A method for accurate landing of unmanned aerial vehicles based on target recognition

Abstract

The invention belongs to the technical field of accurate landing of unmanned aerial vehicles, and specifically relates to an accurate landing method of unmanned aerial vehicles based on target recognition, comprising an unmanned aerial vehicle body, the unmanned aerial vehicle body being composed of a microprocessor flight control module, an unmanned aerial vehicle motion module, a GPS module, a DTU data transmission module, an onboard computer Raspberry Pi, and a downward-looking camera, wherein the unmanned aerial vehicle flight control module is used to generate corresponding real-time flight control instructions according to a flight route calculated by a microprocessor; the GPS module is used to roughly estimate a return position so that a camera can capture landing information; the DTU data transmission module is used for the unmanned aerial vehicle to transmit various parameters of the robot to a central station in real time through 4G, adapt to data set expansion under different surfaces and different viewing angles, play a strong positive feedback on the accuracy of a training network model, achieve better recognition effect, can target recognize and classify four types of aprons, and the four types of aprons can all realize landing by proposing center point estimation values ​​from yolov5.

Description

Technical Field

[0001] The present invention belongs to the technical field of UAV precision landing, and in particular relates to a method for UAV precision landing based on target recognition. Background Art

[0002] With the popularization of automated drones, drone landing has gradually been freed from handheld remote controls, and drone positioning methods have gradually expanded from GNSS to image processing. Image processing applications are a major milestone in drone positioning, and various institutions have conducted extensive attempts in practice on landing targets and algorithms. At present, in the application of domestic drone landing, the most commonly used method is to use GNSS data in GPS and more accurate RTK for real-time positioning and landing. In the visual direction, image recognition is mostly used to change the target pattern, or a visual reference library, which reduces the complexity to meet real-time requirements by designing specific signs similar to QR codes; secondly, concentric circle and concentric square target targets are used, and clustering algorithms are used to remove interference patterns in the image and predict its Center point; There is also a method based on identifying infrared LED lamp beads. It estimates the horizontal relative position by equipping a Z-axis ranging sensor to achieve center point landing. The existing technology has many defects. GPS navigation transmits signals through radio, which contains all the shortcomings of radio. The calculated landing error value is large. Using RTK real-time positioning and landing can greatly improve the positioning accuracy and achieve centimeter-level positioning. However, the components that the system needs to rely on include mobile stations, fixed stations, antennas, wireless data transmission, etc., which are expensive. For the precise landing technology of drones using visible light images, relevant research institutions currently focus on the research of theoretical algorithms for automatic recognition of single targets. Although the precise landing of specific target targets has high landing accuracy, there are strict requirements on the design and production of target targets. Summary of the invention

[0003] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method for accurate landing of UAVs based on target recognition, which solves the problems in the prior art of large landing calculation error, high cost and strict requirements for the design and manufacture of target.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] A method for accurately landing a drone based on target recognition, the method comprising the following steps:

[0006] S1. Preparation of training set: Fly the drone above different landing pads and at different altitudes, and start recording with the downward camera to save the video content; S2. Image enhancement and expansion: Expand the number of data sets through the copy-paste algorithm and perspective transformation, so that the model can train more accurate recognition accuracy and center point prediction values, and at the same time enable the drone to adapt to various terrain target point predictions. The segmentation and filling algorithm can enable the drone to identify the landing point under different surfaces, and the perspective transformation will expand the landing target into various side view shapes; S3. Training weight parameters: In order to meet the real-time center point prediction and for Raspberry Pi 4b, choose a smaller depth and width yolov5 network model nTrain a weight file in the RTX3070 graphics card host; S4. Load the weight file to achieve precise landing: The global positioning system GPS module can provide rough return position information for the drone flight, and use the Raspberry Pi 4b to transmit the position information to the drone flight control AP module to perform the return mission while keeping the drone in a hovering state. Turn on the downward 160-degree wide-angle camera connected to the Raspberry Pi 4b, and pass the image to the yolov5 pre-trained network to output the center point of the target detection. According to the difference between the center of the camera image and the center point of the target detection, the Raspberry Pi sends the PID to adjust the speed to the drone flight control AP module to coordinate with the inertial measurement unit IMU module to control the drone to move to the center point.

[0007] In step S1, frames of the video content are randomly extracted and saved in a picture format.

[0008] In step S2, the data set is enhanced by using a copy-paste algorithm and a perspective transformation.

[0009] In step S3, the drone-mounted Raspberry Pi uses a target detection algorithm based on the YOLOv5 model, uses labelimg software to accurately mark the apron in the data set, and uses kmeans to calculate the parameter value of the anchor to ensure the accuracy of the classified apron target border and center point.

[0010] In step S4, the center point offset threshold is set. If the actual offset is less than the threshold, a fixed altitude landing command is executed. If the actual offset is greater than or equal to the center point offset threshold, the S4 instruction is repeated again to correct the collective posture of the drones and ultimately achieve high-precision landing.

[0011] If the actual offset is less than the threshold, a fixed height descent command is executed, and the fixed height is 10 cm.

[0012] Beneficial effects of the present invention:

[0013] 1. The segmentation and filling algorithm and perspective transformation are used to expand the target data set to adapt to the expansion of data sets under different surfaces and different perspectives, which has a strong positive feedback on the accuracy of the training network model and achieves better recognition effect.

[0014] 2. It can identify and classify four types of aprons. All four types of aprons can be landed by proposing the center point estimation value from yolov5. For the concentric circle and communication square aprons identified, the clustering algorithm can also be used to land. The specific QR code mark can be used for landing, and secondary development can also be completed with Apritag or Aruco visual benchmark library.

[0015] 3. By interpolating the camera center point and the predicted center point to adjust the PID algorithm parameters, not only can the target be accurately landed, but also the dynamic target can be tracked at a certain speed.

[0016] 4. Raspberry Pi 4b, as an onboard computer, can not only achieve precise landing of the drone, but also use external control to complete other project development and save costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 It is a schematic diagram of implementing the copy-paste algorithm to expand the data set according to an embodiment of the present invention;

[0019] Figure 2 is a schematic diagram of training weights according to an embodiment of the present invention;

[0020] Figure 3 It is a physical picture of the drone and a schematic diagram of the power supply line;

[0021] Figure 4 is a schematic diagram of an onboard computer controlled landing according to an embodiment of the present invention;

[0022] Figure 5 It is a flow chart of the steps of landing a drone according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In the description of the present invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inside", "all around" and the like indicating orientation or positional relationship are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the components or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0025] like Figure 1-4 As shown, a method for accurate landing of a drone based on target recognition includes a drone body, a drone body composed of a microprocessor flight control module, a drone motion module, a GPS module, a DTU data transmission module, an onboard computer Raspberry Pi, and a wide-angle 160-degree downward-looking camera. The drone flight control module is used to generate corresponding real-time flight control instructions according to the flight route calculated by the microprocessor; the GPS module is used to roughly estimate the return position so that the camera can capture landing information; the DTU data transmission module is used for the drone to transmit various robot parameters to the central station in real time through 4G, so that the back-end central station can obtain the real-time status of the robot in time; the onboard computer Raspberry Pi uses the ubnutu system to load yolov5 for target detection and speed command transmission to the flight control system, and the drone power distribution board is connected to the external battery separation system to connect the Type-C output stable 5V / 3A to power the Raspberry Pi, and a 15-pin ribbon cable is used with a wide-angle camera. The method includes the following steps:

[0026] S1. Preparation of training set: The drone is flown above different landing pads and at different altitudes, and the downward camera is turned on to record the video content to save the video content. The video content is randomly extracted and saved in image format;

[0027] S2, Image enhancement and expansion: The copy-paste algorithm and perspective transformation are used to expand the number of data sets, so that the model can be trained to have more accurate recognition accuracy and center point prediction values. At the same time, the UAV can adapt to the prediction of target points in various terrains. The segmentation and filling algorithm can enable the UAV to identify the landing points under different surfaces, and the perspective transformation can expand the landing target into various shapes under side view angles.

[0028] S3, training weight parameters: In order to meet the real-time center point prediction, and for Raspberry Pi 4b, select a smaller depth and width yolov5 network model n to train the weight file in the RTX3070 graphics card host. The drone-mounted Raspberry Pi uses the target detection algorithm based on the YOLOv5 model, uses labelimg software to accurately mark the apron in the data set, and uses kmeans to calculate the parameter value of the anchor to ensure the accuracy of the classified apron target border and center point;

[0029] S4. Load the weight file to achieve precise landing: The global positioning system GPS module can provide rough return position information for the drone flight. The Raspberry Pi 4b is used to transmit the position information to the drone flight control AP module to perform the return mission while keeping the drone in a hovering state. The downward 160-degree wide-angle camera connected to the Raspberry Pi 4b is turned on, and the image is transmitted to the yolov5 pre-trained network to output the center point of the target detection. According to the difference between the center of the camera image and the center of the target detection, the Raspberry Pi sends the PID adjustment speed to the drone flight control AP module and the inertial measurement unit IMU module to control the drone to move to the center point. The center point offset threshold is set. If the actual offset is less than the threshold, a fixed height landing command is executed. If the actual offset is greater than or equal to the center point offset threshold, the S4 instruction is repeated again to correct the collective posture of the drone. If the actual offset is less than the threshold, a fixed height landing command is executed, and the fixed height is 10 cm, and finally a high-precision landing is achieved.

[0030] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0031] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for accurate landing of a drone based on target recognition, characterized in that: The drone includes a drone body, which includes a microprocessor flight control module, a drone motion module, a GPS module, a DTU data transmission module, a Raspberry Pi 4b and a downward-looking camera. The microprocessor flight control module is used to generate corresponding real-time flight control instructions according to the flight route calculated by the microprocessor; the GPS module is used to roughly estimate the return position so that the camera can capture landing information; The DTU data transmission module transmits various parameters of the drone to the central station in real time through the 4G network, so that the central station can obtain the real-time status of the drone in time; the Raspberry Pi 4b uses the ubnutu system to load YOLOv5 for target detection and speed command transmission to the microprocessor flight control module, and the drone power distribution board is connected to the external battery separation system to power the Raspberry Pi 4b, and a 15-pin ribbon cable is used to connect the camera to obtain image information. The method includes the following steps: S1. Preparation of training set: Make the drone fly above different landing pads and at different altitudes, and start recording with the downward-looking camera to save the video content; S2, image enhancement and expansion: The copy-paste algorithm and perspective transformation are used to expand the number of data sets, so that the model can be trained to have more accurate recognition accuracy and center point prediction values, and the drone can adapt to the prediction of target points on various terrains. The copy-paste algorithm enables the drone to identify the landing points under different surfaces, and the perspective transformation expands the landing target into the shape under the side view. S3, training weight parameters: In order to meet the real-time center point prediction and for Raspberry Pi 4b, select the appropriate YOLOv5 network model to train the weight file in the host; S4. Load the weight file to achieve precise landing: The GPS module is used to provide rough return position information for the UAV flight. The Raspberry Pi 4b is used to transmit the position information to the microprocessor flight control module to perform the return mission while keeping the UAV in a hovering state. The downward camera connected to the Raspberry Pi 4b is turned on, and the image is transmitted to the YOLOv5 pre-trained network to output the center point of the target detection. According to the difference between the center of the camera image and the center point of the target detection, the Raspberry Pi 4b sends the PID adjustment speed to the microprocessor flight control module to coordinate with the inertial measurement unit IMU module to control the UAV to move to the center point. In step S3, the Raspberry Pi 4b uses the target detection algorithm based on the YOLOv5 model, uses the labelimg software to accurately mark the aprons in the data set, and uses kmeans to calculate the parameter values ​​of the anchor to ensure the accuracy of the classified apron target border and center point; In step S4, a center point offset threshold is set. If the actual offset is less than the threshold, a fixed landing height command is executed, and the fixed height is 10 cm. If the actual offset is greater than or equal to the center point offset threshold, the S4 instruction is repeated again to correct the drone body posture and finally achieve high-precision landing.

2. The method for accurate landing of a drone based on target recognition according to claim 1, characterized in that: In step S1, frames of video content are randomly extracted and saved in image format.

3. The method for accurate landing of a drone based on target recognition according to claim 1 is characterized in that: In step S2, the dataset is enhanced by using a copy-paste algorithm and perspective transformation.

Images