Navigation method, system and equipment for dynamic perception and obstacle avoidance based on unmanned aerial vehicle

Abstract

The invention provides a navigation method, system and device for dynamic perception and obstacle avoidance based on an unmanned aerial vehicle, and the method comprises the steps: obtaining a real-time depth map of a to-be-detected region, and determining the instantaneous positions of a dynamic target and a static target in the real-time depth map; establishing a polar coordinate system by taking the instantaneous position of the dynamic target as an original point, discretizing a peripheral space into a polar unit network, constructing a Markov chain based on probability distribution of polar units, and updating a time-varying transfer matrix of the Markov chain based on a transfer probability between the polar units; and selecting a continuous polar unit sequence based on the updated Markov chain, fitting the continuous polar unit sequence by adopting a polynomial trajectory to obtain a prediction path, and proposing dynamic target adaptive trajectory planning. According to the invention, the perception precision and the state estimation stability of the unmanned aerial vehicle to the dynamic and static targets are obviously improved, the real-time response capability of trajectory planning to the motion change of the dynamic targets is enhanced, and the autonomy and the flight safety of the unmanned aerial vehicle in a dynamic environment are ensured.

Description

Technical Field

[0001] This invention relates to the field of UAV perception and obstacle avoidance technology, specifically to a navigation method, system, and device based on UAV dynamic perception and obstacle avoidance. Background Technology

[0002] In recent years, lightweight autonomous drones have been deployed on a large scale in various industrial applications, such as inspection, exploration, and search and rescue, due to their flexibility, low cost, and versatility. These applications typically operate in highly complex and dynamic environments, involving human workers, static structures, robots, and vehicles. However, in complex environments, drones face intricate navigation and obstacle avoidance problems. The limited computing resources and sensor field of view of small drones, coupled with the uncertainty of dynamic targets in dynamic environments, pose significant challenges to real-time perception, state estimation, and obstacle avoidance planning.

[0003] Therefore, there is an urgent need for a navigation method, system, and device based on UAV dynamic perception and obstacle avoidance to address the shortcomings of existing technologies. Summary of the Invention

[0004] The purpose of this invention is to provide a navigation method, system, and device based on dynamic perception and obstacle avoidance of unmanned aerial vehicles (UAVs) to solve the technical problems mentioned in the background.

[0005] To achieve the above objectives, the first aspect of this invention proposes a navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles (UAVs), comprising the following steps: Step 1: Obtain the real-time depth map of the area to be detected, and determine the instantaneous positions of dynamic and static targets in the real-time depth map; Step 2: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. Step 3: Select a continuous polar unit sequence based on the updated Markov chain, and use polynomial trajectory fitting to obtain the predicted path from the continuous polar unit sequence; Step 4: Propose dynamic target adaptive trajectory planning based on the predicted path.

[0006] Furthermore, determining the instantaneous positions of dynamic and static targets in the real-time depth map in step 1 specifically includes: Step 11: Generate the first bounding box based on the real-time depth map using the U-depth Detector; Step 12: Convert the real-time depth map into a point cloud, and use the DBSCAN clustering algorithm to cluster the point cloud converted from the real-time depth map to obtain point cloud clusters and second bounding boxes. Step 13: Fuse the first bounding box and the second bounding box using IoU thresholding to detect 3D obstacles; Step 14: Distinguish between dynamic and static targets in the 3D obstacle to determine the instantaneous positions of dynamic and static targets in the real-time depth map.

[0007] Furthermore, step 14, which distinguishes between dynamic and static targets for 3D obstacles, specifically includes: Step 141: Based on the point cloud and second bounding box obtained in Step 11, the estimated velocity of each point in the point cloud of the 3D obstacle is obtained through nearest neighbor search; Step 142: If the proportion of effective moving points in the point cloud cluster of a 3D obstacle is greater than the first threshold, then the 3D obstacle is determined to be a candidate dynamic target; Step 143: Calculate the mean displacement and normalized average variance of the adjacent frames of the candidate dynamic target point cloud cluster; when the mean displacement is greater than the second threshold and the normalized average variance is less than the third threshold, the candidate dynamic target is determined to be a dynamic target.

[0008] Furthermore, the mean displacement of the point cloud cluster of the candidate dynamic target in step 143 and normalized mean variance The calculation formula is: in, It represents the displacement between the nearest neighbors in adjacent frames. This represents the number of valid points.

[0009] Further, in step 2, a polar coordinate system is established with the instantaneous position of the dynamic target as the origin, the surrounding space is discretized into a polar unit network, the probability distribution of the polar units and the transition probabilities between polar units are determined, a Markov chain is constructed based on the probability distribution of the polar units, and the time-varying transition matrix of the Markov chain is updated based on the transition probabilities between polar units. Specifically, this includes: Step 21: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, and discretize the surrounding space into a polar element network. angular range radius range ; Step 22: Define the state of the Markov chain as the probability distribution of the polar unit, and the initial probability distribution of the polar unit at the initial time. The expression is: in, Normalization factor; () represents the polar coordinates of the dynamic target; Radial direction in polar coordinates ; The angle covariance in polar coordinates. ; Step 23: For dynamic pole units with dynamic targets polar unit To the dynamic pole unit The transition probability; for static pole units with static targets, their probability is set to zero and diffused to adjacent free units. polar unit To adjacent free units The transition probability; Step 24: Update the time-varying transition matrix of the Markov chain based on the transition probabilities between the polar units, where the probability distribution of the polar units satisfies the normalization constraint. .

[0010] Further, step 231: from pole unit (i,j) to dynamic pole unit The transition probability is: in, Radial velocity component of a dynamic target v is the velocity of the dynamic target estimated by the Kalman filter. The direction of the dynamic target estimated by the Kalman filter; For dynamic pole units The central angle; To predict the time step; This indicates that the dynamic target appears in the dynamic pole unit. The probability density, where s is the dynamic pole unit. area, and These represent the position and covariance estimated by Kalman in polar coordinates, respectively. For dynamic pole units The central location; Step 232: From pole unit (i,j) to adjacent free pole unit The transition probability is: in, Adjacent free pole units The Euclidean distance to the static target; Control the spread; It is the sum of the attenuation factors of all adjacent free pole units.

[0011] Furthermore, step 3, which involves selecting a continuous polar unit sequence based on the updated Markov chain and using polynomial trajectory fitting to obtain the predicted path, specifically includes: Step 31: Based on the updated Markov chain, determine the velocity and radius of motion of the dynamic target; Step 32: Select the continuous pole cell sequence with the highest probability percentage along the velocity direction of the dynamic target within the radius of the dynamic target's motion range; Step 33: Using the current position of the dynamic target as the trajectory starting point, map the continuous polar unit sequence in the global coordinate system to obtain the temporal path points of the dynamic target, and obtain the predicted path of the dynamic target based on the temporal path points.

[0012] Furthermore, step 4, which proposes dynamic target adaptive trajectory planning based on the predicted path, specifically includes: Step 41: Front Trajectory Generation: The front trajectory of the dynamic target is generated by sampling key points of the path using the RRT* algorithm, constructing a polynomial trajectory that minimizes the minimum-snap energy based on corridor constraints, and generating a smooth reference trajectory that satisfies the dynamic constraints. Step 42: Back-end trajectory optimization: Back-end trajectory optimization of dynamic targets utilizes the predicted trajectory of dynamic targets to construct a spatiotemporal probability risk field, and optimizes the trajectory in real time using the gradient descent method.

[0013] A second aspect of this invention proposes a lightweight navigation system for dynamic perception and obstacle avoidance for small unmanned aerial vehicles (UAVs), comprising: The differentiation module is used to acquire a real-time depth map of the area to be detected and determine the instantaneous positions of dynamic and static targets in the real-time depth map; The modeling module is used to establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. The prediction module is used to select a continuous pole unit sequence based on the updated Markov chain and obtain the predicted path by fitting the continuous pole unit sequence with a polynomial trajectory. The planning module is used to propose adaptive trajectory planning for dynamic targets based on predicted paths.

[0014] A third aspect of the present invention provides an electronic device comprising: a memory, a processor, and a computer program, the computer program being stored in the memory and configured to be executed by the processor to implement any of the methods described herein.

[0015] The beneficial effects of the technical solution of this invention include: Based on the theory of multimodal perception fusion and dynamic environment modeling, this invention comprehensively considers the navigation and obstacle avoidance challenges of small UAVs under conditions of limited perception range, scarce computing resources and uncertainty of dynamic target motion, and achieves efficient and robust autonomous flight in complex environments. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the dynamic guide point generation based on Gaussian gradient according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the generation of static guide points based on uniform projection according to an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0018] First, some of the technical terms used in this invention will be explained to help those skilled in the art understand the invention.

[0019] Example 1 The first aspect of this invention proposes a navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles (UAVs), such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the real-time depth map of the area to be detected, and determine the instantaneous positions of dynamic and static targets in the real-time depth map; Furthermore, step 1, determining the dynamic and static targets in the real-time depth map, specifically includes: Step 11: Generate the first bounding box based on the real-time depth map using the U-depth Detector; Step 12: Convert the real-time depth map into a point cloud, and use the DBSCAN clustering algorithm to cluster the point cloud converted from the real-time depth map to obtain point cloud clusters and second bounding boxes. Step 13: Fuse the first bounding box and the second bounding box using IoU thresholding to detect 3D obstacles; Step 14: Distinguish between dynamic and static targets in the 3D obstacle to determine the instantaneous positions of dynamic and static targets in the real-time depth map.

[0020] Furthermore, step 14, which distinguishes between dynamic and static targets for 3D obstacles, specifically includes: Step 141: Based on the point cloud and second bounding box obtained in Step 11, the estimated velocity of each point in the point cloud of the 3D obstacle is obtained through nearest neighbor search; Step 142: If the proportion of effective moving points in the point cloud cluster of a 3D obstacle is greater than the first threshold, then the 3D obstacle is determined to be a candidate dynamic target; Step 143: Calculate the mean displacement and normalized average variance of the adjacent frames of the candidate dynamic target point cloud cluster; when the mean displacement is greater than the second threshold and the normalized average variance is less than the third threshold, the candidate dynamic target is determined to be a dynamic target.

[0021] Furthermore, the mean displacement of the point cloud cluster of the candidate dynamic target in step 143 and normalized mean variance The calculation formula is: in, It represents the displacement between the nearest neighbors in adjacent frames. This represents the number of valid points.

[0022] In this embodiment, all candidate dynamic targets are extracted into a set. And record its bounding box geometric center. It efficiently processes cluttered environments and ultimately outputs the 3D position of each dynamic target. And its 3D bounding box dimensions.

[0023] An adaptive Kalman filter is used to estimate the motion state of a dynamic target. The target's state vector is defined as follows: in, This indicates the position coordinates of the dynamic target at the center of the 3D bounding box. Represents the velocity vector. Based on the time interval between frames. , Represents the acceleration vector; The process model can be represented as: in, for identity matrix Indicates process noise. It is the covariance matrix of the process model.

[0024] The observation model is: in, For the sensor at any time The obtained measurement values, Indicates measurement noise. To measure the noise covariance, the observation matrix It is an identity matrix.

[0025] In this embodiment, to adapt to the diversity of dynamic target motion in complex dynamic environments, an adaptive mechanism based on innovation covariance theory is used to update filter parameters. This method dynamically adjusts the process noise covariance and measurement noise covariance by analyzing the observation residual characteristics, i.e., the statistical characteristics of the innovation sequence, in real time, thereby enhancing the model's robustness to changes in motion patterns while ensuring computational efficiency.

[0026] Covariance matrix of the process model The expression is: in, For the new information sequence, This represents the actual covariance. To improve system resource utilization, a dedicated Kalman filter tracking instance is created only when a new dynamic target is detected, thereby achieving persistent state estimation and trajectory maintenance for the target while minimizing computational overhead.

[0027] In this embodiment, the current time t is... n dynamic goals and the previous time t n-1 dynamic goals To determine the association, the expression for the association similarity score is: in, These are empirical parameters; The Mahalanobis distance, , Represents the covariance matrix updated by the Kalman filter; similarity score , As the normalization factor, the features of each object , Indicates the dimensions of the bezel box. Let the standard deviation of the point cloud be . This is the length of the diagonal of the bounding box.

[0028] A higher value indicates a more similar geometric shape. A smaller value indicates a better kinematic match between the two. If the similarity score... Greater than a fixed threshold The current time t can be used n dynamic goals and the previous time t n-1 dynamic goals Link them together, otherwise refuse to match.

[0029] Step 2: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. Further, in step 2, a polar coordinate system is established with the instantaneous position of the dynamic target as the origin, the surrounding space is discretized into a polar unit network, the probability distribution of the polar units and the transition probabilities between polar units are determined, a Markov chain is constructed based on the probability distribution of the polar units, and the time-varying transition matrix of the Markov chain is updated based on the transition probabilities between polar units. Specifically, this includes: Step 21: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, and discretize the surrounding space into a polar element network. angular range radius range ; Step 22: Define the state of the Markov chain as the probability distribution of the polar unit, and the initial probability distribution of the polar unit at the initial time. The expression is: in, Normalization factor; () represents the polar coordinates of the dynamic target; Radial direction in polar coordinates ; The angle covariance in polar coordinates. ; Step 23: For dynamic pole units with dynamic targets polar unit To the dynamic pole unit The transition probability; for static pole units with static targets, their probability is set to zero and diffused to adjacent free units. Again polar unit To adjacent free units The transition probability; Step 24: Update the time-varying transition matrix of the Markov chain based on the transition probabilities between the polar units, where the probability distribution of the polar units satisfies the normalization constraint. .

[0030] Further, step 231: from pole unit (i,j) to dynamic pole unit The transition probability is: in, Radial velocity component of a dynamic target v is the velocity of the dynamic target estimated by the Kalman filter. The direction of the dynamic target estimated by the Kalman filter; For dynamic pole units The central angle; To predict the time step; This indicates that the dynamic target appears in the dynamic pole unit. The probability density, where s is the dynamic pole unit. area, and These represent the position and covariance estimated by Kalman in polar coordinates, respectively. For dynamic pole units The central location; Step 232: From pole unit (i,j) to adjacent free pole unit The transition probability is: in, Adjacent free pole units The Euclidean distance to the static target; Control the spread; It is the sum of the attenuation factors of all adjacent free pole units.

[0031] Step 3: Select a continuous polar unit sequence based on the updated Markov chain, and use polynomial trajectory fitting to obtain the predicted path from the continuous polar unit sequence; Furthermore, step 3, which involves selecting a continuous polar unit sequence based on the updated Markov chain and using polynomial trajectory fitting to obtain the predicted path, specifically includes: Step 31: Based on the updated Markov chain, determine the velocity and radius of motion of the dynamic target; Step 32: Select the continuous pole cell sequence with the highest probability percentage along the velocity direction of the dynamic target within the radius of the dynamic target's motion range; Step 33: Using the current position of the dynamic target as the trajectory starting point, map the continuous polar unit sequence in the global coordinate system to obtain the temporal path points of the dynamic target, and obtain the predicted path of the dynamic target based on the temporal path points.

[0032] Specifically, based on the speed of dynamic targets and prediction time window Calculate its range radius Then, within the radius of the motion range, along the velocity direction of the dynamic target. Select the continuous polar unit sequence with the highest probability. The current position estimated using Kalman filtering. As the starting point of the trajectory, these polar units are mapped to time path points in the global coordinate system. The predicted path of a dynamic target is obtained based on time-series path points.

[0033] Step 4: Propose dynamic target adaptive trajectory planning based on the predicted path.

[0034] Furthermore, step 4, which proposes dynamic target adaptive trajectory planning based on the predicted path, specifically includes: Step 41: The front-end trajectory of the dynamic target is generated by sampling key points of the path using the RRT* algorithm, constructing a polynomial trajectory with minimum-snap energy based on corridor constraints, and generating a smooth reference trajectory that satisfies the dynamic constraints. Step 42: Back-end trajectory optimization of dynamic targets. Construct a spatiotemporal probability risk field by predicting the trajectory of dynamic targets, and optimize the trajectory in real time using the gradient descent method.

[0035] In this embodiment, based on Markov-based trajectory prediction, a dynamic target object is set. Future period Within, the most probable motion trajectory of each dynamic object is predicted as follows: .

[0036] Accurately describing the uncertainty of dynamic target motion requires the prediction trajectory interval for each dynamic object. Sampling is performed to construct a dynamic risk field, which is an ellipse with an approximately Gaussian distribution. Figure 2 This is a schematic diagram illustrating the generation of dynamic guide points based on Gaussian gradients. Figure 2 In the middle, the principal axis direction is determined by the instantaneous velocity direction of the dynamic target. The decision, expressed as: in, These are the Gaussian kernel coefficients; This represents the position of the m-th dynamic target at time t. Indicates the location to be queried; covariance matrix Characterizing dynamic targets in Given the uncertainty of location at a given time, the expression for the covariance matrix is: ; in, Represents a diagonal matrix; Let V be the position variance of the semi-axis direction at time t. ;in, The position covariance estimated by the main axis Kalman filter at time t The position covariance estimated by the half-axis Kalman filter at time t; Let the radius of the dynamic target envelope circle be denoted by ; the covariance at future time steps can be obtained by . To obtain, The difference between matrix A and matrix A is that in A exist for ; The rotation matrix is ​​expressed as: ; The spatial gradient of the probability field can be expressed as: In terms of navigation and obstacle avoidance, a trajectory generation method is proposed based on the differential flat model of UAV. After giving the start and end points and a few necessary waypoints, the trajectory is parameterized by a k-order B-spline curve on the time node vector. The core of this method is to optimize the N-2(k-1) intermediate control points in the trajectory control points, excluding the fixed points at the beginning and end. The ultimate goal is to generate a smooth B-spline trajectory that satisfies collision-free, dynamic constraints and is smooth.

[0037] Optimization problem The expression for the total cost function is: in, For the cost of kinetic feasibility, For the cost of smoothness, For the cost of static collision, As a cost of dynamic collisions, , , , All are weighting coefficients.

[0038] In this embodiment, regarding static target avoidance: such as Figure 3 As shown, Figure 3 This is a schematic diagram of static guide points generated based on uniform projection. Collision cost estimation for static targets depends on static guide points. These guide points are generated using the uniform projection method, and the initial trajectory is represented by a curve formed by green and red control points. For the control point cost calculation within the collision segment, the A* algorithm is first used to search for a collision-free path that closely follows the surface of the dynamic target within the collision region. Figure 3 The purple curve in the image is used to obtain samples uniformly along this path. Points. For this segment Each of the control points , and send it to the A ray is projected onto each sampling point, and that sampling point is set as a static guide point. Figure 3 The purple dot in the middle.

[0039] To ensure the uniformity of the spatial distribution of control points after penalty optimization, the sampling time should be much longer than the time interval between control points. Based on experiments, the expression for the static collision cost is: in, For a custom static safety distance, the symbolic distance function Positive and negative numbers represent the exterior and interior of the dynamic target, respectively. Based on the above function, control points inside the dynamic target or those at close range can be penalized and pushed away from the collision zone along the direction of the projected ray.

[0040] In terms of dynamic target avoidance: given the motion characteristics of dynamic targets and perception prediction errors, directly calculating collision costs using observation data is unreliable. For example... Figure 3 As shown, a safe and smooth trajectory, i.e., the green curve, is generated by guiding the process using Gaussian gradients. The risk field is constructed as a differentiable Gaussian-like elliptic field. Based on the gradient descent method, the control points in the high-risk area are rapidly pushed along the purple ray (gradient direction) to a safe position with a lower field value. The purple dashed line in the figure represents the trajectory before optimization, and the intersection of the ray and the low-risk field is the dynamic guide point. The step size is denoted as [missing value]. The penalty strategy applies to two types of control points: those with field values ​​exceeding a threshold, and points within a low-risk zone adjacent to a high-risk boundary. Based on experiments, the dynamic collision cost is designed as a cubic shear penalty function, expressed as: in, User-defined dynamic safety distance, sign function ( Positive and negative numbers represent control points in low-risk and high-risk fields, respectively.

[0041] Example 2 A second aspect of this invention proposes a lightweight navigation system for dynamic perception and obstacle avoidance for small unmanned aerial vehicles (UAVs), comprising: The differentiation module is used to acquire a real-time depth map of the area to be detected and determine the instantaneous positions of dynamic and static targets in the real-time depth map; The modeling module is used to establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. The prediction module is used to select a continuous pole unit sequence based on the updated Markov chain and obtain the predicted path by fitting the continuous pole unit sequence with a polynomial trajectory. The planning module is used to propose adaptive trajectory planning for dynamic targets based on predicted paths.

[0042] Example 3 A third aspect of the present invention provides an electronic device comprising: a memory, a processor, and a computer program, the computer program being stored in the memory and configured to be executed by the processor to implement any of the methods described herein.

[0043] The contents not described in detail in this specification are prior art known to those skilled in the art. Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

Claims

1. A navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: Step 1: Obtain the real-time depth map of the area to be detected, and determine the instantaneous positions of dynamic and static targets in the real-time depth map; Step 2: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. Step 3: Select a continuous polar unit sequence based on the updated Markov chain, and use polynomial trajectory fitting to obtain the predicted path from the continuous polar unit sequence; Step 4: Propose dynamic target adaptive trajectory planning based on the predicted path.

2. The navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles according to claim 1, characterized in that, Step 1, determining the instantaneous positions of dynamic and static targets in the real-time depth map, specifically includes: Step 11: Generate the first bounding box based on the real-time depth map using the U-depth Detector; Step 12: Convert the real-time depth map into a point cloud, and use the DBSCAN clustering algorithm to cluster the point cloud converted from the real-time depth map to obtain point cloud clusters and second bounding boxes. Step 13: Fuse the first bounding box and the second bounding box using IoU thresholding to detect 3D obstacles; Step 14: Distinguish between dynamic and static targets in the 3D obstacle to determine the instantaneous positions of dynamic and static targets in the real-time depth map.

3. The navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles according to claim 2, characterized in that, Step 14, which distinguishes between dynamic and static targets for 3D obstacles, specifically includes: Step 141: Based on the point cloud and second bounding box obtained in Step 11, the estimated velocity of each point in the point cloud of the 3D obstacle is obtained through nearest neighbor search; Step 142: If the proportion of effective moving points in the point cloud cluster of a 3D obstacle is greater than the first threshold, then the 3D obstacle is determined to be a candidate dynamic target; Step 143: Calculate the mean displacement and normalized average variance of the adjacent frames of the candidate dynamic target point cloud cluster; when the mean displacement is greater than the second threshold and the normalized average variance is less than the third threshold, the candidate dynamic target is determined to be a dynamic target.

4. The navigation method based on UAV dynamic perception and obstacle avoidance according to claim 3, characterized in that, The mean displacement of the point cloud cluster of candidate dynamic targets in step 143 and normalized mean variance The calculation formula is: in, It represents the displacement between the nearest neighbors in adjacent frames. This represents the number of valid points.

5. The navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles according to claim 1, characterized in that, Step 2 establishes a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretizes the surrounding space into a polar unit network, determines the probability distribution of the polar units and the transition probabilities between polar units, constructs a Markov chain based on the probability distribution of the polar units, and updates the time-varying transition matrix of the Markov chain based on the transition probabilities between polar units. Specifically, this includes: Step 21: Establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, and discretize the surrounding space into a polar element network. angular range radius range ; Step 22: Define the state of the Markov chain as the probability distribution of the polar unit, and the initial probability distribution of the polar unit at the initial time. The expression is: in, Normalization factor; () represents the polar coordinates of the dynamic target; Radial direction in polar coordinates ; The angle covariance in polar coordinates. ; Step 23: For dynamic pole units with dynamic targets polar unit To the dynamic pole unit The transition probability; for static pole units with static targets, their probability is set to zero and diffused to adjacent free units. polar unit To adjacent free units The transition probability; Step 24: Update the time-varying transition matrix of the Markov chain based on the transition probabilities between the polar units, where the probability distribution of the polar units satisfies the normalization constraint. .

6. The navigation method based on UAV dynamic perception and obstacle avoidance according to claim 5, characterized in that, Step 231: From pole unit (i,j) to dynamic pole unit The transition probability is: in, Radial velocity component of a dynamic target v is the velocity of the dynamic target estimated by the Kalman filter. The direction of the dynamic target estimated by the Kalman filter; For dynamic pole units The central angle; To predict the time step; This indicates that the dynamic target appears in the dynamic pole unit. The probability density, where s is the dynamic pole unit. area, and These represent the position and covariance estimated by Kalman in polar coordinates, respectively. For dynamic pole units The central location; Step 232: From pole unit (i,j) to adjacent free pole unit The transition probability is: in, Adjacent free pole units The Euclidean distance to a static target; Control the spread; It is the sum of the attenuation factors of all adjacent free pole units.

7. The navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles according to claim 1, characterized in that, Step 3, which involves selecting a continuous polar unit sequence based on the updated Markov chain and using polynomial trajectory fitting to obtain the predicted path, specifically includes: Step 31: Based on the updated Markov chain, determine the velocity and radius of motion of the dynamic target; Step 32: Select the continuous pole cell sequence with the highest probability percentage along the velocity direction of the dynamic target within the radius of the dynamic target's motion range; Step 33: Using the current position of the dynamic target as the trajectory starting point, map the continuous polar unit sequence in the global coordinate system to obtain the temporal path points of the dynamic target, and obtain the predicted path of the dynamic target based on the temporal path points.

8. The navigation method based on dynamic perception and obstacle avoidance of unmanned aerial vehicles according to claim 1, characterized in that, Step 4, which proposes dynamic target adaptive trajectory planning based on the predicted path, specifically includes: Step 41: Front Trajectory Generation: The front trajectory of the dynamic target is generated by sampling key points of the path using the RRT* algorithm, constructing a polynomial trajectory that minimizes the minimum-snap energy based on corridor constraints, and generating a smooth reference trajectory that satisfies the dynamic constraints. Step 42: Back-end trajectory optimization: Back-end trajectory optimization of dynamic targets utilizes the predicted trajectory of dynamic targets to construct a spatiotemporal probability risk field, and optimizes the trajectory in real time using the gradient descent method.

9. A lightweight navigation system for dynamic perception and obstacle avoidance for small unmanned aerial vehicles (UAVs), characterized in that, include: The differentiation module is used to acquire a real-time depth map of the area to be detected and determine the instantaneous positions of dynamic and static targets in the real-time depth map; The modeling module is used to establish a polar coordinate system with the instantaneous position of the dynamic target as the origin, discretize the surrounding space into a polar unit network, determine the probability distribution of the polar units and the transition probability between the polar units, construct a Markov chain based on the probability distribution of the polar units, and update the time-varying transition matrix of the Markov chain based on the transition probability between the polar units. The Markov chain is used to characterize the state transition characteristics of the dynamic target. The prediction module is used to select a continuous pole unit sequence based on the updated Markov chain and obtain the predicted path by fitting the continuous pole unit sequence with a polynomial trajectory. The planning module is used to propose adaptive trajectory planning for dynamic targets based on predicted paths.

10. An electronic device, comprising: A memory, a processor, and a computer program, characterized in that: the computer program is stored in the memory and configured to be executed by the processor to implement the method of any one of claims 1 to 8.

Images