Online obstacle avoidance route planning method based on random point scattering

Abstract

The invention discloses an online obstacle avoidance route planning method based on random scattering, and belongs to the field of unmanned flight route planning, and the method specifically comprises the steps: firstly, obtaining initial position coordinates of a starting point and a target point in a to-be-measured obstacle region, and converting the initial position coordinates into a Mercator plane coordinate system; then, generating a minimum rectangle or square for the obstacle area; and constructing a key point set Sets by adopting a Gaussian bias random point scattering strategy. Thirdly, calculating a comprehensive evaluation function of each key point, and selecting the key point corresponding to the minimum value as an optimal waypoint at the next moment; and judging the feasibility of the minimum value of the comprehensive evaluation, outputting the waypoint to a flight control system when the minimum value of the comprehensive evaluation is feasible, and starting real-time re-planning of the next period as a new starting point. Otherwise, falling into local deadlock or no solution, returning to increase the standard deviation in the Gaussian bias random point scattering strategy, and forcing the search range to diffuse to the region deviating from the target until a feasible solution is found. The task can be completed safely and efficiently.

Description

Technical Field

[0001] This invention belongs to the field of unmanned flight path planning, specifically involving an online obstacle avoidance path planning method based on random point placement. Background Technology

[0002] As drones are increasingly used in logistics, agricultural monitoring, and disaster relief, the demand for drone flight path planning is also growing. Advances in sensor technology, communication technology, and computing power have provided strong technical support for drone task allocation and flight path planning.

[0003] Unmanned aerial vehicle (UAV) route planning refers to designing a flight path for a UAV from its origin to its destination to ensure that it can complete its mission safely and efficiently. Route planning needs to consider a variety of factors, including environmental constraints, mission requirements, and UAV performance.

[0004] Common UAV path planning methods include global planning algorithms, such as A*, Dijkstra's algorithm, and RRT, used to generate the optimal path from start to finish in a known environment; local planning algorithms (such as dynamic windowing and artificial potential field methods) used for real-time obstacle avoidance and path adjustment; and multi-objective optimization algorithms used to simultaneously optimize path length, energy consumption, and safety. In addition, path smoothing techniques, such as Bézier curves, are often used to optimize flight trajectories and ensure stable UAV flight. These methods, combined with environmental modeling and real-time sensor data, can effectively handle both static and dynamic obstacles, achieving safe and efficient flight.

[0005] The paper "A UAV Obstacle Avoidance Route Planning Method Based on Heading Control Law" (publication number CN110320931A) detects and avoids circular obstacles, taking into account the requirements of the aircraft's lateral control law or actual flight trajectory. The planned route is actually flyable. It uses a hybrid model to model the flight environment, solving the problem that a single environment cannot meet the multiple environmental requirements of obstacle detection and route calculation on the route.

[0006] The patent application CN118655902A, titled "UAV Obstacle Avoidance System Based on Obstacle Behavior Prediction," acquires the changing states of various obstacles in real time through an obstacle state capture module, predicts the behavior logic of obstacles, obtains the current operating speed and operating state of the UAV, predicts the overlap trajectory between the obstacle's operating path and the UAV's operating path, establishes a UAV path avoidance model, establishes an obstacle avoidance operating path for the UAV, and the UAV deviates its operating state according to the operating path to avoid obstacles. It distinguishes different types of obstacles in advance, predicts the operating rules of different types of obstacles, and formulates adaptive avoidance paths for the UAV, ensuring the success rate of avoidance while improving avoidance efficiency and reducing the response time for obstacle avoidance.

[0007] However, existing route planning methods, such as traditional grid-based or uniform sampling methods, often only pursue the shortest geometric path, ignoring the kinematic constraints of the UAV itself. This results in planned paths often containing numerous sharp-angled broken lines, making them difficult for the UAV to track in actual flight, and potentially causing it to stall due to forced turns. Furthermore, when facing complex obstacles, simple uniform discretization strategies lead to an exponential increase in computational load, making it difficult to meet the timeliness requirements of online real-time obstacle avoidance. Summary of the Invention

[0008] To overcome the problems of poor path flyability and low search efficiency in existing technologies, this invention proposes an online obstacle avoidance route planning method based on random point scattering. It comprehensively considers the UAV dynamics model and the environmental potential field distribution. Through a non-uniform probability sampling strategy and a multi-dimensional dynamic cost function, it significantly improves the computational efficiency of UAV mission execution and the stability of path tracking while ensuring obstacle avoidance safety.

[0009] The online obstacle avoidance route planning method based on random point allocation includes the following steps:

[0010] Step 1: Target the obstacle area to be tested. The initial geodetic coordinates of the UAV can be obtained in real time through airborne sensors or data links. and the geodetic coordinates of the target point .

[0011] Initial geodetic coordinates of the UAV Geodetic coordinates of the target point ,

[0012] Represents the initial longitude of the drone; Represents the initial latitude of the drone; This represents the initial altitude of the drone; This represents the initial longitude of the target point; Represents the initial latitude of the target point; This represents the initial altitude of the target point;

[0013] Step 2: Based on the WGS-84 ellipsoid model, the initial geodetic coordinates of the UAV and the geodetic coordinates of the target point are projected and transformed to serve as the starting and ending points of the UAV's planned path.

[0014] The conversion formula is as follows:

[0015]

[0016]

[0017] Corresponding to the horizontal and vertical coordinates under the Mercator projection plane, including the coordinates of the initial plane. Planar coordinates of the target point ; Indicates the initial altitude of the drone; Indicates the initial altitude of the target point;

[0018] The semi-major axis of the Earth's ellipsoid; The first eccentricity of the Earth's ellipsoid is calculated using the following formula: ; Flattening; For the input longitude, including and ; For the input dimensions, including and ;

[0019] Step 3: Based on the obstacle area to be tested Generate a minimum rectangle or square that completely covers the obstacle area.

[0020] The obstacle zones include: polygonal no-fly zones with vertices and typical circular / elliptical no-fly zones.

[0021] When the obstacle region is a polygon with vertices, first traverse and calculate the distance between any two vertices, and select the line connecting the two vertices with the maximum distance as the baseline segment; then discuss different cases:

[0022] (1) If all other vertices are distributed on the same side of the line segment, then each vertex is perpendicular to the baseline. The maximum value of the perpendicular is taken as the length / width of the rectangle, and the baseline segment is taken as the width / length of the rectangle. The range of the minimum rectangle can be obtained.

[0023] (2) If the other vertices are distributed on both sides of the baseline segment, then the baseline segment should be the diagonal of the minimum rectangle. Connect the two vertices on both sides, select the line segment with the maximum distance, and the length is the length / width of the minimum rectangle, and then obtain the range of the minimum rectangle.

[0024] When the obstacle area is a circular / elliptical no-fly zone: For a circular no-fly zone, construct a square with the center of the circle as the center and the diameter as the side length; for an elliptical no-fly zone, construct the smallest rectangle with the center of the ellipse as the center of the rectangle, and the major axis and minor axis of the ellipse as the length and width, respectively.

[0025] Step 4: Construct a set of key points using a Gaussian bias random point-scattering strategy on the four sides of the generated minimum rectangle. .

[0026] The specific steps are as follows:

[0027] Step 401: Set the drone's current position, i.e., its initial coordinates. With the target point Connect the lines between them to form an ideal line of sight. ;

[0028] Step 402, Judgment If there is an intersection point with the boundary of the generated minimum rectangle, proceed to step 403; otherwise, proceed to step 404.

[0029] Step 403 If there is an intersection with the rectangle's boundary, then select the point at a distance from the target point. The nearest intersection point is used as the optimal guiding point. Proceed to step 405;

[0030] Step 404: If the obstacle is located behind the drone or does not obstruct the view, the current flight path is determined to be unobstructed, and the target point is directly targeted. Add to key point set ;

[0031] Step 405, according to the guide point Corresponding position in one-dimensional space Calculate the Gaussian probability density, and then perform Gaussian biased random sampling to obtain the result within a one-dimensional interval. A sequence of samples at randomly scattered locations.

[0032] First, define the position variable in one-dimensional space. ( (This is the total perimeter of the rectangle's boundary), with the top left corner of the rectangle being the origin. );

[0033] Then, with Construct a Gaussian probability density function for the expected mean. Thus obtain A sequence of samples randomly scattered.

[0034]

[0035] in, The standard deviation is used to control the dispersion of the data points.

[0036] Finally, based on this probability distribution, in the interval random sampling Sample sequences at each location.

[0037] Step 406: Using inverse geometric mapping, the generated samples are restored to two-dimensional Mercator coordinates, and all mapped coordinate points form a non-uniform keypoint set. .

[0038] Let the coordinates of the four vertices of the smallest rectangle be the top left corner, bottom left corner, top right corner, bottom left corner, bottom ... Top right corner bottom right corner and bottom left corner ; Rectangle width ,high .

[0039] For the sample The mapping is restored to two-dimensional Mercator coordinates. Starting from the origin and proceeding clockwise, the rectangle is divided into the top segment, right segment, bottom segment, and left segment. The linear function for each segment is calculated as follows:

[0040] The upper section (when) hour):

[0041]

[0042] Right side (when) hour):

[0043]

[0044] The lower section (when) hour):

[0045]

[0046] Left segment (when) hour):

[0047]

[0048] Step 5: Set of key points Each candidate point in Construct a comprehensive evaluation function that includes distance cost, heading angle penalty cost, and safety potential field cost. .

[0049] The comprehensive cost function as follows:

[0050]

[0051] in, These are the dynamic weighting coefficients for each cost item, which are adjusted according to the task stage.

[0052] The distance cost term is calculated using the following formula:

[0053]

[0054] The heading angle penalty term is calculated using the following formula:

[0055]

[0056] The maximum permissible turning angle for the drone. The penalty gain coefficient, The angle between the current heading vector and the desired heading vector;

[0057] For the safety potential field project, the calculation formula is:

[0058]

[0059] in, The shortest Euclidean distance from the key point to the edge of the obstacle. For a safe buffer distance, This is the potential field gradient adjustment coefficient, used to control the decay rate of the repulsive force field of the obstacle.

[0060] Step Six, from Select the minimum value from the comprehensive evaluation functions. The corresponding candidate point is used as the optimal waypoint for the next time step. ;

[0061]

[0062] Step 7: Determine the minimum value of the comprehensive evaluation. Is it a feasible solution, i.e., does it satisfy... If so, proceed to step eight; otherwise, if the comprehensive evaluation of all candidate points tends to infinity, resulting in a local deadlock or no solution, proceed to step nine.

[0063] Step 8: Set waypoints Output to the flight control system and calculate. Distance between current location and target point Therefore, to judge Is it greater than the preset threshold? If so, If the starting point is the new starting point, return to step one and start the real-time replanning for the next cycle; otherwise, the drone has reached the target point.

[0064] Step 9: Return to Step 4 and increase the standard deviation in the Gaussian biased random point distribution strategy. The search range is forced to expand to areas away from the target until a feasible solution is found.

[0065] The advantages of this invention are:

[0066] 1. Improve path search efficiency: This invention adopts a Gaussian bias sampling strategy based on target orientation, replacing the traditional uniform sampling, so that key points are distributed in the optimal path region pointing to the target with a high probability, effectively reducing the amount of computation of invalid nodes and significantly improving the convergence speed of online planning.

[0067] 2. Deadlock escape capability: This invention designs a dynamic variance adjustment mechanism. When the drone is detected to be trapped in a local minimum, it can automatically increase the sampling standard deviation and force the search range to expand to the region away from the target, thereby giving the algorithm the ability to autonomously jump out of deadlock in complex obstacle environments.

[0068] 3. Reduced computational complexity: Direct guidance is achieved in unobstructed environments through line-of-sight intersection detection. Combined with inverse geometric mapping, two-dimensional spatial sampling is transformed into one-dimensional linear spatial sampling, which significantly reduces computational dimensionality and complexity, making it easier to deploy on airborne embedded platforms with limited computing resources.

[0069] 4. Smooth path and satisfy dynamic constraints: A multidimensional comprehensive cost function including a heading angle penalty term was constructed to strictly limit the maximum turning angle of the UAV, ensuring that the generated obstacle avoidance path is smooth and continuous, and conforms to the actual flight kinematic characteristics of fixed-wing or high-speed rotary-wing UAVs. Attached Figure Description

[0070] Figure 1 This is a flowchart of an online obstacle avoidance route planning method based on random point distribution according to the present invention;

[0071] Figure 2 This is a schematic diagram of the Gaussian bias random point distribution of the present invention. Detailed Implementation

[0072] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0073] The online obstacle avoidance route planning method based on random point distribution, such as Figure 1 As shown, the specific steps are as follows:

[0074] Step 1: Data Acquisition and Symbol Definition: Targeting the obstacle area to be tested. The initial geodetic coordinates of the UAV can be obtained in real time through airborne sensors or data links. and the geodetic coordinates of the target point .

[0075] Initial geodetic coordinates of the UAV Geodetic coordinates of the target point ,

[0076] Represents the initial longitude of the drone; Represents the initial latitude of the drone; This represents the initial altitude of the drone; This represents the initial longitude of the target point; Represents the initial latitude of the target point; This represents the initial altitude of the target point;

[0077] Step 2: Based on the WGS-84 ellipsoid model, perform a projection transformation on the initial geodetic coordinates of the UAV and the geodetic coordinates of the target point to obtain the transformed initial plane coordinates of the UAV. Planar coordinates of the target point , serving as the starting and ending points for the drone's planned path.

[0078] To improve the accuracy of coordinate transformation, the nonlinear motion planning problem on the Earth's surface is transformed into a linear programming problem in a Cartesian coordinate system. This embodiment does not simply use a spherical model, but instead performs Mercator projection calculations based on the WGS-84 ellipsoid model.

[0079] First, define the semi-major axis of the Earth's ellipsoid. Flatness The first eccentricity of the Earth's ellipsoid The calculation formula is:

[0080]

[0081] Compared to the simplified formula for a sphere, the introduction of eccentricity... The correction terms effectively reduce projection deformation in high-latitude regions, ensuring the geometric accuracy of subsequent obstacle avoidance calculations. Based on this, the UAV's geographic coordinates... Convert to Mercator projection coordinates The rigorous mathematical expression is:

[0082]

[0083]

[0084]

[0085] Corresponding to the horizontal and vertical coordinates under the Mercator projection plane, including the coordinates of the initial plane. Planar coordinates of the target point ; Indicates the initial altitude of the drone; Indicates the initial altitude of the target point; For the input longitude, including and ; For the input dimensions, including and ;

[0086] Step 3: Based on the obstacle area to be tested Generate a minimum rectangle or square that completely covers the obstacle area.

[0087] The obstacle zones include: polygonal no-fly zones with vertices and typical circular / elliptical no-fly zones.

[0088] When the obstacle region is a polygon with vertices, first traverse and calculate the distance between any two vertices, and select the line connecting the two vertices with the maximum distance as the baseline segment; then discuss different cases:

[0089] (1) If all other vertices are distributed on the same side of the baseline segment, then each vertex is perpendicular to the baseline. The maximum value of the perpendicular is taken as the length / width of the rectangle, and the baseline segment is taken as the width / length of the rectangle. The range of the minimum rectangle can be obtained.

[0090] (2) If the other vertices are distributed on both sides of the baseline segment, then the baseline segment should be the diagonal of the minimum rectangle. Connect the two vertices in pairs, select the line segment with the maximum distance, and the length is the length / width of the minimum rectangle. Then combine this with the diagonal of the minimum rectangle to obtain the range of the minimum rectangle.

[0091] When the obstacle area is a circular / elliptical no-fly zone: For a circular no-fly zone, construct a square with the center of the circle as the center and the diameter as the side length; for an elliptical no-fly zone, construct the smallest rectangle with the center of the ellipse as the center of the rectangle, and the major axis and minor axis of the ellipse as the length and width, respectively.

[0092] Step 4: Constrained random sampling based on Gaussian mixture model: Construct a set of key points using a Gaussian bias random sampling strategy on the four sides of the generated minimum rectangle. .

[0093] To balance the randomness and target orientation of the search, this invention abandons the traditional uniform discretization method and constructs a non-uniform point distribution based on a Gaussian probability density function of the target orientation through three sub-steps: "logical judgment - model construction - geometric mapping". The specific steps are as follows:

[0094] Step 401: Set the drone's current position, i.e., its initial coordinates. With the target point Connect the lines between them to form an ideal line of sight. ;

[0095] Step 402, Judgment If there is an intersection point with the boundary of the smallest rectangle of the generated obstacle, proceed to step 403; otherwise, proceed to step 404.

[0096] Step 403 If there is an intersection with the rectangle's boundary, then select the point at a distance from the target point. The nearest intersection point is used as the optimal guiding point. Proceed to step 405;

[0097] Step 404: If the obstacle is located behind the drone or does not obstruct the view, the current flight path is determined to be unobstructed, and the target point is directly targeted. Add to key point set It skips subsequent sampling steps to maximize flight efficiency.

[0098] Step 405, according to the guide point Corresponding position in one-dimensional space Calculate the Gaussian probability density, and then perform Gaussian biased random sampling to obtain the result within a one-dimensional interval. A sequence of samples at randomly scattered locations.

[0099] To simplify the complex two-dimensional boundary sampling problem into a one-dimensional probability problem, we first define the position variables in one-dimensional space. ( (This is the total perimeter of the rectangle's boundary), with the top left corner of the rectangle being the origin. );

[0100] Then, with Construct a Gaussian probability density function for the expected mean. Thus obtain A sequence of samples randomly scattered.

[0101]

[0102] in, The standard deviation is used to control the dispersion of the data points. A smaller value is set when the drone is in the cruise phase. (like This ensures that the generated keypoints are highly concentrated in the direction pointing to the target, reducing unnecessary computation; when a local minimum is detected, the value is dynamically increased. This allows the range of points to expand to areas away from the target, giving the path planning algorithm the ability to escape deadlock.

[0103] Finally, based on this probability distribution, in the interval random sampling Sample sequences at each location.

[0104] Step 406: Using inverse geometric mapping, the generated samples are restored to two-dimensional Mercator coordinates, and all mapped coordinate points form a non-uniform keypoint set. .

[0105] Let the coordinates of the four vertices of the smallest rectangle be the top left corner, bottom left corner, top right corner, bottom left corner, bottom ... Top right corner bottom right corner and bottom left corner ; Rectangle width ,high .

[0106] For the sample The mapping is restored to two-dimensional Mercator coordinates. From the origin position Located in the top left corner Starting clockwise, the rectangle is divided into the top segment, right segment, bottom segment, and left segment. The linear function calculation for each segment is as follows:

[0107] The upper section (when) hour):

[0108]

[0109] Right side (when) hour):

[0110]

[0111] The lower section (when) hour):

[0112]

[0113] Left segment (when) hour):

[0114]

[0115] This example illustrates the Gaussian biased random point distribution, as shown below. Figure 2 As shown.

[0116] Step 5: Set of key points Each candidate point in Construct a comprehensive evaluation function that incorporates multi-dimensional kinematic constraints, including distance cost, heading angle penalty cost, and safety potential cost. .

[0117] To select the optimal path that satisfies the dynamic characteristics of fixed-wing or high-speed rotary-wing UAVs, a comprehensive cost function is defined. as follows:

[0118]

[0119] in, These are the dynamic weighting coefficients for each cost item, which are adjusted according to the task stage.

[0120] This is the distance cost term, used to ensure the optimality of the path. The calculation formula is as follows:

[0121]

[0122] As a heading angle penalty term, considering the minimum turning radius limit of the UAV, let the current heading vector be... The desired heading vector is The angle between the two The cosine value is:

[0123]

[0124] The formula for calculating the turning penalty function is as follows:

[0125]

[0126] The maximum permissible turning angle for the drone. The penalty gain coefficient, The angle between the current heading vector and the desired heading vector; this formula shows that once the turning angle exceeds the physical limit, the cost will increase exponentially, thus forcing the algorithm to abandon the infeasible point.

[0127] For the safety potential field project, the calculation formula is:

[0128]

[0129] in, The shortest Euclidean distance from the key point to the edge of the obstacle. For a safe buffer distance, This is the potential field gradient adjustment coefficient, used to control the decay rate of the repulsive force field of the obstacle; The larger the value, the "harder" the potential field boundary, and the more violent the reaction to intrusion into the safe zone; The smaller the value, the smoother the change in potential field. This formula ensures that the cost increases smoothly when the drone approaches an obstacle, avoiding abrupt changes in control commands.

[0130] Step Six, from Select the minimum value from the comprehensive evaluation functions. The corresponding candidate point is used as the optimal waypoint for the next time step. ;

[0131]

[0132] This step employs an online rolling planning strategy, using extreme value search and deadlock escape mechanisms to determine the waypoint for the next time step.

[0133] Step 7: Determine the minimum value of the comprehensive evaluation. Is it a feasible solution, i.e., does it satisfy... If so, proceed to step eight; otherwise, if the comprehensive evaluation of all candidate points tends to infinity, resulting in a local deadlock or no solution, proceed to step nine.

[0134] Step 8: Set waypoints Output to the flight control system and calculate. Distance between current location and target point Therefore, to judge Is it greater than the preset threshold? If so, If the starting point is the new starting point, return to step one and start the real-time replanning for the next cycle; otherwise, the drone has reached the target point.

[0135] Step 9: Return to Step 4 and increase the standard deviation in the Gaussian biased random point distribution strategy. The search range is forced to expand to areas away from the target until a feasible solution is found.

[0136] Through the repeated cycle of "perception-sampling-decision-execution", a smooth and complete obstacle avoidance trajectory is eventually formed.

[0137] Example:

[0138] To avoid computational redundancy caused by traditional fixed-step sampling, this embodiment establishes a non-uniform sampling model based on the geometric relationship between the line connecting the current position of the UAV and the target point and the rectangular boundary.

[0139] In practice, the total number of key points is set. First, calculate the intersection point of the line connecting the UAV's position and the target point with the boundary of the rectangle, denoted as... .by The normalized position on the boundary is the mean. Set standard deviation .

[0140] Generate a sequence that follows the Box-Muller transformation The keypoints are randomly distributed. This ensures that approximately 68% of the keypoints automatically fall within the "golden area" pointing towards the target, while the remaining 32% are scattered across other areas to prevent them from getting trapped in local minima. The final generated set... These are the high-value candidate points selected through probability screening.

[0141] Then, the cost function search based on multidimensional dynamic constraints is performed, specifically as follows:

[0142] Traversal calculation: for a set Each candidate point in Calculate the comprehensive cost

[0143]

[0144] Calculate the cost of each item:

[0145] Distance Cost Calculate the total path length between the starting point and the candidate point, and between the candidate point and the ending point, and then normalize the path length.

[0146] Turning Cost : Calculate vectors (Current heading) and vector The angle between the starting point and the candidate point ($) .like (The set maximum turning angle) then let The point is directly removed to ensure the path is flyable.

[0147] Security Cost Introducing a potential field function, if the distance between the candidate point and the obstacle boundary is less than... (Safety threshold), this cost increases exponentially.

[0148] Dynamic weight adjustment: Unlike fixed weights, this embodiment monitors the distance between the drone and obstacles in real time. :

[0149] when (In cruise mode) the weight is automatically adjusted. Prioritize speed; when (In obstacle avoidance mode) the weights are automatically adjusted. Prioritize safety and flight stability. Optimal decision: Select... The point with the minimum value is used as the waypoint for the next time step. It also generates flight control commands.

[0150] In this way, a safe flight route can be planned on a map with no-fly zones, from the starting point to the destination, effectively avoiding obstacles.

Claims

1. A method for online obstacle avoidance route planning based on randomly scattered points, characterized in that, Includes the following steps: Step 1: Target the obstacle area to be tested. The initial geodetic coordinates of the UAV can be obtained in real time through airborne sensors or data links. and the geodetic coordinates of the target point ; Step 2: Based on the WGS-84 ellipsoid model, the initial geodetic coordinates of the UAV and the geodetic coordinates of the target point are projected and transformed to serve as the starting and ending points of the UAV's planned path. Step 3: Based on the obstacle area to be tested Generate a minimum rectangle or square that completely covers the obstacle area; Step 4: Construct a set of key points using a Gaussian bias random point-scattering strategy on the four sides of the generated minimum rectangle. ; The specific steps are as follows: Step 401: Set the drone's current position, i.e., its initial coordinates. With the target point Connect the lines between them to form an ideal line of sight. ; Step 402, Judgment Does it intersect with the boundary of the generated minimum rectangle? If so, proceed to step 403. Otherwise, proceed to step 404; Step 403 If there is an intersection with the rectangle's boundary, then select the point at a distance from the target point. The nearest intersection point is used as the optimal guiding point. Proceed to step 405; Step 404: If the obstacle is located behind the drone or does not obstruct the view, the current flight path is determined to be unobstructed, and the target point is directly targeted. Add to key point set ; Step 405, according to the guide point Corresponding position in one-dimensional space Calculate the Gaussian probability density, and then perform Gaussian biased random sampling to obtain the result within a one-dimensional interval. A sequence of samples at randomly scattered locations; Step 406: Using inverse geometric mapping, the generated samples are restored to two-dimensional Mercator coordinates, and all mapped coordinate points form a non-uniform keypoint set. ; Let the coordinates of the four vertices of the smallest rectangle be the top left corner, bottom left corner, top right corner, bottom left corner, bottom ... Top right corner bottom right corner and bottom left corner ; Rectangle width ,high ; For the sample The mapping is restored to two-dimensional Mercator coordinates. The rectangle's top-left corner is the origin. Starting from the origin and moving clockwise, the rectangle is divided into top, right, bottom, and left segments. The linear function for each segment is calculated as follows: (1) Top paragraph: When hour: (2) Right side: When hour: (3) The lower section: When hour: (4) Left side: When hour: Step 5: Set of key points Each candidate point in Constructing a system that includes distance cost Heading angle penalty cost and the cost of the safe potential field Comprehensive evaluation function ; The comprehensive cost function as follows: in, These are the dynamic weighting coefficients for each cost item, which are adjusted according to the task stage. Step Six, from Select the minimum value from the comprehensive evaluation functions. The corresponding candidate point is used as the optimal waypoint for the next time step. ; Step 7: Determine the minimum value of the comprehensive evaluation. Is it a feasible solution, i.e., does it satisfy... If so, proceed to step eight; otherwise, if the comprehensive evaluation of all candidate points tends to infinity, resulting in a local deadlock or no solution, proceed to step nine. Step 8: Set waypoints Output to the flight control system and calculate. Distance between current location and target point Therefore, to judge Is it greater than the preset threshold? If so, If the starting point is the new starting point, return to step one and start the real-time replanning for the next cycle; otherwise, the drone has reached the target point. Step 9: Return to Step 4 and increase the standard deviation in the Gaussian biased random point distribution strategy. The search range is forced to expand to areas away from the target until a feasible solution is found.

2. The online obstacle avoidance route planning method based on random point allocation as described in claim 1, characterized in that, In step one, the initial geodetic coordinates of the UAV Geodetic coordinates of the target point , Represents the initial longitude of the drone; Represents the initial latitude of the drone; This represents the initial altitude of the drone; This represents the initial longitude of the target point; Represents the initial latitude of the target point; This represents the initial altitude of the target point.

3. The online obstacle avoidance route planning method based on random point allocation as described in claim 2, characterized in that, In step two, the conversion formula is as follows: Corresponding to the horizontal and vertical coordinates under the Mercator projection plane, including the coordinates of the initial plane. Planar coordinates of the target point ; Indicates the initial altitude of the drone; Indicates the initial altitude of the target point; The semi-major axis of the Earth's ellipsoid; The first eccentricity of the Earth's ellipsoid is calculated using the following formula: ; Flattening; For the input longitude, including and ; For the input dimensions, including and .

4. The online obstacle avoidance route planning method based on random point allocation as described in claim 1, characterized in that, In step three, the obstacle areas include: polygonal no-fly zones with vertices and typical circular / elliptical no-fly zones; When the obstacle region is a polygon with vertices, first traverse and calculate the distance between any two vertices, and select the line connecting the two vertices with the maximum distance as the baseline segment; then discuss different cases: (1) If all other vertices are distributed on the same side of the line segment, then each vertex is perpendicular to the baseline. The maximum value of the perpendicular is taken as the length / width of the rectangle, and the baseline segment is taken as the width / length of the rectangle. The range of the minimum rectangle can be obtained. (2) If the other vertices are distributed on both sides of the baseline segment, then the baseline segment should be the diagonal of the minimum rectangle. Connect the two vertices on both sides, select the line segment with the maximum distance, and the length is the length / width of the minimum rectangle, and then obtain the range of the minimum rectangle. When the obstacle area is a circular / elliptical no-fly zone: For a circular no-fly zone, construct a square with the center of the circle as the center and the diameter as the side length; for an elliptical no-fly zone, construct the smallest rectangle with the center of the ellipse as the center of the rectangle, and the major axis and minor axis of the ellipse as the length and width, respectively.

5. The online obstacle avoidance route planning method based on random point allocation as described in claim 1, characterized in that, Step 405 specifically involves: First, define the position variable in one-dimensional space. , This is the total perimeter of the rectangle's boundary. Then, with Construct a Gaussian probability density function for the expected mean. Thus obtain A sample sequence of randomly scattered points; in, Standard deviation is used to control the dispersion of the scattered points; Finally, based on this probability distribution, in the interval random sampling Sample sequences at each location.

6. The online obstacle avoidance route planning method based on random point allocation as described in claim 1, characterized in that, In step five, The distance cost term is calculated using the following formula: The heading angle penalty term is calculated using the following formula: The maximum permissible turning angle for the drone. The penalty gain coefficient, The angle between the current heading vector and the desired heading vector; For the safety potential field project, the calculation formula is: in, The shortest Euclidean distance from the key point to the edge of the obstacle. For a safe buffer distance, This is the potential field gradient adjustment coefficient, used to control the decay rate of the repulsive force field of the obstacle.

Images