Low-altitude inspection unmanned aerial vehicle flight obstacle avoidance method and system

Abstract

The application belongs to the technical field of unmanned aerial vehicle inspection obstacle avoidance, and specifically discloses a low-altitude inspection unmanned aerial vehicle flight obstacle avoidance method and system. The unmanned aerial vehicle task planning information is deeply integrated into the obstacle avoidance decision process, a comprehensive optimal decision of balancing safety, task efficiency and energy consumption is realized, and the completion degree and flight efficiency of the task are improved. A local trajectory generation and real-time optimization mechanism based on an optimization algorithm is adopted, which is beneficial to the continuous adjustment of the unmanned aerial vehicle to adapt to environmental changes, and significantly improves the flight stability and energy consumption efficiency of the unmanned aerial vehicle when obstacles are avoided. Through the closed-loop real-time feedback of environment perception, dynamic modeling, risk assessment, trajectory planning and flight control, the unmanned aerial vehicle can continuously and autonomously perform the inspection task in the dynamically changing low-altitude complex environment, and the demand for manual intervention and the potential risk of flight accidents are reduced.

Description

Technical Field

[0001] This invention belongs to the field of drone inspection and obstacle avoidance technology, specifically relating to a method and system for obstacle avoidance during low-altitude inspection drone flight. Background Technology

[0002] With the deep integration and widespread application of drone technology in fields such as industrial inspection, agricultural plant protection, infrastructure maintenance, and environmental monitoring, its importance as an intelligent aerial operation platform is becoming increasingly prominent. Especially in low-altitude, complex environment tasks such as power line inspection, oil and gas pipeline patrol, and crop pest and disease monitoring, drones have become indispensable technological tools due to their efficiency, flexibility, and low cost. However, with the continuous increase in drone flight density and the increasing diversification of operational scenarios, ensuring the flight safety and mission reliability of drones in low-altitude operating environments has become a core technological bottleneck restricting their further development. Highly efficient and precise obstacle avoidance capabilities are key to overcoming this bottleneck.

[0003] Current drone obstacle avoidance technologies primarily rely on preset safety distances and simple predictive models for reactive avoidance. However, they have certain shortcomings when facing small, high-speed biological targets with complex behavioral patterns: Firstly, existing drone obstacle avoidance technologies generally lack deep learning and predictive capabilities regarding biological behavioral patterns, making it impossible to accurately predict the trajectory and intentions of a living organism within the next few seconds before it takes evasive action. This results in lag and uncertainty in obstacle avoidance decisions, making it difficult for drones to effectively avoid the risk of sudden biological collisions during high-speed flight.

[0004] On the other hand, existing drone obstacle avoidance technologies fail to fully consider the behavioral differences of different species, environmental conditions, and risk levels. Existing strategies often adopt rigid, one-size-fits-all avoidance actions, which may seriously affect the continuity, efficiency, and path optimization of inspection tasks due to excessive avoidance, or cause flight safety accidents due to insufficient avoidance, and even harm the organisms themselves. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a method and system for obstacle avoidance during low-altitude inspection UAVs are proposed.

[0006] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a method for obstacle avoidance during flight of a low-altitude inspection drone, comprising: S1, collecting environmental perception data of several inspection drones through a multimodal sensor array mounted on several inspection drones, and performing time synchronization, spatial registration and multi-scale fusion processing to generate unified environmental fusion perception data.

[0007] S2. Construct a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model.

[0008] S3. Obtain UAV mission planning information, evaluate the collision risk signals of several UAVs based on the UAV dynamic environment model, and select obstacle avoidance strategies.

[0009] S4. Based on the obstacle avoidance strategy, the current position of the inspection drone and the target waypoint, plan several obstacle avoidance trajectories of the inspection drone in the drone dynamic environment model, and optimize them through simulation.

[0010] S5. The optimized obstacle avoidance trajectory is converted into a series of flight control commands and transmitted to the UAV flight controller to drive the UAV to execute them.

[0011] A second aspect of the present invention provides a low-altitude inspection drone flight obstacle avoidance system, comprising: an environmental fusion perception data acquisition module, a dynamic environment module construction module, a collision risk assessment module, an obstacle avoidance trajectory generation and optimization module, and an obstacle avoidance trajectory execution module.

[0012] The environmental fusion perception data acquisition module is connected to the dynamic environment module construction module, the dynamic environment module construction module is connected to the collision risk assessment module, the collision risk assessment module is connected to the obstacle avoidance trajectory generation and optimization module, and the obstacle avoidance trajectory generation and optimization module is connected to the obstacle avoidance trajectory execution module.

[0013] The environmental fusion perception data acquisition module collects environmental perception data from several inspection drones using a multimodal sensor array mounted on several inspection drones, and performs time synchronization, spatial registration, and multi-scale fusion processing to generate unified environmental fusion perception data.

[0014] The dynamic environment module is a construction module that builds a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model.

[0015] The collision risk assessment module acquires drone mission planning information and assesses the collision risk signals of several drones based on the drone dynamic environment model, while selecting obstacle avoidance strategies.

[0016] The obstacle avoidance trajectory generation and optimization module plans several obstacle avoidance trajectories for the inspection drone in the drone dynamic environment model based on the obstacle avoidance strategy, the current position of the inspection drone, and the target waypoint, and optimizes them through simulation.

[0017] The obstacle avoidance trajectory execution module converts the optimized obstacle avoidance trajectory into a series of flight control commands, which are then transmitted to the UAV flight controller to drive the UAV to execute them.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention realizes the identification, tracking and future trajectory prediction of moving obstacles by constructing a dynamic environment model that includes a static obstacle distribution model and a dynamic obstacle prediction model.

[0019] 2. This invention deeply integrates UAV mission planning information into the obstacle avoidance decision-making process, achieving a comprehensive optimal decision that balances safety, mission efficiency, and energy consumption, thereby improving mission completion and flight efficiency.

[0020] 3. This invention adopts a local trajectory generation and real-time optimization mechanism based on optimization algorithms, which helps the UAV to continuously adapt to environmental changes and make adjustments, significantly improving its flight stability and energy efficiency when avoiding obstacles, and reducing energy consumption caused by violent maneuvers.

[0021] 4. This invention, through high-precision, high-frequency closed-loop real-time feedback of environmental perception, dynamic modeling, risk assessment, trajectory planning and flight control, helps UAVs to continuously and autonomously perform inspection tasks in dynamically changing low-altitude complex environments, reducing the need for human intervention and the risk of potential flight accidents. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram illustrating the implementation steps of the method of the present invention.

[0024] Figure 2 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1 Please see Figure 1As shown, the present invention provides a method for obstacle avoidance during flight of a low-altitude inspection drone. The specific steps are as follows: S1, several environmental perception data of several inspection drones are collected by a multi-modal sensor array carried by several inspection drones, and time synchronization, spatial registration and multi-scale fusion processing are performed to generate unified environmental fusion perception data.

[0027] In one feasible example of the above steps, the multimodal sensor array includes a lidar, a vision sensor, a millimeter-wave radar, and an inertial measurement unit.

[0028] The specific process of acquiring several environmental perception data of the inspection drone includes: acquiring several three-dimensional point cloud data around the inspection drone through the lidar, including the spatial location information and geometric shape information of obstacles.

[0029] The visual sensor acquires visible light image data and depth image data of the area in front of the inspection drone. The visible light image data is used for target recognition and semantic segmentation, and the depth image data is used to enhance the distance information of obstacles.

[0030] The millimeter-wave radar acquires information on the distance, speed, and angle of several moving targets around the inspection drone.

[0031] The inertial measurement unit acquires real-time attitude angles, angular velocities, linear velocities, and linear acceleration data of several inspection drones.

[0032] In one feasible embodiment of the present invention, the specific process of generating unified environmental fusion perception data includes: the time synchronization is achieved through precise timestamp alignment.

[0033] It should be noted that the time synchronization is to eliminate the huge spatial errors caused by the different data acquisition times and output frequencies of different sensors when the drone is moving at high speed.

[0034] The spatial registration is achieved by coordinate system transformation through the sensor extrinsic parameter calibration matrix.

[0035] It should be noted that each sensor has its own coordinate system. LiDAR point clouds exist in their own radar coordinate system, camera pixels in their image coordinate system, and millimeter-wave radar also has its own polar coordinate system.

[0036] Specifically, through sensor extrinsic parameter calibration, the position and attitude of each sensor relative to a common coordinate system, such as the UAV body coordinate system or IMU coordinate system, are accurately measured in advance—this is the extrinsic parameter matrix. During processing, this extrinsic parameter matrix is ​​used to transform the data from all sensors to the same unified coordinate system, so that the observation results of an obstacle in different sensors can be aligned.

[0037] The multi-scale fusion processing employs a Kalman filter algorithm to weightedly fuse lidar point cloud data, visual depth data, and millimeter-wave radar data, generating an enhanced local 3D obstacle distribution map. This map is then combined with inertial measurement unit data to estimate the obstacle positions and motion states.

[0038] It should be noted that the multi-scale fusion aims to fuse preprocessed multi-source information into a perception result that is superior to any single information source.

[0039] The term "multi-scale" can be understood as fusing data from different dimensions, such as feature level and data level.

[0040] The weighted fusion typically employs Kalman filtering or its variants, such as the Extended Kalman Filter (EKF) algorithm. The specific process involves: predicting the next state of the inspection drone and obstacles based on data from IMU (Instrument Detection Unit) and other sources; updating this prediction using observation data from LiDAR, vision, and millimeter-wave radar. The algorithm dynamically calculates the degree of trust in each sensor's data, i.e., the weighting coefficient. For example, in strong light, camera data has higher reliability and therefore a larger weight; in adverse weather conditions, the weight of millimeter-wave radar data increases, while the weights of LiDAR and camera data decrease.

[0041] After fusion, an enhanced local 3D obstacle distribution map is finally generated. Furthermore, the local 3D obstacle distribution map includes: obstacle location and shape, semantic labels for obstacles such as cars, pedestrians, and trees, and precise velocity vectors of obstacles.

[0042] S2. Construct a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model.

[0043] In one feasible embodiment of the present invention, the specific construction process of the dynamic obstacle prediction model includes: A1, dynamic target recognition: based on environmental fusion perception data, several dynamic targets around several inspection drones are identified through operational judgment rules, and the categories and initial states of the multiple dynamic targets are output.

[0044] It should be noted that the specific content of the motion determination rule is as follows: if the speed of several moving targets around the inspection drone directly output by the millimeter-wave radar exceeds the preset static target swing speed threshold, such as >0.3m / s, then the moving target is marked as a potential dynamic target.

[0045] Furthermore, based on the spatial location information of obstacles output by the lidar, the inter-frame difference is used to assist in the judgment. That is, if the position of an obstacle at the same location changes more than the preset upper limit of the position change distance in two consecutive frames with an interval of 0.1 seconds, such as >0.1m, then the potential dynamic target is identified as a dynamic target.

[0046] The specific process of outputting the category of dynamic targets includes: combining the visual semantic segmentation results, such as category labels for pedestrians, vehicles, and birds, with the speed / size characteristics of millimeter-wave radar, such as vehicle speed typically being 5-20 m / s and pedestrian speed being 1-5 m / s, to determine the target category. For example, if the speed is 3 m / s and the visual label is pedestrian, then it is determined to be a pedestrian.

[0047] It also outputs the initial state of each dynamic target, including but not limited to position (three-dimensional coordinates), velocity (three-dimensional vector), and dimensions (length, width, and height), for collision volume calculation.

[0048] It needs to be explained that the key is to accurately identify dynamic obstacles that require special attention from a complex environment, and to avoid misjudging tree branches swaying in the wind as dynamic targets.

[0049] A2. Target tracking: Matching and associating several dynamic targets within a continuous time window to form a complete historical trajectory of the dynamic target, which usually includes a state sequence of the past 5-10 seconds.

[0050] It should be noted that the specific process of the association matching is as follows: the Hungarian algorithm is used to match dynamic targets in two consecutive time windows, namely, to calculate the comprehensive score of the position distance, velocity difference and category consistency between a dynamic target in the current frame and each dynamic target in the previous frame in the consecutive time window, and to select the dynamic target in the previous frame with the highest comprehensive score with the current frame dynamic target, and match it with the current frame dynamic target as the same dynamic target, thereby performing association matching on several dynamic targets in the consecutive time window.

[0051] It is also necessary to understand that for newly emerging dynamic targets, such as pedestrians who suddenly appear, a new tracking ID is created; for disappearing dynamic targets, such as those that leave the perception range, their trajectory for the last 3 seconds is retained for prediction and correction.

[0052] The specific process for forming a complete historical motion trajectory of a dynamic target is as follows: Sensor noise is processed using extended Kalman filtering or particle filtering. For example, millimeter-wave radar velocity measurements may have an error of ±0.2 m / s. By filtering and fusing multiple frames of data, a smoother velocity curve is output. For position data from visual / LiDAR, instantaneous jumps, such as sudden position changes caused by occlusion, are eliminated through sliding window averaging. Each dynamic target corresponds to a historical trajectory, containing information such as timestamp, position, and velocity, such as a trajectory sequence. .

[0053] A3. Future trajectory prediction: Based on historical motion trajectories, the prediction algorithm outputs the motion trajectory and possible location distribution of dynamic obstacles within a future time window, which serves as the dynamic obstacle prediction model.

[0054] For example, the future time window can be 3-5 seconds, which corresponds to the response time of the inspection drone's obstacle avoidance decision. Furthermore, the length of the future time window is dynamically adjusted according to the drone's flight speed. For instance, if the drone's speed is high, such as 10 m / s, a 5-second time window covers a range of 50 m; if the speed is slow, such as 3 m / s, a 3-second time window covers a range of 9 m, balancing prediction accuracy and obstacle avoidance lead time.

[0055] It should be noted that the specific selection methods of the prediction algorithm include: (1) For dynamic targets with simple motion patterns, such as vehicles moving in a straight line at a constant speed: use physical models for prediction, such as constant speed model, which assumes that the future speed will remain unchanged, and constant acceleration model, which assumes that the acceleration will remain unchanged.

[0056] (2) For dynamic targets with complex motion patterns, such as pedestrians moving around at will: adopt data-driven models, such as LSTM-based sequence prediction models that learn the nonlinear patterns of historical trajectories, and Transformer models that capture long-term dependencies, such as pedestrians’ habits of avoiding obstacles.

[0057] (3) For dynamic targets with known categories: optimize by combining prior knowledge. For example, the maximum speed of pedestrians is usually <5m / s. Limit the speed range during prediction to avoid outputting unreasonable trajectories.

[0058] In one feasible embodiment of the present invention, the specific construction process of the static obstacle distribution model includes: B1, grid map initialization: dividing a number of inspection drones into a preset three-dimensional space centered on themselves, such as a cylindrical space with a radius of 50m and a height of 0-30m, into uniform three-dimensional grid units. The size and resolution of each grid are set according to the requirements of the inspection scenario. For example, in a complex urban environment, it is 0.1m×0.1m×0.1m, and in an open park, it can be widened to 0.5m×0.5m×0.5m.

[0059] It should be noted that the purpose of discretizing the continuous three-dimensional space is to facilitate the numerical description of whether a spatial location is occupied by an obstacle. Furthermore, low-altitude inspection needs to identify small obstacles such as tree branches and cables, so the grid resolution is usually higher than that of outdoor large-scale navigation, which may use a 1m-level resolution to avoid missing small structures due to excessively large grids.

[0060] B2. Mapping of sensing data to grids: Mapping environmental fusion sensing data to corresponding 3D grid cells and removing 3D grid cells marked as dynamic targets.

[0061] Specifically, for LiDAR point clouds: the three-dimensional coordinates of each point are directly mapped to its corresponding three-dimensional grid cell; for visual depth maps: the depth information of pixels is converted into three-dimensional coordinates through camera intrinsic parameters and then mapped to three-dimensional grid cells; for millimeter-wave radar static targets: if a target with a velocity of 0, such as a wall, is detected, its position information is mapped to a three-dimensional grid cell.

[0062] It should be noted that unifying static obstacle observations from different sensors into a three-dimensional grid cell framework provides a data foundation for subsequent occupancy probability calculations.

[0063] B3. Calculation of grid occupancy probability: For each 3D grid cell after the culling operation, the probability that the 3D grid cell is occupied by an obstacle is calculated based on the environmental fusion perception data mapped into it, and it is recorded as the static obstacle occupancy probability of the 3D grid cell.

[0064] For example, the probability of a three-dimensional grid cell being occupied by an obstacle is calculated using the Bayesian update rule. The specific process is as follows: (1) Initial probability setting: For each three-dimensional grid cell to be calculated, its initial occupancy probability is set to 0.5, which means that it is neither certain that there is an obstacle nor certain that there is no obstacle, thus providing a benchmark for subsequent updates based on sensor data.

[0065] (2) Sensor data weight allocation: Since different sensors have different reliability in different scenarios, such as LiDAR has better resistance to light interference than vision, and millimeter-wave radar has better resistance to rain and fog than LiDAR, it is necessary to first allocate weights to the multi-source sensing data mapped to the grid.

[0066] For example, the sensor data weight settings for normal lighting scenarios include: LiDAR weight 0.7, visual depth map weight 0.3, and millimeter-wave radar weight 0, because static target detection has a lower priority than the former two.

[0067] The sensor data weight settings for adverse rain and fog scenarios include: LiDAR weight 0.2, visual depth map weight 0.1, and millimeter-wave radar weight 0.7.

[0068] (3) Updating occupancy probability based on Bayesian rules: Based on the weighted sensor data, the grid occupancy probability is dynamically adjusted using the Bayesian formula. If the sensor data shows that there is an obstacle in the grid, such as a lidar point falling into the grid or a visual depth map detecting an entity at that location, then the formula is: .

[0069] For example, if the initial occupancy probability is 0.5 and the weight of the lidar sensor data is 0.7, then the updated probability = 0.5 + 0.7 × (1 - 0.5) = 0.8; If sensor data shows no obstacles within the grid, such as when a drone flies over the grid and multiple sensors provide no feedback, then the formula is: .

[0070] For example, if the initial occupancy probability is 0.5 and the visual sensor data weight is 0.3, then the updated probability = 0.5 - 0.3 × 0.5 = 0.3.

[0071] If multiple consecutive frames of sensor data show either obstacles or no obstacles, repeat the above update process.

[0072] B4. Static obstacle distribution model construction: The static obstacle occupancy probabilities of several three-dimensional grid units in the preset three-dimensional space are integrated to obtain a three-dimensional occupancy grid map, which is used as the static obstacle distribution model.

[0073] It should be noted that as the inspection drone flies, the static obstacle occupancy probability of the 3D grid cell is continuously updated with new environmental fusion perception data.

[0074] This invention achieves the identification, tracking, and future trajectory prediction of moving obstacles by constructing a dynamic environment model that includes a static obstacle distribution model and a dynamic obstacle prediction model.

[0075] S3. Obtain UAV mission planning information, evaluate the collision risk signals of several UAVs based on the UAV dynamic environment model, and select obstacle avoidance strategies.

[0076] In one feasible embodiment of the present invention, the UAV mission planning information includes preset inspection paths, target waypoints, no-fly zone information, and inspection mission priorities for a number of inspection UAVs.

[0077] The collision risk signals include the probability of a drone colliding with a static obstacle, the probability of a drone colliding with a dynamic obstacle, and the probability of a drone colliding with another drone.

[0078] In one feasible embodiment of the present invention, the specific assessment process of the collision risk signals of the plurality of UAVs includes: the probability assessment of the collision risk between the UAV and the static obstacle specifically includes the following: based on the linear velocity of the inspection UAV and the preset inspection path, predict the flight trajectory of the inspection UAV within a future time window, such as 5 seconds, and discretize it into a waypoint every 0.5 seconds.

[0079] Each waypoint of the inspection drone's flight trajectory is mapped to a 3D grid cell of a static obstacle distribution model. The static obstacle occupancy probability of the corresponding 3D grid cell is extracted, and the collision risk probability between the drone and the static obstacle is calculated using the standard formula for the collision risk probability between the drone and the static obstacle. .

[0080] It should be noted that the standard formula for the probability of collision between the drone and a static obstacle is as follows: ,in The first step in inspecting the flight path of the drone The static obstacle occupancy probability of a predicted waypoint in a 3D grid cell. The number of waypoints within a future time window, if If the threshold is not met, it is considered a high static risk. The threshold can be adjusted according to the scenario.

[0081] The core of the drone-to-static-obstacle collision risk probability assessment is to determine whether the drone's future path will overlap with static obstacles such as buildings and trees, and to quantify the collision risk as the drone-to-static-obstacle collision risk probability index.

[0082] The probability assessment of the collision risk between the drone and the dynamic obstacle specifically includes the following: after aligning the flight trajectory of the inspection drone in the predicted future time window with the movement trajectory of the dynamic obstacle in the same future time window in the time-space dimension, the spatial distance between the two at each time step in the future time window is calculated. If the spatial distance is less than a preset safety threshold, it indicates that there is a collision risk between the inspection drone and the dynamic obstacle.

[0083] By generating several possible trajectories, such as 1000, through Monte Carlo simulation, the proportion of inspection drones with a collision risk with dynamic obstacles is statistically analyzed and denoted as the probability of collision between the drone and the dynamic obstacle. .

[0084] It should be noted that the core of the aforementioned drone collision risk probability assessment is to determine whether the future trajectories of the drone and dynamic targets such as pedestrians or other vehicles will intersect in space and time. The quantified indicator is the dynamic collision probability. Collision time countdown .

[0085] The drone-to-drone collision risk probability assessment specifically includes the following: after aligning the preset inspection paths of several inspection drones in the time-space dimension, calculate the spatial distance between itself and other inspection drones at each time step in the future time window. If the spatial distance is less than a preset safety threshold, it indicates that there is a collision risk between the inspection drone and other inspection drones.

[0086] By generating 1000 possible trajectories using Monte Carlo simulation, the proportion of inspection drones with a collision risk with other inspection drones is statistically analyzed and denoted as the drone-to-drone collision risk probability. .

[0087] In one feasible embodiment of the present invention, the specific process of selecting the obstacle avoidance strategy includes: predefining a variety of typical obstacle avoidance strategies for the assessed collision risk signal; It should be noted that the aforementioned typical obstacle avoidance strategies specifically include: Strategy 1, deceleration and waiting: reducing flight speed, such as from 5m / s to 2m / s, waiting for dynamic obstacles to leave or passing through static obstacle areas. This strategy is suitable for low-priority tasks and energy-sensitive scenarios.

[0088] Strategy 2, Lateral Flight: Deviate to the left, right or up and down of the preset inspection path, such as by 1-3m, bypass the obstacle and return to the original path. It is suitable for scenarios with small obstacles and time-sensitive tasks.

[0089] Strategy 3, Path Replanning: Based on the A or RRT algorithm, the inspection path from the current location to the target waypoint is replanned to completely avoid high-risk areas. It is suitable for high collision risk and safety-first scenarios.

[0090] Strategy 4: Multi-drone cooperative obstacle avoidance: Multiple inspection drones negotiate through communication to adjust their relative positions or speeds. For example, if one drone temporarily avoids an obstacle, another will pass first. This strategy is suitable for areas with a high density of multiple drones.

[0091] The impact of the collision risk signal on the inspection task completion, energy consumption, and safety of several inspection drones is evaluated.

[0092] It should be noted that the specific analysis process for the indicators affecting the completion of the inspection task is as follows: If the inspection drone needs to deviate from the preset path to avoid risks, the deviation distance between the actual inspection path and the preset inspection path and the delay time to reach the target waypoint are calculated, and weighted by waypoint priority. The calculation formula is as follows: ,in .

[0093] It needs to be explained that for high-priority tasks, their Larger means more attention is paid to time delay.

[0094] The specific analysis process for the energy consumption impact index is as follows: Obstacle avoidance by inspection drones will increase additional energy consumption, calculated based on the inspection drone dynamics model: ,in These are motor power, extra flight time, attitude energy consumption coefficient, and attitude adjustment range, respectively. Sharp turns and rapid ascents and descents will significantly increase energy consumption.

[0095] The specific analysis process for the safety impact indicators is as follows: The overall safety risk is quantified by weighted summation, considering the probabilities of the three types of collision risks. ,in These represent the weights of the three types of collision risk probabilities, with the weight in multi-machine scenarios being... Larger, densely packed static obstacle areas Larger.

[0096] Based on collision risk signals and impact indicators, a multi-objective decision-making algorithm is used to select an obstacle avoidance strategy that meets the priority requirements of several inspection drones' preset inspection tasks.

[0097] It should be noted that the obstacle avoidance strategy is not static and needs to be dynamically updated according to changes in the real-time collision risk probability.

[0098] For example, if a sudden change in the trajectory of a dynamic obstacle is detected during execution, such as a pedestrian suddenly changing direction, the risk is reassessed and the strategy is switched, such as changing from lateral detour to path replanning; if the strategy of one UAV in multi-UAV collaboration is adjusted, other UAVs need to update their own strategies synchronously; if the task priority is temporarily changed, such as an emergency inspection command issued by the ground control console, the weights are immediately recalculated and a new strategy is selected.

[0099] This invention achieves a comprehensive optimal decision that balances safety, mission efficiency, and energy consumption by deeply integrating UAV mission planning information into the obstacle avoidance decision-making process, thereby improving mission completion and flight efficiency.

[0100] S4. Based on the obstacle avoidance strategy, the current position of the inspection drone and the target waypoint, plan several obstacle avoidance trajectories of the inspection drone in the drone dynamic environment model, and optimize them through simulation.

[0101] In one feasible embodiment of the present invention, the specific process of step S4 includes: generating preliminary obstacle avoidance flight trajectories of several inspection drones in a dynamic environment model based on the obstacle avoidance strategy, and performing smoothing processing and dynamic constraint checks on them to ensure that the preliminary obstacle avoidance flight trajectories meet the limits of the inspection drone's maximum speed, maximum acceleration and maximum angular velocity, and maintain stable flight.

[0102] Furthermore, based on multi-objective constraints, the initial obstacle avoidance flight trajectory after smoothing and dynamic constraint checks is optimized in real time to minimize energy consumption, minimize deviation from the preset inspection path, and maximize the safe distance from obstacles.

[0103] This invention employs a local trajectory generation and real-time optimization mechanism based on optimization algorithms, which helps the UAV continuously adapt to environmental changes and significantly improves its flight stability and energy efficiency when avoiding obstacles, while reducing energy consumption caused by violent maneuvers.

[0104] S5. The optimized obstacle avoidance trajectory is converted into a series of flight control commands and transmitted to the UAV flight controller to drive the UAV to execute them.

[0105] In one feasible embodiment of the present invention, the specific process of step S5 includes: the UAV flight controller generating low-level control commands based on the obstacle avoidance trajectory and sending them to the UAV flight controller to drive the UAV to fly according to the obstacle avoidance trajectory.

[0106] It should be noted that the low-level control commands are defined as concrete parameters that directly drive the execution of the UAV hardware.

[0107] For example, the low-level control commands may include, but are not limited to, the following: power system commands: motor speed, such as 2800 rpm for the left front motor and 2600 rpm for the right rear motor, used to adjust flight speed and altitude; attitude control commands: servo angles, such as rudder deflection of 5° and aileron deflection of 3°, used to adjust flight direction and stability; execution timing commands: command transmission frequency, which matches the UAV flight control refresh frequency, typically 10-50Hz, to ensure trajectory following accuracy.

[0108] It should also be noted that the optimized obstacle avoidance trajectory is first input into step S5, and then the low-level control commands are transmitted to the UAV flight controller via the UAV's internal communication bus, such as the CAN bus and UART serial port. After receiving the commands, the UAV flight controller combines its real-time collected flight status data, such as the angular velocity and linear acceleration output by the IMU, to make fine adjustments to the commands, avoiding trajectory deviations caused by airflow interference. In case of crosswinds, it automatically increases the speed of the motor on the windward side to compensate for the impact of wind resistance, and finally drives the UAV to fly along the preset obstacle avoidance trajectory.

[0109] Furthermore, it can be understood that the present invention achieves dynamic obstacle avoidance by real-time monitoring of the flight status and environmental perception data of the inspection drone, continuously updating the dynamic environment model during flight, and repeatedly executing the collision risk assessment, obstacle avoidance strategy selection, and obstacle avoidance trajectory generation and optimization steps based on the updated dynamic environment model.

[0110] This invention, through high-precision, high-frequency closed-loop real-time feedback of environmental perception, dynamic modeling, risk assessment, trajectory planning, and flight control, helps UAVs to continuously and autonomously perform inspection tasks in dynamically changing low-altitude complex environments, reducing the need for human intervention and the risk of potential flight accidents.

[0111] Example 2 Please see Figure 2 As shown, the present invention provides a low-altitude inspection UAV flight obstacle avoidance system, the specific modules of which are distributed as follows: environmental fusion perception data acquisition module, dynamic environment module construction module, collision risk assessment module, obstacle avoidance trajectory generation and optimization module, and obstacle avoidance trajectory execution module.

[0112] The environmental fusion perception data acquisition module is connected to the dynamic environment module construction module, the dynamic environment module construction module is connected to the collision risk assessment module, the collision risk assessment module is connected to the obstacle avoidance trajectory generation and optimization module, and the obstacle avoidance trajectory generation and optimization module is connected to the obstacle avoidance trajectory execution module.

[0113] The environmental fusion perception data acquisition module collects environmental perception data from several inspection drones using a multimodal sensor array mounted on several inspection drones, and performs time synchronization, spatial registration, and multi-scale fusion processing to generate unified environmental fusion perception data.

[0114] The dynamic environment module is a construction module that builds a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model.

[0115] The collision risk assessment module acquires drone mission planning information and assesses the collision risk signals of several drones based on the drone dynamic environment model, while selecting obstacle avoidance strategies.

[0116] The obstacle avoidance trajectory generation and optimization module plans several obstacle avoidance trajectories for the inspection drone in the drone dynamic environment model based on the obstacle avoidance strategy, the current position of the inspection drone, and the target waypoint, and optimizes them through simulation.

[0117] The obstacle avoidance trajectory execution module converts the optimized obstacle avoidance trajectory into a series of flight control commands, which are then transmitted to the UAV flight controller to drive the UAV to execute them.

[0118] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0119] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0120] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0121] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0122] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for obstacle avoidance during low-altitude inspection drone flights, characterized by: include: S1. Several environmental perception data of several inspection drones are collected by multimodal sensor arrays carried by several inspection drones, and time synchronization, spatial registration and multi-scale fusion processing are performed to generate unified environmental fusion perception data. S2. Construct a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model; S3. Obtain UAV mission planning information, evaluate the collision risk signals of several UAVs based on the UAV dynamic environment model, and select obstacle avoidance strategies. S4. Based on the obstacle avoidance strategy, the current position of the inspection drone and the target waypoint, plan several obstacle avoidance trajectories of the inspection drone in the drone dynamic environment model, and optimize them through simulation. S5. The optimized obstacle avoidance trajectory is converted into a series of flight control commands and transmitted to the UAV flight controller to drive the UAV to execute them.

2. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific process for generating unified environmental fusion perception data includes: The time synchronization is achieved through precise timestamp alignment; The spatial registration is achieved by coordinate system transformation through the sensor extrinsic parameter calibration matrix. The multi-scale fusion processing employs a Kalman filter algorithm to weightedly fuse lidar point cloud data, visual depth data, and millimeter-wave radar data, generating an enhanced local 3D obstacle distribution map. This map is then combined with inertial measurement unit data to estimate the obstacle positions and motion states.

3. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific construction process of the dynamic obstacle prediction model includes: A1. Dynamic Target Recognition: Based on environmental fusion perception data, identify several dynamic targets around several inspection drones through operational judgment rules, and output the category and initial state of the multiple dynamic targets; A2. Target tracking: Matching and associating several dynamic targets within a continuous time window to form a complete historical trajectory of the dynamic targets; A3. Future trajectory prediction: Based on historical motion trajectories, the prediction algorithm outputs the motion trajectory and possible location distribution of dynamic obstacles within a future time window, which serves as the dynamic obstacle prediction model.

4. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific construction process of the static obstacle distribution model includes: B1. Grid map initialization: Divide the preset three-dimensional space of several inspection drones with themselves as the center point into uniform three-dimensional grid units, where the size of each grid is set according to the inspection scenario requirements. B2. Mapping of sensing data to grid: Mapping the fused environmental sensing data to the corresponding 3D grid cells and removing 3D grid cells marked as dynamic targets. B3. Calculation of grid occupancy probability: For each 3D grid cell after the culling operation, the probability that the 3D grid cell is occupied by an obstacle is calculated based on the environmental fusion perception data mapped into it, and it is recorded as the static obstacle occupancy probability of the 3D grid cell. B4. Static obstacle distribution model construction: The static obstacle occupancy probabilities of several three-dimensional grid units in the preset three-dimensional space are integrated to obtain a three-dimensional occupancy grid map, which is used as the static obstacle distribution model.

5. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The UAV mission planning information includes preset inspection paths, target waypoints, no-fly zone information, and inspection mission priorities for several inspection UAVs. The collision risk signals include the probability of a drone colliding with a static obstacle, the probability of a drone colliding with a dynamic obstacle, and the probability of a drone colliding with another drone.

6. The obstacle avoidance method for low-altitude inspection UAVs according to claim 5, characterized in that: The specific assessment process for the collision risk signals of the aforementioned drones includes: The probability assessment of the collision risk between drones and static obstacles specifically includes the following: predicting the flight trajectory of the inspection drone within a future time window based on the linear velocity of the inspection drone and the preset inspection path; Each waypoint of the inspection drone's flight trajectory is mapped to a three-dimensional grid cell of the static obstacle distribution model. The static obstacle occupancy probability of the corresponding three-dimensional grid cell is extracted from the grid cell. The collision risk probability between the drone and the static obstacle is calculated using the standard formula for the collision risk probability between the drone and the static obstacle. The probability assessment of the collision risk between the drone and the dynamic obstacle specifically includes the following: After aligning the flight trajectory of the inspection drone in the predicted future time window with the movement trajectory of the dynamic obstacle in the future time window in the time-space dimension, the spatial distance between the two at each time step in the future time window is calculated. If the spatial distance is less than a preset safety threshold, it indicates that there is a collision risk between the inspection drone and the dynamic obstacle. Several possible trajectories are generated through Monte Carlo simulation. The proportion of inspection drones that have a risk of collision with dynamic obstacles is statistically analyzed and denoted as the probability of collision between drones and dynamic obstacles. The drone-to-drone collision risk probability assessment specifically includes the following: after aligning the preset inspection paths of several inspection drones in the time-space dimension, calculate the spatial distance between itself and other inspection drones at each time step in the future time window. If the spatial distance is less than a preset safety threshold, it indicates that there is a collision risk between the inspection drone and other inspection drones. Several possible trajectories are generated through Monte Carlo simulation. The proportion of inspection drones that have a collision risk with other inspection drones is statistically analyzed and denoted as the drone-to-drone collision risk probability.

7. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific process of selecting an obstacle avoidance strategy includes: For the collision risk signals identified in the assessment, several typical obstacle avoidance strategies are predefined; The impact of the collision risk signal on the inspection task completion, energy consumption, and safety of several inspection drones was evaluated. Based on collision risk signals and impact indicators, a multi-objective decision-making algorithm is used to select an obstacle avoidance strategy that meets the priority requirements of several inspection drones' preset inspection tasks.

8. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific process of step S4 includes: Based on the obstacle avoidance strategy, preliminary obstacle avoidance flight trajectories of several inspection drones are generated in the dynamic environment model, and smoothing and dynamic constraint checks are performed on them to ensure that the preliminary obstacle avoidance flight trajectories meet the limits of the inspection drone's maximum speed, maximum acceleration and maximum angular velocity, and maintain stable flight. Furthermore, based on multi-objective constraints, the initial obstacle avoidance flight trajectory after smoothing and dynamic constraint checks is optimized in real time to minimize energy consumption, minimize deviation from the preset inspection path, and maximize the safe distance from obstacles.

9. The obstacle avoidance method for low-altitude inspection UAVs according to claim 1, characterized in that: The specific process of step S5 includes: The UAV flight controller generates low-level control commands based on the obstacle avoidance trajectory and sends them to the UAV flight controller to drive the UAV to fly along the obstacle avoidance trajectory.

10. A system for performing the obstacle avoidance method for low-altitude inspection UAVs according to claims 1-9, characterized in that: include: The environmental fusion perception data acquisition module collects environmental perception data from several inspection drones through a multimodal sensor array mounted on several inspection drones, and performs time synchronization, spatial registration and multi-scale fusion processing to generate unified environmental fusion perception data. The dynamic environment module is a construction module that builds a dynamic environment model for UAVs based on environmental fusion perception data, including a static obstacle distribution model and a dynamic obstacle prediction model. The collision risk assessment module acquires drone mission planning information and assesses the collision risk signals of several drones based on the drone dynamic environment model, while selecting obstacle avoidance strategies. The obstacle avoidance trajectory generation and optimization module plans several obstacle avoidance trajectories for the inspection drone in the drone dynamic environment model based on the obstacle avoidance strategy, the current position of the inspection drone, and the target waypoint, and optimizes them through simulation. The obstacle avoidance trajectory execution module converts the optimized obstacle avoidance trajectory into a series of flight control commands, which are then transmitted to the UAV flight controller to drive the UAV to execute them.

Images