Three-dimensional modeling of power transmission line based on octocopter and laser radar and flight control method

Abstract

The application relates to a power transmission line three-dimensional modeling and flight control method based on an eight-rotor unmanned aerial vehicle and a laser radar, and comprises the following steps: constructing an eight-rotor coaxial contra-rotating unmanned aerial vehicle hardware platform integrated with a three-dimensional laser radar, a GNSS sensor and a wind channel heat dissipation system; realizing unmanned aerial vehicle trajectory optimization in a wind-resistant environment through model predictive control; realizing loop detection, GNSS assistance and point cloud processing by combining laser radar point cloud collection and Fast-LIO2_SLAM, and constructing a power transmission line three-dimensional model. Compared with the prior art, the application has the advantages of improving the efficiency, accuracy and reliability of the unmanned aerial vehicle in power transmission inspection, ensuring the robustness and endurance capacity under complex tasks and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles (UAVs) and power transmission lines, and in particular to a three-dimensional modeling and flight control method for power transmission lines based on an octorotor UAV and a laser radar. Background Art

[0002] With the advancement of smart grid construction, the demand for intelligent inspection and 3D modeling of transmission lines is becoming increasingly urgent. Drones equipped with LiDAR have become a research hotspot in this field due to their efficient data collection capabilities and contactless operation. Among current technologies, octorotor coaxial counter-propeller drones, due to their high lift density and redundant design, are ideal carriers for heavy sensors (such as 3D LiDAR and GNSS). The application of SLAM algorithms such as Fast-LIO2 provides the foundational framework for high-precision 3D modeling.

[0003] Chinese patent application publication number CN111427054A discloses a precise distance measurement system for power transmission and distribution line corridor hazards. By equipping a multi-rotor drone with a lidar and enabling real-time communication and data transmission, combined with RTK positioning technology, this system addresses the low efficiency of lidar control and data transmission in existing technologies. This improves the accuracy and efficiency of distance measurement for power transmission and distribution line corridors, enabling efficient 3D modeling and tree obstacle analysis. However, this application fails to address hardware performance bottlenecks, algorithm efficiency, and environmental adaptability issues.

[0004] In summary, despite the progress made in existing technologies, the following core challenges still exist in transmission line scenarios:

[0005] (1) Hardware performance bottleneck: Transmission line inspections require long hours of work. Sensors such as lidar are prone to heat accumulation due to power consumption. Existing heat dissipation solutions rely on additional equipment, increasing system complexity and weight. Transmission lines are usually installed at high altitudes. Traditional multi-rotor drones have limited lift reserves and are unable to withstand high-altitude wind disturbances, resulting in poor flight stability and affected data collection accuracy.

[0006] (2) Insufficient algorithm efficiency and environmental adaptability: Due to the slender features of conductors and insulators in transmission lines, existing algorithms have difficulty effectively extracting small-scale structural details. Repeated structures such as towers can easily cause SLAM loop detection confusion, resulting in global map drift. LiDAR may experience a surge in point cloud noise or detector saturation under strong sunlight. Traditional filtering algorithms are not robust enough to complex lighting conditions, affecting modeling reliability. Summary of the Invention

[0007] The purpose of the present application is to overcome the defects of the prior art, and provide an eight-rotor unmanned aerial vehicle and laser radar-based power transmission line three-dimensional modeling and flight control method, to solve or partially solve the problems of hardware performance limitations, algorithm robustness deficiencies, safety and efficiency contradictions in power transmission line three-dimensional modeling, and through the deep integration of hardware platform innovation and algorithm process optimization, to build an efficient and safe power transmission line three-dimensional modeling and flight control system.

[0008] The purpose of the present application can be achieved by the following technical solutions:

[0009] In one aspect of the present application, an eight-rotor unmanned aerial vehicle and laser radar-based power transmission line three-dimensional modeling and flight control method is provided, comprising the following steps:

[0010] An eight-rotor coaxial counter-rotor unmanned aerial vehicle hardware platform is constructed, which integrates a three-dimensional laser radar, a GNSS sensor and a wind channel heat dissipation system;

[0011] Unmanned aerial vehicle trajectory optimization in a wind-resistant environment is achieved through model predictive control;

[0012] Through the combination of laser radar point cloud collection and Fast-LIO2_SLAM, loop detection, GNSS assistance and point cloud processing are achieved to construct a power transmission line three-dimensional model.

[0013] As a preferred technical solution, the process of achieving unmanned aerial vehicle trajectory optimization in a wind-resistant environment through model predictive control comprises the following steps:

[0014] Based on computational fluid dynamics (CFD), the coaxial rotor aerodynamic design calculates the lift bias by solving the Navier-Stokes problem, reducing the thrust loss between the upper and lower rotors;

[0015] Through control moment distribution, the motor thrust of the eight-rotor unmanned aerial vehicle is dynamically distributed to achieve anti-gust disturbance and motor fault redundancy control;

[0016] The flight path is optimized through an evolutionary algorithm to reduce the energy consumption and time of the inspection.

[0017] As a preferred technical solution, the wind channel heat dissipation system passively cools the laser radar through the airflow of the unmanned aerial vehicle, maintaining the normal operating temperature of the sensor during long-term operation without additional power supply equipment.

[0018] As a preferred technical solution, after collecting the laser radar point cloud, the following laser radar anti-strong light interference processing process is included:

[0019] Based on statistical filtering and intensity threshold, the original laser radar point cloud is processed to remove noise;

[0020] The UDP algorithm is used to achieve feature extraction through progressive self-attention and bird's-eye view, and robust registration of point clouds under strong light is achieved.

[0021] As a preferred technical solution, the model predictive control is combined with the wind field model to predict the dynamic behavior of the UAV in trajectory planning, and achieve smooth tracking and wind-resistant stable flight in the transmission line corridor by optimizing the control input.

[0022] As a preferred technical solution, the process of loop detection, GNSS assistance and point cloud processing based on Fast-LIO2_SLAM includes the following steps:

[0023] Loop detection is performed based on the density map ORB feature, and anti-aliasing perception in repetitive structure scenes is achieved through 2D bird's-eye view projection and self-similarity pruning;

[0024] Through the tight coupling of GNSS, LiDAR and inertial, the extended Kalman filter is used to achieve global positioning constraints and suppress long-distance drift.

[0025] As a preferred technical solution, in the process of constructing a 3D model of a transmission line, point cloud processing of the 3D model of the transmission line includes the following steps:

[0026] Reconstruct the sag shape of transmission lines based on minimum spanning tree clustering and catenary fitting of iterative K-means;

[0027] The repeated structure of the tower is identified through deep learning semantic segmentation, and high-precision 3D reconstruction is achieved by combining the GroundSLAM featureless matching algorithm.

[0028] As an optimal technical solution, during the state estimation process, the fusion weights of LiDAR, IMU, and GNSS are dynamically adjusted based on real-time indicators such as GNSS signal strength, lidar point cloud density, and IMU noise.

[0029] Another aspect of the present invention provides an electronic device comprising: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the aforementioned three-dimensional modeling and flight control method of power transmission lines based on an octocopter drone and a lidar.

[0030] Another aspect of the present invention provides a computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the aforementioned three-dimensional modeling and flight control method of power transmission lines based on an octocopter drone and a lidar.

[0031] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0032] (1) Hardware performance improvement: On the one hand, the present invention improves the lift density of the UAV through the coaxial counter-propeller design, which can carry a larger sensor payload and meet the requirements of heavy-duty laser radar and cooling systems. On the other hand, the air duct heat dissipation solution can reduce the operating temperature of the sensor, extend the continuous operation time, and improve the algorithm accuracy.

[0033] (2) Improved modeling and flight control efficiency: The present invention effectively reduces the noise rate and catenary modeling error through anti-glare point cloud processing, and the tower reconstruction detail resolution reaches the centimeter level. By adopting an improved SLAM framework to reduce the global drift rate, it is significantly better than the traditional loosely coupled solution.

[0034] (3) Significant engineering application value: The present invention uses wind-resistant control design to enable UAVs to operate stably in strong wind environments and adapt to complex meteorological conditions. In addition, it improves the efficiency of full-process automated modeling, reduces manual post-processing costs, and provides high-precision three-dimensional benchmarks for digital management of transmission lines, such as sag monitoring and tree barrier analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a lidar in an embodiment;

[0036] Figure 2 Schematic diagram of an electronic device in an embodiment. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0038] Example 1

[0039] In order to solve the problems existing in the above-mentioned prior art, this embodiment provides a three-dimensional modeling and flight control method of power transmission lines based on an octorotor drone and a laser radar. Figure 1 The method deeply integrates hardware platform innovation and algorithm process optimization to build an efficient and safe transmission line 3D modeling and flight control system, including:

[0040] (1) Eight-rotor coaxial counter-propeller UAV hardware platform.

[0041] The platform integrates high-lift redundant aerodynamic design, power-saving heat dissipation, anti-glare perception and wind-resistant trajectory optimization design.

[0042] (1) High lift redundant aerodynamic design.

[0043] The platform utilizes an eight-rotor coaxial counter-propeller configuration. Computational fluid dynamics (CFD) simulation optimizes the upper and lower rotor layout, solves the compressible Navier-Stokes equations to analyze aerodynamic interference, and utilizes a lift biasing algorithm to reduce thrust loss, improving the lift-to-drag ratio during high-speed flight, particularly when the ratio of forward speed to blade tip speed is large. Furthermore, a 16-motor redundant system, combined with a control torque distribution algorithm, dynamically adjusts the thrust of each motor, achieving gust resistance and single-motor fault tolerance.

[0044] The eight-rotor coaxial counter-rotating blade design is designed to provide high lift density, enhanced stability, and redundancy, which are crucial for carrying heavy sensors (such as 3D lidar) and complex cooling systems. While the coaxial design improves thrust efficiency and maneuverability, it also introduces complex aerodynamic interference issues.

[0045] To address the issues of insufficient lift efficiency and stability, this application uses computational fluid dynamics (CFD) to analyze the aerodynamic performance of a coaxial rotor at high forward ratios (the ratio of forward flight speed to rotor tip speed). This involves solving the compressible Navier-Stokes equations, employing a Reynolds-averaged turbulence model, and accounting for the laminar / turbulent boundary layer transition. Detailed CFD simulations of the airflow around the rotor and fuselage identify design optimization points to maximize lift and minimize drag. For coaxial systems, CFD helps understand and mitigate the negative interaction between the upper and lower rotors, which can result in thrust losses of up to 25-35%. At high forward ratios (e.g., greater than 0.6), even with lift offset (the degree of rolling moment acting on the rotor thrust), performance can be significantly degraded due to issues on the retreating blade side. Optimizing lift offset through CFD can improve the lift-to-drag ratio and reduce thrust fluctuations at high speeds, which is crucial for rapid inspections of transmission line corridors.

[0046] The platform's rotor layout, optimized fuselage structure, and novel flight mechanism are designed to significantly enhance the performance, versatility, and practicality of drones, overcoming the limitations of traditional designs. This includes optimizing the rotor layout and fuselage shape for superior energy efficiency. The layout and even independent design of each coaxial propeller pair are optimized to minimize interference and maximize total thrust. The fuselage design is streamlined to reduce parasitic drag and meet the endurance requirements of transmission line inspection missions.

[0047] By adopting an evolutionary algorithm for path planning, which is indirectly linked to lift efficiency through mission optimization, optimizing the flight path for transmission line inspections can reduce flight time and energy consumption, thereby indirectly improving the overall mission efficiency, which is particularly critical for battery-powered UAVs.

[0048] (2) No power consumption, heat dissipation and anti-strong light perception design.

[0049] The aircraft's integrated air duct cooling system utilizes airflow to passively dissipate heat from the LiDAR, eliminating the need for additional equipment weight and ensuring the sensor's continued operation in high-temperature environments. Furthermore, the platform features a combined 3D LiDAR and GNSS sensor. The system utilizes a UDP (Unlocking Generalization Power) anti-glare algorithm, combined with progressive self-attention and bird's-eye view feature extraction to mitigate sunlight interference, and statistical filtering and intensity thresholding to remove point cloud noise.

[0050] The drone of this embodiment is provided with an air duct, and the airflow of the drone is used to dissipate heat for the laser radar, so that the radar can operate at a normal temperature without adding any additional equipment.

[0051] The drone is equipped with a 3D LiDAR and GNSS. To address the challenges posed by strong light to the LiDAR, the sensor system in this embodiment is designed to resist strong light (suppress sunlight interference):

[0052] LiDAR-based SLAM is generally robust to lighting, but can still be affected by direct sunlight, while visual SLAM is highly sensitive to lighting changes. Deep learning is revolutionizing point cloud registration by increasing robustness to noise. Some LiDAR models consider direct sunlight as a weather effect. Sunlight, especially at low angles of incidence, can saturate LiDAR detectors or introduce significant noise in the point cloud. Mitigation strategies include:

[0053] 1) Algorithmic filtering: Considering that sunlight may cause abnormally high intensity readings or signal loss, statistical outlier removal, intensity-based filtering, and noise modeling are used.

[0054] 2) UDP (Unlock Generalization Power): This algorithm aims to achieve robust LiDAR point cloud registration under diverse conditions by eliminating cross-attention modules and leveraging progressive self-attention and bird's-eye view (BEV) features to reduce ambiguity. While primarily focused on the generalization of registration, its principles for robust feature extraction under challenging conditions are relevant. Self-supervised methods are also being developed for robust registration of noisy heterogeneous point clouds. This demonstrates that the state-of-the-art involves more than just basic filters, but rather involves robust feature learning and registration processes that can inherently handle noise.

[0055] (3) Optimization design of wind-resistant trajectory.

[0056] The lift and control torque of an octorotor drone are generated by horizontally rotating rotors, whose design principles and flight dynamics have been extensively studied and optimized. The blades are strategically arranged to direct thrust in multiple directions. For an octorotor coaxial drone (with a total of 16 motors), a control distribution algorithm is used to efficiently and robustly distribute the desired forces and torques to the individual motors, particularly in the event of motor failure or to counter specific disturbances such as gusts of wind.

[0057] Considering that transmission line inspection missions often involve flying close to structures under potentially windy conditions and require precise maneuvers. Robust and adaptive flight control is critical to safety and mission success. The platform uses a model predictive control (MPC) algorithm, combined with a real-time wind field model to predict the dynamic behavior of the drone, optimize the control input in the future time domain, and achieve smooth tracking and wind-resistant stable flight in the transmission line corridor. MPC is used for optimized line tracking in autonomous inspection systems. MPC requires an accurate system model and higher computing resources, but has advantages in trajectory planning and obstacle avoidance. For transmission line tracking, MPC optimizes the control input by predicting system behavior, optimizes the smoothness and accuracy of the trajectory in the future time domain, and combines the wind model to consider the impact of wind.

[0058] (2) Transmission line three-dimensional modeling algorithm system.

[0059] The software part mainly includes the enhanced Fast-LIO2_SLAM framework, refined point cloud processing and structure reconstruction, and adaptive multi-sensor fusion design.

[0060] (1) Enhanced Fast-LIO2_SLAM framework.

[0061] The framework includes front-end odometry, back-end optimization, and multi-source fusion. For front-end odometry, the raw LiDAR point cloud is directly used, and an incremental KD tree is used to maintain a global map to avoid feature extraction bias in sparse environments. On the back-end, loop detection based on density map ORB features is introduced to project the 3D point cloud into 2D bird's-eye view density features. A self-similarity pruning algorithm is used to address perceptual confusion caused by the repetitive structure of the tower. For data fusion, this method utilizes tightly coupled GNSS-LiDAR-inertial fusion, fusing raw GNSS pseudorange observations via an extended Kalman filter to suppress long-range drift.

[0062] FAST-LIO2, a robust and computationally efficient baseline, serves as the front-end odometry system. It directly uses raw lidar points and employs an incremental KD tree (iK-D tree) for global map data maintenance. Its direct approach (point-to-plane matching on the raw points) avoids feature extraction issues in sparse environments. The iK-D tree supports efficient map updates.

[0063] Loop closure detection: Loop closure detection is crucial to address accumulated drift in Fast-LIO2, particularly in large-scale outdoor environments such as transmission line corridors, which can be long and contain repetitive structures such as towers and cable spans. This method uses density-map-based ORB features and follows the approach of Gupta et al. for loop closure detection, applying the ORB descriptor on a 2D density projection (bird's-eye view) of the local map. This method is robust, requires no parameter tuning, and can handle a wide range of LiDAR and non-planar motion. Self-similarity pruning addresses perceptual confusion in repetitive environments. This method leverages the speed and robustness of ORB features by transforming the 3D problem into a 2D image matching problem. Density projection provides some insensitivity to exact point locations, while the bird's-eye view (BEV) helps handle viewpoint variations.

[0064] (2) Refined point cloud processing and structural reconstruction design.

[0065] Deep learning noise reduction networks, such as the UGP framework, and self-supervised registration algorithms remove strong light noise, improve point cloud registration accuracy, and enhance point cloud anti-interference capabilities. Minimum spanning tree (MST) clustering and an iterative K-means catenary fitting algorithm accurately reconstruct conductor sag and minimize sag errors. WireNetV2_CNN semantic segmentation is used to identify repeating tower units, combined with GroundSLAM's featureless matching algorithm to address the challenge of self-occlusion reconstruction in lattice structures.

[0066] Based on the characteristics of transmission line catenaries, a method for constructing line features simulating each catenary from point clouds is proposed. This method employs minimum spanning tree (MST) clustering, dynamic programming, and iterative K-means clustering combined with catenary fitting. PL2DM models vertical power lines as catenaries. Power lines sag in a catenary-like manner due to their own weight. Accurately fitting this model is crucial for determining sag, clearance, and other parameters.

[0067] To address the problem of easy misidentification of tower repeats, existing standard photogrammetry processing fails on complex grid-like structures (e.g. radio towers, similar to power transmission towers) because of similar elements with repetitions and self-occlusions. This embodiment uses deep learning, such as WireNetV2 CNN for semantic segmentation. GroundSLAM uses featureless image-level matching for SLAM on surfaces with repetitive patterns. Power transmission towers are typically lattice structures with many repeating units, which pose challenges for both SLAM (loop closure detection) and reconstruction (feature matching). Semantic segmentation provides an understanding of which part of the tower a point belongs to, guiding the reconstruction. Robust SLAM that can handle repetitions is also key.

[0068] (3) Adaptive multi-sensor fusion.

[0069] Develop LSAF-LSTM dynamic weight algorithm, according to GNSS signal strength, laser radar point cloud density, IMU noise and other real-time indexes, through long short-term memory network dynamic adjustment of sensor fusion weight, automatically improve LiDAR weight in GNSS signal weak area, ensure the robustness of state estimation in complex environment.

[0070] Use LSTM to dynamically adjust the fusion weights of LiDAR, IMU and GNSS according to real-time reliability indicators such as GNSS signal strength, visual feature density, LiDAR point cloud density and IMU noise, to achieve adaptive sensor fusion. By learning time dependence and using attention mechanism, to prioritize more reliable sensors in challenging conditions, for example, reduce visual weight when light is insufficient, or reduce GNSS weight when GNSS signal is weak, directly meeting the demand for outdoor environment robustness.

[0071] 1. Navigation and precise positioning in complex scenarios.

[0072] 1.1 Multi-sensor fusion for UAV state estimation

[0073] 1) The UAV body integrates IMU, GNSS, vision and LiDAR sensors, and uses the filtering of extended Kalman filter (EFK) to realize the UAV pose estimation.

[0074] 2) EFK effectively improves tracking accuracy relative to unprocessed optical sensor data. This is crucial for the accuracy of the position of the robotic arm end effector. Sensor fusion improves the robustness of the system to single sensor failure or noise.

[0075] 1.1.1 Inertial Measurement Unit (IMU).

[0076] 1. Working principle and data output:

[0077] The IMU consists of a three-axis accelerometer and a three-axis gyroscope. The accelerometer measures specific force (linear acceleration minus gravity, projected onto the aircraft's coordinate system), while the gyroscope measures angular velocity. The IMU output is raw sensor data representing these physical quantities.

[0078] 2. Error characteristics.

[0079] The IMU's measurement data is affected by a variety of error sources, including:

[0080] 1) Bias Instability: The sensor bias changes slowly over time.

[0081] 2) Initial Bias: The error in sensor sensitivity.

[0082] 3) Scale Factor Errors: Sensor sensitivity error.

[0083] 4) Noise: Random fluctuations in the output signal, usually characterized by noise density and random walk.

[0084] 5) Cross-axis sensitivity and misalignment: Errors caused by sensor manufacturing defects.

[0085] These errors, especially bias and noise, will lead to unbounded error accumulation, namely drift, when IMU data is integrated for a long time to estimate velocity, position and pose. Strict constraints on drone size, weight, power consumption and cost.

[0086] 2. Pose estimation based on EKF.

[0087] The IMU's ability to provide high-frequency motion data makes it an ideal choice for driving the EKF's prediction step. It allows for continuous state propagation between lower-frequency updates from other sensors. Despite the IMU's inherent error accumulation issues, its continuous, high-frequency output makes it an indispensable component of the EKF. It provides relative motion information between updates from external sensors, such as GNSS and vision sensors, whose measurements are then used to correct for the IMU's accumulated drift.

[0088] 3. The importance of body coordinate system representation and IMU calibration.

[0089] For highly dynamic UAV maneuvers, converting raw IMU data to a global coordinate system can impair the observability of critical kinematic information. The typical nonlinear and highly dynamic flight patterns of UAVs make traditional IMU integration in a global coordinate system ineffective. This coordinate transformation, especially when attitude information is inaccurate, can lose or obscure key details of the UAV's motion, particularly attitude changes during intense maneuvers.

[0090] 1.1.2 Global Navigation Satellite System (GNSS).

[0091] 1. Working principle and data output.

[0092] A GNSS receiver calculates its position by triangulating signals from multiple orbiting satellites. Each satellite broadcasts its orbital position and the precise time of its signal transmission. GNSS output includes three-dimensional position (longitude, latitude, altitude) and velocity.

[0093] 2. Sources of error and limitations.

[0094] GNSS positioning accuracy is affected by many factors:

[0095] 1) Atmospheric Distortions: Ionosphere and troposphere delays affect signal propagation speed.

[0096] 2) Multipath propagation: Signals reflect off surfaces like buildings and terrain, taking multiple paths to the receiver, introducing errors. This is particularly serious in urban canyon environments.

[0097] 3) Signal Blockage / Insufficient Satellites: Tall buildings, tunnels, dense vegetation, or underwater / underground operations can block satellite signals, resulting in positioning failure.

[0098] 4) Limited Accuracy: The accuracy of standard GNSS may not meet the requirements of high-precision tasks. Technologies such as Real-Time Kinematic (RTK) can achieve centimeter-level accuracy, but usually require base stations and have high requirements for observation conditions.

[0099] The unreliability and even complete failure of GNSS in many critical UAV operating environments (such as urban areas, indoors, and underwater) has been a major catalyst for the development and application of comprehensive multi-sensor fusion systems. While GNSS is a powerful aid when available, its inherent limitations define the core problems that fusion technology aims to solve.

[0100] 1.1.3 Visual Sensor (Camera)

[0101] 1. Working principle

[0102] Event Cameras / Neuromorphic Sensors: Bio-inspired sensors that asynchronously report pixel-level brightness changes (events) rather than outputting full frames.

[0103] 2. Data characteristics and processing.

[0104] Event cameras output a sparse, asynchronous stream of events (pixel address, timestamp, polarity). They offer high temporal resolution (microseconds), high dynamic range (HDR >120dB), low latency, and low power consumption, while reducing data redundancy. Processing requires specialized feature detection, tracking, and optical flow estimation algorithms tailored to event streams. Event cameras are more than just an incremental improvement over traditional cameras; they represent a potential paradigm shift in drone vision for applications with high dynamic range or extreme lighting conditions. Their asynchronous, sparse data nature, combined with high temporal resolution and dynamic range, directly address key weaknesses of traditional frame-based cameras, such as motion blur and exposure issues.

[0105] 3. Visual Odometry (VO) and Visual Inertial Odometry (VIO)

[0106] 1) VO: Estimates camera motion by analyzing consecutive image frames. Monocular VO suffers from scale ambiguity.

[0107] 2) Visual Input / Output (VIO): Tightly integrate visual data with IMU measurements. The IMU helps resolve scale ambiguity in monocular VIO, improves robustness to fast motion and weakly textured scenes, and provides high-frequency motion priors. VIO techniques include filtering-based methods (such as the EKF-based MSCKF) and optimization-based methods (such as bundle adjustment).

[0108] Visual-inertial odometry (VIO) embodies a deep synergistic relationship. The vision system provides rich environmental features to constrain the drift accumulated by the IMU over time, while the IMU provides high-frequency motion information to resolve visual ambiguities (such as the monocular scale problem), enabling the system to track in feature-sparse areas or rapid motion and helping to predict feature positions. This tight coupling is often the key to achieving robust navigation in GNSS-denied environments. However, while visual features are abundant and readily available in many environments, their reliability is significantly affected by environmental factors (lighting, weather, texture) and changes in viewpoint. This variability requires robust feature detection / matching algorithms and often requires fusion with more consistent sensors (such as IMUs or LiDAR) to maintain continuous operation.

[0109] 1.1.4 LiDAR

[0110] 1. Working principle and data output.

[0111] LiDAR directly measures distance by emitting laser pulses and measuring the time of flight of the pulses from objects back to the sensor. Its main data output is a three-dimensional point cloud, representing the spatial coordinates of a series of points on the surrounding surface. Point cloud data can also contain attributes such as reflection intensity and echo times (for multi-echo LiDAR).

[0112] 2. LiDAR types.

[0113] LiDAR is classified according to its platform (airborne, ground - mobile, fixed), detection technology (photon type, linear array type), scanning method (mechanical, solid - MEMS, OPA, Flash area array type) and output dimension (2D, 3D, 4D). Due to the fewer moving parts, solid-state LiDAR is attracting more and more attention due to its potential low cost and high robustness.

[0114] 2. Adaptive weight generation LSTM weight

[0115] The LSTM module is the core of the entire adaptive fusion system, responsible for parsing real-time reliability indicators from various sensors and generating dynamic fusion weights.

[0116] 3.1 LSTM architecture for sensor fusion

[0117] According to the LASF-LSTM architecture, the LSTM architecture for sensor fusion designed by the embodiment is as follows:

[0118] 1) Input layer: Accept pre-processed and vectorized sensor reliability indicators. The number of feature points and average response value of GNSS and camera, the point cloud density and average intensity of LiDAR, and the sliding window variance of various axis acceleration and angular velocity of IMU, etc. Stack these indicators in the time dimension to form a sequence input to the LSTM.

[0119] 2) LSTM layer: Stack multiple LSTM layers to learn time patterns at different levels of abstraction. This embodiment designs two sequential LSTM layers. The first layer of LSTM processes the original reliability indicator sequence to extract preliminary time features; the second layer of LSTM further learns more complex long-term dependencies based on more abstract features output by the first layer. Each LSTM unit contains a precise gating mechanism.

[0120] 3) Output Layer: The final output of the LSTM layer (typically the hidden state at the last time step, or some pooling of the hidden states across all time steps) is fed into one or more fully connected (Dense) layers, described as a "time-distributed dense layer." The output of this dense layer is the adaptive fusion weights of each sensor. The softmax activation function is used to represent the probability distribution.

[0121] 3.2LSTM core calculation unit formula

[0122] The LSTM unit controls the flow and memory of information through its internal gating structure (forget gate, input gate, and output gate) and a neuron state (cell state). At a time step t, the calculation process of the LSTM unit is as follows:

[0123] Assume that the input vector of the current time step is (composed of the sensor reliability index at that moment), the hidden state of the previous time step is , the neuron state at the previous time step is .

[0124] 1. Forget Gate ): determines what information to discard from the neuron state. and , and is the neuron state Each number in outputs a number between 0 and 1 (1 means completely keep, 0 means completely discard).

[0125]

[0126] 2. Input Gate ): decides what new information to store in the neuron state. It also checks and .

[0127]

[0128] 3. Candidate cell State, ): Creates a new vector of candidate values ​​that may be added to the neuron state. Generated using the tanh activation function.

[0129]

[0130] 4. Cell state update ): The old neuron state Update to the new neuron state This is achieved by selectively forgetting old information through the forget gate and selectively adding new candidate information through the input gate.

[0131]

[0132] 5. Output gate ): determines which parts of the neuron state will be used as output.

[0133]

[0134] 6. Hidden State Update, ): The final output of the LSTM unit, which is also the hidden state input for the next time step. It is obtained by passing the neuron state through tanh (scaling its value to between -1 and 1) and multiplying it with the output of the output gate.

[0135]

[0136] This internal structure of LSTM, especially the neuron state The existence of , gives it a strong memory capacity. Information can be transmitted on it and remain basically unchanged, and only added or deleted through the fine control of the gate. This enables LSTM to capture the long-term trends and patterns of the sensor reliability indicator series. The cell state of LSTM Able to accumulate information about the history of such reliability changes, the forget gate and input gate This mechanism controls which old information is retained and which new information is incorporated. This mechanism enables the LSTM to make decisions based on the context of varying reliability. For example, it might learn that a brief drop in GNSS quality typically recovers quickly, thus avoiding a significant weight reduction; whereas persistent performance degradation requires a decisive weight reduction. This memory-based contextual awareness helps generate more stable and reasonable adaptive weights, avoiding overreaction to temporary sensor disturbances, thereby improving the smoothness and reliability of the overall state estimate.

[0137] In summary, this embodiment combines coaxial rotor design with CFD optimization, taking into account both high lift and anti-interference. The heat dissipation system achieves a balance between energy consumption and performance, improving hardware redundancy and efficiency. The UDP algorithm and tightly coupled GNSS-SLAM solve strong light problems, the repeated structure improves the ability to resist environmental interference, and the catenary geometry modeling and the modeling method for accurately extracting transmission line details are used. The dynamic fusion of MPC and LSAF-LSTM ensures robustness and endurance under complex tasks. Through fluid mechanics and control theory, this solution significantly improves the efficiency, accuracy and reliability of drones in power transmission inspections.

[0138] Example 2

[0139] Based on the above embodiments, this embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the transmission tower bolt detection method based on the improved yolov5s as described in Example 1.

[0140] like Figure 2 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above method. Of course, in addition to software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0141] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0142] Example 3

[0143] This embodiment provides a computer-readable storage medium, including one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the aforementioned transmission tower bolt detection method based on the improved yolov5s.

[0144] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0145] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A three-dimensional modeling and flight control method for power transmission lines based on an octorotor drone and a laser radar, characterized in that: The steps include: Build an eight-rotor coaxial counter-propeller UAV hardware platform integrating 3D lidar, GNSS sensor and air duct cooling system; Optimizing UAV trajectories in wind-resistant environments through model predictive control; By combining LiDAR point cloud acquisition with Fast-LIO2_SLAM, loop detection, GNSS assistance, and point cloud processing are implemented to build a 3D model of the transmission line. The process of loop closure detection, GNSS assistance, and point cloud processing based on Fast-LIO2_SLAM includes the following steps: Loop detection is performed based on the density map ORB feature, and anti-aliasing perception in scenes with repeated structures is achieved through 2D bird's-eye view projection and self-similarity pruning; Through the tight coupling of GNSS, LiDAR and inertial, the extended Kalman filter is used to achieve global positioning constraints and suppress long-distance drift.

2. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1 is characterized in that: The process of optimizing UAV trajectories in wind-resistant environments using model predictive control includes the following steps: Coaxial rotor aerodynamic design based on computational fluid dynamics, calculating lift offset by solving the Navier-Stokes problem; By controlling the torque distribution, the motor thrust of the octorotor UAV is dynamically distributed to achieve gust resistance and motor failure redundancy control; Optimize flight paths through evolutionary algorithms.

3. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1, characterized in that: The air duct cooling system described above passively cools the laser radar through the airflow during the flight of the UAV, and maintains the normal operating temperature of the sensor during long-term operation without the need for additional power supply equipment.

4. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1, characterized in that: After collecting the LiDAR point cloud, the following LiDAR anti-strong light interference processing process is also included: Denoise the original lidar point cloud based on statistical filtering and intensity thresholding; The UDP algorithm is used to achieve feature extraction through progressive self-attention and bird's-eye view, and robust registration of point clouds under strong light is achieved.

5. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1, characterized in that: The model predictive control is combined with a wind field model to predict the UAV's dynamic behavior in trajectory planning, and achieves smooth tracking and wind-resistant stable flight in the transmission line corridor by optimizing the control input.

6. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1, characterized in that: In the process of constructing the 3D model of the transmission line, the point cloud processing of the 3D model of the transmission line includes the following steps: Reconstruct the sag shape of transmission lines based on minimum spanning tree clustering and iterative K-means catenary fitting; The repeated structure of the tower is identified through deep learning semantic segmentation, and high-precision 3D reconstruction is achieved by combining the GroundSLAM featureless matching algorithm.

7. The method for three-dimensional modeling and flight control of power transmission lines based on an octorotor drone and a laser radar according to claim 1, characterized in that: During the state estimation process, the fusion weights of LiDAR, IMU, and GNSS are dynamically adjusted according to the real-time indicators of GNSS signal strength, LiDAR point cloud density, and IMU noise.

8. An electronic device, characterized in that: include: One or more processors and a memory, wherein the memory stores one or more programs, wherein the one or more programs include instructions for executing the three-dimensional modeling and flight control method of the power transmission line based on the octorotor drone and the lidar as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that It includes one or more programs for execution by one or more processors of an electronic device, and the one or more programs include instructions for executing the three-dimensional modeling and flight control method of power transmission lines based on an octorotor drone and a lidar as described in any one of claims 1-7.

Images