A UAV tracking control method after the vehicle target is out of sight

Through pure visual prediction and neural network control, drones can automatically track and search after losing target vehicles, solving the problems of difficulty and high cost of information interaction in the existing technology, and improving the success rate and flexibility of drones following vehicles sex.

CN118519446BActive Publication Date: 2024-11-15CHONGQING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410592987.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-14
Publication Date
2024-11-15
Estimated Expiration
2044-05-14

AI Technical Summary

Technical Problem

The existing drone follow-up technology relies on the positioning device and multiple sensors of the target vehicle, which makes information interaction difficult, costly, and difficult to retrieve when the target vehicle is lost.

Method used

Through pure visual methods, predict the coordinate trajectory and search direction of the target vehicle, combined with LSTM neural network and dynamic model, the drone automatically tracks and searches for the target vehicle, and uses multi-angle image data to train the convolutional neural network for target recognition and calculate expectations Speed ​​and direction, weighted prediction trajectory is performed to improve search success rate.

Benefits of technology

Following can be achieved without installing a positioning communication system for the target vehicle, which significantly improves the search success rate of target vehicle, reduces costs, and effectively retrieves the target vehicle in the event of loss.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118519446B_ABST
    Figure CN118519446B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of drone control technology, and in particular to a drone tracking control method after a vehicle target leaves the field of view. It includes the following of a drone automatically following a target vehicle; predicting the predicted coordinate trajectory of the target vehicle in the world coordinate system based on the change data of the target vehicle in the world coordinate system before it disappears; calculating the drone search speed and search direction based on the change data of the target vehicle in the image frame coordinate system before it disappears; and weighting the predicted coordinate trajectory and the search coordinate trajectory to determine the expected search trajectory of the drone. The present invention can complete drone following a vehicle in a purely visual manner, and can retrieve the target vehicle after it is lost; the present invention can significantly improve the success rate of the target vehicle search by integrating two methods of searching for the target vehicle in a weighted manner according to different situations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle control, and in particular to a method for tracking and controlling an unmanned aerial vehicle after a vehicle target leaves a field of view. Background Art

[0002] In recent years, drone-following car photography has become a new demand. With the continuous advancement of technology, from the initial manual operation of drone following to the automatic following of drones based on multi-sensor combinations, automatic following of drones has become the mainstream demand for drone-following cars.

[0003] However, most current drone-based vehicle following still relies on positioning devices and lidar installed on the target vehicle. The shortcomings are: 1. Before following a vehicle, the drone needs to establish a clear information exchange method with the target vehicle, and cannot follow a target vehicle that cannot exchange information; 2. The use of multiple sensors increases the cost of drone-based vehicle following; 3. During the vehicle following process, if the target vehicle is lost and the positioning interaction signal is weak, there is a lack of specific means to retrieve the target vehicle. Summary of the invention

[0004] The present invention discloses a method for tracking and controlling a UAV after a vehicle target leaves the field of view. The specific method is as follows:

[0005] The drone automatically follows the target vehicle;

[0006] When the target vehicle disappears from the image currently captured by the drone, the image data before disappearance captured by the drone within a first preset time period before disappearance is obtained;

[0007] Predicting the predicted coordinate trajectory of the target vehicle in the world coordinate system according to the change data of the target vehicle in the world coordinate system before it disappears;

[0008] Calculate the UAV search speed and search direction based on the change data of the target vehicle in the image frame coordinate system before it disappears;

[0009] Determining the expected search speed of the drone based on the drone search speed;

[0010] Calculate the search coordinate trajectory of the drone in the world coordinate system based on the drone search direction;

[0011] The weighted predicted coordinate trajectory and search coordinate trajectory are used to determine the expected search trajectory of the UAV.

[0012] The advantages of this embodiment are that drone following can be completed in a purely visual manner, and the target vehicle can be retrieved after being lost; it integrates two methods of searching for target vehicles, one of which is for sudden loss of the target due to shielding and the like, and the future trajectory of the vehicle can be predicted based on the vehicle's driving history data; the other is for gradual loss of the target due to changes in speed or wind direction, and the direction of travel of the vehicle before the loss is used as the search direction; the two methods of searching for target vehicles are integrated by allocating weights according to different situations, which can significantly improve the success rate of searching for the target vehicle.

[0013] Furthermore, the drone automatically follows the target vehicle, and the specific method is as follows:

[0014] Manually control the drone to follow the target vehicle and switch between several shooting angles to obtain multi-angle image data of the target vehicle;

[0015] The target recognition convolutional neural network is trained with multi-angle image data, and the target recognition convolutional neural network is used to recognize the target vehicle in the image data;

[0016] Mark the target coordinates and target size in the image frame coordinate system where the center of the target vehicle in the multi-angle image data is located;

[0017] Selecting one of the multi-angle image data as reference image data;

[0018] Keep the current motion state of the drone, stop manually controlling the drone, use the target coordinates and target size as the desired target, and the drone automatically tracks the target vehicle.

[0019] The advantage of this embodiment is that as long as the multi-angle image data of the target vehicle is obtained, the target vehicle can be followed, and there is no need to install a positioning communication system on the target vehicle, which makes it easy to change the following target of the drone.

[0020] Furthermore, the target coordinates and target size in the image frame coordinate system where the center of the target vehicle in the multi-angle image data is located are marked. The specific method is as follows:

[0021] Calculate the ratio of the target vehicle area to the image frame area as the target size;

[0022] The center of the image frame is used as the origin of the image frame coordinate system, and the target coordinates are determined according to the specific spatial position of the target vehicle center in the image frame coordinate system.

[0023] The advantage of this embodiment is that by using a ratio as the tracking expectation, the flight altitude and the following angle can be adjusted together, resulting in a better following effect.

[0024] Furthermore, the reference image data is selected, and the specific method is as follows:

[0025] The image data in which the proportion of the target vehicle in the multi-angle image data is within a preset range and the center of the target vehicle is closest to the center of the image is selected as the reference image data.

[0026] The advantage of this embodiment is that when the center of the target vehicle is at the center of the image, the tracking margin around it is maximum and the target is not easily lost.

[0027] Furthermore, the drone automatically tracks the target vehicle, and the specific method is as follows:

[0028] Acquire a number of historical image data at equal time intervals when the speed and direction of the drone remain unchanged within a second preset period before the current moment;

[0029] Calculate the size difference between the target vehicle size and the target size in each historical image data to construct a size difference sequence, and calculate the coordinate difference between the target vehicle coordinates and the target coordinates in each historical image data to construct a coordinate difference sequence;

[0030] Taking the size difference sequence and the coordinate difference sequence as input, the first LSTM neural network is used to predict the expected speed variable and expected direction variable of the drone at the next moment;

[0031] The UAV dynamics model is used to solve the rotational speeds of the four motors of the UAV at the next moment, so that the UAV can automatically track the target vehicle.

[0032] The advantage of this embodiment is that the LSTM neural network is used as the main algorithm for controlling the four motors, and there is no need to set up an additional controller. During the control process, the four motors are adjusted according to the image observation results. As the control time increases, the control effect will become better and better.

[0033] Furthermore, the coordinates of the target vehicle in the world coordinate system are calculated as follows:

[0034] Get the coordinates of the drone in the world coordinate system when the image was taken;

[0035] Read the relative distance between the target vehicle and the drone measured by the ranging radar on the drone;

[0036] According to the coordinates of the UAV in the world coordinate system and the relative distance, the coordinates of the target vehicle in the world coordinate system are calculated.

[0037] Furthermore, the predicted coordinate trajectory of the target vehicle in the world coordinate system is predicted, and the specific method is as follows:

[0038] The change data of the target vehicle in the world coordinate system before it disappears is input into the second LSTM neural network to predict the predicted coordinate trajectory in the world coordinate system after the target vehicle disappears.

[0039] The advantage of this embodiment is that the target trajectory of the target vehicle before it disappears is recorded, and the target trajectory of the target vehicle before it disappears is used to predict the future driving trajectory of the target vehicle. For situations where the target vehicle disappears suddenly due to occlusion, the prediction effect is significantly improved.

[0040] Furthermore, the UAV search speed is calculated as follows:

[0041] According to the change data of the target vehicle in the image frame coordinate system before it disappears, the coordinate change data of the target vehicle in the world coordinate system is calculated;

[0042] Calculating the moving speed of the target vehicle in the world coordinate system according to the coordinate change data of the target vehicle in the world coordinate system;

[0043] The target vehicle's moving speed in the world coordinate system plus the preset speed is used as the drone's search speed;

[0044] Calculate the search direction of the drone. The specific method is as follows:

[0045] According to the change data of the target vehicle in the image frame coordinate system before it disappears, the disappearance direction of the target vehicle compared to the UAV is determined, and the disappearance direction of the target vehicle compared to the UAV is extended as the search direction of the UAV.

[0046] The advantage of this embodiment is that the disappearance direction of the target vehicle in the image frame is used as the future driving trajectory of the target vehicle, which can significantly improve the tracking effect in response to the gradual loss of the target due to speed changes; the formulated search speed is based on the speed of the target vehicle before it disappears, which can ensure that the search speed is higher than the driving speed of the target vehicle, and can also ensure that the search effect will not be reduced due to the drone's too fast search speed.

[0047] Furthermore, when weighted predicting coordinate tracks and searching coordinate tracks, the weights are allocated as follows:

[0048] Two sets of allocation weights are preset, in which the weight of the predicted coordinate trajectory in the first set of allocation weights is greater than the weight of the search coordinate trajectory, and in the second set of allocation weights, the weight of the predicted coordinate trajectory is less than the weight of the search coordinate trajectory;

[0049] If the target vehicle gradually disappears in the image frame, the second set of distribution weights is selected;

[0050] If the target vehicle disappears from the image frame for a short time, the first set of allocation weights is selected.

[0051] The advantage of this embodiment is that the two weight allocation methods are respectively aimed at different target loss situations, and the two search strategies are integrated through weights, which can significantly improve the overall search effect.

[0052] Furthermore, the method for obtaining the first LSTM neural network training sample is as follows:

[0053] Release variable speed and direction tracking samples, as well as fixed speed and direction drone samples;

[0054] When the tracking sample is offset / changed in size in the drone camera, the drone sample is manually manipulated to return the tracking sample to its initial state in the drone camera;

[0055] Repeat several times and record the speed and direction variables of the drone during manual control;

[0056] The size difference sequence and coordinate difference sequence of the tracking sample when it is offset / sized change in the drone camera are used as training sample input, and the speed variable and direction variable of the manually controlled drone are used as training sample output.

[0057] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings of the present invention are as follows.

[0059] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0060] Figure 2 Schematic diagram of the process of UAV tracking the target vehicle before the target vehicle disappears.

[0061] Figure 3 Schematic diagram of the automatic tracking process of the drone before the target vehicle disappears. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0063] A UAV tracking control method after the vehicle target is out of sight, such as Figure 1 As shown, the specific steps are as follows:

[0064] S1. The drone automatically follows the target vehicle.

[0065] In step S1, the drone automatically follows the target vehicle. Figure 2 As shown, the specific method is as follows:

[0066] S11, manually control the drone to follow the target vehicle, switch several shooting angles to obtain multi-angle image data of the target vehicle {image1, image2, ..., image n}, n is the image number.

[0067] S12, training a target recognition convolutional neural network with the multi-angle image data, and identifying a target vehicle in the image data with the target recognition convolutional neural network;

[0068] In step S12, the CNN convolutional neural network includes an input layer, a convolutional layer, a pooling layer and a fully connected layer;

[0069] The convolution process formula is as follows:

[0070] X out =f(X in .W+b)

[0071] Among them, X in It is the convolution layer input, that is, image n , X out is the convolution output, W is the convolution layer weight parameter, and b is the bias.

[0072] The formula for the pooling process is as follows:

[0073] X out =pooling(X in )

[0074] Among them, X in is the pooling layer input, X out is the output of the pooling layer.

[0075] The formula for the full connection process is as follows:

[0076] F(X)=f(X.W+b)

[0077] Among them, X is the output of the pooling layer, W is the weight parameter of the fully connected layer, and b is the bias.

[0078] The loss function formula of CNN convolutional neural network is:

[0079]

[0080] Among them, y is the true value, is the predicted value.

[0081] The activation function of the CNN convolutional neural network is ReLU, and the formula is:

[0082] CNNC convolutional neural network optimizer SGD optimization algorithm updates model parameters by updating the gradient of samples in each iteration and setting a fixed learning rate.

[0083] S13, marking the target coordinates and target size in the image frame coordinate system where the center of the target vehicle in the multi-angle image data is located, the specific steps are as follows:

[0084] S131, calculating the ratio of the target vehicle area to the image frame area as the target size, the calculation formula is:

[0085]

[0086] In the formula, size target is the target size, S target is the area occupied by the target vehicle in the image, S whole is the overall area of ​​the image frame.

[0087] S132, taking the center of the image frame as the origin of the image frame coordinate system O image (X image,0 ,Y image,0 ), the target coordinates O are determined according to the specific spatial position of the image frame coordinate system where the center of the target vehicle is located image (X image,tar ,Y image,tar ).

[0088] S14, selecting one of the multi-angle image data as the reference image data image target , the specific method is:

[0089] The image data whose target vehicle proportion is within the preset range and whose center is closest to the image center is selected as the reference image data from the multi-angle image data, that is,

[0090]

[0091] S15, keep the current motion state of the drone, stop manual control of the drone, take the target coordinates and target size as the desired target, and the drone automatically tracks the target vehicle, such as Figure 3 As shown, the specific method is as follows:

[0092] S151, obtaining a number of historical image data of equal time intervals when the speed and direction of the drone remain unchanged within a second preset period before the current moment {image t-1 、image t-2 、…、image t-n}.

[0093] S152, respectively calculate the size difference between the target vehicle size and the target size in each historical image data to construct a size difference sequence {Δsize t-1 , Δsize t-2 ,…,Δsizet-n}, and the coordinate difference between the target vehicle coordinates and the target coordinates in each historical image data to construct a coordinate difference sequence {ΔO t-1 , ΔO t-2 ,…,ΔO t-n}.

[0094] in,

[0095] S153. Using the size difference sequence and the coordinate difference sequence as input, the first LSTM neural network is used to predict the expected speed variable and the expected direction variable of the drone at the next moment.

[0096] In step S153, the steps for obtaining the first LSTM neural network training sample are as follows:

[0097] Release variable speed and direction tracking samples, as well as fixed speed and direction drone samples;

[0098] When the tracking sample is offset / changed in size in the drone camera, the drone sample is manually manipulated to return the tracking sample to its initial state in the drone camera;

[0099] Repeat several times and record the speed and direction variables of the drone during manual control;

[0100] The size difference sequence and coordinate difference sequence of the tracking sample when it is offset / sized change in the drone camera are used as training sample input, and the speed variable and direction variable of the manually controlled drone are used as training sample output.

[0101] S154. Using the UAV dynamics model, the rotation speeds of the four motors of the UAV at the next moment are solved to complete the automatic tracking of the target vehicle by the UAV.

[0102] The LSTM neural network consists of an input gate, a forget gate, and an output gate.

[0103] The update formula of the forget gate is as follows:

[0104] F T =σ(W f .[H T-1 ,X T ]+b f )

[0105] Among them, H T-1 is the state value of the hidden layer at the previous moment, X T is the current input, W f is the weight matrix, b f is the bias matrix, σ is the activation function, F T is the output of the forget gate.

[0106] The input gate update formula is as follows:

[0107] I T =σ(W i .[H T-1 ,X T ]+b i )

[0108] Among them, H T-1 is the state value of the hidden layer at the previous moment, X T is the current input, W i is the weight matrix, b i is the bias matrix, σ is the activation function, I T is the output of the input gate.

[0109] S2. When the target vehicle disappears from the image currently captured by the drone, obtain the image data before disappearance captured by the drone within a first preset time period before disappearance.

[0110] S3. Predict the target vehicle's predicted coordinate trajectory traj in the world coordinate system based on the change data of the target vehicle in the world coordinate system before it disappears. for .

[0111] In step S3, the coordinates of the target vehicle in the world coordinate system are calculated as follows:

[0112] S31, obtain the coordinates O of the drone in the world coordinate system when the image is taken. world (X world,air ,Y world, a ir ).

[0113] S32, reading the relative distance (ΔX, ΔY) between the target vehicle and the UAV measured by the ranging radar on the UAV.

[0114] S33, according to the coordinates and relative distance of the UAV in the world coordinate system, calculate the coordinates of the target vehicle in the world coordinate system world (X world,tar ,Y world,tar ).

[0115] S34, predicting the coordinate trajectory of the target vehicle in the world coordinate system using a second LSTM neural network.

[0116] S4. Calculate the search speed and search direction of the UAV based on the change data of the target vehicle in the image frame coordinate system before it disappears.

[0117] The change data of the target vehicle in the image frame coordinate system before it disappears is:

[0118] O image,t-1 (X image,tar ,Yimage,tar ) t-1 , O image,t-2 (X image,tar ,Y image,tar ) t-2 ,…,O image,t-n (X image,tar ,Y image,tar ) t-n

[0119] The search speed calculation formula is as follows:

[0120]

[0121] Where f is the sampling frequency.

[0122] The search direction uses the coordinates of the target vehicle in the image frame coordinate system before it disappears as nodes, uses interpolation method to form a trajectory curve, and extends the trajectory curve as the search direction.

[0123] S5. Determine the expected search speed of the drone according to the search speed of the drone.

[0124] S6. Calculate the search coordinate trajectory traj of the drone in the world coordinate system based on the drone search direction sea .

[0125] S7. Weighted predicted coordinate trajectory and search coordinate trajectory to determine the expected search trajectory of the UAV.

[0126] In step S7, when weighted prediction coordinate trajectory and search coordinate trajectory are performed, the weight distribution method is as follows:

[0127] Two sets of allocation weights are preset, in which the weight of the predicted coordinate trajectory in the first set of allocation weights T1 (w1, w2) is greater than the weight of the search coordinate trajectory, and in which the weight of the predicted coordinate trajectory in the second set of allocation weights T2 (w1, w2) is less than the weight of the search coordinate trajectory;

[0128] If the target vehicle gradually disappears in the image frame, the second set of distribution weights is selected;

[0129] If the target vehicle disappears from the image frame for a short time, the first set of allocation weights is selected.

[0130] traj exp =traj for .w1+traj sea .w2

[0131] S8. Search for the target vehicle according to the desired search trajectory through the UAV dynamics equation.

[0132]

[0133] Among them, I is the moment of inertia matrix, the position vector is [x, y, z], and the attitude vector is The velocity vector is Euler angular velocity vector U1 is the virtual control amount of height, U2 is the virtual control amount of roll, U3 is the virtual control amount of pitch, U4 is the virtual control amount of yaw channel, J RP is the propeller moment of inertia, Ω is the algebraic sum of the motor speed, k θ , k ψ They are the components of the total resistance torque coefficient in the X, Y, and Z axes respectively.

[0134] Finally, 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 above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for tracking and controlling a drone after a vehicle target leaves the field of view, characterized in that: The specific method is as follows: The drone automatically follows the target vehicle; When the target vehicle disappears from the image currently captured by the drone, the image data before disappearance captured by the drone within a first preset time period before disappearance is obtained; Predicting the predicted coordinate trajectory of the target vehicle in the world coordinate system according to the change data of the target vehicle in the world coordinate system before it disappears; Calculate the UAV search speed and search direction based on the change data of the target vehicle in the image frame coordinate system before it disappears; Determining the expected search speed of the drone based on the drone search speed; Calculate the search coordinate trajectory of the drone in the world coordinate system based on the drone search direction; The weighted predicted coordinate trajectory and search coordinate trajectory are used to determine the expected search trajectory of the UAV.

2. The method for tracking and controlling a drone after a vehicle target leaves the field of view according to claim 1, characterized in that: The drone automatically follows the target vehicle. The specific method is as follows: Manually control the drone to follow the target vehicle and switch between several shooting angles to obtain multi-angle image data of the target vehicle; The target recognition convolutional neural network is trained with multi-angle image data, and the target recognition convolutional neural network is used to recognize the target vehicle in the image data; Mark the target coordinates and target size in the image frame coordinate system where the center of the target vehicle in the multi-angle image data is located; Selecting one of the multi-angle image data as reference image data; Keep the current motion state of the drone, stop manually controlling the drone, use the target coordinates and target size as the desired target, and the drone automatically tracks the target vehicle.

3. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 2, characterized in that: Mark the target coordinates and target size in the image frame coordinate system where the target vehicle center is located in the multi-angle image data. The specific method is as follows: Calculate the ratio of the target vehicle area to the image frame area as the target size; The center of the image frame is used as the origin of the image frame coordinate system, and the target coordinates are determined according to the specific spatial position of the target vehicle center in the image frame coordinate system.

4. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 2, characterized in that: Select the benchmark image data. The specific method is as follows: The image data in which the proportion of the target vehicle in the multi-angle image data is within a preset range and the center of the target vehicle is closest to the center of the image is selected as the reference image data.

5. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 2, characterized in that: The drone automatically tracks the target vehicle. The specific method is as follows: Acquire a number of historical image data at equal time intervals when the speed and direction of the drone remain unchanged within a second preset period before the current moment; The size difference between the target vehicle size and the target size in each historical image data is calculated to construct a size difference sequence, and the coordinate difference between the target vehicle coordinate and the target coordinate in each historical image data is calculated to construct a coordinate difference sequence; Taking the size difference sequence and the coordinate difference sequence as input, the first LSTM neural network is used to predict the expected speed variable and expected direction variable of the drone at the next moment; The UAV dynamics model is used to solve the rotational speeds of the four motors of the UAV at the next moment, so that the UAV can automatically track the target vehicle.

6. The method for tracking and controlling a drone after a vehicle target leaves the field of view according to claim 1, characterized in that: The coordinates of the target vehicle in the world coordinate system are calculated as follows: Get the coordinates of the drone in the world coordinate system when the image was taken; Read the relative distance between the target vehicle and the drone measured by the ranging radar on the drone; According to the coordinates of the UAV in the world coordinate system and the relative distance, the coordinates of the target vehicle in the world coordinate system are calculated.

7. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 6, characterized in that: Predict the target vehicle's predicted coordinate trajectory in the world coordinate system. The specific method is as follows: The change data of the target vehicle in the world coordinate system before it disappears is input into the second LSTM neural network to predict the predicted coordinate trajectory in the world coordinate system after the target vehicle disappears.

8. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 6, characterized in that: Calculate the UAV search speed as follows: According to the change data of the target vehicle in the image frame coordinate system before it disappears, the coordinate change data of the target vehicle in the world coordinate system is calculated; Calculating the moving speed of the target vehicle in the world coordinate system according to the coordinate change data of the target vehicle in the world coordinate system; The target vehicle's moving speed in the world coordinate system plus the preset speed is used as the drone's search speed; Calculate the search direction of the drone. The specific method is as follows: According to the change data of the target vehicle in the image frame coordinate system before it disappears, the disappearance direction of the target vehicle compared to the UAV is determined, and the disappearance direction of the target vehicle compared to the UAV is extended as the search direction of the UAV.

9. The method for tracking and controlling a drone after a vehicle target leaves the field of view as claimed in claim 1, characterized in that: When weighted predicting coordinate tracks and searching coordinate tracks, the weights are distributed as follows: Two sets of allocation weights are preset, in which the weight of the predicted coordinate trajectory in the first set of allocation weights is greater than the weight of the search coordinate trajectory, and in the second set of allocation weights, the weight of the predicted coordinate trajectory is less than the weight of the search coordinate trajectory; If the target vehicle gradually disappears in the image frame, the second set of distribution weights is selected; If the target vehicle disappears from the image frame for a short time, the first set of allocation weights is selected.

10. The method for tracking and controlling a UAV after a vehicle target leaves the field of view as claimed in claim 5, characterized in that: The method for obtaining the first LSTM neural network training sample is as follows: Release variable speed and direction tracking samples, as well as fixed speed and direction drone samples; When the tracking sample is offset / changed in size in the drone camera, the drone sample is manually manipulated to return the tracking sample to its initial state in the drone camera; Repeat several times and record the speed and direction variables of the drone during manual control; The size difference sequence and coordinate difference sequence of the tracking sample when it is offset / sized change in the drone camera are used as training sample input, and the speed variable and direction variable of the manually controlled drone are used as training sample output.