Method of vision-based long-range and short-range guidance for autonomous UAV landing

1 . A method of vision-based long-range and short-range guidance for autonomous unmanned aerial vehicle (UAV) landing, comprising the following steps:

S 1 : collecting an unmanned vehicle image dataset in advance;

S 2 : training an unmanned vehicle detection model by using the unmanned vehicle image dataset combined with a YOLOv5 neural network;

S 3 : loading the unmanned vehicle detection model onto a control module of a UAV;

S 4 : collecting, by the UAV during a landing process, images of an area below the UAV at specified time intervals, and inputting the collected images into the unmanned vehicle detection model for recognition and detection;

S 5 : if an unmanned vehicle is recognized in the area below the UAV, determining position information of the unmanned vehicle through image analysis, and then proceeding to step S 6 ; otherwise, returning to step S 4 ;

S 6 : outputting, by the control module, a long-range guidance control instruction based on the position information of the unmanned vehicle, to instruct the UAV to fly to a specified distance position above the unmanned vehicle;

S 7 : capturing, by the UAV, an image of a target installed on the unmanned vehicle, and identifying position information of the target; and

S 8 : outputting, by the control module, a short-range guidance control instruction based on the position information of the target, to instruct the UAV to land on an unmanned vehicle platform;

wherein step S 2 comprises the following sub-steps:

S 21 : preprocessing the unmanned vehicle image dataset and dividing the unmanned vehicle image dataset into a training set and a validation set according to a preset ratio;

S 22 : annotating images in the training set with corresponding labels by using an annotation tool, and recording categories for cluster analysis, to build the YOLOv5 neural network; and

S 23 : training and validating the YOLOv5 neural network based on the training set and the validation set, to obtain the unmanned vehicle detection model, wherein in step S 21 , the preprocessing specifically adopts Mosaic data augmentation, comprising but not limited to random horizontal or vertical flipping, cropping, and scale transformation operations.

2 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein in step S 23 , before the training and validation of the YOLOv5 neural network, image sizes and resolutions are standardized: first, scaling down the images based on an input size required by the YOLOv5 neural network, and then adding black bars to shorter sides to form a square, thereby meeting input specifications of 608 pixels*608 pixels.

3 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein during the training of the YOLOv5 neural network in step S 23 , a Generalized Intersection over Union (GIOU) loss function is specifically used to calculate a loss of a bounding box:

GIOU=| A∩B∥A∪B|−|C \( A∪B∥C|=IoU−|C \( A∪B∥C|

that is, for any two arbitrary boxes A and B, a smallest enclosed shape C is found, wherein C contains both A and B; then a ratio of an area of C outside A and B to a total area of C is calculated, and the ratio is subtracted from an Intersection over Union (IoU) of A and B to obtain a loss value of the bounding box.

4 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein the target installed on the unmanned vehicle is a pattern comprising two AprilTags and an H-shaped geometric pattern, with the two AprilTags located in upper and lower grooves of the H-shaped geometric pattern.

5 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 4 , wherein the target has a size of 0.5 m*0.5 m.

6 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 4 , wherein step S 7 comprises the following sub-steps:

S 71 : capturing, by the UAV, an image of the target installed on the unmanned vehicle;

S 72 : analyzing and processing the image of the target to obtain position information of the AprilTags in the image of the target; and

S 73 : performing coordinate system transformation on the position information of the April Tags in the image of the target, and then obtaining Euler angles through a homography matrix, to obtain real-time positions of the AprilTags.

7 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

8 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 7 , wherein adjusting a flight status of the UAV comprises adjusting a flight direction, a flight altitude, a flight speed, hovering, a landing speed, and a landing altitude of the UAV.

9 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

10 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 1 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

11 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 2 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

12 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 3 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

13 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 4 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

14 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 5 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

15 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 6 , wherein the control module of the UAV comprises a recognition unit and a processing unit; the recognition unit is equipped with the unmanned vehicle detection model;

the processing unit is configured to calculate distance information between the UAV and the unmanned vehicle or the target, and output a control signal

for adjusting a flight status of the UAV.

16 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 9 , wherein adjusting a flight status of the UAV comprises adjusting a flight direction, a flight altitude, a flight speed, hovering, a landing speed, and a landing altitude of the UAV.

17 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 10 , wherein adjusting a flight status of the UAV comprises adjusting a flight direction, a flight altitude, a flight speed, hovering, a landing speed, and a landing altitude of the UAV.

18 . The method of vision-based long-range and short-range guidance for autonomous UAV landing according to claim 11 , wherein adjusting a flight status of the UAV comprises adjusting a flight direction, a flight altitude, a flight speed, hovering, a landing speed, and a landing altitude of the UAV.