Unmanned aerial vehicle deviation reduction correction method based on visual identification

Abstract

An unmanned aerial vehicle landing deviation correction method based on visual identification comprises the steps of obtaining a shaft mouth area image in a landing mode of an unmanned aerial vehicle through a visual sensor, and identifying a preset landing mark in the shaft mouth area image to output a center pixel position corresponding to the landing mark. Based on attitude information when the unmanned aerial vehicle descends and calibration parameters of the visual sensor, the geometric center position of the unmanned aerial vehicle is projected to an image coordinate system unified with a landing mark, and body center pixel coordinates are obtained; calculating the pixel deviation of the central pixel coordinate of the body and the central pixel position in the image coordinate system, and converting the pixel deviation into space horizontal deviation and space vertical deviation. And determining a near-earth pneumatic compensation coefficient corresponding to the descending height of the unmanned aerial vehicle according to the ratio of the diameter of the shaft mouth to the diameter of the unmanned aerial vehicle propeller disc and the descending height. And generating a flight control execution instruction set based on the space horizontal deviation, the space vertical deviation and the near-earth pneumatic compensation coefficient so as to control the unmanned aerial vehicle to correct the descending action.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) control, and more specifically, relates to a UAV deviation reduction correction method based on visual recognition. Background Art

[0002] When operating a drone in a narrow or high-risk area (for example, the top of a high-voltage transmission tower, a building facade maintenance platform, a chimney, or the top of a wellhead), the drone needs to be continuously corrected for landing deviation. This is mainly because the available space in these landing areas is extremely small, and there may be high-voltage electricity, thermal airflow, or vortex airflow around them. If the landing position of the drone deviates by a small distance, it may touch an obstacle, causing a crash or equipment damage.

[0003] Currently, during terminal descent of drones within chimneys, shafts, or other annular, enclosed roof areas, propeller downwash is constrained by the radial constraints of the annular wall, resulting in significant annular recirculation and longitudinal vortex structures. This recirculation is repeatedly reflected and directed between the shaft's inner wall and the lower edge of the rotor disk, resulting in significant asymmetry in the induced velocity distribution across different rotor disk regions. During vertical descent to a height range of approximately 1 to 1.5 rotor disk diameters, this asymmetric flow field triggers a typical alternating process of localized ground effect enhancement and lateral ground effect reduction, significantly increasing the amplitude of lift fluctuations on one side of the rotor disk and directly inducing short-period coupled roll and pitch oscillations of the aircraft. Furthermore, the turbulent components carried by the radial recirculation, superimposed on the vertical plume, can rapidly shift the overall center of lift position, causing the aircraft to experience repeated alternations of height overshoot (transient lift) and height collapse (transient sinking).

[0004] However, in the existing technology, when the flight control system senses the terminal attitude and altitude errors, if it only distributes thrust and corrects the attitude according to the aerodynamic assumptions of an open field, it will cause a phase mismatch between the correction action and the actual aerodynamic environment. This mismatch will significantly amplify the lateral drift and altitude fluctuations in the last approximately 1 meter of descent, which can easily lead to the risk of the drone touching the shaft wall or falling when landing at a narrow shaft entrance. Summary of the Invention

[0005] In order to address the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a UAV deflection correction method based on visual recognition.

[0006] The present invention adopts the following technical solutions.

[0007] The first aspect of the present invention discloses a method for correcting the deviation of an unmanned aerial vehicle (UAV) based on visual recognition, the method comprising: Acquire an image of the shaft opening area in the landing mode of the UAV through a visual sensor, identify a preset landing mark in the shaft opening area image, and output a center pixel position corresponding to the landing mark; Based on the attitude information of the UAV during descent and the calibration parameters of the visual sensor, the geometric center position of the UAV body in the real space is projected into the image coordinate system that is consistent with the landing mark to obtain the pixel coordinates of the center of the body; Calculating the pixel deviation between the center pixel coordinates of the body and the center pixel position in the image coordinate system, and converting the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation; According to the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the UAV descent height, the aerodynamic compensation coefficient corresponding to the UAV descent height is determined; A flight control execution instruction set is generated based on the spatial horizontal deviation, the spatial vertical deviation and the ground proximity aerodynamic compensation coefficient, and the UAV is controlled to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set.

[0008] Furthermore, the method of acquiring an image of the shaft opening area in the landing mode of the drone through a visual sensor and identifying a preset landing mark in the shaft opening area image to output a center pixel position corresponding to the landing mark includes: When the UAV flight control system detects a landing trigger signal, it controls the UAV to enter a landing mode in response to the landing trigger signal, and simultaneously calls the onboard visual sensor to capture an image of the shaft opening area at a fixed frame rate; An original image matrix is ​​constructed based on multiple shaft opening area images acquired at a fixed frame rate, and each image element in the original image matrix is ​​binarized according to preset black and white contrast and shape features.

[0009] Furthermore, the method of acquiring an image of the shaft opening area in the landing mode of the drone through a visual sensor and identifying a preset landing mark in the shaft opening area image to output a center pixel position corresponding to the landing mark further includes: Performing an opening operation on the original image matrix to remove noise from image elements, identify all connected pixel regions, calculate the area and aspect ratio of the connected pixel regions, and screen out candidate pixel regions that meet preset sign specifications; The pixel coordinates of the enclosing rectangular bounding box corresponding to each candidate pixel area are calculated, and each candidate pixel area is normalized and matched with the preset landing mark template to output the coordinates of the pixel area with the highest matching degree as the center pixel position of the landing mark.

[0010] Furthermore, based on the attitude information of the drone during descent and the calibration parameters of the visual sensor, the geometric center position of the drone body in the actual space is projected into an image coordinate system that is unified with the landing mark to obtain the pixel coordinates of the center of the drone body, including: Acquire the attitude information of the drone during descent from the drone flight control system, the attitude information including the roll angle, pitch angle, and yaw angle of the drone, and synchronize the attitude information with the calibration coefficients to output combined data of the attitude information and calibration coefficients with a unified timestamp; The geometric center position of the drone body in the actual space is determined according to the drone body structural parameters, and the geometric center position is mapped to the image coordinate system monitored by the visual sensor in combination with the combined data to determine the pixel coordinates of the body center.

[0011] Furthermore, the calculating of the pixel deviation between the center pixel coordinates of the body and the center pixel position in the image coordinate system, and converting the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation, includes: Calculating a horizontal pixel deviation and a vertical pixel deviation between the center pixel coordinates of the body and the center pixel position coordinates in the unified image coordinate system, and outputting the pixel deviation composed of the horizontal pixel deviation and the vertical pixel deviation; The pixel deviation is converted into actual distance according to the drone height through the imaging ratio relationship of the visual sensor to output the spatial horizontal deviation and the spatial vertical deviation, and the spatial horizontal deviation and the spatial vertical deviation are combined into a binary vector, which is a relative deviation vector.

[0012] Furthermore, the method of determining the ground-proximate aerodynamic compensation coefficient corresponding to the descent height of the UAV according to the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the descent height of the UAV includes: Calculate a dimensionless index and a deviation amplitude based on the relative deviation vector, the propeller disk diameter, the shaft opening diameter, and the drone descent height, wherein the dimensionless index includes the ratio of the drone descent height to the propeller disk diameter and the ratio of the shaft opening diameter to the propeller disk diameter, and the deviation amplitude is the Euclidean norm of each component in the relative deviation vector; A two-dimensional table is constructed offline according to the dimensionless index, and an axial ground effect amplification factor is calculated by bilinear interpolation. At the same time, an asymmetric lift compensation coefficient and a lateral spoiler suppression coefficient corresponding to the linear scaling of the deviation amplitude are calculated online. The ground-proximity aerodynamic compensation coefficient includes the axial ground-effect amplification coefficient, the asymmetric lift compensation coefficient and the lateral disturbance suppression coefficient.

[0013] Furthermore, generating a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation, and the ground proximity aerodynamic compensation coefficient, and controlling the drone to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set, includes: Performing a weighted correction on the spatial vertical deviation according to the axial near-ground effect magnification factor to output a vertical correction value after axial compensation; The horizontal deviation amplitude corresponding to the spatial horizontal deviation is calculated, and the correction direction angle is determined according to the rotor position angle of the UAV. The rotor thrust is adjusted in combination with the asymmetric lift compensation coefficient to output the thrust adjustment amount corresponding to each rotor.

[0014] Furthermore, the generating of a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation, and the ground proximity aerodynamic compensation coefficient, and controlling the UAV to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set, further includes: Calculating a target roll angle and a target pitch angle for attitude correction of the UAV based on the spatial horizontal deviation and the lateral spoiler suppression coefficient, and setting limits corresponding to the target roll angle and the target pitch angle; When the axial ground effect magnification factor exceeds a first threshold or the ratio of the descent height of the UAV to the propeller disk diameter is lower than a second threshold, the preset limited descent rate is executed, otherwise the terminal rate is executed; The flight control execution instruction set is obtained by packaging the flight control execution instructions corresponding to the thrust adjustment amount of each rotor, the target roll angle and the target pitch angle, the limited descent rate and the terminal rate.

[0015] A second aspect of the present invention discloses a UAV deflection correction system based on visual recognition, which is used to implement the UAV deflection correction method based on visual recognition described in the first aspect. The system includes: A landing mark recognition module is used to obtain an image of the shaft opening area in the landing mode of the UAV through a visual sensor, identify a preset landing mark in the shaft opening area image, and output the center pixel position corresponding to the landing mark; The body center mapping module is used to project the geometric center position of the UAV body in the real space into the image coordinate system unified with the landing mark based on the attitude information of the UAV during descent and the calibration parameters of the visual sensor, and obtain the pixel coordinates of the body center; a pixel deviation calculation module, configured to calculate the pixel deviation between the central pixel coordinates of the body and the central pixel position in the image coordinate system, and convert the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation; The compensation coefficient determination module is used to determine the ground-level aerodynamic compensation coefficient corresponding to the descent height of the UAV based on the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the descent height of the UAV; The UAV descent deviation correction module is used to generate a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation and the near-ground aerodynamic compensation coefficient, and respond to the flight control execution instruction set to coordinate the aerodynamic disturbance to control the UAV to correct the descent action.

[0016] A third aspect of the present invention discloses a terminal, comprising a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method of the first aspect.

[0017] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the first aspect when executed by a processor.

[0018] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages: (1) After the UAV enters the landing mode, the present invention uses the onboard visual sensor to collect images of the shaft opening area at a fixed frame rate, identifies a preset landing mark (such as a high-contrast geometric shape) in the image, and obtains the center pixel position of the landing mark in the image coordinate system through image processing, providing a target reference point for subsequent relative deviation calculation. At the same time, combined with the attitude information of the flight control system and the camera calibration parameters, the position of the geometric center of the aircraft in the actual space is projected into the image coordinate system, and a comparable coordinate system between the landing point reference and the center of the aircraft is established, which facilitates the calculation of the horizontal and vertical pixel differences between the center of the landing mark and the center of the aircraft in the image coordinate system. The pixel difference is then converted into spatial horizontal deviation and vertical deviation using the camera calibration coefficient and the current height to obtain a quantized error that can be used for attitude and position correction.

[0019] (2) The present invention searches for or calculates the ground-level aerodynamic compensation coefficients, including the asymmetric lift compensation coefficient and the lateral disturbance suppression coefficient, based on the ratio of the shaft diameter to the rotor blade diameter and the current descent altitude. This quantifies the asymmetric aerodynamic characteristics of the terminal phase and provides a correction factor for generating correction instructions. Finally, the relative deviation is combined with the aerodynamic compensation coefficient to generate a flight control execution instruction set that includes thrust distribution, attitude correction, and descent rate limitation. This synchronizes the correction action with the changing trend of the aerodynamic disturbance, offsets the asymmetric lift and turbulence interference caused by the wall-ground effect coupling, and prevents phase mismatch from amplifying drift. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the process of the UAV deflection correction method based on visual recognition provided by the present invention; Figure 2 It is a structural schematic diagram of the UAV deviation reduction correction system based on visual recognition provided by the present invention. DETAILED DESCRIPTION

[0021] The present application will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present application.

[0022] like Figure 1 As shown, in one embodiment, a method for correcting the deviation of a UAV based on visual recognition includes the following steps: In step S110 , a visual sensor is used to obtain an image of the shaft opening area in the landing mode of the UAV, and a preset landing mark is identified in the image of the shaft opening area to output a center pixel position corresponding to the landing mark.

[0023] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S110 specifically includes the following steps: Step S111: When the UAV flight control system detects a landing trigger signal, it controls the UAV to enter a landing mode in response to the landing trigger signal, and at the same time calls the onboard visual sensor to collect images of the shaft opening area at a fixed frame rate.

[0024] Step S112: constructing an original image matrix based on multiple shaft opening area images acquired at a fixed frame rate, and performing binarization processing on each image element in the original image matrix according to preset black and white contrast and shape features.

[0025] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S110 specifically further includes the following steps: In step S113, an opening operation is performed on the original image matrix to remove noise from the image elements, identify all connected pixel regions, calculate the area and aspect ratio of the connected pixel regions, and screen out candidate pixel regions that meet the preset mark specifications.

[0026] Step S114 , calculating the pixel coordinates of the corresponding enclosing rectangular bounding box of each candidate pixel region, and performing normalized matching on each candidate pixel region with the preset landing mark template to output the pixel region coordinates with the highest matching degree as the center pixel position of the landing mark.

[0027] In a specific embodiment, the present invention provides a method for reducing deviation of a UAV based on visual recognition, comprising steps 1 to 5: Step 1: Landing area visual acquisition and landmark recognition.

[0028] After the drone enters landing mode, the onboard downward-looking visual sensor collects images of the shaft opening area at a fixed frame rate and identifies a preset landing mark (such as a high-contrast geometric shape) in the image. Through image processing, the center pixel position of the landing mark in the image coordinate system is obtained, providing a target reference point for subsequent relative deviation calculations. This includes the following sub-steps: Sub-step 1.1: Landing mode triggering and sensor initialization.

[0029] Specifically, when the UAV flight control system detects a landing trigger signal, it sends a start command to the onboard downward-looking camera (i.e., visual sensor) and reads or sets the camera's calibration parameters, including the camera's pixel focal length in the horizontal or vertical direction, the pixel position of the principal point (optical axis center) in the image coordinate system, and the acquisition frame rate (frames / second, range 15-60). The above parameters are obtained through the camera calibration process, stored in the onboard storage unit, and sent to the visual processing unit through the digital interface (MIPI / USB).

[0030] Sub-step 1.2: Acquire the original image of the landing area.

[0031] Specifically, the camera captures downward-looking images of the shaft opening and surrounding areas at a set acquisition frame rate with a resolution of W×H (pixels), where W is the image width and H is the height. Finally, an original image matrix is ​​constructed based on the captured downward-looking images. The original image matrix is ​​composed of the grayscale values ​​of the red channel, the green channel, and the blue channel (0-255) of the pixel points, and each grayscale value has horizontal and vertical pixel coordinates as indexes.

[0032] Sub-step 1.3: Extract landing mark candidate areas.

[0033] Specifically, binarization is performed according to the preset color range (black and white contrast) or shape features (circle or cross), and an opening operation is performed to remove noise and small-area targets. All connected pixel areas are identified, and their areas and aspect ratios are calculated. Candidate areas that meet the preset specifications of the logo are screened out, and the four boundary pixel coordinates of the enclosing rectangle are calculated for each candidate area. Finally, a set of candidate areas is output, and each candidate area in the set is represented by a rectangular bounding box.

[0034] In sub-step 1.4, each candidate region is further verified by performing normalized cross-correlation matching between the candidate region image and a pre-stored landing point mark template (e.g., a high-contrast cross). The region with the highest matching score is selected as the final target region, and the coordinates of the target region are output as the coordinates of the center pixel position of the landing mark.

[0035] In step S120 , based on the attitude information of the drone during descent and the calibration parameters of the visual sensor, the geometric center position of the drone body in the actual space is projected into an image coordinate system that is unified with the landing mark to obtain the pixel coordinates of the center of the body.

[0036] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S120 specifically includes the following steps: Step S121: Obtain the attitude information of the drone during descent from the drone flight control system. The attitude information includes the roll angle, pitch angle, and yaw angle of the drone. The attitude information and calibration coefficients are synchronized in time to output combined data of the attitude information and calibration coefficients with a unified timestamp.

[0037] Step S122: determine the geometric center position of the drone body in the actual space based on the drone body structural parameters, and map the geometric center position to the image coordinate system monitored by the visual sensor in combination with the combined data to determine the pixel coordinates of the body center.

[0038] In a specific embodiment, the present invention provides a method for correcting the deviation of a drone based on visual recognition. Step 2 is to calculate the projection position of the center of the drone body. Combining the attitude information of the flight control system and the camera calibration parameters, the position of the drone's geometric center in real space is projected into the image coordinate system to obtain the pixel coordinates of the drone center, and a comparable coordinate system is established between the landing point reference and the drone center. This includes the following sub-steps: Sub-step 2.1: Synchronously obtain the posture and camera parameters.

[0039] Specifically, during the drone landing control process, the flight control system has a built-in time synchronization unit that triggers a synchronous data acquisition task at a fixed period (10-20ms). This task simultaneously reads the body attitude information from the inertial measurement unit (IMU) and obtains calibration parameters from the camera control unit. The body attitude information and camera calibration parameters are packaged into data frames within the same task cycle and attached with a unified timestamp. Finally, the combined data of attitude and camera parameters with a unified timestamp is obtained.

[0040] Among them, attitude information includes roll angle (describing the rotation state of the body around the longitudinal axis), pitch angle (describing the rotation state of the body around the transverse axis), and yaw angle (describing the rotation state of the body around the vertical axis). Camera calibration parameters include pixel focal length (divided into horizontal and vertical directions, obtained by the camera imaging system during the calibration phase, used to describe the proportional relationship between spatial points and pixel coordinates), optical center position (referring to the pixel coordinate corresponding to the intersection of the camera optical axis in the image coordinate system), and extrinsic parameters (the camera installation position and installation angle, describing the relative position relationship between the camera coordinate system and the body coordinate system).

[0041] Sub-step 2.2, determine the position of the geometric center of the body in the body coordinate system.

[0042] Specifically, the fuselage's geometric center is defined as the intersection of the arms. This intersection, fixed at the origin of the fuselage coordinate system during design and manufacturing, serves as both the drone's geometric center and the reference point for center of gravity design. It remains constant despite the drone's flight attitude or environmental conditions. To ensure the accuracy of subsequent calculations, the fuselage's center position is confirmed during manufacturing using a coordinate measuring machine and stored as a constant in the flight control system.

[0043] Sub-step 2.3: Convert the center position of the body to the monitoring direction of the camera.

[0044] Specifically, camera calibration is used to obtain the spatial transformation parameters between the body coordinate system and the camera coordinate system, including the rotation matrix (which describes the directional relationship) and the translation vector (which describes the position offset). During calculation, the position of the body center in the body coordinate system is mapped to the camera coordinate system through the spatial transformation relationship, thereby obtaining the three-dimensional position of the body center in the camera monitoring direction. This process is used to convert the coordinates from the body's perspective into the coordinates from the camera's perspective to ensure that the body center position in the camera monitoring image coordinate system is consistent with the actual body center position.

[0045] Sub-step 2.4, project the body center position onto the image plane.

[0046] Specifically, the camera's perspective imaging principle is used to convert the 3D position of the drone's center in the camera coordinate system into a 2D pixel position on the image plane. This process first converts the 3D spatial coordinates into pixel plane coordinates proportionally based on the focal length and optical center position parameters. The conversion result and the landing mark pixel coordinates are then kept in the same coordinate system (i.e., the unified image coordinate system) for direct comparison. For mass-produced drones, if the camera's mounting position and angle are highly consistent, a mapping table between the drone's center position and pixel coordinates can be created after initial calibration. During operation, the drone's center pixel position can be directly retrieved from the table, reducing the real-time computational burden.

[0047] Step S130 , calculating the pixel deviation between the center pixel coordinates of the object and the center pixel position in the image coordinate system, and converting the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation.

[0048] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S130 specifically includes the following steps: Step S131 , calculating the horizontal pixel deviation and the vertical pixel deviation between the center pixel coordinates of the object and the center pixel position coordinates in a unified image coordinate system, and outputting a pixel deviation composed of the horizontal pixel deviation and the vertical pixel deviation.

[0049] In step S132, the pixel deviation is converted into the actual distance according to the drone height through the imaging ratio relationship of the visual sensor to output the spatial horizontal deviation and the spatial vertical deviation, and the spatial horizontal deviation and the spatial vertical deviation are combined into a binary vector, which is a relative deviation vector.

[0050] In a specific embodiment, the present invention provides a method for correcting the deviation of a drone based on visual recognition. Step 3, relative deviation calculation, calculates the horizontal and vertical pixel differences between the center of the landing mark and the center of the drone in the image coordinate system. Using the camera calibration coefficient and the current altitude, the pixel differences are converted into spatial horizontal deviation and vertical deviation to obtain a quantized error that can be used for attitude and position correction. This includes the following sub-steps: Sub-step 3.1, obtain the pixel positions of the landing mark and the center of the aircraft.

[0051] Specifically, the center pixel position coordinates corresponding to the landing mark and the center pixel position coordinates of the body are placed in a unified image coordinate system (including horizontal and vertical axes, in pixels). The two center pixel position coordinates are based on the same resolution, the same optical center position, and the same acquisition frame rate.

[0052] Sub-step 3.2, calculate the horizontal and vertical pixel differences.

[0053] Specifically, the horizontal pixel difference is the difference between the horizontal pixel position of the landing mark center and the horizontal pixel position of the body center, and the vertical pixel difference is the difference between the vertical pixel position of the landing mark center and the vertical pixel position of the body center.

[0054] Sub-step 3.3, convert the pixel difference into the actual space difference.

[0055] Specifically, through the imaging ratio relationship of the camera, the horizontal and vertical pixel differences are converted into actual spatial distances according to the current height of the UAV, which are divided into the actual deviation of the UAV in the horizontal plane and the actual deviation in the horizontal plane. The actual deviation of the UAV in the horizontal plane is equal to the ratio of the horizontal pixel difference to the horizontal pixel focal length in the camera calibration parameters multiplied by the vertical height of the UAV from the landing plane; the actual deviation of the UAV in the horizontal plane is equal to the ratio of the vertical pixel difference to the vertical pixel focal length in the camera calibration parameters multiplied by the vertical height of the UAV from the landing plane.

[0056] Sub-step 3.4, output the quantized error that can be used for attitude and position correction.

[0057] Specifically, the actual deviation of the UAV in the horizontal plane and the actual deviation in the horizontal plane are combined into a binary vector as the relative position error between the current UAV and the landing mark, and are packaged into a data frame and sent to the UAV flight control system to provide input for subsequent aerodynamic compensation and correction command generation.

[0058] Step S140 , determining a ground-level aerodynamic compensation coefficient corresponding to the descent height of the UAV based on the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the descent height of the UAV.

[0059] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S140 specifically includes the following steps: Step S141, calculate the dimensionless index and deviation amplitude based on the relative deviation vector, the propeller disk diameter, the shaft opening diameter, and the drone descent height. The dimensionless index includes the ratio of the drone descent height to the propeller disk diameter and the ratio of the shaft opening diameter to the propeller disk diameter. The deviation amplitude is the Euclidean norm of each component in the relative deviation vector.

[0060] In step S142, a two-dimensional table is constructed offline based on the dimensionless index, and the axial ground effect amplification factor is calculated by bilinear interpolation. At the same time, the asymmetric lift compensation coefficient and the lateral spoiler suppression coefficient corresponding to the linear scaling of the deviation amplitude are calculated online.

[0061] Among them, the near-ground aerodynamic compensation coefficient includes the axial near-ground effect amplification coefficient, the asymmetric lift compensation coefficient and the lateral disturbance suppression coefficient.

[0062] In a specific embodiment, the visual recognition-based UAV descent correction method provided by the present invention includes step 4, determining the aerodynamic compensation coefficient. Based on the ratio of the shaft opening diameter to the rotor blade diameter, as well as the current descent altitude, the ground-level aerodynamic compensation coefficients, including the asymmetric lift compensation coefficient and the lateral turbulence suppression coefficient, are searched or calculated. This quantifies the asymmetric aerodynamic characteristics of the terminal phase and provides a correction factor for generating correction instructions. This includes the following sub-steps: Sub-step 4.1, determination of aerodynamic compensation coefficients based on two-dimensional table interpolation method.

[0063] Specifically, based on the relative position error obtained in step 3, the drone's body geometry (propeller disk diameter, shaft opening diameter), and the drone's descent height, the dimensionless index (height ratio and diameter ratio) and the deviation amplitude are calculated. The height ratio is the ratio of the drone's descent height to the propeller disk diameter, and the diameter ratio is the ratio of the shaft opening diameter to the propeller disk diameter. The deviation amplitude is equal to the square root of the sum of the actual deviation of the drone in the horizontal plane and the actual deviation in the horizontal plane, that is, the Euclidean norm of these two.

[0064] Sub-step 4.2: Obtain the axial ground effect magnification factor.

[0065] Specifically, a two-dimensional table is constructed during the offline calibration phase to record the total thrust amplification required under different height-to-diameter ratios, and bilinear interpolation is used during runtime: Set height ratio Falling on the grid , diameter ratio Falling on the grid , the four table point values ​​are ,make: ; but ; Where, , the boundary is extrapolated nearby and the overall amplitude is limited to the empirical safety range [1.0,1.8]; is the interpolation weight in the height direction, indicating the current height is greater than the height of the two adjacent table height nodes. and The position ratio between When , it means it just falls on a low-height node and no upward interpolation is required; when When , it means it just falls on the high altitude node and no downward interpolation is needed; when When , it means it is in the middle of the two nodes, and the weight is half each; is the axial ground effect amplification factor.

[0066] It should be noted that during the offline two-dimensional table construction process, two key factors are gradually varied through ground test platforms or physical flight tests under typical UAV shaft landing conditions: the altitude ratio (the ratio of the current descent altitude to the propeller disk diameter) and the diameter ratio (the ratio of the shaft diameter to the propeller disk diameter). For each altitude / diameter ratio combination, the aircraft is kept in a horizontal, fixed-point hovering state and a normalized horizontal deviation (for example, 0.2 times the propeller disk diameter) is artificially applied. The change in thrust distribution required by the flight control system to maintain attitude stability under this disturbance is recorded. This thrust change is converted to the unit normalized deviation to obtain the corresponding asymmetric lift base response value. These measurements are repeated at different altitude / diameter ratio combinations until the preset operating range is covered. Finally, all measured base response values ​​are entered into a two-dimensional table with altitude ratio as the vertical axis and diameter ratio as the horizontal axis. This serves as the basis for the online compensation lookup table. Areas outside the table are extrapolated using boundary values ​​and clipped to a safe range.

[0067] It should be noted that These are the axial thrust amplification factors pre-measured in the offline test, corresponding to the four corner points that fall within the current index interval when looking up the table. They serve as the basic data points for bilinear interpolation, equivalent to the four corners of the map. The value of the middle position needs to be calculated proportionally based on these four corners. is the predicted value for low height ratio and low diameter ratio; is the predicted value for low height ratio and high diameter ratio; is the predicted value for high height ratio and low diameter ratio; It is the predicted value of high height ratio and high diameter ratio.

[0068] Sub-step 4.3: Obtain the asymmetric lift compensation coefficient based on linear scaling of the deviation amplitude.

[0069] Specifically, create a two-dimensional table offline The values ​​in the two-dimensional table represent the asymmetric lift base response caused by the unit normalized deviation (dimensionless, typically in the range of 0.3-1.0). During online calculation, they are linearly scaled and limited according to the current deviation amplitude. The expression is: ; Where, Indicates the intensity of the deviation relative to the disk scale, is the horizontal deviation amplitude, is the paddle disc diameter; The minimum and maximum values ​​of the engineering limit are in the range of [0.85, 1.25]. If the minimum value of the engineering limit is too small, the correction force will be insufficient, resulting in the attitude drift not being able to converge in time. If the maximum value is too large, the propeller thrust on one side may be too high, causing the aircraft to shake or even become unstable. is the asymmetric lift compensation coefficient.

[0070] It should be noted that It is a truncation function used to limit a value between the specified upper and lower limits, expressed as: ; Where, ; .

[0071] Sub-step 4.4, obtain the lateral disturbance suppression coefficient based on attitude and lateral channel desensitization.

[0072] Specifically, based on the two-dimensional table established offline The values ​​in the two-dimensional table describe the attitude of the vertical shaft lateral backflow intensity and the desensitization requirements of the lateral channel (dimensionless, range 0.0-0.8). The deviation amplitude is slightly enhanced and limited online. The expression is: ; Where, is the linear enhancement coefficient, ranging from [0.2, 0.6]; is the upper limit of the limit, and the value is 1.5; is the lateral disturbance suppression coefficient.

[0073] It should be noted that the linear enhancement factor is used to amplify or weaken the flight control's ability to offset lateral disturbances. The lateral correction (roll or yaw command) is multiplied by the linear enhancement factor before being issued to the actuator. When the linear enhancement factor is 0.2, only 20% of the original correction is used to control the lateral channel, preventing oscillation caused by excessive correction. This is suitable for the terminal stage of the shaft, where the wall turbulence is extremely unstable. When the linear enhancement factor is 0.6, only 60% of the original correction is used by the flight control, which is suitable for the descent stage with a relatively stable aerodynamic environment.

[0074] Step S150: Generate a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation, and the ground proximity aerodynamic compensation coefficient, and control the drone to correct the descent action in response to the flight control execution instruction set in coordination with the aerodynamic disturbance.

[0075] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S150 specifically includes the following steps: Step S151 , performing weighted correction on the spatial vertical deviation according to the axial near-ground effect magnification factor to output a vertical correction value after axial compensation.

[0076] Step S152, calculate the horizontal deviation amplitude corresponding to the spatial horizontal deviation, and determine the correction direction angle according to the UAV rotor position angle, and adjust the rotor thrust in combination with the asymmetric lift compensation coefficient to output the thrust adjustment amount corresponding to each rotor.

[0077] In some embodiments, the method for reducing deviation of a drone based on visual recognition provided by the present invention, step S150 specifically further includes the following steps: Step S153: Calculate the target roll angle and target pitch angle for the attitude correction of the UAV based on the spatial horizontal deviation and the lateral disturbance suppression coefficient, and set the limits corresponding to the target roll angle and target pitch angle.

[0078] Step S154: When the axial ground effect magnification factor exceeds the first threshold or the ratio of the drone's descent height to the propeller disk diameter is lower than the second threshold, the preset limited descent rate is executed; otherwise, the terminal rate is executed.

[0079] The flight control execution instruction set is obtained by packaging the flight control execution instructions corresponding to the thrust adjustment amount of each rotor, the target roll angle and the target pitch angle, the limited descent rate and the terminal rate.

[0080] In a specific embodiment, the present invention provides a method for correcting the deviation of a UAV based on visual recognition. Step 5 is to generate correction instructions. The relative deviation obtained in step 3 is combined with the aerodynamic compensation coefficient in step 4 to generate a flight control execution instruction set including thrust distribution, attitude correction, and descent rate limit. The correction action is synchronized with the changing trend of the aerodynamic disturbance, offsetting the asymmetric lift and turbulence interference caused by the wall-ground effect coupling, and preventing phase mismatch from amplifying drift. The method includes the following sub-steps: Sub-step 5.1, aerodynamic compensation weighted deviation solution.

[0081] Specifically, according to the axial ground effect amplification factor Vertical deviation in space Perform weighted correction to obtain the vertical correction amount after axial compensation , the expression is: ; when ≥0, indicating the amplification degree of the vertical aerodynamic response to the near ground effect. When it is large, it means that the ground effect under the current altitude of the UAV and the geometric conditions of the wellhead is significant, and the vertical correction amplitude needs to be increased in advance to offset the altitude fluctuation.

[0082] Sub-step 5.2, calculation of asymmetric thrust distribution.

[0083] Specifically, the actual deviation of the UAV in the horizontal plane obtained in step 3 is and the actual longitudinal deviation of the horizontal plane , combined with the asymmetric lift compensation coefficient obtained in step 4 and the rotor position angle of the UAV, calculate the horizontal deviation amplitude, the correction direction angle, and the adjustment amount of each rotor thrust.

[0084] Among them, the horizontal deviation amplitude is calculated The expression is: ; Calculate the corrected bearing angle The expression is: ; Thrust adjustment for each rotor The expression is: ; Where, This is the thrust proportional gain, ranging from 0.1-0.5N / m. When the thrust adjustment is added to the basic hover thrust, a new thrust control command is obtained.

[0085] Sub-step 5.3: Generate posture correction instructions.

[0086] Specifically, the actual deviation of the UAV in the horizontal plane obtained in step 3 is and the actual longitudinal deviation of the horizontal plane , combined with the lateral disturbance suppression coefficient obtained in step 4 , calculate the target roll angle and target pitch angle. Among them, the target roll angle is the corresponding attitude proportional coefficient and lateral disturbance suppression coefficient Actual deviation in the longitudinal direction from the horizontal plane The product of the three; the target pitch angle is the inverse of the corresponding attitude proportional coefficient and the lateral spoiler suppression coefficient Actual deviation from the horizontal plane The product of the three; the attitude scale coefficient corresponding to the target roll angle and the target pitch angle is in the range of 2-5° per meter. After calculating the target roll angle and the target pitch angle, set the engineering limit (±8°) to prevent attitude oscillation caused by over-correction.

[0087] Sub-step 5.4, drop rate limit and instruction encapsulation.

[0088] Specifically, according to the axial ground effect amplification factor The target descent rate is determined by the current altitude of the drone: when >0.5 or height ratio When the descent rate is limited to below 0.3m / s; otherwise, it is executed according to the normal terminal rate (0.4-0.6m / s).

[0089] Finally, the flight control execution instructions corresponding to the aforementioned rotor thrust adjustment amounts, target roll angle and target pitch angle, limited descent rate and terminal rate are packaged into an instruction set, and sent out at one time through the flight control bus. The instruction set is executed to perform descent correction control on the drone.

[0090] The following describes the drone deflection correction system based on visual recognition provided by the present invention. The drone deflection correction system based on visual recognition described below and the drone deflection correction method based on visual recognition described above can refer to each other.

[0091] like Figure 2 As shown, in one embodiment, a UAV deviation correction system based on visual recognition includes a landing mark recognition module, an aircraft center mapping module, a pixel deviation calculation module, a compensation coefficient determination module and a UAV deviation correction module.

[0092] The landing mark recognition module is used to obtain the shaft opening area image in the UAV landing mode through the visual sensor, and identify the preset landing mark in the shaft opening area image to output the center pixel position corresponding to the landing mark.

[0093] The body center mapping module is used to project the geometric center position of the drone body in the actual space to the image coordinate system unified with the landing mark based on the attitude information of the drone during descent and the calibration parameters of the visual sensor, and obtain the pixel coordinates of the body center.

[0094] The pixel deviation calculation module is used to calculate the pixel deviation between the center pixel coordinates of the body and the center pixel position in the image coordinate system, and convert the pixel deviation into spatial horizontal deviation and spatial vertical deviation.

[0095] The compensation coefficient determination module is used to determine the near-ground aerodynamic compensation coefficient corresponding to the drone's descent height based on the ratio of the shaft opening diameter to the drone's propeller disc diameter and the drone's descent height.

[0096] The drone descent correction module is used to generate a flight control execution instruction set based on the spatial horizontal deviation, spatial vertical deviation and proximity aerodynamic compensation coefficient, and respond to the flight control execution instruction set to coordinate aerodynamic disturbance control to correct the drone's descent action.

[0097] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation plans of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and is not a limitation on the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for reducing deviation of unmanned aerial vehicles based on visual recognition, characterized in that: The method comprises: Acquire an image of the shaft opening area in the landing mode of the UAV through a visual sensor, identify a preset landing mark in the shaft opening area image, and output a center pixel position corresponding to the landing mark; Based on the attitude information of the UAV during descent and the calibration parameters of the visual sensor, the geometric center position of the UAV body in the real space is projected into the image coordinate system that is consistent with the landing mark to obtain the pixel coordinates of the center of the body; Calculating the pixel deviation between the center pixel coordinates of the body and the center pixel position in the image coordinate system, and converting the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation; According to the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the UAV descent height, the aerodynamic compensation coefficient corresponding to the UAV descent height is determined; A flight control execution instruction set is generated based on the spatial horizontal deviation, the spatial vertical deviation and the ground proximity aerodynamic compensation coefficient, and the UAV is controlled to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set.

2. The method for reducing deviation of a UAV based on visual recognition according to claim 1, characterized in that: The method of acquiring an image of the shaft opening area in the landing mode of the UAV by a visual sensor and identifying a preset landing mark in the shaft opening area image to output a center pixel position corresponding to the landing mark includes: When the UAV flight control system detects a landing trigger signal, it controls the UAV to enter a landing mode in response to the landing trigger signal, and simultaneously calls the onboard visual sensor to capture an image of the shaft opening area at a fixed frame rate; An original image matrix is ​​constructed based on multiple shaft opening area images acquired at a fixed frame rate, and each image element in the original image matrix is ​​binarized according to preset black and white contrast and shape features.

3. The method for reducing deviation of a UAV based on visual recognition according to claim 2, characterized in that: The method further includes: acquiring an image of the shaft opening area in the landing mode of the drone through a visual sensor, identifying a preset landing mark in the shaft opening area image, and outputting a center pixel position corresponding to the landing mark. Performing an opening operation on the original image matrix to remove noise from image elements, identify all connected pixel regions, calculate the area and aspect ratio of the connected pixel regions, and screen out candidate pixel regions that meet preset sign specifications; The pixel coordinates of the enclosing rectangular bounding box corresponding to each candidate pixel area are calculated, and each candidate pixel area is normalized and matched with the preset landing mark template to output the coordinates of the pixel area with the highest matching degree as the center pixel position of the landing mark.

4. The method for reducing deviation of a UAV based on visual recognition according to claim 1, characterized in that: The method of projecting the geometric center position of the drone body in the actual space to the image coordinate system unified with the landing mark based on the attitude information of the drone during descent and the calibration parameters of the visual sensor to obtain the pixel coordinates of the center of the drone body includes: Acquire the attitude information of the drone during descent from the drone flight control system, the attitude information including the roll angle, pitch angle, and yaw angle of the drone, and synchronize the attitude information with the calibration coefficients to output combined data of the attitude information and calibration coefficients with a unified timestamp; The geometric center position of the drone body in the actual space is determined according to the drone body structural parameters, and the geometric center position is mapped to the image coordinate system monitored by the visual sensor in combination with the combined data to determine the pixel coordinates of the body center.

5. The method for reducing deviation of a UAV based on visual recognition according to claim 1, characterized in that: Calculating the pixel deviation between the center pixel coordinates of the body and the center pixel position in the image coordinate system, and converting the pixel deviation into a spatial horizontal deviation and a spatial vertical deviation, includes: Calculating a horizontal pixel deviation and a vertical pixel deviation between the center pixel coordinates of the body and the center pixel position coordinates in the unified image coordinate system, and outputting the pixel deviation composed of the horizontal pixel deviation and the vertical pixel deviation; The pixel deviation is converted into actual distance according to the drone height through the imaging ratio relationship of the visual sensor to output the spatial horizontal deviation and the spatial vertical deviation, and the spatial horizontal deviation and the spatial vertical deviation are combined into a binary vector, which is a relative deviation vector.

6. The method for reducing deviation of a UAV based on visual recognition according to claim 5, characterized in that: Determining the ground proximity aerodynamic compensation coefficient corresponding to the descent height of the UAV according to the ratio of the shaft opening diameter to the propeller disk diameter of the UAV body and the descent height of the UAV includes: Calculate a dimensionless index and a deviation amplitude based on the relative deviation vector, the propeller disk diameter, the shaft opening diameter, and the drone descent height, wherein the dimensionless index includes the ratio of the drone descent height to the propeller disk diameter and the ratio of the shaft opening diameter to the propeller disk diameter, and the deviation amplitude is the Euclidean norm of each component in the relative deviation vector; A two-dimensional table is constructed offline according to the dimensionless index, and an axial ground effect amplification factor is calculated by bilinear interpolation. At the same time, an asymmetric lift compensation coefficient and a lateral spoiler suppression coefficient corresponding to the linear scaling of the deviation amplitude are calculated online. The ground-proximity aerodynamic compensation coefficient includes the axial ground-effect amplification coefficient, the asymmetric lift compensation coefficient and the lateral disturbance suppression coefficient.

7. The method for reducing deviation of a UAV based on visual recognition according to claim 6, characterized in that: The generating of a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation, and the ground proximity aerodynamic compensation coefficient, and controlling the UAV to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set, includes: Performing a weighted correction on the spatial vertical deviation according to the axial near-ground effect magnification factor to output a vertical correction value after axial compensation; The horizontal deviation amplitude corresponding to the spatial horizontal deviation is calculated, and the correction direction angle is determined according to the rotor position angle of the UAV. The rotor thrust is adjusted in combination with the asymmetric lift compensation coefficient to output the thrust adjustment amount corresponding to each rotor.

8. The method for reducing deviation of a UAV based on visual recognition according to claim 7, characterized in that: The method further comprises: generating a flight control execution instruction set based on the spatial horizontal deviation, the spatial vertical deviation, and the ground proximity aerodynamic compensation coefficient, and controlling the drone to correct the descent action in coordination with the aerodynamic disturbance in response to the flight control execution instruction set. Calculating a target roll angle and a target pitch angle for attitude correction of the UAV based on the spatial horizontal deviation and the lateral spoiler suppression coefficient, and setting limits corresponding to the target roll angle and the target pitch angle; When the axial ground effect magnification factor exceeds a first threshold or the ratio of the descent height of the UAV to the propeller disk diameter is lower than a second threshold, the preset limited descent rate is executed, otherwise the terminal rate is executed; The flight control execution instruction set is obtained by packaging the flight control execution instructions corresponding to the thrust adjustment amount of each rotor, the target roll angle and the target pitch angle, the limited descent rate and the terminal rate.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images