An autonomous obstacle avoidance method for drones based on AI algorithm

Abstract

The present invention relates to the field of data processing technology, and more specifically, to an autonomous obstacle avoidance method for a drone based on an AI algorithm. The method comprises: obtaining point cloud data at each moment during the flight of the drone; clustering the point cloud data to obtain a point cloud area at each moment; obtaining each speed-similar time period according to the speed of the drone, and obtaining the speed consistency of each point cloud area within each speed-similar time period; correcting the speed consistency according to the speed change trend of the point cloud area and the drone, obtaining the dynamic possibility of each point cloud area, and then obtaining a dynamic point cloud area; obtaining the collision possibility of each dynamic point cloud area, enhancing the point cloud data in the dynamic point cloud area according to the collision possibility, identifying the obstacle area based on the enhanced point cloud data, and planning the path. The present invention enhances the point cloud area corresponding to the dynamic obstacle, thereby improving the accuracy of subsequent obstacle avoidance.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology. More specifically, the present invention relates to an autonomous obstacle avoidance method for a drone based on an AI algorithm. Background Art

[0002] With the rapid development of drone technology, drones have become increasingly common across a wide range of industries, including logistics and distribution, environmental monitoring, agricultural inspections, security patrols, and geographic surveys. These industries are placing increasing demands on drone performance, particularly in autonomous flight. These requirements require drones to not only accurately fly along predetermined routes but also flexibly adjust their flight paths based on environmental changes. Furthermore, complex and dynamic flight environments present additional challenges, particularly in environments with numerous obstacles. Drones must be able to operate efficiently and safely. Therefore, ensuring the safe and effective operation of drones in these complex environments has become a core issue in the development of drone technology.

[0003] Therefore, drone obstacle avoidance technology is key to ensuring flight safety and efficiency. Existing drone obstacle avoidance methods mostly focus on identifying and avoiding static obstacles (such as buildings, trees, and hills). These methods typically use lidar, vision sensors, and other methods to detect surrounding static objects and avoid collisions.

[0004] However, existing drone obstacle avoidance methods usually identify static obstacles but ignore the existence of dynamic obstacles in the environment. The movement trajectory and speed of dynamic obstacles are unpredictable, especially in complex environments (such as cities, rural areas, or natural environments). If the drone cannot perceive and avoid these obstacles in real time, collisions are likely to occur during flight, resulting in equipment damage or even mission failure. Summary of the Invention

[0005] In order to solve the problem that drone obstacle avoidance methods usually identify static obstacles but ignore the existence of dynamic obstacles in the environment, resulting in the drone being unable to perceive and avoid these obstacles in real time, which may cause collisions, the present invention proposes an autonomous obstacle avoidance method for drones based on an AI algorithm, which includes the following steps:

[0006] During the flight of the drone, the point cloud data and the speed of the drone at each moment are obtained; the point cloud data at each moment are clustered to obtain the point cloud area at each moment;

[0007] Obtain each speed-similar time period based on the drone's speed; obtain the speed consistency of each point cloud area at the current moment in each speed-similar time period based on the speed and speed direction differences of the point cloud area in the speed-similar time period; correct the speed consistency based on the speed change trend of the point cloud area and the drone in different speed-similar time periods to obtain the dynamic possibility of each point cloud area at the current moment; filter the point cloud areas based on the dynamic possibility to obtain the dynamic point cloud area at the current moment;

[0008] Based on the speed change uncertainty and motion change uncertainty on the motion trajectory of the dynamic point cloud area, the collision possibility of each dynamic point cloud area at the current moment is obtained; based on the collision possibility, the point cloud data in each dynamic point cloud area is enhanced to obtain the enhanced point cloud data at the current moment; based on the enhanced point cloud data at the current moment, the obstacle area is identified and the path is planned.

[0009] The innovation of the present invention lies in obtaining the speed consistency of each point cloud area in each speed-similar time period based on the speed and speed direction differences in the speed similar time period of the drone flight in the point cloud area, and then correcting the speed consistency according to the speed change trend of the point cloud area and the drone in different speed similar time periods to obtain the dynamic possibility of the point cloud area, which can effectively distinguish the dynamic area from the static area during the drone flight process; finally, after identifying the dynamic point cloud area, the point cloud data is enhanced to different degrees according to the collision possibility of the dynamic point cloud area, which can improve the point cloud density and accuracy in the high collision probability area, so that the enhanced data can more finely represent the shape and position of the obstacle, improve the accuracy of recognition and obstacle avoidance, and then identify the obstacle area and perform path planning based on the enhanced point cloud data.

[0010] Preferably, obtaining each speed similarity time period according to the speed of the drone includes:

[0011] Arrange the drone speed values ​​at all times before the current time in chronological order to generate a historical speed sequence. For any data in the historical speed sequence, if the difference between the data and its right adjacent data is greater than the difference threshold, use it as a segmentation point to obtain all segmentation points in the historical speed sequence. Divide the historical speed sequence into several continuous segments based on the segmentation points. The time interval corresponding to each continuous segment is recorded as each speed similarity time period.

[0012] The changes in speed and speed direction of the point cloud area are analyzed during the time period when the speed of the drone is similar, which facilitates the subsequent distinction between dynamic and static areas.

[0013] Preferably, obtaining the speed consistency of each point cloud area at the current moment in each speed-similar time period includes:

[0014] The spatial velocity and spatial velocity direction of the i-th point cloud area at the current moment in the j-th velocity similarity time period constitute the spatial velocity vector at the k-th moment; the Jaccard coefficient between the spatial velocity vector at the k-th moment and the spatial velocity vector at the k+1-th moment is recorded as the velocity similarity of the i-th point cloud area at the current moment in the j-th velocity similarity time period, recorded as ;

[0015] , is the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period; is the number of moments in the j-th speed similarity time period.

[0016] Preferably, obtaining the spatial velocity and spatial velocity direction of the i-th point cloud region at the current moment at the k-th moment within the j-th velocity similarity time period includes:

[0017] The least squares method is used to perform curve fitting on the spatial coordinate positions of the i-th point cloud area at all times within the j-th speed similarity time period at the current moment, and the motion trajectory fitting curve of the i-th point cloud area at the current moment within the j-th speed similarity time period is obtained; the Euclidean distance between the spatial position coordinates of the i-th point cloud area at the current moment at the k-th moment and the k+1-th moment within the j-th speed similarity time period is recorded as the spatial velocity at the k-th moment; the tangent direction of the point cloud data at the k-th moment on the motion trajectory fitting curve of the i-th point cloud area at the current moment within the j-th speed similarity time period is recorded as the spatial velocity direction at the k-th moment.

[0018] Preferably, the obtaining of the dynamic possibility of each point cloud area at the current moment includes:

[0019] ;

[0020] Where, Represents the dynamic possibility of the i-th point cloud area at the current moment; Represents the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period; as well as They represent the mean spatial velocity of the UAV at all times in the j-th speed similarity time period and the j+1-th speed similarity time period respectively; as well as They represent the mean spatial velocity of the i-th point cloud area at all times within the j-th speed similarity time period and the j+1-th speed similarity time period at the current moment; J represents the number of speed similarity time periods; exp() represents an exponential function with a natural constant as the base; || represents the absolute value symbol; and norm() represents the normalization function.

[0021] It is easy to distinguish between dynamic areas and static areas, thereby improving the ability to identify moving obstacles

[0022] Preferably, obtaining the collision probability of each dynamic point cloud area at the current moment includes:

[0023] , Represents the collision possibility of the ath dynamic point cloud area at the current moment; Represents the velocity uncertainty of the ath dynamic point cloud area at the current moment; Represents the uncertainty of the motion change of the ath dynamic point cloud area at the current moment; represents the Euclidean distance between the spatial coordinate position of the ath dynamic point cloud area and the drone at the current moment; B represents the number of all historical moments; Represents the collision trend factor of the ath dynamic point cloud area at the bth historical moment at the current moment; norm() represents the normalization function.

[0024] By enhancing point cloud data to varying degrees based on collision probability, the resolution and integrity of the point cloud data can be improved, thereby improving the accuracy of subsequent path planning and obstacle avoidance.

[0025] Preferably, obtaining the velocity change uncertainty of the ath dynamic point cloud region at the current moment includes:

[0026] The least squares method is used to perform curve fitting on the spatial position of the a-th dynamic point cloud region at the current moment at all historical moments to obtain the motion trajectory fitting curve of the a-th dynamic point cloud region at the current moment. The variance of the slope of all point cloud data on the motion trajectory fitting curve is recorded as the velocity change uncertainty of the a-th dynamic point cloud region at the current moment.

[0027] Preferably, obtaining the motion uncertainty of the ath dynamic point cloud region includes:

[0028] The direction of the straight line connecting the spatial coordinate position of the ath dynamic point cloud area at the bth historical moment and the spatial coordinate position at the b+1th historical moment is recorded as the target direction of the ath dynamic point cloud area at the bth historical moment; the angle between the target direction and the horizontal direction is recorded as the movement direction of the ath dynamic point cloud area at the bth historical moment; the angle between the mean of the tangent direction of all point cloud data on the motion trajectory fitting curve of the ath dynamic point cloud area and the horizontal direction is recorded as the target mean direction of the ath dynamic point cloud area; the sine value of the difference between the mean movement direction of the ath dynamic point cloud area at all historical moments and the target mean direction is recorded as the movement change uncertainty of the ath dynamic point cloud area at the current moment.

[0029] Preferably, obtaining the collision trend factor of the ath dynamic point cloud region at the current moment at the bth historical moment includes:

[0030] Obtain the spatial velocity vector of the UAV and each dynamic point cloud area at the current moment at each historical moment. Based on the spatial velocity vector of the UAV and each dynamic point cloud area at each historical moment, obtain the relative speed of the UAV and each dynamic point cloud area at each historical moment, and use it as the collision trend factor of each dynamic point cloud area at the current moment at each historical moment.

[0031] Preferably, based on the collision possibility, enhancing the point cloud data in each dynamic point cloud area to obtain enhanced point cloud data at the current moment; identifying the obstacle area based on the enhanced point cloud data at the current moment and planning the path includes:

[0032] According to the collision probability of each dynamic point cloud area at the current moment, the point cloud data in each dynamic point cloud area is enhanced in combination with the fast point cloud interpolation algorithm to obtain the enhanced point cloud data at the current moment; based on the point cloud segmentation algorithm, the enhanced point cloud data at the current moment is segmented to identify the obstacle area; based on the identified obstacle area, the fast expanding random tree algorithm is used to plan an obstacle avoidance path for the drone at the current moment.

[0033] By enhancing the point cloud data to varying degrees, the point cloud density and accuracy in areas with high collision probability can be improved, and obstacles can be identified more accurately.

[0034] The present invention has the following beneficial effects: the purpose of the present invention is to analyze the dynamic possibility of the point cloud area according to the speed and speed direction differences within a similar time period of the drone flight speed in the point cloud area, and can effectively distinguish the dynamic area from the static area during the drone flight process; finally, after identifying the dynamic point cloud area, the point cloud data is enhanced to varying degrees according to the collision possibility of the dynamic point cloud area, which can improve the point cloud density and accuracy in the high collision probability area, so that the enhanced data can more finely represent the shape and position of the obstacle, thereby improving the accuracy of recognition and obstacle avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flowchart of the steps of an autonomous obstacle avoidance method for a drone based on an AI algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0037] See also Figure 1 , which shows a flowchart of a method for autonomous obstacle avoidance of a drone based on an AI algorithm provided by one embodiment of the present invention, the method comprising the following steps:

[0038] S001. During the flight of the UAV, obtain point cloud data at each moment.

[0039] In an embodiment of the present invention, the preset acquisition frequency is 1 second / time, and a laser radar is mounted on the drone. During the flight of the drone, the laser radar is used to obtain point cloud data at each moment and the speed of the drone at each moment.

[0040] S002. Cluster the point cloud data at each moment to obtain the point cloud area at each moment; obtain each speed similarity time period based on the drone speed, and obtain the speed consistency of each point cloud area at the current moment within each speed similarity time period; correct the speed consistency based on the consistency of the speed change trend between the point cloud area and the drone, and obtain the dynamic possibility of each point cloud area at the current moment.

[0041] It should be noted that point cloud data (such as lidar scanning results) is essentially a collection of a large number of discrete points in three-dimensional space. Direct processing cannot identify the boundaries or types of objects. Therefore, through Euclidean distance clustering, adjacent point cloud data in space can be aggregated into "point cloud areas", corresponding to specific objects in reality (such as trees, buildings, other drones, etc.). After clustering, each point cloud area can be used as an independent unit, and its dynamic characteristics can be determined by trajectory tracking in a time series.

[0042] In an embodiment of the present invention, the preset distance threshold T=0.5 meters. In other embodiments, the implementer may preset the value of T according to the specific implementation situation, and cluster the point cloud data at each moment using the Euclidean distance clustering algorithm based on the distance threshold to obtain the point cloud area at each moment.

[0043] It should be noted that the point cloud area corresponding to the static object usually exhibits relatively stable position and speed characteristics. Therefore, within the time period of similar flight speed of the UAV, the speed and direction of the static object change less. If any point cloud area at the current moment belongs to a static object, the speed and speed direction difference of the point cloud area at the current moment within the time period of similar speed of the UAV will be smaller. Therefore, the present invention first divides several speed-similar time periods according to the flight speed of the UAV, and then obtains the speed consistency of each point cloud area at the current moment in each speed-similar time period.

[0044] In an embodiment of the present invention, the speed values ​​of the drone at all times before the current time are arranged in chronological order to generate a historical speed sequence. For any data in the historical speed sequence, if the difference between the data and its right adjacent data is greater than a difference threshold, the data is used as a segmentation point, and all segmentation points in the historical speed sequence are obtained. The historical speed sequence is divided into a number of continuous segments based on the segmentation points. The time interval corresponding to each continuous segment is recorded as each speed similarity time period.

[0045] Using the optical flow method to track, all moments before the current moment are recorded as historical moments, and the spatial coordinate position of each point cloud area at the current moment at each historical moment is obtained;

[0046] The least squares method is used to perform curve fitting on the spatial coordinate positions of the i-th point cloud region at the current moment at all times within the j-th speed similarity time period, thereby obtaining a motion trajectory fitting curve of the i-th point cloud region at the current moment within the j-th speed similarity time period. It should be noted that the least squares method is a prior art and will not be described in detail in the embodiments of the present invention.

[0047] The Euclidean distance between the spatial position coordinates of the i-th point cloud region at the current moment and the spatial position coordinates at the k-th moment in the j-th speed similarity time period is recorded as the spatial velocity at the k-th moment; the tangent direction of the point cloud data at the k-th moment on the motion trajectory fitting curve of the i-th point cloud region at the current moment in the j-th speed similarity time period is recorded as the spatial velocity direction at the k-th moment;

[0048] According to the spatial velocity and spatial velocity direction at the kth moment, the spatial velocity vector at the kth moment is constructed; the Jaccard coefficient between the spatial velocity vector at the kth moment and the spatial velocity vector at the k+1th moment is recorded as the velocity similarity at the kth moment;

[0049] The specific formula for obtaining the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period is:

[0050] ;

[0051] Where, Represents the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period; represents the number of moments in the jth time period with similar speed; Represents the speed similarity of the i-th point cloud area at the current moment at the k-th moment within the j-th speed similarity time period; if the speed and speed direction similarity of the i-th point cloud area at the current moment in the UAV speed similarity time period is greater, that is, The larger the value of , the more the i-th point cloud area at the current moment belongs to a static object; conversely, the i-th point cloud area at the current moment belongs to a dynamic object.

[0052] It should be noted that if any point cloud area has been moving at a uniform speed, it will also have speed consistency within the time period with similar speed of the UAV. When the speed change trend of the point cloud area is consistent with the speed change trend of the UAV within the time period with different similar speeds of the UAV, the possibility that the point cloud area belongs to a static area is greater. In other words, the greater the speed consistency and the more similar the speed change trends between the UAV and the point cloud area, the greater the possibility that the point cloud area belongs to a static area. On the contrary, the dynamic possibility of the point cloud area is smaller. Therefore, the speed consistency is further corrected according to the speed change trend between the UAV and the point cloud area to obtain the dynamic possibility of the point cloud data.

[0053] In the embodiment of the present invention, the spatial velocity of the UAV at each moment in each speed similar time period is obtained according to the method of obtaining the spatial velocity of each point cloud area at each moment in each speed similar time period at the current moment;

[0054] The specific formula for obtaining the dynamic possibility of the i-th point cloud area at the current moment is:

[0055] ;

[0056] Where, Represents the dynamic possibility of the i-th point cloud area at the current moment; Represents the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period;

[0057] represents the mean spatial velocity of the UAV at all times during the j-th time period of similar velocity; represents the mean spatial velocity of the UAV at all times during the j+1th time period of similar velocity; Represents the mean spatial velocity of the i-th point cloud area at the current moment in the j-th velocity similarity time period; Represents the mean spatial velocity of the i-th point cloud area at all times within the j+1-th velocity similarity time period at the current moment; J represents the number of velocity similarity time periods; exp() represents an exponential function with a natural constant as the base; || represents the absolute value symbol;

[0058] The smaller the value of , the more likely the i-th point cloud area at the current moment belongs to the dynamic area; The smaller the value of , the more likely it is that the speed change trend of the i-th point cloud area at the current moment is consistent with the speed change trend of the drone, and the more likely it is that the i-th point cloud area at the current moment belongs to the static area; otherwise, The larger the value of , the more likely it is that the speed change trend of the i-th point cloud area at the current moment is inconsistent with the speed change trend of the drone, and the greater the possibility that the i-th point cloud area at the current moment belongs to the dynamic area.

[0059] S003. Obtain the dynamic point cloud area at the current moment according to the dynamic possibility of each point cloud area at the current moment; and obtain the collision possibility of each dynamic point cloud area at the current moment.

[0060] It should be noted that by setting a dynamic possibility threshold, point cloud areas with higher dynamic possibility are screened out, thereby defining the dynamic point cloud area at the current moment.

[0061] In an embodiment of the present invention, a dynamic possibility threshold T=0.7 is preset. In other embodiments, implementers may preset the value of T according to the specific implementation method; the point cloud area at the current moment whose dynamic possibility is greater than the dynamic possibility threshold T is recorded as the dynamic point cloud area at the current moment.

[0062] It should be noted that accurate obstacle identification is key to ensuring flight safety. Therefore, if point cloud data enhancement is insufficient, especially in high-threat areas, the position or shape of obstacles may be misjudged, leading to obstacle avoidance failure. Therefore, it is necessary to obtain the collision probability of each dynamic point cloud area and enhance each dynamic point cloud area to varying degrees. This allows for more accurate identification of the obstacle's position, shape, and motion trajectory, thereby improving the reliability of obstacle avoidance.

[0063] First, by obtaining the motion trajectory of the dynamic point cloud area, the speed change uncertainty and motion change uncertainty of the dynamic point cloud data are analyzed based on the motion trajectory to evaluate the stability of the speed change and the regularity of the motion direction of the dynamic point cloud area. The greater the speed change uncertainty and motion change uncertainty of the dynamic point cloud data, the greater the collision possibility of the dynamic point cloud area; then, the relative motion between the UAV and the dynamic point cloud area can be analyzed to further evaluate the collision risk. If the relative speed between the UAV and the dynamic point cloud area is large, it means that the two are approaching faster and the collision risk increases accordingly. Finally, the collision possibility of each dynamic point cloud area is obtained by combining the evaluation of motion change uncertainty, speed uncertainty and relative speed.

[0064] In this embodiment of the present invention, a least squares method is used to perform curve fitting on the spatial position of the ath dynamic point cloud region at the current moment at all historical moments to obtain a motion trajectory fitting curve of the ath dynamic point cloud region at the current moment. The variance of the slope of all point cloud data on the motion trajectory fitting curve is recorded as the velocity change uncertainty of the ath dynamic point cloud region at the current moment.

[0065] The direction of the straight line connecting the spatial coordinate position of the ath dynamic point cloud region at the bth historical moment and the spatial coordinate position at the b+1th historical moment is recorded as the target direction of the ath dynamic point cloud region at the bth historical moment; the angle between the target direction of the ath dynamic point cloud region at the bth historical moment and the horizontal direction is recorded as the movement direction of the ath dynamic point cloud region at the bth historical moment;

[0066] The angle between the mean of the tangent direction of all point cloud data on the motion trajectory fitting curve of the a-th dynamic point cloud region at the current moment and the horizontal direction is recorded as the target mean direction of the a-th dynamic point cloud region at the current moment; the sine value of the difference between the mean of the motion direction of the a-th dynamic point cloud region at all historical moments at the current moment and the target mean direction is recorded as the motion change uncertainty of the a-th dynamic point cloud region at the current moment;

[0067] According to the above-mentioned method for obtaining the spatial velocity vector, the spatial velocity vector of the UAV and each dynamic point cloud area at the current moment is obtained at each historical moment. Based on the spatial velocity vector of the UAV and each dynamic point cloud area at each historical moment, the relative speed of the UAV and each dynamic point cloud area at each historical moment is obtained, and used as the collision tendency factor of each dynamic point cloud area at the current moment at each historical moment;

[0068] The specific formula for obtaining the collision probability of each dynamic point cloud area at the current moment is:

[0069] ;

[0070] Where, Represents the collision possibility of the ath dynamic point cloud area at the current moment; Represents the velocity uncertainty of the ath dynamic point cloud area at the current moment; Represents the uncertainty of the motion change of the ath dynamic point cloud area at the current moment; represents the Euclidean distance between the spatial coordinate position of the ath dynamic point cloud area and the drone at the current moment; B represents the number of all historical moments; Represents the collision trend factor of the ath dynamic point cloud area at the bth historical moment at the current moment; norm() represents the normalization function;

[0071] The larger the value of , the more unstable the speed change of the a-th dynamic point cloud area is, indicating that its motion pattern is complex and the risk of the drone colliding with it is more difficult to predict. In this case, the collision possibility of the a-th dynamic point cloud area is higher. The larger the value of , the more irregular the movement direction of the a-th dynamic point cloud area is, and the higher the collision possibility of the a-th dynamic point cloud area is.

[0072] The larger the value of , the greater the relative speed between the UAV and the ath dynamic point cloud area at the current moment at the bth historical moment. The closer the UAV and the dynamic point cloud area are to each other per unit time, which means that the time interval from the current position to the possible collision between the two is shorter, and the collision risk is higher. The smaller the value of , the closer the distance between the ath dynamic point cloud area and the drone is, and the more likely a collision will occur.

[0073] S004. Enhance the point cloud data in each dynamic point cloud area according to the collision probability of each dynamic point cloud area at the current moment to obtain the enhanced point cloud data at the current moment. Identify the obstacle area based on the enhanced point cloud data at the current moment and plan the path.

[0074] It should be noted that according to the possibility of collision, the dynamic point cloud area is enhanced to varying degrees to improve the point cloud density and accuracy in the high collision probability area, so that the enhanced data can more finely represent the shape and position of the obstacle, thereby improving the accuracy of recognition and obstacle avoidance. Then, based on the enhanced point cloud data, the obstacle area is identified and path planning is performed.

[0075] In an embodiment of the present invention, based on the collision probability of each dynamic point cloud region at the current moment, the point cloud data in each dynamic point cloud region is enhanced in combination with a fast point cloud interpolation algorithm to obtain enhanced point cloud data at the current moment; the enhanced point cloud data at the current moment is segmented using a point cloud segmentation algorithm to identify obstacle areas;

[0076] According to the identified obstacle area, a fast expanding random tree algorithm is used to plan an obstacle avoidance path for the UAV at the current moment.

[0077] It should be noted that the fast point cloud interpolation algorithm, the point cloud segmentation algorithm and the fast expanding random tree algorithm are all existing technologies and will not be described in detail in the embodiments of the present invention.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for autonomous obstacle avoidance of UAV based on AI algorithm, characterized in that: include: During the flight of the drone, obtain the point cloud data and drone speed at each moment; Cluster the point cloud data at each moment to obtain the point cloud area at each moment; Obtain each speed similarity time period based on the UAV speed; obtain the speed consistency of each point cloud area in each speed similarity time period at the current moment based on the speed and speed direction differences of the point cloud area in the speed similarity time period; correct the speed consistency based on the ratio of the difference between the spatial speed mean values ​​of the UAV at all times in different speed similarity time periods to the difference between the spatial speed mean values ​​of the point cloud area at all times in different speed similarity time periods, and obtain the dynamic possibility of each point cloud area at the current moment, including: Where, Represents the dynamic possibility of the i-th point cloud area at the current moment; Represents the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period; as well as They represent the mean spatial velocity of the UAV at all times in the j-th speed similarity time period and the j+1-th speed similarity time period respectively; as well as They represent the spatial velocity mean of the i-th point cloud area at the current moment in the j-th speed similarity time period and the j+1-th speed similarity time period; J represents the number of speed similarity time periods; exp() represents an exponential function with a natural constant as the base; || represents the absolute value symbol; norm() represents the normalization function; the point cloud area is screened based on the dynamic possibility to obtain the dynamic point cloud area at the current moment; Based on the speed change uncertainty and motion change uncertainty on the motion trajectory of the dynamic point cloud area, the collision possibility of each dynamic point cloud area at the current moment is obtained; based on the collision possibility, the point cloud data in each dynamic point cloud area is enhanced to obtain the enhanced point cloud data at the current moment; based on the enhanced point cloud data at the current moment, the obstacle area is identified and the path is planned.

2. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 1, characterized in that: The method of obtaining each speed similarity time period according to the speed of the drone includes: Arrange the drone speed values ​​at all times before the current time in chronological order to generate a historical speed sequence. For any data in the historical speed sequence, if the difference between the data and its right adjacent data is greater than the difference threshold, use it as a segmentation point to obtain all segmentation points in the historical speed sequence. Divide the historical speed sequence into several continuous segments based on the segmentation points. The time interval corresponding to each continuous segment is recorded as each speed similarity time period.

3. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 1, characterized in that: The obtaining of the speed consistency of each point cloud region at the current moment in each speed similarity time period includes: The spatial velocity and spatial velocity direction of the i-th point cloud area at the current moment in the j-th velocity similarity time period constitute the spatial velocity vector at the k-th moment; the Jaccard coefficient between the spatial velocity vector at the k-th moment and the spatial velocity vector at the k+1-th moment is recorded as the velocity similarity of the i-th point cloud area at the current moment in the j-th velocity similarity time period, recorded as ; , is the velocity consistency of the i-th point cloud area at the current moment in the j-th velocity similarity time period; is the number of moments in the j-th speed similarity time period.

4. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 3, characterized in that: The acquisition of the spatial velocity and spatial velocity direction of the i-th point cloud region at the current moment at the k-th moment within the j-th velocity similarity time period includes: The least squares method is used to perform curve fitting on the spatial coordinate positions of the i-th point cloud area at all times within the j-th speed similarity time period at the current moment, and the motion trajectory fitting curve of the i-th point cloud area at the current moment within the j-th speed similarity time period is obtained; the Euclidean distance between the spatial position coordinates of the i-th point cloud area at the current moment at the k-th moment and the k+1-th moment within the j-th speed similarity time period is recorded as the spatial velocity at the k-th moment; the tangent direction of the point cloud data at the k-th moment on the motion trajectory fitting curve of the i-th point cloud area at the current moment within the j-th speed similarity time period is recorded as the spatial velocity direction at the k-th moment.

5. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 1, characterized in that: The obtaining of the collision probability of each dynamic point cloud region at the current moment includes: , Represents the collision possibility of the ath dynamic point cloud area at the current moment; Represents the velocity uncertainty of the ath dynamic point cloud area at the current moment; Represents the uncertainty of the motion change of the ath dynamic point cloud area at the current moment; represents the Euclidean distance between the spatial coordinate position of the ath dynamic point cloud area and the drone at the current moment; B represents the number of all historical moments; Represents the collision trend factor of the ath dynamic point cloud area at the bth historical moment at the current moment; norm() represents the normalization function.

6. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 5, characterized in that: The acquisition of the velocity change uncertainty of the ath dynamic point cloud region at the current moment includes: The least squares method is used to perform curve fitting on the spatial position of the a-th dynamic point cloud region at the current moment at all historical moments to obtain the motion trajectory fitting curve of the a-th dynamic point cloud region at the current moment; the variance of the slope of all point cloud data on the motion trajectory fitting curve is recorded as the velocity change uncertainty of the a-th dynamic point cloud region at the current moment.

7. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 5, characterized in that: The acquisition of the motion uncertainty of the ath dynamic point cloud region includes: The direction of the straight line connecting the spatial coordinate position of the ath dynamic point cloud area at the bth historical moment and the spatial coordinate position at the b+1th historical moment is recorded as the target direction of the ath dynamic point cloud area at the bth historical moment; the angle between the target direction and the horizontal direction is recorded as the movement direction of the ath dynamic point cloud area at the bth historical moment; the angle between the mean of the tangent direction of all point cloud data on the motion trajectory fitting curve of the ath dynamic point cloud area and the horizontal direction is recorded as the target mean direction of the ath dynamic point cloud area; the sine value of the difference between the mean movement direction of the ath dynamic point cloud area at all historical moments and the target mean direction is recorded as the movement change uncertainty of the ath dynamic point cloud area at the current moment.

8. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 5, characterized in that: Acquisition of the collision trend factor of the ath dynamic point cloud region at the bth historical moment at the current moment includes: Obtain the spatial velocity vector of the UAV and each dynamic point cloud area at the current moment at each historical moment. Based on the spatial velocity vector of the UAV and each dynamic point cloud area at each historical moment, obtain the relative speed of the UAV and each dynamic point cloud area at each historical moment, and use it as the collision trend factor of each dynamic point cloud area at the current moment at each historical moment.

9. The autonomous obstacle avoidance method for a UAV based on an AI algorithm according to claim 1, characterized in that: Based on the collision possibility, the point cloud data in each dynamic point cloud area is enhanced to obtain enhanced point cloud data at the current moment; Based on the enhanced point cloud data at the current moment, the obstacle area is identified and the path is planned, including: According to the collision possibility of each dynamic point cloud area at the current moment, the point cloud data in each dynamic point cloud area is enhanced by combining the fast point cloud interpolation algorithm to obtain the enhanced point cloud data at the current moment; the enhanced point cloud data at the current moment is segmented based on the point cloud segmentation algorithm to identify the obstacle area; According to the identified obstacle area, a fast expanding random tree algorithm is used to plan an obstacle avoidance path for the UAV at the current moment.

Images