Precise intelligent landing method for unmanned aerial vehicle in complex environment

Abstract

The invention provides a precise intelligent landing method for an unmanned aerial vehicle in a complex environment, and the method comprises the steps: carrying out the stage division of a landing process according to the terrain clearance of the unmanned aerial vehicle, and obtaining different landing stages; the landing stage comprises a high-altitude positioning stage, a middle-distance adjusting stage and a short-distance precise landing stage; an AprilTag code is obtained through a precise landing algorithm based on vision; dynamic control is carried out according to the AprilTag code in the landing stage; the dynamic control comprises wind speed compensation, illumination change compensation, filtering algorithm optimization and illumination adaptability adjustment. According to the method, the adaptability to signal shielding, environment interference and dynamic target change conditions is improved, and the safety and reliability of autonomous landing of the unmanned aerial vehicle in various complex and changeable scenes are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) positioning, and in particular to a method for precise and intelligent landing of an unmanned aerial vehicle (UAV) in a complex environment. Background Art

[0002] With the rapid development of drone technology, the application of drones in logistics distribution, farmland inspection, disaster relief and other fields has been expanding. In these applications, the precise landing of drones has become an important factor in determining the efficiency of operations and the success rate of missions. At present, most drones on the market rely on GNSS (Global Navigation Satellite System) or RTK (Real-time Kinematic Differential) navigation technology to achieve positioning and ensure accurate landing.

[0003] GNSS uses satellite signals to calculate the geographic coordinates (longitude, latitude, altitude), speed and heading of the receiving device in real time. When the satellite signal is blocked (for example, by tall buildings, trees, mountains and other objects), the receiving device cannot obtain enough satellite data, reducing the geometric precision factor (GDOP), resulting in reduced positioning accuracy. When the field of view is blocked, the visible satellites may be concentrated in the zenith direction or on one side, forming a poor geometric structure (for example, arranged in the same straight line), which will increase the GDOP value and reduce positioning accuracy.

[0004] RTK (Real-Time Kinematic) is a high-precision positioning enhancement technology based on the satellite navigation system (GNSS). It mainly uses the differential signal between the base station and the mobile station to correct the errors in the satellite signal in real time. The base station calculates the errors of the satellite signal (such as atmospheric delay, satellite position error, etc.) and sends this error information to the mobile station. After receiving the correction information, the mobile station can correct its position in real time to achieve centimeter-level accuracy. However, RTK technology is highly dependent on the base station. If the drone flies beyond the effective range of the base station, the positioning accuracy of RTK will drop significantly or even fail completely. In addition, the position of the base station must be known and fixed, which limits its application scenarios, especially when the drone is required to land in an area that is not covered by the base station (such as the ocean or remote areas), the positioning performance of the RTK system will be greatly reduced.

[0005] At the same time, both GNSS and RTK technologies rely on preset target positions for navigation and determine the landing point based on static coordinates. However, this method cannot meet the needs of precise landing in dynamic environments. Especially in densely built-up areas, bad weather, or rapidly changing ground conditions, the target position may be constantly changing or obstructed. Traditional GNSS and RTK technologies are difficult to adapt to these dynamic changes in real time, resulting in unstable positioning accuracy. Therefore, it is very necessary to design a precise and intelligent landing method for drones in complex environments. Summary of the invention

[0006] The purpose of the present invention is to provide a method for precise intelligent landing of UAVs in complex environments, by improving the adaptability to signal blocking, environmental interference and dynamic target changes, thereby ensuring that UAVs can achieve safe and reliable autonomous landing in various complex and changeable scenarios.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for precise intelligent landing of a drone in a complex environment comprises the following steps:

[0009] The landing process is divided into stages according to the height of the drone from the ground, and different landing stages are obtained; the landing stages include: high-altitude positioning stage, mid-range adjustment stage and close-range precision landing stage;

[0010] Get the AprilTag code through the vision-based precision landing algorithm;

[0011] During the landing phase, dynamic control is performed according to the AprilTag code; dynamic control includes wind speed compensation, light change compensation, filtering algorithm optimization and light adaptability adjustment.

[0012] Optionally, the landing process is divided into stages according to the height of the UAV above the ground to obtain different landing stages, specifically: when the height of the UAV above the ground is greater than 10 meters, the landing stage is divided into a high-altitude positioning stage; when the height of the UAV above the ground is less than or equal to 10 meters and greater than or equal to 2 meters, the landing stage is divided into a medium-range adjustment stage; when the height of the UAV above the ground is less than 2 meters, the landing stage is divided into a close-range precision landing stage.

[0013] Optionally, the AprilTag code is obtained through a vision-based precision landing algorithm, specifically: the quadrilateral area is detected through edge detection and quadrilateral detection algorithms, and the quadrilateral area is used as the AprilTag code; the coding pattern and corresponding ID of the AprilTag code are identified; and the position offset of the drone relative to the AprilTag code is output according to the ID.

[0014] Optionally, the expression for dynamic control is: Among them, u(t) is the control output value, e(t) is the difference between the current value and the target value, K p is the proportional gain, K i is the integral gain, K d is the differential gain.

[0015] Optionally, the wind speed compensation is: adjusting the proportional gain and the differential gain in real time according to the change of wind speed to obtain an updated proportional gain and an updated differential gain;

[0016] The calculation formula for updating the proportional gain is: K p1 =K p (1+αV wind ), where K p1 is the updated proportional gain, α is the wind speed adjustment coefficient, V wind is the wind speed; the calculation formula for updating the differential gain is: K d1 =K d (1+αV wind ), where K d1 To update the derivative gain.

[0017] Optionally, the illumination change compensation is: adjusting the integral gain in real time according to the current illumination intensity measured by the illumination sensor to obtain an updated integral gain; the integral gain is calculated as follows: Among them, K i1 is to update the integral gain, γ is the illumination adjustment coefficient, L lux is the current light intensity, L ref is the reference value of light intensity.

[0018] Optionally, the filtering algorithm is optimized to be a Kalman filtering algorithm.

[0019] Optionally, the illumination adaptation adjustment includes: automatic exposure control mechanism, dynamic adjustment of grayscale value and image filtering; the image filtering includes median filtering and Gaussian filtering;

[0020] The calculation formula of the automatic exposure control mechanism is: in, To update the exposure time, is the current exposure time, L target is the expected brightness value, L avg is the average brightness of the image;

[0021] The calculation formula for dynamic adjustment of gray value is: Among them, I new (x, y) is the grayscale value of the adjusted image, I(x, y) is the initial grayscale value of the image, CDF is the cumulative distribution function, M and N are the width and height of the image respectively, CDF min is the minimum value of the cumulative distribution function, L max is the maximum dynamic range of grayscale values.

[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the method for precise intelligent landing of a UAV in a complex environment provided by the present invention comprises: dividing the landing process into stages according to the height of the UAV above the ground to obtain different landing stages; the landing stages comprise: a high-altitude positioning stage, a mid-range adjustment stage and a close-range precise landing stage; obtaining the AprilTag code through a vision-based precise landing algorithm; performing dynamic control according to the AprilTag code during the landing stage; the dynamic control comprises wind speed compensation, illumination change compensation, filtering algorithm optimization and illumination adaptability adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 This is a flow chart of the precise intelligent landing method for a UAV of the present invention. DETAILED DESCRIPTION

[0025] 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.

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] like Figure 1 As shown, the present invention provides a method for precise intelligent landing of a UAV in a complex environment, comprising the following steps:

[0028] Step 100: Divide the landing process into stages according to the height of the drone above the ground to obtain different landing stages; the landing stages include: a high-altitude positioning stage, a mid-range adjustment stage, and a close-range precision landing stage;

[0029] Specifically, when the UAV's height from the ground is greater than 10 meters, the landing phase is divided into the high-altitude positioning phase; when the UAV's height from the ground is less than or equal to 10 meters and greater than or equal to 2 meters, the landing phase is divided into the medium-range adjustment phase; when the UAV's height from the ground is less than 2 meters, the landing phase is divided into the close-range precision landing phase.

[0030] Step 200: Obtaining the AprilTag code through a vision-based precision landing algorithm;

[0031] Specifically, first find a quadrilateral area similar to a rectangle through edge detection and quadrilateral detection. Then, resolve the corresponding ID by identifying the coding pattern of the AprilTag code. Next, calculate the 3D pose of the drone relative to the AprilTag code through the intrinsic parameter matrix of the camera on the drone and the physical size of the known AprilTag code, and obtain the pose data structure of the AprilTag code relative to the camera, including the position offset of the drone's camera relative to the AprilTag code, that is, the translation vector and the rotation relationship between the drone's camera coordinate system and the AprilTag code coordinate system, that is, the rotation matrix.

[0032] More specifically, assuming the camera's intrinsic parameter matrix is ​​K, Among them, f x and f y Refers to the focal length of the drone's camera on the X and Y axes (in pixels). x and C y Refers to the coordinates of the optical center (principal point) of the drone's camera. The translation vector that causes the position shift is t, t = t x ,t y ,t z ] T ; where t x Refers to the position of the drone's camera in the X-axis direction (left-right offset); t y Refers to the position of the drone's camera in the Y-axis direction (up and down offset); t z Refers to the position of the drone's camera in the Z-axis direction (front and back offset), that is, the vertical distance from the drone's camera to the AprilTag code. The Euclidean distance d between the drone's camera and the AprilTag code can be calculated by the translation vector t.

[0033] Step 300: Perform dynamic control according to the AprilTag code during the landing phase; the dynamic control includes wind speed compensation, light change compensation, filter algorithm optimization and light adaptability adjustment.

[0034] Specifically, AprilTag codes of different sizes are used to adjust the positioning of the drone in different landing stages. In the high-altitude positioning stage, the drone system preferentially identifies a large-sized AprilTag code, measures the pose information (position and direction) of the AprilTag code, preliminarily calculates the relative distance to the ground, and adjusts the flight path. First, adjust the attitude according to the obtained translation vector t to ensure that the drone is aligned with the large AprilTag code, so that t x and ty Close to zero to ensure that the drone lands vertically; at the same time, it is further combined with the rotation matrix to ensure that the orientation of the drone is aligned with the orientation of the AprilTag code.

[0035] In the mid-range adjustment phase, the drone switches to recognize three medium-sized AprilTag codes to obtain more accurate position information. By using multiple control algorithms to adjust the lateral and height positions of the drone, the drone is kept within an acceptable error range and gradually approaches the target point, allowing the drone to further correct its position and attitude and improve landing accuracy.

[0036] During the close-range precision landing phase, the drone begins to recognize 16 small-sized AprilTag codes and based on the t x ,t y and t z The data is used to fine-tune the drone’s attitude and align it with the landing mark.

[0037] It should be noted that by identifying AprilTag codes of different sizes and quantities in stages, the drone gradually improved its positioning accuracy during the entire landing process, ensuring a smooth and precise landing from high altitude to the ground.

[0038] Specifically, the dynamic control uses the PID control algorithm to control the attitude and trajectory of the drone by adjusting the three parameters of proportion (P), integral (I) and differential (D) to ensure its stability and accuracy. The expression of the PID control algorithm is:

[0039]

[0040] Among them, u(t) is the control output value, e(t) is the difference between the current value and the target value, K p is the proportional gain, which determines the rapidity of response, K i is the integral gain, which determines the ability to eliminate steady-state errors, K d is the differential gain, which determines the stability of the response.

[0041] More specifically, in complex environments, the UAV system needs to consider the impact of wind speed changes on the attitude and position of the UAV. In order to adapt to the change in wind speed, the system adjusts K in real time. p and K d Parameters are used to compensate for the disturbance caused by wind speed to achieve wind speed compensation. When the wind speed increases, the system increases K p To speed up the response to the deviation; at the same time, increase K d To reduce overshoot and stabilize the landing process. p and K d The adjusted calculation formula is:

[0042] K p1 =K p (1+αV wind );

[0043] K d1 =K d (1+αV wind );

[0044] Among them, K p1 is the adjusted proportional gain, α is the wind speed adjustment coefficient, V wind is the wind speed, K d1 is the adjusted differential gain.

[0045] More specifically, changes in light intensity can affect the recognition accuracy of the drone’s visual sensor, especially in strong or low light environments. To ensure that the drone can stably recognize the AprilTag code, the system adjusts the parameter K i To reduce the impact of light changes and achieve light change compensation. The drone measures the current light intensity through the light sensor and adjusts the K i Make adjustments. i The adjusted calculation formula is:

[0046]

[0047] Among them, K i1 is the adjusted integral gain, γ is the illumination adjustment coefficient, L lux is the current light intensity, L ref is the reference value of light intensity.

[0048] More specifically, the Kalman filter algorithm is used for filtering algorithm optimization. The current position and attitude of the drone are predicted using the state prediction formula of the Kalman filter. Combined with the adjusted formula, the position and attitude (position t and rotation matrix R) of the AprilTag code recognition are corrected to output a smooth relative position information to help the drone maintain a stable landing path. The weights of the predicted value and the measured value are then adjusted using the Kalman gain formula. If a sudden shift in the position of the AprilTag code is detected, the Kalman filter will determine whether this is noise caused by environmental interference and make appropriate corrections.

[0049] It should be noted that when the drone is descending rapidly, it may encounter a brief loss of AprilTag code recognition (such as being blocked). In this case, the Kalman filter state prediction formula is used to predict the AprilTag code position and attitude at the next moment, ensuring that even if the visual system fails temporarily, the attitude and position of the drone can still be adjusted based on the predicted value, thereby maintaining a smooth landing.

[0050] Specifically, when the drone uses visual sensors to land, lighting conditions will greatly affect the accuracy of visual recognition. Too strong or too weak light will cause visual recognition to fail, thereby affecting the accuracy of landing. For example, strong direct sunlight will cause the camera image to be overexposed, making it difficult to recognize visual markers such as the AprilTag code; in the evening or when there is insufficient light indoors, the image will become darker, increasing the difficulty of visual recognition; when the drone flies in an environment where shadows and sunlight alternate, the light intensity will change rapidly, causing the visual system recognition effect to fluctuate. This embodiment solves the above problems through light adaptability adjustment. Light adaptability adjustment includes: automatic exposure control mechanism, dynamic adjustment of grayscale values, and image filtering; and image filtering includes median filtering and Gaussian filtering.

[0051] The automatic exposure control mechanism is the key mechanism for adjusting the imaging brightness of the drone camera. It dynamically adjusts the exposure time and gain of the camera to adapt to different lighting environments to ensure that the image brightness output by the camera is at the optimal level. exp Controls the sensor's sensitivity to light. Increasing exposure time can increase image brightness, while reducing exposure time can prevent overexposure. Gain is used to adjust the amplification of the light signal received by the sensor. Increasing gain in low-light environments helps improve image brightness. The calculation formula for the automatic exposure control mechanism is:

[0052]

[0053] in, To update the exposure time, is the current exposure time, L target is the expected brightness value, L avg is the average image brightness.

[0054] The dynamic adjustment of grayscale value adopts the histogram equalization algorithm to adjust the pixel grayscale value by calculating the cumulative distribution function of the image. This adjustment method can stretch the grayscale distribution of the image, thereby enhancing the visibility of visual features, which in turn helps to maintain the stability of visual recognition in strong light or low light environments and improve the detection accuracy of AprilTag codes. The calculation formula for dynamic adjustment of grayscale value is:

[0055]

[0056] Among them, I new (x, y) is the grayscale value of the adjusted image, I(x, y) is the initial grayscale value of the image, CDF is the cumulative distribution function, M and N are the width and height of the image respectively, CDF min is the minimum value of the cumulative distribution function, L max is the maximum dynamic range of grayscale values, L maxis 255.

[0057] Median filtering is used to remove salt and pepper noise in the image recorded during the recognition process. By replacing each pixel in each image with the median value of the pixel value in its neighborhood during the recognition process, prominent noise points are eliminated without significantly blurring the image edge. The expression of median filtering is:

[0058] I new (x,y)=median({I(i,j)}for(i,j)∈window);

[0059] Among them, I new (x, y) is the grayscale value of the pixel at position (x, y), median represents the median operation, which is used to calculate the median of all pixel grayscale values ​​in a given area, I(i, j) represents the grayscale value of the pixel at position (i, j) in the image, and window represents the filtering window area, which is usually a rectangle of fixed size.

[0060] Gaussian filtering smoothes images through Gaussian kernel convolution. It is mainly used to remove Gaussian noise and smooth the overall image. It effectively reduces random noise in the image, but at the same time blurs the edge information of the image. The expression of Gaussian filtering is:

[0061]

[0062] Among them, G(i,j) is the Gaussian kernel function, defined as: σ is the smoothness of the control filter, and k is the radius of the Gaussian kernel.

[0063] It should be noted that during the landing process, a variety of advanced algorithms are integrated, including: dynamic identification of the posture information of the AprilTag code (large, medium, and small), obtaining the posture relationship between the AprilTag code and the drone, and adjusting the flight path in real time. At the same time, environmental factors such as wind speed and light are taken into account, and the landing accuracy is optimized by dynamically adjusting the PID control parameters. Kalman filtering technology is also introduced to perform real-time fusion and filtering of sensor data to improve positioning stability. In addition, the drone system has dynamic exposure adjustment and grayscale processing functions to adapt to different lighting conditions and ensure stable recognition and tracking of the AprilTag code. The fused algorithm can achieve high-precision and autonomous landing of drones in complex environments, thereby significantly improving the safety and reliability of drone landing.

[0064] The beneficial effects of the present invention are as follows:

[0065] 1) The use of AprilTag code detection and pose estimation algorithm ensures that the drone can be accurately guided to land at different stages;

[0066] 2) Adjust dynamic control parameters in real time according to real-time wind speed and light changes to ensure stable landing of the drone in a changing environment;

[0067] 3) The Kalman filter algorithm reduces noise interference and improves positioning accuracy; at the same time, it predicts the target position when the visual signal is temporarily lost, maintaining a stable landing path;

[0068] 4) The automatic exposure control mechanism and image enhancement algorithm ensure that visual recognition can operate stably in both strong light and low light environments, significantly improving the landing accuracy and robustness of the UAV in dynamic environments.

[0069] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0070] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for precise intelligent landing of a drone in a complex environment, characterized in that: The steps include: The landing process is divided into stages according to the height of the UAV from the ground, and different landing stages are obtained; the landing stages include: a high-altitude positioning stage, a mid-range adjustment stage, and a close-range precision landing stage; Get the AprilTag code through the vision-based precision landing algorithm; During the landing phase, dynamic control is performed according to the AprilTag code; the dynamic control includes wind speed compensation, illumination change compensation, filtering algorithm optimization and illumination adaptability adjustment.

2. The method for precise intelligent landing of a drone in a complex environment according to claim 1, characterized in that: The landing process is divided into stages according to the height of the UAV above the ground to obtain different landing stages, specifically: when the height of the UAV above the ground is greater than 10 meters, the landing stage is divided into the high-altitude positioning stage; when the height of the UAV above the ground is less than or equal to 10 meters and greater than or equal to 2 meters, the landing stage is divided into the medium-range adjustment stage; when the height of the UAV above the ground is less than 2 meters, the landing stage is divided into the close-range precision landing stage.

3. The method for precise intelligent landing of a drone in a complex environment according to claim 1, characterized in that: The AprilTag code is obtained through a vision-based precision landing algorithm, specifically: a quadrilateral area is detected through edge detection and quadrilateral detection algorithms, and the quadrilateral area is used as the AprilTag code; the coding pattern and the corresponding ID of the AprilTag code are identified; and the position offset of the drone relative to the AprilTag code is output according to the ID.

4. The method for precise intelligent landing of a drone in a complex environment according to claim 1, characterized in that: The expression of the dynamic control is: Among them, u(t) is the control output value, e(t) is the difference between the current value and the target value, K p is the proportional gain, K i is the integral gain, K d is the differential gain.

5. The method for precise intelligent landing of a drone in a complex environment according to claim 4, characterized in that: The wind speed compensation is: adjusting the proportional gain and the differential gain in real time according to the wind speed change to obtain an updated proportional gain and an updated differential gain; The calculation formula for the updated proportional gain is: K p1 =K p (1+αV wind ), where K p1 is the updated proportional gain, α is the wind speed adjustment coefficient, V wind is the wind speed; the calculation formula for the updated differential gain is: K d1 =K d (1+αV wind ), where K d1 To update the derivative gain.

6. The method for precise intelligent landing of a drone in a complex environment according to claim 4, characterized in that: The illumination change compensation is: adjusting the integral gain in real time according to the current illumination intensity measured by the illumination sensor to obtain an updated integral gain; the integral gain is calculated as follows: Among them, K i1 is to update the integral gain, γ is the illumination adjustment coefficient, L lux is the current light intensity, L ref is the reference value of light intensity.

7. The method for precise intelligent landing of a drone in a complex environment according to claim 4, characterized in that: The filtering algorithm is optimized as a Kalman filtering algorithm.

8. The method for precise intelligent landing of a drone in a complex environment according to claim 4, characterized in that: The illumination adaptability adjustment includes: automatic exposure control mechanism, dynamic adjustment of grayscale value and image filtering; the image filtering includes median filtering and Gaussian filtering; The calculation formula of the automatic exposure control mechanism is: in, To update the exposure time, is the current exposure time, L target is the expected brightness value, L avg is the average brightness of the image; The calculation formula for the dynamic adjustment of the gray value is: Among them, I new (x, y) is the grayscale value of the adjusted image, I(x, y) is the initial grayscale value of the image, CDF is the cumulative distribution function, M and N are the width and height of the image respectively, CDF min is the minimum value of the cumulative distribution function, L max is the maximum dynamic range of grayscale values.

Images