has high accuracy, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for determining pose information, including:

[0007] Multiple inspection images of the transmission line collected by the inspection equipment are acquired, and image enhancement processing is performed on each inspection image to obtain the image enhancement processed inspection images.

[0008] Using a pre-configured feature extraction network, features are extracted from each inspection image after image enhancement to obtain multiple feature points;

[0009] Multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0010] In one embodiment, image enhancement processing is performed on each inspection image to obtain the enhanced inspection images, including: performing color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; performing multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; and performing color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain the enhanced inspection images.

[0011] In one embodiment, determining multiple inter-frame feature matching pairs based on each feature point includes: inputting each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and the cross-attention information.

[0012] In one embodiment, the pose information of the inspection device is determined based on multiple inter-frame feature matching pairs, including: acquiring the camera matrix and distortion coefficients of the inspection device, as well as the spatial association information of feature points in each inter-frame feature matching pair, and constructing a pose calculation model based on the camera matrix, distortion coefficients, and spatial association information; determining rotation parameters and translation parameters based on the pose calculation model, and determining the pose information of the inspection device based on the rotation parameters and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection device, and the translation parameters are used to characterize the spatial position of the inspection device.

[0013] In one embodiment, a pose calculation model is constructed based on the camera matrix, distortion coefficients, and spatial correlation information, including: constructing a reprojection function based on the camera matrix and distortion coefficients; determining the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information; and constructing a pose calculation model based on the reprojection function and depth data.

[0014] In one embodiment, the feature extraction network is a SuperPoint network that is modified based on the GhostNet network.

[0015] Secondly, this application also provides a pose information determination device, comprising:

[0016] The acquisition module is used to acquire multiple inspection images collected by the inspection equipment of the transmission line, and to perform image enhancement processing on each inspection image to obtain the image enhancement processed inspection images.

[0017] The execution module is used to extract features from each inspection image after image enhancement processing using a pre-configured feature extraction network to obtain multiple feature points.

[0018] The determination module is used to determine multiple inter-frame feature matching pairs based on each feature point, and to determine the pose information of the inspection equipment based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0019] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the embodiments of the first aspect above.

[0020] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the embodiments of the first aspect above.

[0021] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the embodiments of the first aspect above.

[0022] The aforementioned pose information determination method, apparatus, computer equipment, computer-readable storage medium, and computer program product first acquire multiple inspection images of the transmission line collected by the inspection equipment, and then perform image enhancement processing on each inspection image to obtain image-enhanced inspection images. Next, using a pre-configured feature extraction network, features are extracted from each image-enhanced inspection image to obtain multiple feature points. Then, multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined according to these inter-frame feature matching pairs. Each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images. The pose information determination method provided in this application determines the pose information of the inspection equipment through multiple inspection images of the transmission line collected by the inspection equipment. Because the determined pose information of the inspection equipment is adapted to the complex scene of the transmission line and has high accuracy, it can achieve accurate construction of a three-dimensional environmental map of the transmission line. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a pose information determination method in one embodiment;

[0025] Figure 2 This is a flowchart illustrating a method for obtaining enhanced inspection images in one embodiment.

[0026] Figure 3 This is a schematic diagram of a feature matching network in one embodiment;

[0027] Figure 4 This is a flowchart illustrating a method for determining the pose information of an inspection device in one embodiment.

[0028] Figure 5 This is a flowchart illustrating a method for constructing a pose calculation model in one embodiment;

[0029] Figure 6 This is a flowchart illustrating a pose information determination method in another embodiment;

[0030] Figure 7 This is a structural block diagram of a pose information determination device in one embodiment;

[0031] Figure 8 This is an internal structural diagram of a computer device in one embodiment;

[0032] Figure 9 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0034] It should be noted that the terms "comprising" and "having," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusion. The term "multiple" as used in this application refers to two or more. The term "and / or" as used in this application refers to one of the solutions, or any combination of multiple solutions.

[0035] With the continuous expansion of my country's power grid, the total length of transmission lines has exceeded one million kilometers. Traditional manual inspection methods suffer from low efficiency, high risk, and limited coverage. Intelligent inspection equipment has become the core development direction for intelligent operation and maintenance of the power grid. Specifically, inspection equipment can collect images of transmission lines and construct a three-dimensional environmental map of the transmission lines based on these images, thereby enabling the inspection of transmission lines based on the three-dimensional environmental map.

[0036] However, the accurate construction of 3D environmental maps of power transmission lines heavily relies on the real-time pose information of inspection equipment. If the pose information is inaccurate, problems such as map stitching misalignment and map distortion can easily occur. Therefore, there is an urgent need for a method to determine the pose information of inspection equipment that is adaptable to the complex scenarios of power transmission lines and has high accuracy.

[0037] In view of this, this application provides a pose information determination method. First, multiple inspection images of the transmission line collected by the inspection equipment are acquired, and image enhancement processing is performed on each inspection image to obtain enhanced inspection images. Then, a pre-configured feature extraction network is used to extract features from each enhanced inspection image to obtain multiple feature points. Next, multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined according to these inter-frame feature matching pairs. Each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images. The pose information determination method provided in this application determines the pose information of the inspection equipment through multiple inspection images of the transmission line collected by the inspection equipment. Because the determined pose information of the inspection equipment is adaptable to complex scenarios of transmission lines and has high accuracy, it can achieve accurate construction of a three-dimensional environmental map of the transmission line.

[0038] The pose information determination method provided in this application can be executed by a computer device, which can be a terminal or a server.

[0039] In one exemplary embodiment, such as Figure 1 As shown, a method for determining pose information is provided, which includes the following steps:

[0040] Step 101: Acquire multiple inspection images of the transmission line collected by the inspection equipment, and perform image enhancement processing on each inspection image to obtain the image enhancement processed inspection images.

[0041] Optionally, inspection equipment refers to equipment used for the operation and maintenance inspection of power transmission lines. Examples include drones equipped with cameras and ground inspection robots.

[0042] For example, in the embodiments of this application, the inspection device can be a drone equipped with a camera, which can be a binocular camera. The binocular camera group can have a resolution of 1280×720, a field of view of 69°×42°, a maximum frame rate of 90fps, and a depth accuracy of ±2%.

[0043] Furthermore, the camera can be fixed to the top of the drone using a custom aluminum alloy bracket. The bracket features a lightweight, 5mm thick PA plastic with a hollow design. The camera's optical axis is tilted upwards at a 5° angle to the top plane of the drone, ensuring full coverage of power line towers and conductors during inspections. The camera lens is secured to the drone body with shock-absorbing rubber pads to prevent blurry inspection images caused by flight vibrations.

[0044] Transmission lines refer to power facilities that transmit electrical energy from power plants to substations or between substations. Their core components include key parts such as conductors, towers, insulators, and fittings.

[0045] Inspection images refer to image data of power transmission line components and their surrounding environment collected by inspection equipment during the operation and maintenance of power transmission lines.

[0046] In some exemplary embodiments, a computer device can acquire multiple inspection images of a power transmission line collected by an inspection device.

[0047] Specifically, the computer equipment can communicate with the inspection equipment and acquire multiple inspection images of the transmission lines collected by the inspection equipment through this communication connection.

[0048] For example, the inspection equipment can fly along a 60kV transmission line, at a height of 5m above the conductor, following an S-shaped trajectory around the line, passing 3 towers, 6 conductor sections, and 12 sets of insulators, with a total line length of 80m. The data acquisition interval is 0.1s, acquiring a total of 2000 inspection images, including 800 images under low-light conditions (cloudy days) and 700 images under normal lighting conditions. Optionally, GPS data synchronization can be used during acquisition, recording GPS ground truth values ​​every 10 frames.

[0049] Furthermore, after acquiring multiple inspection images of the transmission line collected by the inspection equipment, the computer equipment can perform image enhancement processing on each inspection image to obtain each inspection image after image enhancement processing.

[0050] Specifically, computer equipment can input multiple inspection images into a pre-trained image enhancement processing model to obtain the image enhancement processed images of each inspection image output by the image enhancement processing model.

[0051] Step 102: Using a pre-configured feature extraction network, perform feature extraction on each inspection image after image enhancement to obtain multiple feature points.

[0052] The feature extraction network is a SuperPoint network adapted from GhostNet. For example, the input layer (layer 1) and the last six GhostNet bottleneck modules of the lightweight GhostNet network replace the VGG-like encoding layers of the SuperPoint network. A network width coefficient of 1.25 is set to increase the receptive field while reducing network parameters, thus improving the feature representation ability of weakly textured regions such as conductors and insulators in transmission lines. A feature extraction threshold of 0.01 is set to filter low-response feature points, outputting feature point coordinates and a 128-dimensional feature description vector. The feature extraction network is trained using grayscale images converted from a self-built dataset under self-supervised conditions. The optimizer is Adam, with a learning rate of 0.001 and a batch size of 128. During training, the hyperparameters are maintained consistent with the original SuperPoint network.

[0053] Optionally, feature points can be key image points extracted from the enhanced inspection images, typically corresponding to the core components of the transmission line or prominent markers in the surrounding environment.

[0054] In some exemplary embodiments, a computer device may use a pre-configured feature extraction network to extract features from each inspection image after image enhancement processing to obtain multiple feature points.

[0055] Specifically, the computer equipment can input the image-enhanced inspection images into a pre-configured feature extraction network to obtain multiple feature points output by the pre-configured feature extraction network. In one optional approach, for each inspection image, the pre-configured feature extraction network can output approximately 350 feature points and a 128-dimensional feature description vector.

[0056] In an optional embodiment of this application, in practical applications, if the image size input to the feature matching network is too large or too small, it may cause obstacles to feature extraction or excessive memory requirements. Therefore, each inspection image after image enhancement can be scaled to 400×300 before being input into the feature extraction network.

[0057] Step 103: Determine multiple inter-frame feature matching pairs based on each feature point, and determine the pose information of the inspection equipment based on the multiple inter-frame feature matching pairs.

[0058] Each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0059] For example, inter-frame feature matching pairs refer to the combination of feature points corresponding to the same physical point in different inspection images, determined through feature similarity comparison and spatial correlation verification. For instance, the feature point of an insulator of a transmission line is P1 in the first inspection image and P2 in the second inspection image. After matching verification, (P1, P2) constitutes a set of inter-frame feature matching pairs. Its core function is to establish the spatial correspondence between different inspection image frames.

[0060] Pose information refers to the spatial state parameters of the inspection equipment when acquiring corresponding inspection images. For example, it can include position information and attitude information. Position information is the three-dimensional spatial coordinates, the specific orientation of the inspection equipment along the X / Y / Z axes; attitude information is the three-dimensional attitude angles, including the pitch angle, roll angle, and yaw angle of the inspection equipment.

[0061] In some exemplary embodiments, after obtaining multiple feature points, the computer device can determine multiple inter-frame feature matching pairs based on each feature point.

[0062] Specifically, computer devices can input multiple feature points into a pre-trained matching model to obtain multiple inter-frame feature matching pairs output by the matching model.

[0063] Furthermore, after determining multiple inter-frame feature matching pairs, the computer equipment can determine the pose information of the inspection equipment based on the multiple inter-frame feature matching pairs.

[0064] Specifically, computer equipment can input multiple inter-frame feature matching pairs into the pose information determination model to obtain the pose information of the inspection equipment output by the pose information determination model.

[0065] The aforementioned pose information determination method first acquires multiple inspection images of the transmission line collected by the inspection equipment, and performs image enhancement processing on each inspection image to obtain enhanced inspection images. Then, using a pre-configured feature extraction network, features are extracted from each enhanced inspection image to obtain multiple feature points. Next, multiple inter-frame feature matching pairs are determined based on these feature points, and the pose information of the inspection equipment is determined according to these inter-frame feature matching pairs. Each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images. The pose information determination method provided in this application determines the pose information of the inspection equipment through multiple inspection images of the transmission line collected by the inspection equipment. Because the determined pose information of the inspection equipment is adapted to the complex scene of the transmission line and has high accuracy, it can achieve accurate construction of a three-dimensional environmental map of the transmission line.

[0066] In one exemplary embodiment, such as Figure 2As shown, image enhancement processing is performed on each inspection image to obtain enhanced inspection images, including the following steps:

[0067] Step 201: Perform color space conversion processing on each inspection image to obtain the luminance component, hue component and saturation component of each inspection image in the HSV color space.

[0068] In some exemplary embodiments, the computer device can perform color space conversion processing on each inspection image to obtain the luminance component, hue component and saturation component of each inspection image in the HSV color space.

[0069] Specifically, each inspection image is an RGB image. The computer device can call the OpenCV function cv2.cvtColor(rgb_image, cv2.COLOR_RGB2HSV) to convert the inspection image to HSV space. The hue component H, saturation component S, and luminance component V are obtained by separating the image using cv2.split(hsv_image). The conversion time is ≤3ms / frame, and the color space conversion error is ≤2 pixels of gray value.

[0070] The expression for color space conversion is as follows:

[0071] ;

[0072] ;

[0073] Where max is the maximum pixel value among the R, G, and B channels, and min is the minimum pixel value among the R, G, and B channels. H ranges from 0 to 360 degrees, S ranges from 0 to 1, and V ranges from 0 to 1.

[0074] Step 202: Perform multi-scale fast guided filtering denoising and bilinear interpolation on the luminance component to obtain the processed luminance component.

[0075] In some exemplary embodiments, after obtaining the luminance component, hue component and saturation component of each inspection image in the HSV color space, the computer device can perform multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component.

[0076] Specifically, a fast guided filter function is implemented based on a custom function, with three filter scales and filter window radii of 4×4, 8×8, and 12×12, and regularization parameters. The weights of the filtering results at each scale are allocated as follows: , , By reducing computational complexity with a downsampling rate of 3, the filtering processing time is ≤4ms / frame, which improves the algorithm's running speed by nearly ten times, while maintaining a processing effect essentially the same as ordinary guided filtering. It removes noise from the luminance V component while preserving image edge details, outputting a smoothed luminance V component.

[0077] The expression for a multi-scale fast guided filter is as follows:

[0078] ;

[0079] Where Q(x, y) is the pixel value output by the multi-scale fast guided filter at (x, y), i.e., the illumination map. P(x, y) and I(x, y) are the original image and the guided map, respectively, and they are equal. N is the number of scales. It is a fast guided filter function. It is in the filter window The center pixel value within, and These are the coefficients of a linear function related to the window. The filter radius is used to calculate the filtered pixel values ​​based on the input image, the guide image, the filter radius, and the regularization parameters. It assigns weights to the filtering results at different scales to optimize the overall filtering effect.

[0080] Furthermore, the smoothed luminance V component is divided into 8×8 rectangular sub-images, a contrast threshold of 40 is set, and grayscale mapping is achieved for edge sub-image pixels through bilinear interpolation to avoid overexposure problems caused by traditional histogram equalization. The expression is as follows:

[0081] ;

[0082] in, and These are the ratios of the horizontal and vertical distances of the target pixel relative to the origin of the pixel plane to the window size, respectively.

[0083] Step 203: Perform color space conversion based on the processed luminance, hue, and saturation components to obtain the inspection images after image enhancement.

[0084] In some exemplary embodiments, after obtaining the processed luminance component, the computer device can perform color space conversion processing based on the processed luminance component, hue component and saturation component to obtain each inspection image after image enhancement processing.

[0085] Specifically, the processed HSV image is converted back to RGB space to obtain the enhanced inspection images. The color space conversion expression is as follows:

[0086] ;

[0087] .

[0088] Through image enhancement processing, in low-light scenes, the average brightness of the enhanced inspection image increased from 52 to 128, the contrast increased from 0.32 to 0.78, and the noise standard deviation decreased from 18.6 to 7.2.

[0089] In an exemplary embodiment, determining multiple inter-frame feature matching pairs based on each feature point includes: inputting each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs.

[0090] The feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module. The self-attention unit determines attention information based on each feature point, the cross-attention unit determines cross-attention information based on each feature point, and the dynamic adjustment module determines multiple inter-frame feature matching pairs based on the attention information and cross-attention information.

[0091] In some exemplary embodiments, a computer device can input each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs.

[0092] Specifically, feature matching networks can be like... Figure 3 As shown, in the feature matching network, the self-attention unit introduces the relative position information of feature points through rotation encoding, and the attention score is calculated using the following formula:

[0093] ;

[0094] Where W is the projection matrix, This is the attention score between the feature points of the image and the input / output. The calculation method differs for self-attention units and cross-attention units. In a self-attention unit, for each feature point i, the current state is... Decomposed into key vectors and query vector Based on this, the attention score between a feature point and different feature points on the same image is defined using rotation encoding R, which can be expressed as: .

[0095] The expression for the relative position rotation code R is as follows:

[0096] ; ;

[0097] After calculating the self-attention scores for the two images, the cross-attention unit is input to calculate the cross-attention score: ;

[0098] To avoid unnecessary computation and save inference time, the dynamic adjustment module can dynamically adjust the network depth based on the difficulty of the input image pairs, reducing the number of network layers. Feature points that do not meet the matching score are pre-cropped, and the network width is dynamically adjusted.

[0099] At each layer of the backbone network, the confidence score of the feature points is calculated as follows: .

[0100] The higher the confidence score, the more reliable the representation of the feature point is, that is, the higher the confidence that the feature point is classified as a matching point or an unmatchable point.

[0101] In a given network layer 1, when When a feature point is considered to have high confidence, the threshold is used. The threshold gradually decreases as the number of network layers increases.

[0102] exist The inference of the feature matching network layer will end prematurely. The parameter 'a' needs to balance accuracy and inference time, and is usually set to 95%.

[0103] When the inference process executes normally, computational resources are saved by pruning feature points that do not meet the matching degree. The matching degree score for each feature point is calculated as follows: .

[0104] Feature matching networks only match feature points with high confidence, and the quality of the matching process is represented by the matching score. Therefore, the strategy of dynamically adjusting the network width is to use feature points with high confidence but low matching scores as targets for pruning, which can be expressed as: When b is 0.01, the network can be more effective at cropping target feature points.

[0105] Furthermore, by employing a lightweight head, state prediction and updates can be performed at any layer. Specifically, the point-to-point similarity matrix is ​​first calculated, which can be represented as... .

[0106] in, The transformation is a biased linear transformation, and the similarity is defined as the probability that a pair of feature points is projected from the same point in 3D space. Combining the similarity and matching scores into a soft biased assignment matrix can be expressed as follows: .

[0107] A feature point pair is considered a match when both feature points are predicted to be matchable and their similarity is higher than that of any other point in the two images. Feature points with a matching score below 0.1 are cropped to reduce computational cost. For feature matching pairs output by the network, a RANSAC algorithm with an inlier threshold of 1.0px is used to further eliminate false matches, retaining valid matching pairs for subsequent pose estimation.

[0108] In one exemplary embodiment, such as Figure 4 As shown, the pose information of the inspection equipment is determined based on multiple inter-frame feature matching pairs, including the following steps:

[0109] Step 401: Obtain the camera matrix and distortion coefficients of the inspection equipment, as well as the spatial association information of feature points in each frame feature matching pair, and construct a pose calculation model based on the camera matrix, distortion coefficients and spatial association information.

[0110] Optionally, the camera matrix refers to the matrix describing the internal optical characteristics and imaging geometry of the camera in the inspection equipment. Distortion coefficients refer to the set of parameters used to correct optical distortion of the camera lens, which may include radial distortion coefficients and tangential distortion coefficients.

[0111] Spatial correlation information refers to the spatial relationship data contained in the feature points corresponding to the same physical point in different frame images in the inter-frame feature matching pair.

[0112] In some exemplary embodiments, the computer device can first obtain the camera matrix and distortion coefficients of the inspection device, as well as the spatial association information of feature points in the feature matching pairs between each frame, and then construct a pose calculation model based on the camera matrix, distortion coefficients and spatial association information.

[0113] Specifically, computer equipment can input camera matrix, distortion coefficients and spatial correlation information into the pose calculation model construction model to obtain the pose calculation model output by the pose calculation model construction model.

[0114] Step 402: Determine the rotation and translation parameters based on the pose calculation model, and determine the pose information of the inspection equipment based on the rotation and translation parameters.

[0115] Among them, the rotation parameter is used to characterize the spatial attitude of the inspection equipment, and the translation parameter is used to characterize the spatial position of the inspection equipment.

[0116] In some exemplary embodiments, the computer device can determine rotation and translation parameters based on a pose calculation model, and determine the pose information of the inspection device based on the rotation and translation parameters.

[0117] In one exemplary embodiment, such as Figure 5As shown, a pose calculation model is constructed based on the camera matrix, distortion coefficients, and spatial correlation information, including the following steps:

[0118] Step 501: Construct the reprojection function based on the camera matrix and distortion coefficients;

[0119] Step 502: Based on spatial correlation information, determine the depth data of feature points in each inter-frame feature matching pair;

[0120] Step 503: Construct a pose calculation model based on the reprojection function and depth data.

[0121] In some exemplary embodiments, the computer device can construct a reprojection function based on the camera matrix and distortion coefficients, determine the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information, and then construct a pose calculation model based on the reprojection function and the depth data.

[0122] Furthermore, the computer equipment can determine the rotation and translation parameters based on the pose calculation model, and determine the pose information of the inspection equipment based on the rotation and translation parameters.

[0123] Specifically, the camera intrinsic parameter matrix K is read from the camera calibration result file, and the camera distortion coefficients are loaded for subsequent reprojection error calculation. A maximum likelihood function is constructed based on multiple inter-frame feature matching pairs and the stereo depth information acquired by the camera.

[0124] Solving for the rotation matrix R and translation matrix t of vi relative to pi, the camera pose can be expressed as: .in, For the depth information of spatial points, To set the spatial points of the calibrated camera in the global coordinate system.

[0125] right The uncertain observations obtained on the image plane can be expressed as: , ,in, It is a two-dimensional covariance matrix used to describe the uncertainty of the observation points.

[0126] via orthographic projection Image points Transforming to the 3D direction in the camera coordinate system, we obtain the expression:

[0127] ;

[0128] in, It is the Jacobian matrix of the orthographic projection matrix.

[0129] Using the uncertainty of image points, the covariance matrix of camera points is obtained as shown in the following expression:

[0130] ;

[0131] Wherein, the covariance matrix Its rank is 2, therefore it is singular and irreversible.

[0132] Furthermore, spherical normalization yields the final observation expression as follows, which is referred to as the direction vector:

[0133] ;

[0134] Based on the propagation properties of covariance, the expression can be obtained as follows: ;

[0135] At this point, the covariance matrix is ​​still singular, and the maximum likelihood estimate based on the three residual components of the bearing vector is invalid.

[0136] Therefore, for the redundant representation of the homogeneous vector v, the minimum representation of the covariance information is desirable. To obtain the initial estimate of the camera's absolute direction, the null space of the vector is used for the solution. The null space is defined as follows: Given A as an n-dimensional matrix, the null space (Null) of A is an n-dimensional vector defined by the following expression: ;

[0137] The null space of v forms a two-dimensional coordinate system, whose axes are represented by r and s. r and s are perpendicular to v and lie in the tangent space of v: ;

[0138] The function Null computes the singular value decomposition (SVD) of v and takes the two eigenvectors corresponding to the two zero eigenvalues, which can be expressed as follows: ; It is an orthogonal matrix, that is , Additionally, the Jacobian matrix represents the transformation from the tangent space to the original vector.

[0139] therefore, This yields a simplified equivalence from the original homogeneous vector v to its original homogeneous vector v. The transformation of can be expressed as: Covariance with non-singular values: .

[0140] In summary, for an image point, we obtain a 3D point in the camera coordinate system using forward projection, then obtain the direction vector v using the 3D point, and finally obtain the tangent space of v through v. Furthermore, v is perpendicular to its tangent space. Therefore, the ideal projection of v onto its tangent space should be onto the origin of the tangent space. However, due to errors in the rotation matrix R and the translation matrix t, the projection will always have a deviation. So the goal is to minimize this error and project v onto the origin of the tangent space.

[0141] Based on the transformation formula from homogeneous vector v to simplified equivalent and the covariance formula for non-singular values, substituting the calculation results of these two formulas into the covariance formula for non-singular values, a formula for characterizing the relationship between projection deviation and pose parameters can be derived: .

[0142] Wherein, due to the formula Since it is not equal to 0, if the absolute orientation of the camera in the global coordinate system is known, and the projected coordinates of the global 3D space point in the tangent space corresponding to the direction vector v are obtained, the coordinate derivation formula can be simplified using this direction vector to obtain the coordinates after simplification of the direction vector. The specific derivation relationship is as follows:

[0143] ;

[0144] ;

[0145] because In the two derivations of the simplified coordinate derivation formula for the direction vector above, the unknown camera pose values ​​to be solved are both linearly distributed. Superimposing them into a matrix A yields a homogeneous system of linear equations, that is... .

[0146] in, Since each observation produces two residuals, solving the above expression requires at least 5 points. Using uncorrelated observations, a stochastic model is given:

[0147] ;

[0148] The final normal equation is obtained as follows: .exist Under the given conditions, find u that minimizes the above expression, and solve using SVD to obtain... .

[0149] The solution to the above equation is a specific column in V corresponding to the minimum singular value of D, that is...

[0150] ;

[0151] The above expression needs to be determined by a scale factor because the translation part only points in the correct direction, which is determined by the rotation matrix. , , The scaling factor is obtained by constraining the norm to be 1, that is, the modulus of each column to be 1. Therefore, the final translation is... .

[0152] The constraints used indicate that the nine rotation parameters in the rotation matrix R do not define a correct rotation matrix. This can be calculated using SVD decomposition. .

[0153] The optimal rotation matrix that minimizes the Frobenius norm is shown in the following expression: .

[0154] The linear equations corresponding to the likelihood function are solved using Singular Value Decomposition (SVD), outputting the rotation matrix R and translation vector t, with a solution time ≤ 1ms / frame. After pose estimation, the reprojection error is calculated, and outliers with a reprojection error > 2.0px are removed, ultimately retaining inliers with a proportion ≥ 90%.

[0155] Implementation effect verification:

[0156] In real-world power transmission line scenarios, the closed-loop positioning error decreased from 0.21m to 0.07m, a 63.6% improvement. In the low-light sequences (V101-V203) of the EuRoc dataset, the method provided in this application achieved a root mean square error (RMSE) of 0.08m for the absolute trajectory error (ATE), with errors of 0.06m in the X direction, 0.05m in the Y direction, and 0.07m in the Z direction. In backlit scenes, the feature point repetition rate reached 71%, a 91.9% improvement over traditional ORB features, with no positioning interruptions due to matching failures. In cloudy, low-light scenes, the feature point extraction success rate was 98.7%, a 16.4% improvement over the traditional method (82.3%), with a continuous positioning stabilization time ≥30 minutes. The entire single-frame positioning process took 14ms (image enhancement 12ms + feature extraction 3ms + matching 2ms + pose estimation 1ms - parallel optimization 4ms), achieving 22fps on Jetson XavierNX.

[0157] In one exemplary embodiment, such as Figure 6 As shown, another method for determining pose information is provided, which includes the following steps:

[0158] Step 601: Acquire multiple inspection images of the transmission line collected by the inspection equipment; perform color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; perform multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; perform color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain each inspection image after image enhancement processing.

[0159] Step 602: Using a pre-configured feature extraction network, feature extraction is performed on each inspection image after image enhancement processing to obtain multiple feature points; the feature extraction network is a SuperPoint network adjusted based on the GhostNet network; each feature point is input into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and cross-attention information;

[0160] Step 603: Obtain the camera matrix and distortion coefficients of the inspection equipment, as well as the spatial association information of feature points in each inter-frame feature matching pair, and construct a reprojection function based on the camera matrix and distortion coefficients; determine the depth data of feature points in each inter-frame feature matching pair based on the spatial association information; construct a pose calculation model based on the reprojection function and depth data; determine the rotation and translation parameters based on the pose calculation model, and determine the pose information of the inspection equipment based on the rotation and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection equipment, and the translation parameters are used to characterize the spatial position of the inspection equipment; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0161] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0162] Based on the same inventive concept, this application also provides a pose information determining device for implementing the pose information determining method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more pose information determining device embodiments provided below can be found in the limitations of the pose information determining method described above, and will not be repeated here.

[0163] In one exemplary embodiment, such as Figure 7 As shown, a pose information determination device 700 is provided, including: an acquisition module 701, an execution module 702, and a determination module 703, wherein:

[0164] The acquisition module 701 is used to acquire multiple inspection images collected by the inspection equipment of the transmission line, and to perform image enhancement processing on each inspection image to obtain each inspection image after image enhancement processing.

[0165] The execution module 702 is used to extract features from each inspection image after image enhancement processing using a pre-configured feature extraction network to obtain multiple feature points.

[0166] The determination module 703 is used to determine multiple inter-frame feature matching pairs based on each feature point, and to determine the pose information of the inspection equipment based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0167] In one embodiment, the acquisition module 701 is specifically used to perform color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; to perform multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; and to perform color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain each inspection image after image enhancement processing.

[0168] In one embodiment, the determining module 703 is specifically used to input each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and the cross-attention information.

[0169] In one embodiment, the determining module 703 is specifically used to acquire the camera matrix and distortion coefficients of the inspection equipment, as well as the spatial association information of feature points in each frame feature matching pair, and to construct a pose calculation model based on the camera matrix, distortion coefficients, and spatial association information; to determine rotation parameters and translation parameters based on the pose calculation model, and to determine the pose information of the inspection equipment based on the rotation parameters and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection equipment, and the translation parameters are used to characterize the spatial position of the inspection equipment.

[0170] In one embodiment, the determining module 703 is specifically used to construct a reprojection function based on the camera matrix and distortion coefficients; determine the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information; and construct a pose calculation model based on the reprojection function and depth data.

[0171] In one embodiment, the feature extraction network is a SuperPoint network that is modified and processed based on the GhostNet network.

[0172] The various modules in the aforementioned pose information determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0173] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a pose information determination method.

[0174] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a pose information determination method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0175] Those skilled in the art will understand that Figure 8 and Figure 9The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0176] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0177] Multiple inspection images of the transmission line collected by the inspection equipment are acquired, and image enhancement processing is performed on each inspection image to obtain the image enhancement processed inspection images.

[0178] Using a pre-configured feature extraction network, features are extracted from each inspection image after image enhancement to obtain multiple feature points;

[0179] Multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0180] In one embodiment, when the processor executes the computer program, it further performs the following steps: performing color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; performing multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; and performing color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain each inspection image after image enhancement processing.

[0181] In one embodiment, when the processor executes the computer program, it further implements the following steps: inputting each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and the cross-attention information.

[0182] In one embodiment, when the processor executes the computer program, it further implements the following steps: acquiring the camera matrix and distortion coefficients of the inspection device, as well as the spatial association information of feature points in each frame feature matching pair, and constructing a pose calculation model based on the camera matrix, distortion coefficients, and spatial association information; determining rotation parameters and translation parameters based on the pose calculation model, and determining the pose information of the inspection device according to the rotation parameters and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection device, and the translation parameters are used to characterize the spatial position of the inspection device.

[0183] In one embodiment, when the processor executes the computer program, it also performs the following steps: constructing a reprojection function based on the camera matrix and distortion coefficients; determining the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information; and constructing a pose calculation model based on the reprojection function and the depth data.

[0184] In one embodiment, the feature extraction network is a SuperPoint network that is modified based on the GhostNet network.

[0185] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0186] Multiple inspection images of the transmission line collected by the inspection equipment are acquired, and image enhancement processing is performed on each inspection image to obtain the image enhancement processed inspection images.

[0187] Using a pre-configured feature extraction network, features are extracted from each inspection image after image enhancement to obtain multiple feature points;

[0188] Multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0189] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: performing color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; performing multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; and performing color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain each inspection image after image enhancement processing.

[0190] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: inputting each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and the cross-attention information.

[0191] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: acquiring the camera matrix and distortion coefficients of the inspection device, as well as the spatial association information of feature points in each inter-frame feature matching pair, and constructing a pose calculation model based on the camera matrix, distortion coefficients, and spatial association information; determining rotation parameters and translation parameters based on the pose calculation model, and determining the pose information of the inspection device based on the rotation parameters and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection device, and the translation parameters are used to characterize the spatial position of the inspection device.

[0192] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: constructing a reprojection function based on the camera matrix and distortion coefficients; determining the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information; and constructing a pose calculation model based on the reprojection function and the depth data.

[0193] In one embodiment, the feature extraction network is a SuperPoint network that is modified based on the GhostNet network.

[0194] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0195] Multiple inspection images of the transmission line collected by the inspection equipment are acquired, and image enhancement processing is performed on each inspection image to obtain the image enhancement processed inspection images.

[0196] Using a pre-configured feature extraction network, features are extracted from each inspection image after image enhancement to obtain multiple feature points;

[0197] Multiple inter-frame feature matching pairs are determined based on each feature point, and the pose information of the inspection equipment is determined based on the multiple inter-frame feature matching pairs; each inter-frame feature matching pair includes two feature points, and the two feature points belong to different inspection images.

[0198] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: performing color space conversion processing on each inspection image to obtain the luminance component, hue component, and saturation component of each inspection image in the HSV color space; performing multi-scale fast guided filtering denoising processing and bilinear interpolation processing on the luminance component to obtain the processed luminance component; and performing color space conversion processing based on the processed luminance component, hue component, and saturation component to obtain each inspection image after image enhancement processing.

[0199] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: inputting each feature point into a pre-configured feature matching network to obtain multiple inter-frame feature matching pairs; wherein, the feature matching network includes a self-attention unit, a cross-attention unit, and a dynamic adjustment module; the self-attention unit is used to determine attention information based on each feature point, the cross-attention unit is used to determine cross-attention information based on each feature point, and the dynamic adjustment module is used to determine multiple inter-frame feature matching pairs based on the attention information and the cross-attention information.

[0200] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: acquiring the camera matrix and distortion coefficients of the inspection device, as well as the spatial association information of feature points in each inter-frame feature matching pair, and constructing a pose calculation model based on the camera matrix, distortion coefficients, and spatial association information; determining rotation parameters and translation parameters based on the pose calculation model, and determining the pose information of the inspection device based on the rotation parameters and translation parameters; the rotation parameters are used to characterize the spatial attitude of the inspection device, and the translation parameters are used to characterize the spatial position of the inspection device.

[0201] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: constructing a reprojection function based on the camera matrix and distortion coefficients; determining the depth data of feature points in each inter-frame feature matching pair based on spatial correlation information; and constructing a pose calculation model based on the reprojection function and the depth data.

[0202] In one embodiment, the feature extraction network is a SuperPoint network that is modified based on the GhostNet network.

[0203] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0204] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0205] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of determining pose information, characterized by, The method comprises: acquiring a plurality of inspection images of a power transmission line collected by an inspection device, and performing image enhancement processing on each of the inspection images to obtain each of the inspection images after image enhancement processing; using a pre-configured feature extraction network to perform feature extraction on each of the inspection images after image enhancement processing to obtain a plurality of feature points; determining a plurality of inter-frame feature matching pairs based on each of the feature points, and determining pose information of the inspection device according to the plurality of inter-frame feature matching pairs; each of the inter-frame feature matching pairs comprises two feature points, and the two feature points belong to different inspection images.

2. The method of claim 1, wherein, The image enhancement processing on each of the inspection images to obtain each of the inspection images after image enhancement processing comprises: performing color space conversion processing on each of the inspection images to obtain a luminance component, a hue component and a saturation component of each of the inspection images in an HSV color space; performing multi-scale fast guided filter denoising processing and bilinear interpolation processing on the luminance component to obtain a processed luminance component; performing color space conversion processing based on the processed luminance component, the hue component and the saturation component to obtain each of the inspection images after image enhancement processing.

3. The method of claim 1, wherein, The determination of a plurality of inter-frame feature matching pairs based on each of the feature points comprises: inputting each of the feature points into a pre-configured feature matching network to obtain a plurality of inter-frame feature matching pairs; wherein the feature matching network comprises a self-attention unit, a cross-attention unit and a dynamic adjustment module; the self-attention unit is used to determine attention information according to each of the feature points, the cross-attention unit is used to determine cross-attention information according to each of the feature points, and the dynamic adjustment module is used to determine the plurality of inter-frame feature matching pairs based on the attention information and the cross-attention information.

4. The method according to any one of claims 1 to 3, characterized in that, The determination of the pose information of the inspection device according to the plurality of inter-frame feature matching pairs comprises: acquiring a camera matrix and a distortion coefficient of the inspection device, and spatial correlation information of feature points in each of the inter-frame feature matching pairs, and constructing a pose calculation model based on the camera matrix, the distortion coefficient and the spatial correlation information; determining a rotation parameter and a translation parameter based on the pose calculation model, and determining the pose information of the inspection device according to the rotation parameter and the translation parameter; the rotation parameter is used to represent the spatial attitude of the inspection device, and the translation parameter is used to represent the spatial position of the inspection device.

5. The method of claim 4, wherein, The construction of the pose calculation model based on the camera matrix, the distortion coefficient and the spatial correlation information comprises: constructing a re-projection function based on the camera matrix and the distortion coefficient; determining depth data of feature points in each of the inter-frame feature matching pairs based on the spatial correlation information; constructing the pose calculation model based on the re-projection function and the depth data.

6. The method of claim 1, wherein, The feature extraction network is a SuperPoint network adjusted based on a GhostNet network.

7. A pose information determination apparatus characterized by comprising: The device comprises: An acquisition module is configured to acquire a plurality of inspection images collected by an inspection device of a power transmission line, and perform image enhancement processing on each of the inspection images to obtain each of the inspection images after image enhancement processing. An execution module is configured to perform feature extraction on each of the inspection images after image enhancement processing by using a pre-configured feature extraction network to obtain a plurality of feature points. A determination module is configured to determine a plurality of inter-frame feature matching pairs based on the feature points, and determine pose information of the inspection device according to the plurality of inter-frame feature matching pairs. Each of the inter-frame feature matching pairs includes two feature points, and the two feature points belong to different inspection images.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 6.

Images