Unmanned aerial vehicle landing control method based on AI vision

Abstract

The invention discloses an AI vision-based unmanned aerial vehicle landing control method, and relates to the technical field of computer vision, and the method comprises the steps: collecting ground comprehensive data and current attitude data of an unmanned aerial vehicle; inputting the collected ground comprehensive data into a deep learning model for ground feature analysis, and generating an optimal landing point path planning result; predicting the current position and the current attitude of the unmanned aerial vehicle based on the optimal landing point path planning result and the ground comprehensive data; calculating position information error and attitude error information according to the predicted current position and current attitude of the unmanned aerial vehicle; adjusting the attitude and descent speed of the unmanned aerial vehicle based on the position information error and attitude error information, and generating optimized attitude information and descent trajectory; and adjusting motor thrust distribution according to the optimized attitude information and the descending trajectory in combination with the optimal landing point. According to the invention, by combining the optimized attitude information with the descending trajectory, the overall landing process is ensured to be more stable and accurate, and the safe landing capability is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a control method for unmanned aerial vehicle landing based on AI vision. Background Art

[0002] With the rapid development of drone technology, its application scenarios are becoming increasingly diverse. In the actual use of drones, accurate and safe autonomous landing technology is a key challenge. Traditional drone landing technology mainly relies on GPS data and preset fixed track planning. Although these methods perform well in open and flat environments, they often cannot guarantee the accuracy and safety of landing in complex ground environments.

[0003] The existing AI visual landing technology still has certain shortcomings. On the one hand, the existing methods are usually based on a single visual data source, lacking the fusion analysis of multi-source ground comprehensive data. The existing technology is still based on simple PID control for UAV attitude adjustment and descent trajectory optimization, and cannot fully utilize AI technology for real-time and adaptive control optimization. The distribution of motor thrust during UAV landing is often based on a fixed model, lacking dynamic adjustment capabilities, and it is difficult to meet the needs of real-time attitude error correction and multi-scenario adaptation. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a UAV landing control method based on AI vision to solve the path planning, attitude control and thrust optimization problems of autonomous landing of UAV in complex dynamic terrain.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a control method for landing of a UAV based on AI vision, which includes collecting ground comprehensive data and current posture data of the UAV;

[0008] Input the collected ground comprehensive data into the deep learning model to analyze ground features and generate the optimal landing point path planning results;

[0009] Predict the current position and attitude of the UAV based on the optimal landing point path planning results and ground comprehensive data;

[0010] According to the predicted current position and current attitude of the UAV, the position information error and attitude error information are calculated;

[0011] Based on the position information error and attitude error information, the attitude and descent speed of the UAV are adjusted to generate optimized attitude information and descent trajectory;

[0012] Adjust the motor thrust distribution according to the optimized attitude information and descent trajectory combined with the optimal landing point.

[0013] As a preferred solution of the control method for landing of a UAV based on AI vision described in the present invention, the acquisition of ground comprehensive data and current posture data of the UAV includes:

[0014] The collected ground comprehensive data includes terrain and obstacle distribution data, and the current posture data of the UAV includes the acceleration, angular velocity and current posture information of the UAV, and unified coordinate alignment and normalization processing are performed.

[0015] As a preferred solution of the control method for landing of a UAV based on AI vision described in the present invention, the following steps are included: inputting the collected ground comprehensive data into the deep learning model for ground feature analysis:

[0016] The convolutional neural network model is used to extract the features of the processed ground comprehensive data to obtain the ground feature map, which is combined with the current posture information of the drone to construct a comprehensive feature vector, which is expressed as:

[0017] V=[a,ω,ε,F ′ ];

[0018] Among them, a is the acceleration of the drone, ω is the angular velocity of the drone, ε is the current attitude information of the drone, and F ′ Ground feature map.

[0019] As a preferred solution of the control method for landing of a drone based on AI vision described in the present invention, generating the optimal landing point and path planning results according to the ground feature analysis results includes the following steps:

[0020] The multi-layer perceptron model receives the comprehensive feature vector as input and predicts the safety score of the landing point. The expression is:

[0021]

[0022] Among them, v j is the jth element in the eigenvector V, W j is the weight matrix W corresponding to v j The weight of

[0023] m is the dimension of the feature vector V, b is the bias term, and S is the safety score of the landing point;

[0024] Based on the safety score, the location with the highest score is selected as the optimal landing point. Based on the optimal landing point, the Dijkstra algorithm is used to find the best path from the current location to the optimal landing point. The expression is,

[0025]

[0026] Among them, P is the best path, is the set of all paths, p is a path from the starting point to the target point, k is the number of steps in path p, L k is the straight-line distance of the kth step, E k is the energy consumption in the kth step, T k is the terrain complexity factor in the kth step, S k is the safety score of the landing point at the kth step, α is the weight coefficient of the energy consumption item, β is the weight coefficient of the terrain complexity item, and γ is the weight coefficient of the safety score item of the landing point.

[0027] As a preferred solution of the control method for landing of a drone based on AI vision of the present invention, the following steps are included in predicting the current position and current posture of the drone based on the optimal landing point path planning result and the ground comprehensive data:

[0028] The Kalman filter is used in combination with the speed of the UAV to predict the current position of the UAV as it moves along the optimal path. The expression is:

[0029] x t =x t-1 +Δt(v x cosθ-v y sinθ);

[0030] y t =y t-1 +Δt(v x sinθ+v y cosθ);

[0031] z t =z t-1 +Δytv z ;

[0032] Among them, x t is the x-axis position coordinate predicted by the drone at the current time y, x t-1 is the x-axis position coordinate predicted by the drone at the previous moment t-1, y t is the predicted y-axis position coordinate of the drone at the current time t, y t-1 is the predicted y-axis position coordinate of the drone at the previous moment t-1, z t is the z-axis position coordinate predicted by the drone at the current time t, z t-1 is the predicted z-axis position coordinate of the drone at the previous moment t-1, v x is the velocity component of the drone along the x-axis, v y is the velocity component of the drone along the y-axis, vz is the velocity component of the UAV along the z-axis, and Δt is the time step.

[0033] The attitude is composed of pitch angle, roll angle and yaw angle. The current attitude of the drone is predicted by IMU and gyroscope. The expression is:

[0034]

[0035] φ=arctan 2(a y ,a z );

[0036] ψ=ψ prev +∫ω z dt;

[0037] Among them, μ is the pitch angle of the UAV, φ is the roll angle of the UAV, ψ is the yaw angle of the UAV, and ψ prev is the yaw angle of the previous step, ω z is the change in yaw angle, a x is the acceleration of the drone along the x-axis in the body coordinate system, a y is the acceleration of the drone along the y-axis in the body coordinate system, a z is the acceleration of the UAV along the z-axis in the body coordinate system, x is the x-axis direction of the UAV in the body coordinate system, y is the y-axis direction of the UAV in the body coordinate system, and z is the z-axis direction of the UAV in the body coordinate system.

[0038] As a preferred solution of the control method for landing of a drone based on AI vision described in the present invention, wherein: according to the predicted current position and current attitude of the drone, calculating the position information error and attitude error information includes the following steps:

[0039] From the path planning results, find the path point sequence from the current position of the drone to the optimal landing point, extract the nearest target point as the current expected position, and calculate the error between the current position and the expected position. The expression is:

[0040]

[0041] Among them, Δ pos is the current position error, x O Predict the x-axis expected position coordinates for the drone, y O Predict the desired position coordinates of the y-axis for the drone; O Predict the desired z-axis position coordinates for the drone;

[0042] According to the direction of path planning, the expected attitude angle is calculated and compared with the current attitude angle to obtain the attitude error. The expression is:

[0043]

[0044] in, is the desired pitch angle of the drone, is the desired roll angle of the drone, For drones

[0045] The desired yaw angle, Δ attitude is the attitude error.

[0046] As a preferred solution of the control method for landing of a drone based on AI vision described in the present invention, wherein: based on the position information error and the attitude error information, the attitude and descent speed of the drone are adjusted, and the optimized attitude information and descent trajectory are generated, including the following steps:

[0047] According to the position information error and attitude error information, a comprehensive error model is constructed for subsequent attitude and speed adjustment. The comprehensive error expression is:

[0048]

[0049] Among them, E total is the comprehensive error, λ1 is the weight coefficient of position information error, and λ2 is the weight coefficient of attitude error;

[0050] Based on the comprehensive error term, a model for dynamically adjusting the attitude angle of the UAV is constructed, and the expression is:

[0051]

[0052] Among them, (μ adj ,φ adj ,ψ adj ) is the adjusted attitude angle, η is the adjustment step length;

[0053] While adjusting the attitude, the descent speed of the UAV is dynamically adjusted according to the comprehensive error term, and the expression is:

[0054] v ι,adj =v ι,max ·(1-tanh(κ·E total ));

[0055] Among them, v ι,adj is the adjusted descent speed, v ι,max is the maximum descent speed, κ is the error amplification factor;

[0056] The adjusted attitude angle and descent speed are combined to generate the descent trajectory of the UAV. The trajectory expression is:

[0057]

[0058] Among them, (x n+1,y n+1 ,z n+1 ) is the position coordinate of the drone at the next moment, (x n ,y n ,z n ) is the current position coordinate of the UAV, Δt is the time step, is the unit vector of the current posture direction.

[0059] As a preferred solution of the control method for landing of a drone based on AI vision described in the present invention, wherein: adjusting the motor thrust distribution according to the optimized attitude information and the descent trajectory combined with the optimal landing point includes the following steps:

[0060] Combining the descent trajectory and the thrust requirement, we get the total thrust requirement;

[0061] Allocate motor thrust direction factors according to the optimized attitude and position of the drone;

[0062] Combining the total thrust demand and the thrust direction factor, the actual thrust output of each motor is calculated as follows:

[0063]

[0064] Among them, U h is the thrust of the hth motor, R h is the thrust direction factor of the hth motor, δ τ is the offset between the current time and the optimal thrust adjustment time, τ max is the maximum allowable adjustment time, U total is the total thrust requirement of the UAV, τ is the current time, and h is the motor number of the UAV;

[0065] After the thrust distribution is completed, the acceleration, angular velocity and attitude information of the UAV are collected in real time to dynamically adjust the motor thrust distribution. The expression is:

[0066]

[0067] is the motor thrust after adjustment, q2 is the attitude error adjustment coefficient, is the rate of change of attitude error over time.

[0068] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the AI ​​vision-based drone landing control method as described in the first aspect of the present invention is implemented.

[0069] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the AI ​​vision-based drone landing control method as described in the first aspect of the present invention.

[0070] The beneficial effects of the present invention are as follows: by extracting features from ground comprehensive data through a convolutional neural network, a ground feature map is generated, the accuracy of the UAV in selecting a landing point in complex terrain and the efficiency of path planning are improved, the current position of the UAV in the process of moving along the optimal path is predicted, and its state can be dynamically evaluated during the movement of the UAV to ensure the execution accuracy of path planning; by extracting the nearest target point in the path planning result as the current expected position, the deviation information of the UAV can be effectively captured to ensure that the UAV can quickly identify the degree to which its state deviates from the target; by dynamically adjusting the attitude angle and descent speed of the UAV to generate optimized attitude information and descent trajectory, the risk of loss of control or deviation during landing can be effectively avoided; by combining the optimized attitude information and descent trajectory, the overall landing process is ensured to be smoother and more accurate, thereby improving the safe landing capability in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0072] Figure 1 This is a flow chart of the control method for drone landing based on AI vision in Example 1.

[0073] Figure 2 This is a flow chart of the position error and attitude error information of the control method for landing a drone based on AI vision in Example 1. DETAILED DESCRIPTION

[0074] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0075] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0076] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0077] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a control method for landing of a drone based on AI vision, comprising the following steps:

[0078] Collect comprehensive ground data and the current attitude data of the UAV;

[0079] Input the collected ground comprehensive data into the deep learning model to analyze ground features and generate the optimal landing point path planning results;

[0080] Predict the current position and attitude of the UAV based on the optimal landing point path planning results and ground comprehensive data;

[0081] According to the predicted current position and current attitude of the UAV, the position information error and attitude error information are calculated;

[0082] Based on the position information error and attitude error information, the attitude and descent speed of the UAV are adjusted to generate optimized attitude information and descent trajectory;

[0083] Adjust the motor thrust distribution according to the optimized attitude information and descent trajectory combined with the optimal landing point.

[0084] S1. Collecting ground comprehensive data and the current posture data of the UAV includes:

[0085] Collect comprehensive ground data including terrain and obstacle distribution data, and the current attitude data of the drone including the acceleration, angular velocity and current attitude information of the drone, and perform unified coordinate alignment and normalization processing;

[0086] Furthermore, the ground information is collected through the fusion of multiple sensors, combined with normalization and coordinate alignment processing to ensure the high accuracy and consistency of ground data;

[0087] Use IMU, gyroscope and other multi-sensor fusion algorithms to reduce noise interference and ensure the stability of drone attitude information.

[0088] S2. Inputting the collected ground comprehensive data into the deep learning model for ground feature analysis includes the following steps:

[0089] The convolutional neural network model is used to extract the features of the processed ground comprehensive data to obtain the ground feature map, which is combined with the current posture information of the drone to construct a comprehensive feature vector, which is expressed as:

[0090] V=[a,ω,ε,F ′ ];

[0091] Among them, a is the acceleration of the drone, ω is the angular velocity of the drone, ε is the current attitude information of the drone, and F ′ Ground feature map.

[0092] Furthermore, the acceleration, angular velocity and current attitude information of the drone are obtained from the IMU and gyroscope;

[0093] Combine terrain and geomorphic features with obstacle distribution features to generate a comprehensive ground feature map;

[0094] The state features of the UAV and the ground feature map are spliced ​​into a comprehensive feature vector;

[0095] It achieves efficient fusion of ground environment and UAV status, providing comprehensive input data for subsequent path planning and landing control.

[0096] S3. Use the multi-layer perceptron model to receive the comprehensive feature vector as input and predict the safety score of the landing point. The expression is:

[0097]

[0098] Among them, v j is the jth element in the eigenvector V, W j is the weight matrix W corresponding to v j , m is the dimension of the feature vector V, b is the bias term, and S is the safety score of the landing point;

[0099] Based on the safety score, the location with the highest score is selected as the optimal landing point. Based on the optimal landing point, the Dijkstra algorithm is used to find the best path from the current location to the optimal landing point. The expression is,

[0100]

[0101] Among them, P is the best path, is the set of all paths, p is a path from the starting point to the target point, k is the number of steps in path p, L k is the straight-line distance of the kth step, E k is the energy consumption in the kth step, T k is the terrain complexity factor in the kth step, S kis the safety score of the landing point at the kth step, α is the weight coefficient of the energy consumption item, β is the weight coefficient of the terrain complexity item, and γ is the weight coefficient of the safety score item of the landing point.

[0102] Furthermore, the value range of the safety score of the landing point is [0, 1]. When the value of the safety score of the landing point is closer to 1, it means that the landing point is safer, and when it is closer to 0, it means that the point has a higher risk;

[0103] By selecting the landing point with the highest safety score, the drone can be prioritized to land in the area with the lowest risk and the best conditions;

[0104] The scoring takes into account the complexity of the terrain and the distribution of obstacles, avoiding the shortcomings of the traditional method of selecting landing points with a single indicator;

[0105] Set the current position of the drone as the starting point, the optimal landing point as the end point, and select the path with the lowest comprehensive weight value as the best path;

[0106] By considering multiple factors such as energy consumption, terrain complexity and safety, an optimal path is generated that balances various indicators.

[0107] S4, based on the optimal landing point path planning result and the ground comprehensive data, predicting the current position and current attitude of the UAV includes the following steps:

[0108] The Kalman filter is used in combination with the speed of the UAV to predict the current position of the UAV as it moves along the optimal path. The expression is:

[0109] x t =x t-1 +Δt(v x cosθ-v y sinθ);

[0110] y t =y t-1 +Δt(v x sinθ+v y cosθ);

[0111] z t =z t-1 +Δtv z ;

[0112] Among them, x t is the x-axis position coordinate predicted by the drone at the current time t, x t-1 is the x-axis position coordinate predicted by the drone at the previous moment t-1, y t is the predicted y-axis position coordinate of the drone at the current time t, y t-1is the predicted y-axis position coordinate of the drone at the previous moment t-1, z t is the z-axis position coordinate predicted by the drone at the current time t, z t-1 is the predicted z-axis position coordinate of the drone at the previous moment t-1, v x is the velocity component of the drone along the x-axis, v y is the velocity component of the drone along the y-axis, v z is the velocity component of the UAV along the z-axis, and Δt is the time step.

[0113] The attitude is composed of pitch angle, roll angle and yaw angle. The current attitude of the drone is predicted by IMU and gyroscope. The expression is:

[0114]

[0115] φ=arctan 2(a y ,a z );

[0116] ψ=ψ prev +∫ω z dt;

[0117] Among them, μ is the pitch angle of the UAV, φ is the roll angle of the UAV, ψ is the yaw angle of the UAV, and ψ prev is the yaw angle of the previous step, ω z is the change in yaw angle, a x is the acceleration of the drone along the x-axis in the body coordinate system, a y is the acceleration of the drone along the y-axis in the body coordinate system, a z is the acceleration of the UAV along the z-axis in the body coordinate system, x is the x-axis direction of the UAV in the body coordinate system, y is the y-axis direction of the UAV in the body coordinate system, and z is the z-axis direction of the UAV in the body coordinate system.

[0118] Furthermore, the pitch angle is calculated by the acceleration a of the drone along the x-axis in the body coordinate system. x and total acceleration It is calculated to reflect the tilt of the drone relative to the horizontal plane;

[0119] The roll angle is calculated by the arctan 2 function, which accepts two parameters a x and a y , returns an angle value between -π and π, avoiding the four-quadrant ambiguity problem that may be caused by the traditional atan function;

[0120] The yaw angle is obtained by integrating the angular velocity ω z To calculate, this method can continuously track the changes in the steering angle of the drone;

[0121] Get the initial position and initial velocity of the drone, and get the current acceleration and angular velocity from the IMU;

[0122] Get the latest position from the GPS sensor, use the acceleration data to update the speed of the drone, and use the updated speed data to predict the new position.

[0123] S5. Find the path point sequence from the current position of the drone to the optimal landing point from the path planning results, and extract

[0124] The nearest target point is taken as the current expected position, and the error between the current position and the expected position is calculated. The expression is:

[0125]

[0126] Among them, Δ pos is the current position error, x O Predict the x-axis expected position coordinates for the drone, y O Predict the desired position coordinates of the y-axis for the drone; O Predict the desired z-axis position coordinates for the drone;

[0127] According to the direction of path planning, the expected attitude angle is calculated and compared with the current attitude angle to obtain the attitude error. The expression is:

[0128]

[0129] in, is the desired pitch angle of the drone, is the desired roll angle of the drone, For drones

[0130] The desired yaw angle, Δ attitude is the attitude error;

[0131] Furthermore, the square of the difference between the desired pitch angle and the current pitch angle is

[0132] The square of the difference between the desired roll angle and the current roll angle,

[0133] The square of the difference between the desired yaw angle and the current yaw angle,

[0134] The attitude error is synthesized based on the square root of the sum of the squares of the above three errors;

[0135] The best path from the current position to the optimal landing point is obtained through the path planning algorithm. The path consists of a series of path points P = {p1, p2, ..., p n}, where each point pi =(x i ,y i ,z i ) represents a three-dimensional coordinate;

[0136] Among them, p i is the i-th path point on the optimal path, ranging from 1 to n, x i represents the coordinate value of the optimal path point on the x-axis, y i is the coordinate value of the optimal path point on the y axis, z i To represent the coordinate value of the optimal path point on the z-axis;

[0137] According to the extracted nearest target point, the error between the current position and the expected position is calculated;

[0138] It can accurately extract the nearest target point from the path planning results as the current desired position, and can also calculate the error between the current position and the desired position.

[0139] S6, based on the position information error and attitude error information, adjusting the attitude and descent speed of the UAV, generating optimized attitude information and descent trajectory includes the following steps:

[0140] According to the position information error and attitude error information, a comprehensive error model is constructed for subsequent attitude and speed adjustment. The comprehensive error expression is:

[0141]

[0142] Among them, E total is the comprehensive error, λ1 is the weight coefficient of position information error, and λ2 is the weight coefficient of attitude error;

[0143] Based on the comprehensive error term, a model for dynamically adjusting the attitude angle of the UAV is constructed, and the expression is:

[0144]

[0145] Among them, (μ adj ,φ adj ,ψ adj ) is the adjusted attitude angle, η is the adjustment step length;

[0146] While adjusting the attitude, the descent speed of the UAV is dynamically adjusted according to the comprehensive error term, and the expression is:

[0147] v ι,adj =v ι,max ·(1-tanh(κ·E total ));

[0148] Among them, v ι,adj is the adjusted descent speed, vι,max is the maximum descent speed, κ is the error amplification factor;

[0149] The adjusted attitude angle and descent speed are combined to generate the descent trajectory of the UAV. The trajectory expression is:

[0150]

[0151] Among them, (x n+1 ,y n+1 ,z n+1 ) is the position coordinate of the drone at the next moment, (x n ,y n ,z n ) is the current position coordinate of the UAV, Δt is the time step, is the unit vector of the current posture direction;

[0152] Furthermore, calculation The comprehensive error needs to be differentiated, and the expression is,

[0153]

[0154] is the partial derivative of the comprehensive error with respect to the pitch angle μ, is the partial derivative of the comprehensive error with respect to the roll angle φ, is the partial derivative of the comprehensive error with respect to the yaw angle ψ;

[0155] Dynamically adjust the descent speed of the drone by introducing a function to achieve nonlinear adjustment, so that when the comprehensive error is large, the descent speed is significantly reduced, thereby preventing the drone from approaching the ground too quickly or deviating from the path; when the comprehensive error is small, the descent speed is close to the maximum value, improving the landing efficiency;

[0156] Get the initial position and adjusted descent speed of the drone;

[0157] According to the adjusted attitude angle, calculate the unit vector of the current attitude direction of the drone;

[0158] Update the drone's position coordinates using the adjusted descent speed and time step.

[0159] S7, adjusting the motor thrust distribution according to the optimized attitude information and descent trajectory combined with the optimal landing point includes the following steps:

[0160] Combining the descent trajectory and the thrust requirement, we get the total thrust requirement;

[0161] Allocate motor thrust direction factors according to the optimized attitude and position of the drone;

[0162] Combining the total thrust demand and the thrust direction factor, the actual thrust output of each motor is calculated as follows:

[0163]

[0164] Among them, U h is the thrust of the hth motor, R h is the thrust direction factor of the hth motor, δ τ is the offset between the current time and the optimal thrust adjustment time, τ max is the maximum allowable adjustment time, U total is the total thrust requirement of the UAV, τ is the current time, and h is the motor number of the UAV;

[0165] After the thrust distribution is completed, the acceleration, angular velocity and attitude information of the UAV are collected in real time to dynamically adjust the motor thrust distribution. The expression is:

[0166]

[0167] is the motor thrust after adjustment, q2 is the attitude error adjustment coefficient, is the rate of change of attitude error over time.

[0168] Further, the total thrust required to calculate the UAV involves the mass, gravity, and descent acceleration of the UAV, and the expression is,

[0169] Among them, T total is the total thrust requirement of the UAV, m is the mass of the UAV, g is the acceleration of gravity, E att To quantify the overall posture deviation, is the thrust correction coefficient of attitude error;

[0170] Multiply the attitude error change rate by the adjustment coefficient q2 and add it to the original thrust output U h Get the adjusted thrust output

[0171] This embodiment also provides a computer device, which is suitable for the control method of drone landing based on AI vision, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the control method of drone landing based on AI vision as proposed in the above embodiment.

[0172] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0173] The present embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the control method for realizing the landing of a drone based on AI vision as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a disk or an optical disk.

[0174] In summary, the present invention uses a convolutional neural network to extract features from ground comprehensive data and generate a ground feature map, which improves the accuracy of the drone's landing point selection in complex terrain and the efficiency of path planning, predicts the current position of the drone during its movement along the optimal path, and can dynamically evaluate its state during the movement of the drone to ensure the execution accuracy of path planning. By extracting the nearest target point in the path planning result as the current expected position, the deviation information of the drone can be effectively captured to ensure that the drone can quickly identify the degree to which its state deviates from the target. By dynamically adjusting the attitude angle and descent speed of the drone to generate optimized attitude information and descent trajectory, the risk of loss of control or deviation during landing can be effectively avoided. By combining the optimized attitude information and descent trajectory, the overall landing process is ensured to be smoother and more accurate, thereby improving the safe landing capability.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A control method for drone landing based on AI vision, characterized in that: include, Collect comprehensive ground data and the current attitude data of the UAV; Input the collected ground comprehensive data into the deep learning model to analyze ground features and generate the optimal landing point path planning results; Predict the current position and attitude of the UAV based on the optimal landing point path planning results and ground comprehensive data; According to the predicted current position and current attitude of the UAV, the position information error and attitude error information are calculated; Based on the position information error and attitude error information, the attitude and descent speed of the UAV are adjusted to generate optimized attitude information and descent trajectory; Adjust the motor thrust distribution according to the optimized attitude information and descent trajectory combined with the optimal landing point.

2. The AI ​​vision-based UAV landing control method according to claim 1, characterized in that: Collecting comprehensive ground data and the current posture data of the UAV includes: The collected ground comprehensive data includes terrain and obstacle distribution data, and the current posture data of the UAV includes the acceleration, angular velocity and current posture information of the UAV, and unified coordinate alignment and normalization processing are performed.

3. The AI ​​vision-based UAV landing control method according to claim 2, characterized in that: Input the collected ground comprehensive data into the deep learning model for ground feature analysis The following steps are included: The convolutional neural network model is used to extract the features of the processed ground comprehensive data to obtain the ground feature map, which is combined with the current posture information of the drone to construct a comprehensive feature vector, which is expressed as: V=[a,ω,ε,F ′ ]; Among them, α is the acceleration of the drone, ω is the angular velocity of the drone, and ε is the current attitude information of the drone. F ′ Ground feature map.

4. The AI ​​vision-based UAV landing control method according to claim 3, characterized in that: According to the ground feature analysis results, generating the optimal landing point and path planning results includes the following steps: The multi-layer perceptron model receives the comprehensive feature vector as input and predicts the safety score of the landing point. The expression is: Among them, v j is the jth element in the eigenvector V, W j is the weight matrix W corresponding to v j The right weight, m ​​is the dimension of the feature vector V, b is the bias term, and S is the safety score of the landing point; Based on the safety score, the location with the highest score is selected as the optimal landing point. Based on the optimal landing point, the Dijkstra algorithm is used to find the best path from the current location to the optimal landing point. The expression is, Among them, P is the best path, is the set of all paths, p is a path from the starting point to the target point, k is the number of steps in path p, L k is the straight-line distance of the kth step, E k is the energy consumption in the kth step, T k is the terrain complexity factor in the kth step, S k is the safety score of the landing point at the kth step, α is the weight coefficient of the energy consumption item, β is the weight coefficient of the terrain complexity item, and γ is the weight coefficient of the safety score item of the landing point.

5. The AI ​​vision-based UAV landing control method according to claim 4, characterized in that: Based on the optimal landing point path planning results and ground comprehensive data, predicting the current position and current attitude of the drone includes the following steps: The Kalman filter is used in combination with the speed of the UAV to predict the current position of the UAV as it moves along the optimal path. The expression is: x t =x t-1 +Δt(v x cosθ-v y sinθ); y t =y t-1 +Δt(v x sinθ+v y cosθ); with t =with t-1 +Δtv z ; Among them, x t is the x-axis position coordinate predicted by the drone at the current time t, x t-1 is the x-axis position coordinate predicted by the drone at the previous moment t-1, yt is the y-axis position coordinate predicted by the drone at the current moment t, t-1 is the predicted y-axis position coordinate of the drone at the previous moment t-1, z t is the z-axis position coordinate predicted by the drone at the current time t, z t-1 is the predicted z-axis position coordinate of the drone at the previous moment t-1, v x is the velocity component of the drone along the x-axis, v y is the velocity component of the drone along the y-axis, v z is the velocity component of the UAV along the z-axis, Δt is the time step; The attitude is composed of pitch angle, roll angle and yaw angle. The current attitude of the drone is predicted by IMU and gyroscope. The expression is: φ=arctan 2(a y ,a z ); ψ=ψ prev +∫ω z dt; Among them, μ is the pitch angle of the UAV, φ is the roll angle of the UAV, ψ is the yaw angle of the UAV, and ψ prev is the yaw angle of the previous step, ω z is the change in yaw angle, a x is the acceleration of the drone along the x-axis in the body coordinate system, a y is the acceleration of the drone along the y-axis in the body coordinate system, a z is the acceleration of the UAV along the z-axis in the body coordinate system, x is the x-axis direction of the UAV in the body coordinate system, y is the y-axis direction of the UAV in the body coordinate system, and z is the z-axis direction of the UAV in the body coordinate system.

6. The AI ​​vision-based UAV landing control method according to claim 5, characterized in that: According to the predicted current position and current attitude of the UAV, calculating the position information error and attitude error information includes the following steps: From the path planning results, find the path point sequence from the current position of the drone to the optimal landing point, extract the nearest target point as the current expected position, and calculate the error between the current position and the expected position. The expression is: Among them, Δ pos is the current position error, x O Predict the x-axis expected position coordinates for the drone, y O Predict the desired position coordinates of the y-axis for the drone; O Predict the desired z-axis position coordinates for the drone; According to the direction of path planning, the expected attitude angle is calculated and compared with the current attitude angle to obtain the attitude error. The expression is: in, is the desired pitch angle of the drone, is the desired roll angle of the drone, is the desired yaw angle of the UAV, Δ attitude is the attitude error.

7. The AI ​​vision-based UAV landing control method according to claim 6, characterized in that: Based on the position information error and attitude error information, the attitude and descent speed of the UAV are adjusted to generate the optimized attitude information and descent trajectory, including the following steps: According to the position information error and attitude error information, a comprehensive error model is constructed for subsequent attitude and speed adjustment. The comprehensive error expression is: Among them, E total is the comprehensive error, λ1 is the weight coefficient of position information error, and λ2 is the weight coefficient of attitude error; Based on the comprehensive error term, a model for dynamically adjusting the attitude angle of the UAV is constructed, and the expression is: Among them, (μ adj ,φ adj , ψ adj ) is the adjusted attitude angle, η is the adjustment step length; While adjusting the attitude, the descent speed of the UAV is dynamically adjusted according to the comprehensive error term, and the expression is: v ι，adj =v ι，max ·(1-tanh(κ·E total )); Among them, v ι，adj is the adjusted descent speed, v ι，max is the maximum descent speed, κ is the error amplification factor; The adjusted attitude angle and descent speed are combined to generate the descent trajectory of the UAV. The trajectory expression is: Among them, (x t+1 ,y t+1 , z t+1 ) is the position coordinate of the drone at the next moment, (x t ,y t , z t ) is the current position coordinate of the UAV, Δt is the time step, is the unit vector of the current posture direction.

8. The AI ​​vision-based UAV landing control method according to claim 7, characterized in that: According to the optimized attitude information and descent trajectory combined with the optimal landing point, adjusting the motor thrust distribution includes the following steps: Combining the descent trajectory and the thrust requirement, we get the total thrust requirement; Allocate motor thrust direction factors according to the optimized attitude and position of the drone; Combining the total thrust demand and the thrust direction factor, the actual thrust output of each motor is calculated as follows: Among them, U h is the thrust of the hth motor, R h is the thrust direction factor of the hth motor, δ τ is the offset between the current time and the optimal thrust adjustment time, τ max is the maximum allowable adjustment time, U total is the total thrust requirement of the UAV, τ is the current time, and h is the motor number of the UAV; After the thrust distribution is completed, the acceleration, angular velocity and attitude information of the UAV are collected in real time to dynamically adjust the motor thrust distribution. The expression is: is the motor thrust after adjustment, q2 is the attitude error adjustment coefficient, is the rate of change of attitude error over time.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the AI ​​vision-based drone landing control method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AI ​​vision-based drone landing control method according to any one of claims 1 to 7 are implemented.

Images