Intelligent obstacle avoidance method and control system for unmanned aerial vehicle patrol along traffic line

Abstract

The application provides an intelligent obstacle avoidance method and control system for unmanned aerial vehicle patrol along a traffic route, and belongs to the technical field of intelligent unmanned aerial vehicle patrol. The method generates electronic fence data by defining a flight corridor and its boundary, and constructs an environment cognition model by fusing airborne sensor data collected by a patrol unmanned aerial vehicle with the electronic fence data, positioning data and external environment data. Based on the model, a hybrid decision algorithm is used to synchronously process preset flight rules and real-time perceived obstacle information, generate flight control instructions and execute them, and the environment cognition model is updated in real time based on perception feedback. The system of the application comprises a patrol unmanned aerial vehicle, a ground control center and a distributed autonomous cabin station. The intelligent obstacle avoidance method and control system for unmanned aerial vehicle patrol along a traffic route can effectively improve the obstacle avoidance capability and patrol efficiency of the unmanned aerial vehicle in a complex environment, ensure flight safety, and is suitable for efficient patrol tasks along a traffic route.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle intelligent patrol, in particular to an intelligent obstacle avoidance method and control system for unmanned aerial vehicle patrol along a traffic route. BACKGROUND

[0002] With the continuous development of transportation infrastructure, the patrol work of traffic routes becomes more and more important. The traditional patrol method mainly relies on manual inspection, which is not only low in efficiency and high in cost, but also difficult to implement in some complex terrain or adverse weather conditions. In recent years, the rapid development of unmanned aerial vehicle technology has provided a new solution for traffic route patrol. Unmanned aerial vehicles have high flexibility, wide coverage, low cost and other advantages, which can effectively improve the patrol efficiency and quality.

[0003] However, unmanned aerial vehicles face many challenges in the process of traffic route patrol, the most important of which is obstacle avoidance. The environment along the traffic route is complex, and there are various static and dynamic obstacles such as trees, buildings, vehicles, pedestrians, etc. Unmanned aerial vehicles need to perceive these obstacles in real time and perform effective obstacle avoidance during flight to ensure flight safety and the smooth completion of patrol tasks. In addition, unmanned aerial vehicles also need to comply with the pre-set flight rules, such as flying within the designated flight corridor and avoiding entering the no-fly area.

[0004] Existing unmanned aerial vehicle obstacle avoidance technologies mainly rely on single sensors or simple obstacle avoidance algorithms, which have limitations in complex environments. For example, the obstacle avoidance method based on visual sensors is not effective in poor lighting conditions or low visibility; the obstacle avoidance method based on laser radar has high precision, but the cost is high and it is easily disturbed in some cases. In addition, most of the existing obstacle avoidance algorithms can only deal with static obstacles, and the obstacle avoidance ability for dynamic obstacles is weak.

[0005] Therefore, an intelligent obstacle avoidance method and control system that can integrate multiple sensor data, perceive complex environments in real time and effectively avoid obstacles is needed to meet the needs of unmanned aerial vehicle patrol along the traffic route. SUMMARY

[0006] The purpose of the present application is to provide an intelligent obstacle avoidance method and control system for unmanned aerial vehicle patrol along a traffic route, which can effectively improve the obstacle avoidance ability and patrol efficiency of unmanned aerial vehicles in complex environments, ensure flight safety, and be suitable for efficient patrol tasks along the traffic route.

[0007] To achieve the above purpose, the present application provides an intelligent obstacle avoidance method for unmanned aerial vehicle patrol along a traffic route, comprising the following steps:

[0008] Step S1, defining a flight corridor and its boundary for the target traffic route, and generating electronic fence data;

[0009] Step S2, the patrol unmanned aerial vehicle collects the on-board sensor data, and fuses the electronic fence data, the positioning data and the external environment data to construct an environment cognition model;

[0010] Step S3, based on the environment cognition model, a hybrid decision algorithm is adopted to synchronously process the preset flight rules and the real-time perceived obstacle information, and flight control instructions are generated;

[0011] Step S4, the flight control instructions are executed, and the environment cognition model is updated in real time based on the perception feedback.

[0012] Preferably, in step S1, the flight corridor and its boundary for the target traffic line are defined, and the electronic fence data are generated, including:

[0013] The flight corridor is located within a range of 200 meters extending outward from the roadbed of the traffic line;

[0014] The center line of the traffic line is taken as a reference, and the reference line is generated at a position 100 meters away from the patrol side;

[0015] The outer boundary line of the corridor is generated by offsetting 90 meters away from the traffic line direction from the reference line;

[0016] The inner boundary line of the corridor is generated by offsetting 90 meters close to the traffic line direction from the reference line;

[0017] The nearest distance between the inner boundary line and the traffic line is 10 meters;

[0018] Thus, a flight corridor with a width of 180 meters is formed;

[0019] The functional expression of the reference line is constructed based on the survey data of the traffic line:

[0020]

[0021] Wherein, m represents the slope of the straight line, x represents the horizontal axis coordinate, b represents the vertical axis intercept of the straight line, C represents the control point, n represents the total number of control points, xi represents the i-th sampling point, B represents the cubic B-spline basis function, and t represents the parameter variable.

[0022] ​​​​​​​​​​​​​​​​​The geometric information of the flight corridor is electronic fence data, including a function expression of a reference line , position parameters of an outer boundary line and an inner boundary line , and boundary tolerances .

[0023] Preferably, in step S2, the airborne sensor data includes: an image sequence collected by a visual sensor , point cloud data collected by a laser radar , radar point traces collected by a millimeter wave radar , and body angular velocity and acceleration collected by an inertial measurement unit;

[0024] The positioning data includes: a UAV geodetic coordinate , an altitude , and pose information , wherein represents a latitude, represents a longitude, , y , z respectively represent three-dimensional space coordinates, represents a roll angle, represents a pitch angle, represents a yaw angle;

[0025] The external environment data includes: monitoring data uploaded by a fixed perception node , a wind speed vector , and precipitation data , the monitoring data including real-time video frames taken by a line camera, heat source information detected by an infrared sensor, abnormal vibration signals collected by a vibration sensor, and line deformation data returned by a fiber strain sensor.

[0026] Preferably, in step S2, the specific process of constructing the environment cognition model is as follows:

[0027] Through a multi-source data fusion filter, the airborne sensor data, the positioning data, and the external environment data are time-space synchronized and unified in a coordinate system;

[0028] The fused perception data are clustered and feature-extracted to identify a static obstacle set and a dynamic obstacle set , and the specific process is as follows:

[0029] Clustering processing: a density clustering algorithm DBSCAN is used to cluster the three-dimensional point cloud data in the geodetic coordinate system, and a neighborhood radius = 0.5 meter, minimum number of points MinPts = 5, and the points in the space are aggregated into candidate obstacle clusters;

[0030] Feature extraction: calculate the volume , surface area , centroid coordinates , and average speed of each obstacle cluster;

[0031] Dynamic judgment: when the average speed of the obstacle cluster is greater than 0.3 m / s and the speed change rate is less than 5% in the last 3 sampling periods, it is determined as a dynamic obstacle ; otherwise, it is determined as a static obstacle ;

[0032] For each obstacle in the environment recognition model , its speed vector is estimated by combining its historical motion trajectory and the body motion parameters of the inertial measurement unit, and its future trajectory is predicted within a certain time period;

[0033] Finally, the environment recognition model is expressed as a tuple containing the following elements:

[0034] .

[0035] Preferably, in step S3, based on the environment recognition model, a hybrid decision algorithm is used to process the preset flight rules and the real-time perceived obstacle information simultaneously to generate flight control instructions, including the following steps:

[0036] Step S31, calculate the Euclidean distance from the current body position to the nearest corridor boundary , and define the rule compliance cost function :

[0037] ;

[0038] wherein , are scaling coefficients;

[0039] Step S32, for each obstacle in the environment recognition model , calculate its collision risk with the body, and obtain the total obstacle avoidance cost :

[0040] ;​

[0041] wherein, represents the current distance between the aircraft and the first obstacle, represents a risk coefficient set according to the obstacle type and the relative speed, represents a positive value;

[0042] Step S23, the generation of flight control instructions is converted into a multi-objective optimization problem, whose cost function is:

[0043] ;

[0044] wherein, represents the cost of measuring flight comfort, represents the weight of the rule compliance cost, represents the weight of the obstacle avoidance cost, represents the weight of the flight comfort cost;

[0045] The optimal control instruction sequence that minimizes the cost function is obtained by gradient descent method or model predictive control framework online, and the first instruction is taken as the flight control instruction at the current time, and when the wind force reaches 7 and above, an automatic landing instruction or an instruction that only allows forward flight is generated, and flight towards the direction close to the traffic line is prohibited.

[0046] Preferably, in step S3, the weight of the rule compliance cost and the weight of the obstacle avoidance cost are dynamically adjusted based on the environmental risk factor:

[0047] ;

[0048] ;

[0049] wherein, represents the reference weight of the rule compliance cost, represents the reference weight of the obstacle avoidance cost, , both represent adjustment coefficients, represents the environmental risk factor;

[0050] The environmental risk factor is calculated as follows:

[0051] ;

[0052] wherein, , represent weighting coefficients.

[0053] Preferably, when solving the cost function, the trajectory​ As a constraint condition into the optimization framework, the specific process is:

[0054] Set the prediction horizon , for each dynamic obstacle , extract its trajectory point set in the prediction horizon , denotes the time variable, denotes the current time;

[0055] Define the candidate trajectory point set of the unmanned aerial vehicle in the prediction horizon as , denotes The candidate position of the unmanned aerial vehicle at the moment;

[0056] Establish collision constraints: for all and , require The Euclidean distance between ≥ , denotes the preset safety distance;

[0057] Establish corridor constraints: for all , require Located between and , and the distance from is not less than 0;

[0058] The collision constraint, corridor constraint is embedded in the optimization solving process of the multi-objective cost function, the constraint condition is converted into the cost item through the Lagrange multiplier method, or the candidate solution that violates the constraint is punished in the gradient descent iteration, Ensure that the generated future instruction sequence meets the safety and rule requirements in the entire prediction horizon.

[0059] Preferably, in step S4, the environment cognition model is updated in real time based on the perception feedback, including:

[0060] Match new observation data with existing obstacles through data association algorithm, and update obstacle state estimation using Bayesian filtering;

[0061] After refreshing the environment cognition model, recalculate the cost function and solve the new control instruction to realize closed-loop decision update.

[0062] The application also provides an intelligent patrol control system for implementing the above method, comprising:

[0063] The patrol unmanned aerial vehicle is equipped with a sensor unit, a computing unit, a positioning unit and a communication unit;

[0064] ​A sensor unit is configured to collect airborne sensor data.

[0065] A positioning unit is configured to obtain geodetic coordinates, altitude and pose information of the UAV.

[0066] A communication unit is configured to interact with the ground control center and the distributed autonomous cabin station.

[0067] A computing unit is configured to run a hybrid decision algorithm, generate flight control instructions in real time according to the environmental cognition model, and control the UAV to automatically land or only allow forward flight when the wind reaches 7 or above, and prohibit flight towards the direction close to the traffic line.

[0068] A ground control center is configured to configure and issue electronic fence data and task instructions.

[0069] A distributed autonomous cabin station provides take-off and landing, charging and data interaction services.

[0070] Therefore, the intelligent obstacle avoidance method and control system for the UAV patrol along the traffic line have the following beneficial technical effects:

[0071] (1) By fusing multiple airborne sensor data (such as vision sensor, laser radar, millimeter wave radar, inertial measurement unit, etc.), the application can construct a high-precision environmental cognition model to realize real-time perception of static and dynamic obstacles along the traffic line. Compared with the obstacle avoidance method of a single sensor, multi-source data fusion significantly improves the accuracy and reliability of obstacle identification, especially in complex environments and adverse weather conditions.

[0072] (2) The application adopts a hybrid decision algorithm to simultaneously process the preset flight rules and the real-time perceived obstacle information to generate optimal flight control instructions. This algorithm not only considers flight safety, but also takes into account flight efficiency and comfort. Through a multi-objective optimization framework, the system can dynamically adjust the obstacle avoidance strategy while complying with the flight rules to ensure safe flight of the UAV in complex environments. For example, when the wind reaches 7 or above, the system will automatically generate automatic landing instructions or instructions that only allow forward flight, prohibiting flight towards the direction close to the traffic line, significantly improving flight safety.

[0073] In the hybrid decision algorithm of the application, the weights of rule compliance cost and obstacle avoidance cost can be dynamically adjusted based on environmental risk factors. This dynamic adjustment mechanism enables the system to maintain optimal obstacle avoidance performance in different environmental conditions. In high-risk environments, the system will tend to comply with flight rules and avoid high-risk maneuvers, thereby improving flight safety and reliability.

[0074] (3) Through data association algorithms and Bayesian filtering, this invention can update the environmental cognition model in real time, ensuring that the UAV's perception of the surrounding environment remains up-to-date. This closed-loop decision-making update mechanism enables the UAV to respond quickly to environmental changes, adjust its flight path in a timely manner, and avoid collision risks. For example, when a newly observed obstacle matches an existing obstacle, the system updates the obstacle's state estimate, recalculates the cost function, and generates new control commands to ensure the safety and effectiveness of the flight path.

[0075] (4) This invention generates electronic fence data by defining flight corridors and their boundaries, ensuring that drones fly within safe areas. These preset flight rules not only improve flight safety but also optimize flight paths, reduce unnecessary detours, and improve patrol efficiency. For example, drones can automatically avoid no-fly zones near traffic routes during flight and can quickly adjust their paths when encountering obstacles to continue completing patrol tasks. Attached Figure Description

[0076] Figure 1 This is a flowchart of the intelligent obstacle avoidance method for unmanned aerial vehicle (UAV) patrol along traffic routes according to the present invention.

[0077] Figure 2 This is a schematic diagram of an environmental cognition model. Detailed Implementation

[0078] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0079] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0080] Example 1

[0081] like Figure 1 As shown, the intelligent obstacle avoidance method for drone patrols along transportation routes includes the following steps:

[0082] Step S1: Define the flight corridor and its boundaries for the target traffic route, and generate electronic fence data, including:

[0083] The flight corridor is located within 200 meters outward from the roadbed of the transportation route (i.e., the drone flies within 200 meters along one side of the transportation route).

[0084] Using the centerline of the traffic route as a reference, extend 100 meters to the patrol side (the side furthest from the core area of ​​the traffic route) as the baseline. ;

[0085] From the baseline Offset 90 meters away from the traffic route direction to generate the outer boundary line of the corridor ;

[0086] Offset 90 meters away from the traffic route direction to generate the outer boundary line of the corridor ; ;

[0087] Offset 90 meters away from the traffic route direction to generate the outer boundary line of the corridor ; The nearest distance from the traffic route is 10 meters (i.e. the UAV is prohibited to enter the area within 10 meters from the traffic route);

[0088] Thus, a flight corridor with a width of 180 meters (the distance from the outer boundary to the inner boundary) is formed.

[0089] The function expression of the reference line is constructed based on the survey data of the traffic route:

[0090] ;

[0091] wherein, represents the slope of the straight line, represents the horizontal axis coordinate, represents the vertical axis intercept of the straight line, represents the control point, represents the total number of control points, represents the th sampling point, represents the cubic B-spline basis function, represents the parameter variable.

[0092] The geometric information of the flight corridor is the electronic fence data, including the function expression of the reference line , the position parameters of the outer boundary line and the inner boundary line , and the boundary tolerance .

[0093] The UAV can fly away from the traffic route direction (outer side), but is prohibited to fly towards the traffic route direction ; when encountering a wind of 7 levels or above, it can automatically land or fly forward, and is prohibited to fly towards the traffic route direction.

[0094] Step S2, the patrol UAV collects the on-board sensor data, and fuses the electronic fence data, positioning data and external environment data to construct an environment cognition model, such as Figure 2 .

[0095] The on-board sensor data includes: image sequence collected by the vision sensor, radar point tracks collected by millimeter wave radar and body angular velocity collected by inertial measurement unit and acceleration .

[0096] The positioning data includes: the unmanned aerial vehicle geodetic coordinates obtained through global navigation satellite system real-time differential technology , altitude , and high-frequency and high-precision pose information calculated after fusing inertial measurement unit data , represents latitude, represents longitude, , y , z respectively represent three-dimensional space coordinates, represents the roll angle, represents the pitch angle, represents the yaw angle.

[0097] The external environment data includes: monitoring data uploaded by fixed sensing nodes along the line received from the ground control center , the monitoring data specifically includes real-time video frames shot by cameras along the line, heat source information detected by infrared sensors, abnormal vibration signals collected by vibration sensors, and line deformation data returned by fiber strain sensors; and refined wind speed vectors and precipitation data covering the patrol path obtained from a weather data service interface .

[0098] The specific process of constructing the environment cognition model is as follows:

[0099] Through a multi-source data fusion filter, the airborne sensor data, the positioning data and the external environment data are time and space synchronized and the coordinate systems are unified.

[0100] Time synchronization: taking the PPS second pulse signal output by the unmanned aerial vehicle global navigation satellite system module as the reference time mark, the time stamps of various sensor data are calibrated, and through linear interpolation compensation, the sampling delays of different devices are compensated, so that all data are aligned to the same time axis.

[0101] Coordinate unification: the point cloud data collected by the laser radar and the radar point tracks collected by the millimeter wave radar are converted from the sensor coordinate system to the body coordinate system through the rigid transformation matrix, and then combined with the pose information to be converted to the geodetic coordinate system WGS84; the image sequence is projected to the geodetic coordinate system through the conversion relationship between the camera intrinsic matrix and the camera extrinsic matrix and the body coordinate system; the monitoring data is converted to the geodetic coordinate system according to the installation coordinates of the fixed sensing nodes.

[0102] The fused perception data is clustered and features are extracted to identify a set of static obstacles and a set of dynamic obstacles The specific process is as follows:

[0103] Clustering: The DBSCAN density clustering algorithm is used to cluster the three-dimensional point cloud data in the geodetic coordinate system, the neighborhood radius is set to 0.5 meters, and the minimum number of points MinPts is set to 5, and the adjacent points in space are aggregated into a candidate obstacle cluster.

[0104] Feature extraction: The volume, surface area, centroid coordinates, and average speed of each obstacle cluster are calculated.

[0105] Dynamic judgment: When the average speed of the obstacle cluster is greater than 0.3 meters per second and the speed change rate is less than 5% in the last three sampling periods, it is determined to be a dynamic obstacle ; otherwise, it is determined to be a static obstacle .

[0106] For each obstacle in , the speed vector is estimated based on its historical motion trajectory and the body motion parameters of the inertial measurement unit, and the future trajectory of the obstacle in the next time is predicted.

[0107] Finally, the environment perception model is expressed as a tuple containing the following elements:

[0108] .

[0109] Step S3, based on the environment perception model, a hybrid decision algorithm is used to process the preset flight rules and real-time perceived obstacle information simultaneously to generate flight control instructions, including the following steps:

[0110] Step S31, calculate the rule compliance cost: calculate the Euclidean distance between the current position of the body and the nearest corridor boundary ( denotes the corridor boundary, which is or ), and define a rule compliance cost function :

[0111] ​​​​​​​​ ;

[0112] wherein, , both represent scaling factors;

[0113] This function ensures that when the UAV is close to or beyond the boundary, the value increases sharply, and a larger value is set for the case of close to the traffic line to strictly limit the flight in the direction close to the traffic line.

[0114] Step S32, calculate obstacle avoidance cost: for each obstacle in the environment awareness model , calculate its collision risk with the body, and the total obstacle avoidance cost is obtained by integration:

[0115] ;

[0116] wherein, represents the current distance between the body and the th obstacle, represents the risk coefficient set according to the obstacle type and relative speed, represents a very small positive value, used to prevent division by zero error.

[0117] Step S23, multi-objective cost function optimization: the generation problem of flight control instruction is converted into a multi-objective optimization problem, and the cost function of which is:

[0118] ;

[0119] wherein, represents the cost of measuring flight comfort, represents the weight of rule compliance cost, represents the weight of obstacle avoidance cost, represents the weight of flight comfort cost.

[0120] By gradient descent method or model predictive control framework, the optimal control instruction sequence that minimizes is obtained online, and the first instruction is taken as the flight control instruction at the current time, and when it is detected that the wind force reaches 7 or above, an automatic landing instruction or an instruction that only allows forward flight is generated, and flight in the direction close to the traffic line is prohibited.

[0121] The weight of rule compliance cost and the weight of obstacle avoidance cost are dynamically adjusted based on the environmental risk factor:

[0122] ​;

[0123] ;

[0124] wherein, denotes the baseline weight of the rule compliance cost, denotes the baseline weight of the obstacle avoidance cost, , both denote the adjustment coefficient, denotes the environmental risk factor;

[0125] The environmental risk factor is calculated as follows:

[0126] ;

[0127] wherein, , denote the weighting coefficient.

[0128] This strategy makes the system more inclined to comply with the rules and stick to the preset path in harsh environments, rather than performing high-risk maneuver to avoid obstacles, especially when the wind reaches 7 levels and above, significantly improving to strictly limit the flight direction close to the traffic line.

[0129] When solving the cost function, the trajectory is taken as a constraint condition into the optimization framework, and the specific process is as follows:

[0130] Set the prediction time domain , and for each dynamic obstacle , extract its trajectory point set , denotes the time variable, denotes the current time;

[0131] Define the candidate trajectory point set of the unmanned aerial vehicle in the prediction time domain as , denotes the candidate position of the unmanned aerial vehicle at time, which is generated by the control instruction sequence to be solved.

[0132] Establish collision constraints: for all and , require and Euclidean distance ≥ , denotes the preset safety distance.

[0133] Establish corridor constraints: for all , require to be located and Between, and distance The distance must be no less than 0 (i.e., crossing towards the direction closest to the traffic line is prohibited). ).

[0134] Collision constraints and corridor constraints are embedded in the optimization process of the multi-objective cost function. The constraints are transformed into cost terms through the Lagrange multiplier method, or penalties are imposed on candidate solutions that violate the constraints during gradient descent iterations, to ensure that the generated future instruction sequence meets safety and rule requirements throughout the entire prediction time domain.

[0135] Step S4: Execute flight control commands and update the environmental cognition model in real time based on perception feedback.

[0136] The environmental cognition model is updated in real time based on perceptual feedback, including:

[0137] New observation data is matched with existing obstacles using a data association algorithm, and Bayesian filtering is used to update the obstacle state estimate.

[0138] After updating the environmental cognition model, the cost function is recalculated and new control commands are solved to achieve closed-loop decision update.

[0139] An intelligent patrol control system for implementing the above method includes:

[0140] Patrol drones are equipped with sensor units, computing units, positioning units, and communication units;

[0141] The sensor unit is used to collect data from airborne sensors;

[0142] The positioning unit is used to acquire the UAV's geodetic coordinates, altitude, and pose information.

[0143] The communication unit is used for data interaction with the ground control center and distributed autonomous modules.

[0144] The computing unit is used to run the hybrid decision-making algorithm, generate flight control commands in real time based on the environmental cognition model, and control the drone to automatically land or only allow it to fly forward when the wind force reaches level 7 or above, and prohibit it from flying in the direction of traffic lines.

[0145] The ground control center is used to configure and issue electronic fence data and task instructions;

[0146] Distributed autonomous module stations provide takeoff and landing, charging, and data exchange services.

[0147] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0148] Therefore, the intelligent obstacle avoidance method and control system for the unmanned aerial vehicle to patrol along the traffic line can effectively improve the obstacle avoidance capability and patrol efficiency of the unmanned aerial vehicle in a complex environment, ensure flight safety, and are suitable for efficient patrol tasks along the traffic line.

[0149] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. An intelligent obstacle avoidance method for unmanned aerial vehicle patrol along a traffic route, characterized in that, The method comprises the following steps: Step S1, defining a flight corridor and its boundary for a target traffic line, and generating electronic fence data; Step S2, collecting airborne sensor data by a patrol unmanned aerial vehicle, and fusing the electronic fence data, positioning data and external environment data to construct an environment cognition model; Step S3, based on the environment cognition model, using a hybrid decision algorithm to synchronously process preset flight rules and real-time perceived obstacle information, and generating flight control instructions; Step S4, executing the flight control instructions, and updating the environment cognition model in real time based on perception feedback; Step S3 comprises the following steps: Step S31, Current position of the computer body To the nearest corridor boundary Euclidean distance and define the rules to follow the cost function. : ； wherein , each represent a scaling factor; Step S32, for each obstacle in the environment cognition model , calculate its collision risk with the body , and get the total obstacle avoidance cost : ； wherein, represents the current distance of the machine from the first obstacle, represents a risk factor set according to the obstacle type and the relative speed, represents a positive value; Step S23, the generation problem of flight control instructions is converted into a multi-objective optimization problem, whose cost function is: ； wherein, represents a cost of flight comfort, represents a weight of the cost of rule compliance, represents a weight of the cost of obstacle avoidance, represents a weight of the cost of flight comfort; By solving online using gradient descent or model predictive control framework, we can obtain... Minimize the optimal control command sequence and take the first command as the flight control command at the current moment. When the wind force is detected to reach level 7 or above, generate an automatic landing command or a command that only allows forward flight and prohibits flight towards traffic routes. The weight of the rule compliance cost and the weight of the obstacle avoidance cost are dynamically adjusted based on an environment risk factor: ； ； wherein, represents a reference weight of a rule compliance cost, represents a reference weight of an obstacle avoidance cost, , each represents a regulation coefficient, represents an environmental risk factor; The environment risk factor is calculated as follows: ； wherein , denotes a weighting factor, denotes a wind speed vector, denotes precipitation data. 2.The intelligent obstacle avoidance method for UAV patrol along a traffic route according to claim 1, wherein, In step S1, the flight corridor and its boundary are defined for the target traffic line, and the electronic fence data is generated, comprising: The flight corridor is located within a range of 200 meters extending outward from the roadbed of the traffic line; The center line of the traffic line is taken as the reference, and the reference line is extended 100 meters to the patrol side ; from the datum offset 90 meters from the traffic route direction, generating the outer boundary line of the corridor ; from the datum offset 90 meters in the direction of the traffic route, generating the inner boundary line of the corridor ; and the inner boundary line The nearest distance to the traffic route is 10 meters; Thus, a flight corridor with a width of 180 meters is formed; reference line The functional expression of the reference line is constructed based on survey data of the traffic route: ； wherein, represents a straight line slope, represents a horizontal axis coordinate, represents a straight line vertical axis intercept, represents a control point, represents a total number of control points, represents a th sampling point, represents a cubic B-spline basis function, represents a parameter variable; The geometry information of the flight corridor is the electronic fence data, including the function expression of the reference line , the position parameters and boundary tolerance of the outer boundary line and the inner boundary line . . 3.The intelligent obstacle avoidance method for UAV patrol along a traffic route according to claim 1, wherein, In step S2, the on-board sensor data comprises: image sequences captured by vision sensors , point cloud data captured by lidar , radar tracks captured by millimeter wave radar , and body angular velocity and acceleration captured by inertial measurement units The positioning data includes: unmanned aerial vehicle geodetic coordinates , altitude , pose information , wherein, indicates latitude, indicates longitude, respectively indicates three-dimensional space coordinates, indicates the roll angle, indicates the pitch angle, indicates the yaw angle; The external environment data includes monitoring data uploaded by the fixed perception nodes , wind speed vectors , and precipitation data The monitoring data includes real-time video frames captured by cameras along the line, heat source information detected by infrared sensors, abnormal vibration signals collected by vibration sensors, and line deformation data returned by fiber strain sensors. 4.The intelligent obstacle avoidance method for UAV patrol along a traffic route according to claim 1, wherein, In step S2, the specific process of constructing the environment cognition model is as follows: Through a multi-source data fusion filter, the airborne sensor data, positioning data and external environment data are synchronized in time and space and unified in the coordinate system; The fused perception data is clustered and features are extracted to identify a set of static obstacles and a set of dynamic obstacles The specific process is as follows: Clustering processing: the density clustering algorithm DBSCAN is used to cluster the three-dimensional point cloud data in the geodetic coordinate system, the neighborhood radius is set to 0.5 meters, the minimum number of points MinPts is set to 5, and the points adjacent in space are aggregated into a candidate obstacle cluster. =0.5 meters, the minimum number of points MinPts=5, and the points adjacent in space are aggregated into a candidate obstacle cluster. Feature extraction: compute volume , surface area , centroid coordinates , and average velocity for each obstacle cluster; Dynamicity judgment: when the average speed of the cluster of obstacles > 0.3 m / s and the speed change rate is less than 5% in the last 3 sampling periods, it is determined as a dynamic obstacle ; otherwise, it is determined as a static obstacle ; For each obstacle in , its velocity vector is estimated by combining its historical trajectory with the body motion parameters of the inertial measurement unit and its future trajectory is predicted for a certain time horizon; Ultimately, the environmental awareness model is expressed as a tuple comprising the following elements: 。 5.The intelligent obstacle avoidance method for UAV patrol along a traffic route according to claim 1, wherein, When solving the cost function, the trajectory The constraints are incorporated into the optimization framework as follows: Setting a prediction horizon , for each dynamic obstacle , extracting a set of trajectory points thereof within the prediction horizon , denotes a time variable, denotes the current time instant; The candidate trajectory point set of the UAV in the prediction time domain is defined as , represents the candidate position of the UAV at the moment Establish collision constraints: for all and , require Euclidean distance of ≥ , denotes a preset safety distance; Establish corridor constraints: for all , require to be between and , and distance to is not less than 0; The collision constraint and the corridor constraint are embedded in the optimization solving process of the multi-objective cost function, and the constraint conditions are converted into cost items by the Lagrange multiplier method, or the candidate solutions that violate the constraints are punished in the gradient descent iteration, to ensure that the generated future instruction sequence meets the safety and rule requirements in the entire prediction time domain. 6.The intelligent obstacle avoidance method for UAV patrol along a traffic route according to claim 1, wherein, In step S4, the environment cognition model is updated in real time based on perception feedback, comprising: Through a data association algorithm, new observation data is matched with existing obstacles, and a Bayesian filter is used to update the obstacle state estimation; After refreshing the environment cognition model, the cost function is recalculated and a new control instruction is solved to realize closed-loop decision updating.

7. An intelligent guard control system for implementing the method of any one of claims 1 to 6, characterized by Comprise: The patrol unmanned aerial vehicle is equipped with a sensor unit, a computing unit, a positioning unit and a communication unit; The sensor unit is used to collect airborne sensor data; The positioning unit is used to obtain the geodetic coordinates, altitude and pose information of the unmanned aerial vehicle; The communication unit is used for data interaction with the ground control center and the distributed autonomous cabin station; The computing unit is used to run a hybrid decision algorithm to generate flight control instructions in real time according to the environment cognition model, and when wind force reaches 7 or above, the unmanned aerial vehicle is controlled to automatically land or only allowed to fly forward, and prohibited to fly towards the direction close to the traffic line; The ground control center is used to configure and issue electronic fence data and task instructions; The distributed autonomous cabin station provides take-off and landing, charging and data interaction services.

Images