A paragliding posture detection method, flight control method and system

Abstract

The application relates to a paragliding posture detection method, a flight control method and a system. First, an image containing a paraglider is acquired and a canopy mask is extracted, and then a canopy edge contour is obtained through contour tracking; based on the contour, edge curvature is calculated, and curve fitting is performed on the upper edge of the canopy to extract canopy curvature parameters; at the same time, the canopy is divided into left and right through the establishment of a contour center axis, mirror error is calculated to obtain a symmetry index. Further, the pilot position is normalized in the canopy coordinate system to obtain relative position information. Finally, the edge curvature, the canopy curvature parameters and the symmetry index are mapped into the posture parameters of the paraglider, and quantitative determination of abnormal postures such as collapse, yaw and pitch is realized. The application constructs a continuous conversion from pixel features to control variables through a visual image link, effectively making up for the defect that traditional methods lack direct perception of the towed object.

Description

Technical Field

[0001] This application relates to the fields of computer vision, UAV flight control and intelligent sensing technology, and in particular to a paraglider attitude detection method, flight control method and system. Background Technology

[0002] The core technological bottleneck of current drone-guided paragliding systems lies in the fact that their control decisions rely solely on information sources limited to the pilot's visual observation or indirect physical quantities such as tension and IMU (Inductively Coupled Measure). They cannot directly perceive the paraglider's spatial attitude (such as angle of attack, yaw, and canopy collapse), resulting in a blind spot in the system's perception of the paraglider's actual motion. This deficiency directly leads to control lag, over-reliance on human experience, and difficulty in responding promptly to sudden anomalies such as canopy folding or lateral deviation. Summary of the Invention

[0003] Therefore, it is necessary to provide a paraglider attitude detection method, flight control method, and system to address the technical problem that current UAV-guided paraglider systems lack paraglider attitude perception capabilities.

[0004] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: The first aspect of the present invention provides a paraglider attitude detection method for detecting the attitude of a paraglider towed by a drone, comprising the following steps: Acquire an image including a paraglider, and obtain a wing mask for the paraglider based on the image; Based on contour tracking, the edge contour of the parachute is obtained from the parachute mask; The curvature of the parachute edge is calculated based on the parachute edge profile. Based on the outline of the parachute wing edge, curve fitting is performed on the upper edge of the parachute wing to obtain the parachute surface curvature parameters; Based on the outline of the parachute edge, the central axis of the curvature of the parachute edge is established, and the outline of the parachute edge is divided into left and right sections according to the central axis to obtain the symmetry index. The attitude of the paraglider is determined based on the curvature of the wing edge, the arc parameters of the canopy surface, and the symmetry index.

[0005] Furthermore, the parachute mask is a binary image, wherein pixels with a value of 1 belong to the parachute wing, and pixels with a value of 0 belong to the background. The step of obtaining the parachute wing edge contour from the parachute mask based on contour tracking includes: The parachute mask is subjected to noise reduction and hole filling operations to obtain a pre-processed parachute mask; The preprocessed parachute mask is scanned from the top left corner to the bottom right corner to find the first pixel that meets the preset conditions as the boundary starting point. The preset conditions include that the pixel value is 1 and there are pixels with a pixel value of 0 in the neighborhood of the pixel. Using the boundary starting point as the current point, execute the following loop steps: Starting from the current point, the neighborhood of the current point is searched in a fixed order to find the first pixel that meets the preset conditions, and a new pixel is obtained. The new pixel is then used as the current point to perform the loop step until the current point coincides with the boundary starting point, thus obtaining the set of parachute edge contour points, which serves as the parachute edge contour.

[0006] Furthermore, the calculation of the wing edge curvature based on the wing edge profile includes: The upper edges of the parachute wing edge profile are sorted and parameterized according to their arc lengths; Calculate the wing edge curvature based on the upper edge of the sorted and parameterized wing edge profile:

[0007] In the formula, Indicates along the arc length The curvature of the parachute edge, These represent the pixel coordinates of the upper edge of the sorted and parameterized parachute edge contours, respectively.

[0008] Furthermore, the process of curve fitting the upper edge of the umbrella wing to obtain the umbrella surface curvature parameters includes: The upper edge of the parachute wing is fitted using a quadratic polynomial, a cubic polynomial, or a spline to obtain the fitted curve function; The umbrella surface curvature parameters are obtained based on the fitted curve function:

[0009] In the formula, This indicates the curvature parameter of the umbrella surface. Indicates the height of the bow. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing. This represents the ordinate of the highest point of the contour obtained based on the fitted curve function. This indicates the height difference between the left and right endpoints of the upper edge of the parachute wing.

[0010] Furthermore, establishing the central axis of the curvature of the parachute edge includes any of the following steps: S1: Calculate the centroid of the parachute edge profile based on the parachute edge profile, calculate the principal direction of the parachute edge profile using the PCA algorithm, and establish the central axis of the parachute edge curvature based on the centroid and principal direction of the parachute edge profile. S2: Based on the image including the paraglider, the parachute line convergence point is obtained by key point detection or line segment clustering, and the central axis of the paraglider edge curvature is established according to the parachute line convergence point and the highest point of the contour.

[0011] Furthermore, the symmetry index includes mirror error or area difference, wherein: The method for calculating the mirror error is as follows:

[0012] In the formula, This indicates the left side of the parachute edge profile after dividing the parachute edge profile left and right according to the central axis. Or the outline of the right wing edge The number of pixels, Indicates the first 1 pixel Indicates mirror transformation; The method for calculating the area difference is as follows:

[0013] In the formula, , These represent the left and right areas respectively after dividing the edge contour of the parachute wing according to the central axis.

[0014] Furthermore, the determination of the paraglider's attitude based on the wing edge curvature, canopy arc parameters, and symmetry index includes: When the symmetry index is greater than the symmetry threshold, the paraglider is determined to be in a unilateral collapse posture. When local curvature is abnormal and contour continuity is disrupted, the paraglider's attitude is determined to be leading-edge collapse, wherein: The criteria for determining the local curvature anomaly are as follows:

[0015]

[0016] In the formula, The first-order difference representing the curvature of the parachute edge. Indicates unit arc length, This represents the first curvature change threshold; Alternatively, the criteria for determining the local curvature anomaly are:

[0017]

[0018] In the formula, The second difference representing the curvature of the parachute edge. This represents the threshold for the second curvature change; The criteria for determining the disruption of the contour continuity are as follows:

[0019] In the formula, This represents the distance between two adjacent pixels. This represents two adjacent pixels on the curvature of the umbrella wing edge. Indicates the distance threshold; Alternatively, the condition for determining the disruption of the contour continuity is:

[0020]

[0021] In the formula, , Represents pixels tangent angle, Indicates the tangent angle threshold. , Representing pixels x and y coordinates , Representing pixels The x and y coordinates.

[0022] A second aspect of the present invention provides a flight control method for controlling a drone towing a paraglider, comprising the following steps: Based on images including drones and paragliders, the paraglider attitude is obtained based on the paraglider attitude detection method described in the first aspect of the present invention, and the pilot position is obtained based on target detection. Based on the centroid of the pilot's position and the edge profile of the parachute, calculate the lateral offset of the pilot relative to the center of the parachute; Based on the lateral offset divided by the wing width, the normalized relative position of the pilot with respect to the wing center is obtained; Based on the paraglider attitude and the normalized relative position of the pilot with respect to the center of the wing, calculate the current roll angle, pitch angle and yaw angle of the paraglider; The flight attitude of the UAV is controlled based on the current roll, pitch, and yaw angles of the paraglider and the target roll, pitch, and yaw angles.

[0023] Further calculations are made of the current roll angle, pitch angle, and yaw angle of the paraglider, including: Roll angle :

[0024] In the formula, This indicates the height of the leftmost point of the upper edge of the parachute wing. This indicates the height of the rightmost point of the upper edge of the parachute wing. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing; Pitch angle :

[0025] In the formula, Indicates the umbrella surface curvature parameter; Yaw angle :

[0026] In the formula, This indicates the normalized relative position of the pilot with respect to the center of the parachute.

[0027] A third aspect of the present invention provides a flight control system, comprising: A visual acquisition module, used to acquire images; A target recognition module, which is used to identify paragliders and pilot positions based on the image; An attitude estimation module, wherein the attitude estimation module obtains the paraglider attitude based on the paraglider attitude detection method described in the first aspect of the present invention; The control decision module calculates the lateral offset of the pilot relative to the center of the wing based on the centroid of the pilot's position and the edge contour of the wing; it obtains the normalized relative position of the pilot relative to the center of the wing by dividing the lateral offset by the wing width; it calculates the roll angle, pitch angle, and yaw angle of the current paraglider based on the paraglider attitude and the normalized relative position of the pilot relative to the center of the wing; and it outputs the UAV attitude correction amount based on the roll angle, pitch angle, and yaw angle of the current paraglider and the target roll angle, pitch angle, and yaw angle. The flight control execution module controls the flight attitude of the UAV based on the UAV attitude correction amount.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention extracts basic features such as wing contour, line convergence point, and pilot position from paraglider images, and then constructs a geometric feature model based on these features: curve fitting is performed on the upper edge of the wing to obtain the canopy curvature parameters; the contour is divided into left and right sections by establishing a central axis and the mirror error is calculated to obtain a symmetry index; and the pilot position is normalized in the paraglider coordinate system to obtain relative position information. Furthermore, the above geometric features are mapped to the paraglider's attitude parameters and flight state variables, thereby providing quantifiable visual input for UAV control and realizing continuous conversion from image to control variables. Attached Figure Description

[0029] Figure 1 A flowchart illustrating a paraglider attitude detection method provided in an embodiment of the present invention; Figure 2 A schematic flowchart of a flight control method provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a flight control system provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0031] The first embodiment of the present invention provides a paraglider attitude detection method, such as... Figure 1 As shown, attitude detection for drone-tethered paragliders includes the following steps: Acquire an image including a paraglider, and obtain a wing mask for the paraglider based on the image; Based on contour tracking, the edge contour of the parachute is obtained from the parachute mask; The curvature of the parachute edge is calculated based on the parachute edge profile. Based on the outline of the parachute wing edge, curve fitting is performed on the upper edge of the parachute wing to obtain the parachute surface curvature parameters; Based on the outline of the parachute edge, the central axis of the curvature of the parachute edge is established, and the outline of the parachute edge is divided into left and right sections according to the central axis to obtain the symmetry index. The attitude of the paraglider is determined based on the curvature of the wing edge, the arc parameters of the canopy surface, and the symmetry index.

[0032] In this embodiment, the wing mask of the paraglider can be obtained based on the image by semantic segmentation. Semantic segmentation is performed using a model built on DeepLabV3+ to obtain the wing mask of the paraglider.

[0033] In a further embodiment, the parachute mask is a binary image, wherein pixels with a value of 1 belong to the parachute wing, and pixels with a value of 0 belong to the background. The step of obtaining the parachute wing edge contour from the parachute mask based on contour tracking includes: The umbrella wing mask is subjected to noise reduction using opening operations and hole filling operations using closing operations to ensure boundary connectivity, thereby obtaining a clean area and a pre-processed umbrella wing mask. The preprocessed parachute mask is scanned from the top left corner to the bottom right corner to find the first pixel that meets the preset conditions as the boundary starting point. The preset conditions include that the pixel value is 1 and there are pixels with a pixel value of 0 in the neighborhood of the pixel. The specific order from the top left corner to the bottom right corner described in this embodiment is as follows: starting from the top left corner origin O(0,0), group all pixels according to their coordinates and s values, with s increasing from 0 to (W... 1)+(H 1) Each layer s corresponds to a stepped line segment that slopes diagonally from the lower left to the upper right or from the upper left to the lower right; during scanning, pixels within each layer are read sequentially in ascending order of their horizontal coordinate x, at which point the vertical coordinate y = s. x decreases accordingly. After processing all pixels in the current s layer, the entire process moves to the next layer s+1. This path ensures that the scan front remains a stepped diagonal line moving towards the lower right corner until the final layer reaches the lower right corner (W). 1,H 1) This ensures full mask coverage; Using the boundary starting point as the current point, execute the following loop steps: Starting from the current point, the neighborhood of the current point is searched in a fixed order to find the first pixel that meets the preset conditions, and a new pixel is obtained. The new pixel is then used as the current point to perform the loop step until the current point coincides with the boundary starting point, thus obtaining the set of parachute edge contour points, which serves as the parachute edge contour.

[0034] This embodiment refers to the current point. The neighborhood is defined as:

[0035] in, .

[0036] In this embodiment, the fixed order is used to search the neighborhood in either clockwise or counterclockwise order.

[0037] In this embodiment, each point in the generated parachute edge contour point set is arranged in an orderly manner along the boundary, forming a closed curve. The parachute edge contour point set generated by the method in this embodiment can distinguish between the outer contour and the inner contour, while maintaining the hierarchical relationship. The contour is accurate to the pixel level and there will be no breakage. It is very suitable for subsequent curvature calculation and curve fitting. Moreover, when performing left and right segmentation, it is also necessary to know the order of the points on the curve. Through the generated parachute edge contour point set, the parachute region is transformed into a circle of orderly boundary points.

[0038] In a further embodiment, calculating the wing edge curvature based on the wing edge profile includes: The upper edges of the parachute wing edge profile are sorted and parameterized according to their arc lengths; Calculate the wing edge curvature based on the upper edge of the sorted and parameterized wing edge profile:

[0039] In the formula, Indicates along the arc length The curvature of the parachute edge, These represent the pixel coordinates of the upper edge of the sorted and parameterized parachute edge contours, respectively.

[0040] In a further embodiment, the step of curve fitting the upper edge of the umbrella wing to obtain the umbrella surface curvature parameters includes: The upper edge of the parachute wing is fitted using a quadratic polynomial, a cubic polynomial, or splines to obtain a fitted curve function. As an example, the following polynomial can be used for fitting:

[0041] The umbrella surface curvature parameters are obtained based on the fitted curve function:

[0042] In the formula, This indicates the curvature parameter of the umbrella surface. Indicates the height of the bow. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing. This represents the ordinate of the highest point of the contour obtained based on the fitted curve function. This indicates the height difference between the left and right endpoints of the upper edge of the parachute wing.

[0043] In this embodiment, the upper edge of the paraglider refers to the leading edge, which is the transverse arched outline at the frontmost point of the wing's windward surface and the location of the air inlets for each chamber. This edge directly bears the impact of the incoming airflow, determining the angle of attack and the airflow separation state. It is the primary geometric reference for judging whether the wing has collapsed, veered, or exhibited abnormal attitudes. In a coordinate system with the upper left corner of the image as the origin, the upper edge refers to the top contour boundary formed by the group of consecutive foreground pixels with the smallest ordinate (y-value) in the pre-processed wing mask.

[0044] In a further embodiment, establishing the central axis of the curvature of the parachute edge includes any of the following steps: S1: Calculate the centroid of the parachute edge contour based on the parachute edge contour, calculate the principal direction of the parachute edge contour using the PCA algorithm, and establish the central axis of the parachute edge curvature based on the centroid and principal direction. Specifically, after extracting the foreground pixel set of the parachute mask, first calculate its contour centroid using geometric moments (zero-order moment and first-order moment), which serves as the spatial positioning base point for the central axis. Then, perform principal component analysis (PCA) on the two-dimensional coordinates of all foreground pixels, construct the covariance matrix, and solve for its eigenvalues ​​and eigenvectors. The eigenvector corresponding to the largest eigenvalue is the direction with the largest variance in the parachute pixel distribution, which is also the principal extension direction (major axis direction) of the parachute. Extending this principal direction vector spatially from the centroid uniquely determines the central axis of the parachute. S2: Based on the image including the paraglider, the parachute line convergence point is obtained by key point detection or line segment clustering, and the central axis of the paraglider edge curvature is established according to the parachute line convergence point and the highest point of the contour. In this embodiment, the paracord convergence point is obtained using key point detection or line segment clustering. Specifically, this can be achieved by detecting multiple paracord line segments and then finding the least-squares intersection point of these segments to obtain the paracord convergence point.

[0045] In a further embodiment, the symmetry index includes mirror error or area difference, wherein: The method for calculating the mirror error is as follows:

[0046] In the formula, This indicates the left side of the parachute edge profile after dividing the parachute edge profile left and right according to the central axis. Or the outline of the right wing edge The number of pixels, Indicates the first 1 pixel Indicates mirror transformation; The method for calculating the area difference is as follows:

[0047] In the formula, , These represent the left and right areas respectively after dividing the edge contour of the parachute wing according to the central axis.

[0048] In a further embodiment, determining the attitude of the paraglider based on the wing edge curvature, canopy arc parameters, and symmetry index includes: When the symmetry index is greater than the symmetry threshold, the paraglider's attitude is determined to be a unilateral collapse, specifically:

[0049] In the formula, Including symmetry threshold; In addition, to prevent false positives, enhanced criteria can be set:

[0050] In the formula, , These represent the area threshold and height threshold, respectively. , These represent the average height of the left and right halves of the parachute, respectively.

[0051] When local curvature is abnormal and contour continuity is disrupted, the paraglider's attitude is determined to be leading-edge collapse, wherein: The criteria for determining the local curvature anomaly are as follows:

[0052]

[0053] In the formula, The first-order difference representing the curvature of the parachute edge. Indicates unit arc length, This represents the first curvature change threshold, which physically means that the local curvature changes too rapidly. Alternatively, the criteria for determining the local curvature anomaly are:

[0054]

[0055] In the formula, The second difference representing the curvature of the parachute edge. This represents the threshold for the second curvature change, which is determined using a more sensitive second-order difference. The criteria for determining the disruption of the contour continuity are as follows:

[0056] In the formula, This represents the distance between two adjacent pixels. This represents two adjacent pixels on the curvature of the umbrella wing edge. Indicates the distance threshold; under normal circumstances A pixel value significantly larger than this range indicates a broken edge / gap; Alternatively, the condition for determining the disruption of the contour continuity is:

[0057]

[0058] In the formula, , Represents pixels tangent angle, Indicates the tangent angle threshold. , Representing pixels x and y coordinates , Representing pixels The horizontal and vertical coordinates of the line represent, in physical terms, the sharp broken lines that appear at the boundary.

[0059] Furthermore, to adapt to scale variations, curvature-related quantities are preferentially expressed in normalized radians or relative to the scale. For scale normalization, the distance threshold is expressed as a relative value: Angle threshold Can be set to (Based on data calibration) Obtained through calibration or learning.

[0060] Curvature and its difference are calculated from the point set of the paraglider profile. If the curvature gradient or second-order difference exceeds a preset threshold, or if statistical outliers appear in the curvature distribution, it is determined to be a curvature abrupt change. Simultaneously, if the distance between adjacent profile points exceeds a threshold, or the change in tangent direction exceeds an angle threshold, or the local point density is below a threshold, it is determined to be an edge discontinuity. When both curvature abrupt changes and edge discontinuities are satisfied simultaneously, it is determined to be a leading edge collapse. Curvature abrupt changes are determined using the curvature derivative / outlier criteria, while edge discontinuities are determined using distance / angle / density criteria. The combined criteria provide a quantitative standard for leading edge collapse.

[0061] A second embodiment of the present invention provides a flight control method, such as... Figure 2 As shown, the drone control for drone-to-paraglider operation includes the following steps: Based on images including drones and paragliders, the paraglider attitude is obtained using the paraglider attitude detection method described in the first embodiment of the present invention, and the pilot position is obtained based on target detection. Based on the centroid of the pilot's position and the edge profile of the parachute, calculate the lateral offset of the pilot relative to the center of the parachute; Based on the lateral offset divided by the wing width, the normalized relative position of the pilot with respect to the wing center is obtained; Based on the paraglider attitude and the normalized relative position of the pilot with respect to the center of the wing, calculate the current roll angle, pitch angle and yaw angle of the paraglider; The flight attitude of the UAV is controlled based on the current roll, pitch, and yaw angles of the paraglider and the target roll, pitch, and yaw angles.

[0062] In this embodiment, the center point of the drone pilot in the image is obtained through target detection or human keypoint modeling:

[0063] The centroid of the wing edge profile is Calculate the relative displacement, representing the lateral offset of the pilot relative to the center of the parachute:

[0064] Normalization yields:

[0065] The final normalized relative position of the pilot with respect to the center of the parachute is a dimensionless parameter that can withstand changes in distance.

[0066] In a further embodiment, calculating the current paraglider's roll angle, pitch angle, and yaw angle includes: Roll angle :

[0067] In the formula, This indicates the height of the leftmost point of the upper edge of the parachute wing. This indicates the height of the rightmost point of the upper edge of the parachute wing. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing; Pitch angle :

[0068] In the formula, Indicates the umbrella surface curvature parameter; Yaw angle :

[0069] In the formula, This indicates the normalized relative position of the pilot with respect to the center of the parachute.

[0070] In a further embodiment, controlling the flight attitude of the drone can be: If the parachute veers to the left, control the drone to correct it to the right; if the angle of attack is too large, reduce the traction; if it collapses, quickly reduce the load / release it.

[0071] A third embodiment of the present invention provides a flight control system, such as Figure 3 As shown, it includes: A visual acquisition module, used to acquire images; A target recognition module, which is used to identify paragliders and pilot positions based on the image; Attitude estimation module, wherein the attitude estimation module obtains the paraglider attitude based on the paraglider attitude detection method described in the first embodiment of the present invention; The control decision module calculates the lateral offset of the pilot relative to the center of the wing based on the centroid of the pilot's position and the edge contour of the wing; it obtains the normalized relative position of the pilot relative to the center of the wing by dividing the lateral offset by the wing width; it calculates the roll angle, pitch angle, and yaw angle of the current paraglider based on the paraglider attitude and the normalized relative position of the pilot relative to the center of the wing; and it outputs the UAV attitude correction amount based on the roll angle, pitch angle, and yaw angle of the current paraglider and the target roll angle, pitch angle, and yaw angle. The flight control execution module controls the flight attitude of the UAV based on the UAV attitude correction amount.

[0072] In an optional embodiment, the visual recognition module of this invention detects and segments paraglider targets based on a deep learning model. Further improvements are made to address the structural characteristics of the paraglider: First, a multi-task learning network is constructed, enabling the model to simultaneously output the wing region, key points of the parachute lines, and the pilot's position, improving the consistency of feature extraction. Second, a structural constraint loss function, including symmetry and shape constraints, is introduced during model training to enhance the model's ability to represent the paraglider's curved structure. Further, the deep learning output is input into a geometric feature calculation module, where parameters such as canopy curvature, left-right symmetry, and the pilot's relative position are obtained through curve fitting and spatial relationship analysis. Additionally, temporal information is introduced to smooth and predict the recognition results, improving stability and robustness in dynamic flight environments.

Claims

1. A paraglider attitude detection method for detecting the attitude of a paraglider towed by a drone, characterized in that, Includes the following steps: Acquire an image including a paraglider, and obtain a wing mask for the paraglider based on the image; Based on contour tracking, the edge contour of the parachute is obtained from the parachute mask; The curvature of the parachute edge is calculated based on the parachute edge profile. Based on the outline of the parachute wing edge, curve fitting is performed on the upper edge of the parachute wing to obtain the parachute surface curvature parameters; Based on the outline of the parachute edge, the central axis of the curvature of the parachute edge is established, and the outline of the parachute edge is divided into left and right sections according to the central axis to obtain the symmetry index. The attitude of the paraglider is determined based on the curvature of the wing edge, the arc parameters of the canopy surface, and the symmetry index. The upper edges of the parachute wing edge profile are sorted and parameterized according to their arc lengths; Calculate the wing edge curvature based on the upper edge of the sorted and parameterized wing edge profile: In the formula, Indicates along the arc length The curvature of the parachute edge, These represent the pixel coordinates of the upper edge of the sorted and parameterized parachute edge contours, respectively. The process of curve fitting to the upper edge of the parachute wing to obtain the parachute surface curvature parameters includes: The upper edge of the parachute wing is fitted using a quadratic polynomial, a cubic polynomial, or a spline to obtain the fitted curve function; The umbrella surface curvature parameters are obtained based on the fitted curve function: In the formula, This indicates the curvature parameter of the umbrella surface. Indicates the height of the bow. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing. This represents the ordinate of the highest point of the contour obtained based on the fitted curve function. This indicates the height difference between the left and right endpoints of the upper edge of the parachute wing; The symmetry index includes mirror error or area difference, wherein: The method for calculating the mirror error is as follows: In the formula, This indicates the left side of the parachute edge profile after dividing the parachute edge profile left and right according to the central axis. Or the outline of the right wing edge The number of pixels, Indicates the first 1 pixel Indicates mirror transformation; The method for calculating the area difference is as follows: In the formula, , These represent the left and right areas respectively after dividing the edge contour of the parachute wing according to the central axis; The determination of the paraglider's attitude based on the wing edge curvature, canopy arc parameters, and symmetry index includes: When the symmetry index is greater than the symmetry threshold, the paraglider is determined to be in a unilateral collapse posture. When local curvature is abnormal and contour continuity is disrupted, the paraglider's attitude is determined to be leading-edge collapse, wherein: The criteria for determining the local curvature anomaly are as follows: In the formula, The first-order difference representing the curvature of the parachute edge. Indicates unit arc length, This represents the first curvature change threshold; Alternatively, the criteria for determining the local curvature anomaly are: In the formula, The second difference representing the curvature of the parachute edge. This represents the threshold for the second curvature change; The criteria for determining the disruption of the contour continuity are as follows: In the formula, This represents the distance between two adjacent pixels. This represents two adjacent pixels on the curvature of the umbrella wing edge. Indicates the distance threshold; Alternatively, the condition for determining the disruption of the contour continuity is: In the formula, , Represents pixels tangent angle, Indicates the tangent angle threshold. , Representing pixels x and y coordinates , Representing pixels The x and y coordinates.

2. The paraglider attitude detection method according to claim 1, characterized in that, The parachute mask is a binary image, where pixels with a value of 1 belong to the parachute wing, and pixels with a value of 0 belong to the background. The step of obtaining the parachute wing edge contour from the parachute mask based on contour tracking includes: The parachute mask is subjected to noise reduction and hole filling operations to obtain a pre-processed parachute mask; The preprocessed parachute mask is scanned from the top left corner to the bottom right corner to find the first pixel that meets the preset conditions as the boundary starting point. The preset conditions include that the pixel value is 1 and there are pixels with a pixel value of 0 in the neighborhood of the pixel. Using the boundary starting point as the current point, execute the following loop steps: Starting from the current point, the neighborhood of the current point is searched in a fixed order to find the first pixel that meets the preset conditions, and a new pixel is obtained. The new pixel is then used as the current point to perform the loop step until the current point coincides with the boundary starting point, thus obtaining the set of parachute edge contour points, which serves as the parachute edge contour.

3. The paraglider attitude detection method according to claim 2, characterized in that, Establishing the central axis of the curvature of the parachute edge includes any of the following steps: S1: Calculate the centroid of the parachute edge profile based on the parachute edge profile, calculate the principal direction of the parachute edge profile using the PCA algorithm, and establish the central axis of the parachute edge curvature based on the centroid and principal direction of the parachute edge profile. S2: Based on the image including the paraglider, the parachute line convergence point is obtained by key point detection or line segment clustering, and the central axis of the paraglider edge curvature is established according to the parachute line convergence point and the highest point of the contour.

4. A flight control method for controlling a drone towing a paraglider, characterized in that, Includes the following steps: Based on images including drones and paragliders, the paraglider attitude is obtained based on the paraglider attitude detection method according to any one of claims 1 to 3, and the pilot position is obtained based on target detection; Based on the centroid of the pilot's position and the edge profile of the parachute, calculate the lateral offset of the pilot relative to the center of the parachute; Based on the lateral offset divided by the wing width, the normalized relative position of the pilot with respect to the wing center is obtained; Based on the paraglider attitude and the normalized relative position of the pilot with respect to the center of the wing, calculate the current roll angle, pitch angle and yaw angle of the paraglider; The flight attitude of the UAV is controlled based on the current roll, pitch, and yaw angles of the paraglider and the target roll, pitch, and yaw angles.

5. The flight control method according to claim 4, characterized in that, Calculate the current roll, pitch, and yaw angles of the paraglider, including: Roll angle : In the formula, This indicates the height of the leftmost point of the upper edge of the parachute wing. This indicates the height of the rightmost point of the upper edge of the parachute wing. This indicates the distance between the left and right endpoints of the upper edge of the parachute wing; Pitch angle : In the formula, Indicates the umbrella surface curvature parameter; Yaw angle : In the formula, This indicates the normalized relative position of the pilot with respect to the center of the parachute.

6. A flight control system, characterized in that, include: A visual acquisition module, used to acquire images; A target recognition module, which is used to identify paragliders and pilot positions based on the image; An attitude estimation module, wherein the attitude estimation module obtains the paraglider attitude based on the paraglider attitude detection method according to any one of claims 1 to 3; The control decision module calculates the lateral offset of the pilot relative to the center of the wing based on the centroid of the pilot's position and the edge contour of the wing; it obtains the normalized relative position of the pilot relative to the center of the wing by dividing the lateral offset by the wing width; it calculates the roll angle, pitch angle, and yaw angle of the current paraglider based on the paraglider attitude and the normalized relative position of the pilot relative to the center of the wing; and it outputs the UAV attitude correction amount based on the roll angle, pitch angle, and yaw angle of the current paraglider and the target roll angle, pitch angle, and yaw angle. The flight control execution module controls the flight attitude of the UAV based on the UAV attitude correction amount.

Images