CN-121999151-A - Power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology

CN121999151ACN 121999151 ACN121999151 ACN 121999151ACN-121999151-A

Abstract

The invention relates to a power transmission corridor parameterization reconstruction method integrating nonlinear geometric prior and semantic topology, which comprises the steps of obtaining point cloud data, image data and IMU inertial data of a power transmission corridor, carrying out self-adaptive enhancement to obtain high dynamic range images and time-synchronous point cloud data, carrying out semantic segmentation by adopting a Wire-Aware BiSeNetV network based on the high dynamic range images to obtain pixel-level semantic segmentation results, dividing the time-synchronous point cloud data into multiple types of semantic tags, carrying out processing by adopting a semantic-guided decoupling state estimation method based on the point cloud data with the semantic tags to obtain six-degree-of-freedom pose of an unmanned aerial vehicle, carrying out fitting by adopting a catenary constraint parameterization map building method based on the six-degree-of-freedom pose of the unmanned aerial vehicle, and carrying out drift correction to obtain parameters of a power transmission corridor Wire. Compared with the prior art, the invention has the advantages of high precision, high robustness and the like.

Inventors

WU CHUNYAN
HUANG YUNWEI
YANG YIZHONG
FU YONG
WEI HAO
ZHU WENHUAN

Assignees

合肥工业大学
上海交通大学

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (10)

1. The utility model provides a transmission corridor parameterization reconstruction method integrating nonlinear geometric prior and semantic topology, which is characterized by comprising the following steps: acquiring point cloud data, image data and IMU inertial data of a power transmission corridor, and performing self-adaptive enhancement to obtain high dynamic range images and time-synchronous point cloud data; based on the high dynamic range image, performing semantic segmentation by adopting a Wire-Aware BiSeNetV network to obtain a pixel-level semantic segmentation result, and dividing the time-synchronized point cloud data into multiple types of semantic tags; Processing by adopting a semantic guided decoupling state estimation method based on point cloud data with semantic tags to obtain six-degree-of-freedom pose of the unmanned aerial vehicle; Fitting by adopting a catenary constraint parameterization mapping method based on the six-degree-of-freedom pose of the unmanned aerial vehicle, and carrying out drift correction to obtain parameters of the transmission corridor conductors.
2. The method for power transmission corridor parameterized reconstruction of fusion of nonlinear geometric prior and semantic topology according to claim 1, wherein the step of adaptively enhancing comprises: Combining the IMU inertial data to obtain a point cloud of the previous frame Conversion from radar coordinate system to current camera coordinate system Obtaining a three-dimensional coordinate under the current camera coordinate, which is expressed as: , in the formula, Is at present Three-dimensional coordinates of moment three-dimensional point cloud under current camera coordinate system , Is used as an external reference matrix, and the external reference matrix is used as a reference matrix, The pose at the current moment is obtained by predicting inertial data of the IMU, For the pose at the moment of the previous frame, Is an extrinsic matrix; based on the three-dimensional coordinates, utilizing a pinhole model to cloud the last frame of points Projecting to a two-dimensional image plane to obtain projected pixel coordinates, and forming a sparse mask map consistent with the resolution of the image data Wherein, the operation expression of the projected pixel coordinates is: , in the formula, For the projected pixel coordinates, As a function of the projection, Is a camera internal reference matrix; the sparse mask pattern According to the reflectivity intensity of the projection point Determining a valid statistical domain ROI: And calculate the weighted average reflectivity of the effective statistical domain ROI Wherein the effective statistical field represents a tower region; calculating the optimal exposure time of the next frame by adopting a reflectivity-exposure feedforward control loop The computational expression is: , in the formula, For the optimal exposure time for the next frame, For the exposure time being used for the current frame, For a preset target brightness value, For only sparse mask image in previous frame image The average gray level in the interior of the frame, / The LiDAR for the previous and current frames is weighted to average the reflectivity, Is a sensitivity coefficient; optimal exposure time based on the next frame And acquiring data to obtain the point cloud data of the high dynamic range image and the time synchronization.
3. The power transmission corridor parameterization reconstruction method integrating nonlinear geometric prior and semantic topology according to claim 1, wherein the semantic tags are respectively rigid tower points, semi-rigid power line points and invalid background points.
4. The method for parameterized reconstruction of a power transmission corridor with fusion of nonlinear geometric prior and semantic topology according to claim 3, wherein the step of obtaining the six degrees of freedom pose of the unmanned aerial vehicle comprises the following steps: 1) Constructing a corresponding observation model according to the point cloud data with the semantic tags and different semantic tags; 2) For the observation model, performing iterative loop in a decoupled error state iterative Kalman filtering frame, and calculating Kalman gain Wherein the Kalman gain The calculated expression of (2) is: , Wherein: , in the formula, For the a priori error state covariance matrix, In order to measure the jacobian matrix, In the form of a diagonal matrix, As a function of the variance of the values, Is the first A semantic tag is planted; 3) Based on the Kalman gain Fusing the observation residual errors calculated by the observation model, calculating the error state update quantity, and correcting the state of the unmanned aerial vehicle; 4) Judging whether the iteration termination condition is met, if so, the corrected unmanned aerial vehicle state is the six-degree-of-freedom pose of the unmanned aerial vehicle, and if not, returning to the step 1) to update the observation model for iterative computation until the termination condition is met.
5. The method for parameterized reconstruction of a power transmission corridor with fusion of nonlinear geometric prior and semantic topology according to claim 4, wherein the observation model comprises two parts, namely a residual model and a variance function, For a rigid tower point, its residual model is a point-plane residual model, expressed as: , in the formula, For the point-plane residual model, the point-to-plane distance residual is represented, Is a plane normal vector which is a plane normal vector, As the point of the current observation, Is the corresponding point in the map; The variance function is expressed as: , in the formula, In order to observe the variance of the image, Is the fundamental physical ranging noise of the LiDAR sensor, Is the first The semantic tags are seeded with the seed semantic tags, Is a rigid tower point; For a semi-rigid power line point, its residual model is a dotted line residual model, expressed as: , in the formula, Is a dotted line residual model, representing a vertical distance residual from a point to a straight line, Is a matrix of units which is a matrix of units, Is a unit direction vector of a straight line, For the currently measured power line point cloud coordinates, To correspond to any point on a straight line, Is a projection matrix; The variance function is expressed as: , in the formula, In order to observe the variance of the image, Is the fundamental physical ranging noise of the LiDAR sensor, In order for the coefficient of expansion to be the same, For the wind speed of the wind, As a function of the wind speed, Is a semi-rigid power line point; for invalid background points, the variance function is set to be To thoroughly eliminate invalid background points, wherein Is an invalid background point.
6. The method for power transmission corridor parameterized reconstruction of fused nonlinear geometric prior and semantic topology of claim 5, further comprising, for semi-rigid power line points, introducing a robust kernel function to re-weight the variance function, wherein the re-weighted residual function is expressed as: , in the formula, As a function of the residual after the weight is applied, As the value of the original residual value, Is a threshold parameter.
7. A method of power transmission corridor parameterization reconstruction incorporating nonlinear geometric priors and semantic topologies according to claim 1, wherein the step of obtaining parameters of the power transmission corridor conductors comprises: setting a sliding window Wherein For the length of the sliding window, The power line point cloud data frame is the t moment; Using a priori pose at the current time Uniformly converting all the power line point clouds in the sliding window into a world coordinate system to obtain: , in the formula, For conversion into the ith powerline point cloud in the world coordinate system in the sliding window, Converting the ith power line point cloud before conversion in a sliding window; constructing an objective function based on nonlinear least squares based on all power line point clouds to a world coordinate system to minimize the sum of orthogonal distances from all observation points to a catenary model, wherein the expression of the objective function is as follows: , Wherein: , , , in the formula, For an optimal catenary parameter vector, As a state variable to be optimized, Is a three-dimensional coordinate of the lowest point of the catenary in the world coordinate system, Is the azimuth angle of the vertical plane in which the catenary lies relative to the X-axis of the world coordinate system, As a shape parameter of the catenary, Is a horizontal tension force, and is a horizontal tension force, Is the gravity force per unit length of the device, For the total number of power line point clouds contained within the sliding window, As a function of the residual error, For the perpendicular distance of a point to a two-dimensional vertical plane, For the altitude residual between the observed altitude and the theoretical catenary model at the current unmanned pose in ESIKF updates, The horizontal projection distance of the point in the vertical plane where the lead is located; wherein, the expression of the catenary model is: , in the formula, In the horizontal coordinate The theoretical calculated height of the wire is that, Is a hyperbolic cosine function; in the updating of the error state iterative Kalman filtering framework, the nonlinear least square-based objective function is solved in an iterative mode to obtain an optimal catenary parameter vector ; Wherein, still include the following: based on the height residual The jacobian matrix of its position in the line direction relative to the drone is calculated, expressed as: , introducing ESIKF the jacobian matrix as an observation constraint into an update process: , in the formula, In the form of a jacobian matrix, As a residual of the in-plane height, Is the position of the unmanned plane in the line direction, To correct the amount of position correction in the direction of the power line, In order to correct the gain of the gain-adjusting device, For the actual altitude of the unmanned plane or the observation point in the world coordinate system at the current moment, To be according to the fitted optimal catenary parameters 。
8. The power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology according to claim 1, further comprising the steps of closed loop detection and global optimization based on topology fingerprints, and specifically comprising the following steps: Extracting an insulator central axis from the insulator point cloud by utilizing a RANSAC algorithm and combining surface normal direction constraint of the insulator point cloud point as an insulator framework; Projecting the insulator point cloud to an insulator central axis coordinate system, and calculating a radius distribution function along the insulator central axis Expressed as: , in the formula, The projection coordinates of the point cloud point in a plane perpendicular to the central axis; For the radius distribution function Performing fast Fourier transform to obtain frequency domain fingerprint including main frequency Phase of Amplitude spectrum, wherein the dominant frequency Corresponding to the disc spacing of the insulator Phase, phase Corresponding to the axial offset of the insulator string relative to the starting point, the amplitude spectrum is determined by the specific shape of the disc and is used as the unique fingerprint of the insulator string; constructing an insulator topological graph, wherein an insulator framework is used as a node, and the node also comprises the frequency domain fingerprint and a connection relationship is used as an edge; When the unmanned aerial vehicle flies through an area, recording the current position and an insulator skeleton sequence of a series of insulators, and when the unmanned aerial vehicle flies to a new position, taking the current skeleton sequence as a query sequence; comparing the query sequence with all historical insulator skeleton sequences in the insulator topological graph by using a Needleman-Wunsch sequence comparison algorithm, if the comparison similarity score of a certain historical insulator skeleton sequence and the query sequence exceeds a threshold value, judging that loop-back occurs, performing global optimization through geometric correction, and completing a closed loop detection process, wherein the geometric correction is performed through the following formula: , in the formula, For accurate axial displacement after phase lock correction, For the coarse displacement estimated by the front-end odometer, In order to round up the rounding function, For the physical spacing between the insulator disks, For the current observed phase angle extracted by the FFT, To convert the phase angle to a scaling factor for the physical distance, Is a displacement in a period.
9. The power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology according to claim 1, wherein a lightweight coordinate attention mechanism is embedded in the tail end of a standard detail branch by the Wire-Aware BiSeNetV network, and feature maps obtained in the tail end of the standard detail branch are respectively subjected to global pooling along an X/Y axis.
10. The power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology according to claim 1, wherein the Wire-Aware BiSeNetV network is trained by adopting a boundary-category weighted mixed loss function, and the expression of the boundary-category weighted mixed loss function is as follows: , in the formula, The mixing loss function is weighted for the boundary-class, In order to mine the loss of the online difficult cases, In order for the coefficient of balance to be present, Is boundary loss; Wherein: , , in the formula, As a total number of difficult-to-handle pixels, For the filtered set of difficult-to-sample pixels, Is a pixel Belongs to the category of Is a real tag of the (c) in the (c), Predicting pixels for a network Belongs to the category of Is a function of the probability of (1), For the predicted edge probability map of the network, Is a true edge label graph.

Description

Power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology Technical Field The invention relates to the technical field of parameter reconstruction, in particular to a power transmission corridor parameterization reconstruction method integrating nonlinear geometric prior and semantic topology. Background With the development of unmanned aerial vehicle technology, autonomous navigation and environment reconstruction technology based on multi-sensor fusion (LiDAR, vision and IMU) is widely applied to power inspection of a power transmission corridor. However, the power transmission corridor belongs to a typical unstructured, high-dynamic and geometric degradation environment, and the existing SLAM (synchronous positioning and mapping) technology still faces serious challenges in practical application. First, in visual perception, power inspection generally uses a look-up view, and there is an extreme dynamic range contradiction (strong background sky and weak foreground backlight towers) in the field of view. The traditional automatic exposure algorithm is based on full graph statistics, is easy to be misled by a highlight sky to reduce exposure, so that a key tower dark area falls into a sensor dead zone, texture features are lost, and further visual mileage technology unlocking is caused. Second, in terms of geometric modeling, existing SLAM systems are generally based on "environmental rigid body assumptions". However, the transmission line belongs to a semi-rigid object, and is influenced by wind power to generate galloping, so that rigid constraint is broken. At the same time, the elongated wire geometrically has translational invariance in the axial direction, resulting in an insignificant state along the line direction (geometrical degradation), and is very prone to an unbounded cumulative drift (corridor effect). Finally, in closed loop detection, insulator strings and towers with highly repeated profiles are widely present in the transmission corridor. The repeated textures can cause serious ' perceived aliasing ' (Perceptual Aliasing) ', so that a loop detection algorithm based on a traditional word bag model is extremely easy to generate mismatching, the back end optimization is failed, and a globally consistent high-precision map is difficult to construct. Therefore, there is a need for a parameterized reconstruction method for a power transmission corridor that can accommodate a strong backlight environment, solve the problems of non-rigid body and geometric degradation, and effectively resist aliasing. Disclosure of Invention The invention aims to provide a power transmission corridor parameterized reconstruction method for improving parameter accuracy and fusing nonlinear geometric prior and semantic topology. The aim of the invention can be achieved by the following technical scheme: A power transmission corridor parameterized reconstruction method integrating nonlinear geometric prior and semantic topology comprises the following steps: acquiring point cloud data, image data and IMU inertial data of a power transmission corridor, and performing self-adaptive enhancement to obtain high dynamic range images and time-synchronous point cloud data; based on the high dynamic range image, performing semantic segmentation by adopting a Wire-Aware BiSeNetV network to obtain a pixel-level semantic segmentation result, and dividing the time-synchronized point cloud data into multiple types of semantic tags; Processing by adopting a semantic guided decoupling state estimation method based on point cloud data with semantic tags to obtain six-degree-of-freedom pose of the unmanned aerial vehicle; Fitting by adopting a catenary constraint parameterization mapping method based on the six-degree-of-freedom pose of the unmanned aerial vehicle, and carrying out drift correction to obtain parameters of the transmission corridor conductors. Further, the step of adaptively enhancing includes: Combining the IMU inertial data to obtain a point cloud of the previous frame Conversion from radar coordinate system to current camera coordinate systemObtaining a three-dimensional coordinate under the current camera coordinate, which is expressed as: , in the formula, Is at presentThree-dimensional coordinates of moment three-dimensional point cloud under current camera coordinate system,Is used as an external reference matrix, and the external reference matrix is used as a reference matrix,The pose at the current moment is obtained by predicting inertial data of the IMU,For the pose at the moment of the previous frame,Is an extrinsic matrix; based on the three-dimensional coordinates, utilizing a pinhole model to cloud the last frame of points Projecting to a two-dimensional image plane to obtain projected pixel coordinates, and forming a sparse mask map consistent with the resolution of the image dataWherein, the operation expression of the projected pixel