A method for measuring the posture of rotating aircraft based on contour features

Abstract

The present invention relates to a method for measuring the posture of a rotating aircraft based on contour features. The method predicts posture parameters based on prior information about the target's motion, thereby narrowing the target range and reducing the amount of data to be processed. Based on the predicted posture parameter values, the minimum envelope rectangle of the contour edge points is extracted, and the predicted posture parameter values ​​are corrected. A real-time method for generating a target contour template is designed based on the imaging characteristics of the rotating object and the camera imaging model. A similarity measurement function is constructed based on the difference between the eigenvalues ​​of the images inside and outside the template, as well as the difference between the template eigenvalues ​​and the observed values, and a method for optimizing the target posture parameters is established, thereby achieving high-precision measurement of the posture parameters of the rotating object under low signal-to-noise ratio conditions.

Description

Technical Field

[0001] The present invention relates to the field of computer image processing, and in particular to a method for measuring the position and posture of a rotating aircraft based on contour features. Background Art

[0002] The position and attitude of spacecraft, such as rocket boosters, docking rings, and conical decoys, are crucial kinematic parameters. Measuring these parameters is crucial for monitoring their condition and improving their design. While these spacecraft vary greatly in appearance, they all share the structural characteristics of a body of revolution. Therefore, by measuring the position and attitude of a body of revolution attached to a rigid target, the kinematic parameters of these targets can be obtained.

[0003] Visual pose measurement technology has high measurement accuracy and can adapt to measurement scenarios where the target moves at a fast speed, has a large range of motion, and has a complex and changeable measurement environment. Therefore, it is widely used in the fields of aircraft pose parameter measurement and accident cause determination. According to the number of sensors, visual measurement can be divided into monocular measurement, binocular measurement, and multi-eye measurement. Compared with the other two measurement methods, the monocular measurement system has the advantages of simple structure, simple camera calibration method, and large measurement field of view, and is more widely used. According to different measurement principles, aircraft pose parameter measurement methods can be divided into two categories: pose solution methods based on geometric features and pose estimation methods based on models:

[0004] (1) The pose calculation algorithm based on geometric features refers to establishing a functional relationship between image features and pose parameters (including displacement vectors and rotation matrices) based on the target's two-dimensional image features, three-dimensional geometric features, and camera model, and calculating the target pose parameters. This method is suitable for measurement scenarios (close-range measurement) where the imaging quality is high and the image features used for pose calculation are stable within the field of view. Depending on the different image features used, pose parameter calculation algorithms are divided into pose calculation algorithms based on point features, pose calculation algorithms based on line features, and pose calculation algorithms based on circle features.

[0005] (2) Model-based pose estimation methods assume that the target's three-dimensional model is known and estimate the pose parameters based on the correspondence between the target's three-dimensional model and the two-dimensional image features. Compared with pose solution algorithms based on geometric features, model-based methods utilize the target's prior shape knowledge and are less affected by occlusion and complex backgrounds. They have higher measurement accuracy and stability and have been widely used in spatial vision tasks to measure the pose of cooperative targets and non-cooperative known targets. According to the type of features used, model-based pose estimation methods can be divided into contour feature-based, feature point-based, and multi-feature fusion methods.

[0006] When measuring the posture parameters of an aircraft over long distances, the target image size is small and the quality is poor, resulting in the following problems with existing measurement methods:

[0007] (1) The pose solution algorithm based on geometric features cannot extract image features stably and reliably, and the algorithm has poor stability and large measurement errors. (2) The model-based method estimates the pose parameters according to the correspondence between the target three-dimensional model and the target image features, which has higher accuracy and reliability. However, this type of method has high requirements on the accuracy of the initial values ​​of the pose parameters. If the initial value accuracy is too low, the algorithm will fall into a local optimal solution or filter divergence. Summary of the Invention

[0008] In order to solve the defects of poor measurement stability and low precision in the prior art, the purpose of the present invention is to provide a method for measuring the position and posture of a rotating aircraft based on contour features.

[0009] In order to achieve the above object, the technical solution provided by the present invention is:

[0010] A method for measuring the position and posture of a rotating aircraft based on contour features comprises the following steps:

[0011] Step 1, image preprocessing: input the image to be processed and perform image preprocessing, mainly including distortion correction and Gaussian filtering; Step 2, pose prediction: predict pose parameters based on prior information of target motion and acquired measurement data;

[0012] Step 3, pose correction: correct the pose parameter prediction value according to the observation information (contour features);

[0013] Step 4, pose optimization: Generate a contour template in real time according to the structural characteristics of the rotating body, and build a similarity measurement function based on this, establish a target pose parameter optimization method, and achieve high-precision measurement of pose parameters.

[0014] Due to the adoption of the above-mentioned technical solution, the beneficial effects of the present invention are: through the contour feature-based rotating aircraft posture measurement method of the present invention, the amount of data to be processed is effectively improved and reduced, and compared with the existing method, it has higher posture measurement accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a method for measuring the position and posture of a rotating aircraft based on contour features of the present invention.

[0016] Figure 2 It is a schematic diagram of the cylindrical imaging model.

[0017] Figure 3 It is a schematic diagram of the frustum imaging model.

[0018] Figure 4It is a schematic diagram of pose prior distribution and corrected distribution.

[0019] Figure 5 It is a schematic diagram of the pose sampling range.

[0020] Figure 6 It is a schematic diagram of sample similarity distribution.

[0021] Figure 7 It is a schematic diagram of the posterior distribution of pose parameters. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] A method for measuring the position and posture of a rotating aircraft based on contour features comprises the following steps:

[0024] Step 1, image preprocessing: input the image to be processed and perform image preprocessing, mainly including distortion correction and Gaussian filtering; Step 2, pose prediction: predict pose parameters based on prior information of target motion and acquired measurement data;

[0025] Step 3, pose correction: correct the pose parameter prediction value according to the observation information (contour features);

[0026] Step 4, pose optimization: Generate a contour template in real time according to the structural characteristics of the rotating body, and build a similarity measurement function based on this, establish a target pose parameter optimization method, and achieve high-precision measurement of pose parameters.

[0027] The following describes in detail a method for measuring the position and posture of a rotating aircraft based on contour features of the present invention in conjunction with the accompanying drawings and a typical specific embodiment.

[0028] The specific implementation of step 1 is as follows:

[0029] First, the captured image is distortion-corrected based on the camera's intrinsic parameter matrix and distortion coefficients. Then, a Gaussian filter algorithm is used to smooth the image to reduce the impact of noise on measurement accuracy. Both the distortion correction and Gaussian filter algorithms use existing algorithms.

[0030] The specific implementation of step 2 is as follows:

[0031] Taking the measurement scenario of a typical rotating aircraft - booster separation parameter measurement as an example, we first use the prior information of the target motion to predict the target's pose parameters, narrow the target's imaging area, reduce the computational complexity of subsequent algorithms, and improve the measurement accuracy and stability. The booster pose parameter P is defined as:

[0032] (1) Position parameters: expressed as the centroid coordinates of the booster (X0, Y0, Z0);

[0033] (2) Attitude parameters: The unit normal vector V of the booster axis is V = (V x ,V y ,V z ), can also be equivalently expressed as the rotation angle (φ) of the booster axis normal vector (the initial normal vector is defined as [0,0,1]) around the X, Y, and Z axes x ,φ y ,φ z ). The conversion relationship between the axis normal vector and the rotation angle can be expressed as:

[0034]

[0035] Step 2.1: Pose Prediction

[0036] Predicted values ​​of target pose parameters It can be expressed as:

[0037]

[0038] In formula (2), represents the predicted value of the target pose parameter at the moment corresponding to the n+1 frame image, P n ,ApInf n where represents the measured and motion parameter values ​​of the target pose at the time corresponding to the nth frame, respectively, and Δt represents the time difference between two adjacent frames. Compared to velocity parameters, motion parameters such as acceleration are less reliable, and the amplitude of the pose change caused by acceleration during Δt is relatively small. Therefore, for the motion parameter prior information, only the velocity parameter is retained, and ApInf is the set of expected values ​​of the booster velocity and angular velocity:

[0039] ApInf0={v x ,v y ,v z ,ω x ,ω y ,ω z} (3)

[0040] In formula (3), v x ,v y ,v z ,ω x ,ω y ,ω z They represent the speed of the target in the X, Y, and Z directions and the angular velocity of the target normal vector around the X, Y, and Z axes respectively.

[0041] Generally speaking, the prior information of the booster motion parameters ApInf approximately follows a Gaussian distribution, with the velocity in the X direction For example, v xThe probability density function f(v x ) can be expressed as:

[0042]

[0043] In formula (4), is the standard deviation, is the expected value.

[0044] Step 2.2: Motion parameter update

[0045] As measurement data accumulates, the algorithm updates the prior information about the motion (ApInf and f(·)). The update criteria take into account both the prior information and the measurement information to avoid prediction errors caused by abnormal target motion or large measurement errors. The update criteria for the expected value of the motion parameters are as follows:

[0046]

[0047] In formula (5), Δ n represents the difference in pose parameters between two adjacent frames (frame n and frame n-1). The expected value of the probability density function f(·) is replaced by the updated motion parameter value, and the variance remains unchanged.

[0048] The specific implementation of step 3 is as follows:

[0049] The algorithm uses the eigenvalues ​​of the contour to correct the predicted values ​​of the pose parameters. The pose parameter correction mainly includes two parts: target positioning and parameter correction: first, the minimum envelope rectangle (prediction rectangle) of the target imaging contour is calculated based on the predicted values ​​of the pose parameters, and the target area R is determined based on this. T ; By extracting region R T The minimum envelope rectangle (observation rectangle) of the target is used to correct the predicted value of the pose parameters.

[0050] Step 3.1: Target positioning

[0051] Booster profile feature F c is the combination of imaging features of each component, which can be expressed as:

[0052] F c ={F main ,F top ,F nozzle} (6)

[0053] In formula (6), F main ,F top ,F nozzle They represent the imaging features of the booster body, head, and nozzle respectively.

[0054] The length of the main part of a typical booster accounts for more than 80%, which is the main factor determining the minimum envelope rectangle of the contour feature. Therefore, in the target positioning stage, in order to simplify the calculation, the booster is simplified to a cylindrical T cy : The height is equivalent to the booster axis length L, and the radius R is the same as the main part. According to the size of the equivalent target and the camera imaging model, the minimum envelope rectangle EnRec of the imaging feature can be expressed as:

[0055] EnRec=F(T cy ) (7)

[0056] Function F represents the imaging model of the camera, and the cylinder T cy The imaging characteristics of are approximately represented by a rectangle with a width of w and a length of l. The characteristic values ​​of the minimum envelope rectangle mainly include size, center coordinates, slope, etc. The specific calculation method is as follows:

[0057] When measuring at long distances, the value of w is less affected by the target posture and is mainly determined by the Z value, which can be approximately expressed as:

[0058]

[0059] In formula (8), f represents the focal length of the camera.

[0060] The value of l is determined by the position parameter and the posture parameter, which can be approximately expressed as:

[0061]

[0062] The center coordinates of the rectangle (x r ,y r ) is expressed as:

[0063]

[0064] The slope k of the rectangle r Expressed as:

[0065]

[0066] In formula (11), θ r Represents the angle between the major axis of the rectangle and the x-axis of the image coordinate system.

[0067] After extracting the minimum envelope rectangle, it is adaptively expanded / optimized to increase the stability of target positioning, and the expanded rectangle is defined as the target area R T Rectangle expansion / optimization method: the center coordinates and slope of the rectangle remain unchanged, and the length and width are multiplied by the magnification factor s respectively. w ,s l Taking into account the increased amount of data after expansion and the stability of target positioning, s w,s l The general value is 4, 2;

[0068] Step 3.2: Minimum envelope rectangle extraction

[0069] The algorithm sets an adaptive gradient threshold based on the gradient difference between the target contour edge point and other edge points to extract the target area R T Edges with large inner gradient values ​​and spatial continuity.

[0070] Typically, when noise interference is minimal, the minimum enveloping rectangle can well reflect the characteristic values ​​of the target's contour. However, when noise interference is present, the minimum enveloping rectangle cannot accurately represent the size and other parameters of the contour features in at least one dimension (especially the dimension perpendicular to the target's long axis). Therefore, when the pose parameters corresponding to the rectangle differ significantly from the predicted pose parameters, the extracted rectangle needs to be optimized to reduce noise interference and ensure measurement accuracy.

[0071] Since the number of noise points is small and has a certain continuity with the target contour, the main direction of the edge point (the slope of the rectangle's major axis) has little effect; the uncertainty of the spatial distribution of noise points increases the uncertainty of the target contour envelope rectangle, which greatly affects the rectangle parameters. Based on the distribution characteristics of noise, the rate of change of the number of edge points R is used to calculate the value of the edge point. w ,R l Optimize the envelope rectangle for evaluation index, R w ,R l The definition is as follows:

[0072]

[0073] In formula (12), N r is the number of edge points inside the rectangle, ΔN r Indicates the change in the number of edge points when the length or width of the rectangle changes. w ,R l The smaller the value of ΔN r The greater the probability that the corresponding edge point is a noise point.

[0074] Minimum envelope rectangle correction: Adjust the size of the rectangle EnRec along the axis direction of the major / minor axis respectively. During the correction process, calculate R according to formula (12) w ,R l Value: If R w >T w or R l >T l (T w ,T lIf is an adaptive threshold, positively correlated with the width and length of the rectangle, the rectangle size continues to decrease in that direction; otherwise, the process stops until all four sides of the rectangle are corrected. The corrected minimum envelope rectangle can well reflect the characteristic values ​​of the target imaging characteristics (parameters such as center, size, and slope).

[0075] Step 3.3: Parameter correction After the rectangle correction is completed, the pose parameter prediction value is corrected according to the characteristic value of the rectangle (center coordinates, size, slope, etc.). The specific correction method is as follows:

[0076] (1) Position parameter correction

[0077] From equations (8) to (11), we can see that the size w,l and center coordinates (x r ,y r ) and the slope k r Both can reflect the position parameters of the target. Among them, the Z value is obtained according to formula (8) and the w value:

[0078]

[0079] According to formula (10), the ratio of position parameters can be obtained:

[0080]

[0081] From formula (9) and (10), we can see that l and k r It is a combined function of position parameters and attitude parameters, which can be used for joint correction of posture parameters, but it cannot correct position parameters alone. Therefore, the algorithm corrects the position parameters according to equations (13) and (14). First, according to the estimated value of Z and Z w And its corresponding prior probability obtains the corrected value Z0:

[0082]

[0083] In formula (15), for and Z w The prior probability of (the probability distribution function is updated in real time according to the measurement value and motion parameter value of the previous frame).

[0084] According to equations (14) and (15), the corrected values ​​of X and Y are obtained:

[0085]

[0086] (2) Posture parameter correction

[0087] From equations (9) and (11), we can see that the length l and slope k of rectangle EnRec are rThe posture parameters that can reflect the target are the basic constraints for posture parameter correction. The posture parameter correction problem is expressed as an optimization problem: r Under the constraint of , the angle between the target normal vectors before and after correction is minimized. Denote the target normal vectors before and after correction respectively, then the optimization problem of posture parameter correction is recorded as:

[0088]

[0089]

[0090] In fact, after the booster is separated, it usually rotates around a fixed axis, that is, it has only one attitude angle (in φ x The other two attitude angles are basically unchanged or have small changes. The attitude parameters can be expressed as:

[0091]

[0092] In formula (18), ω is the angular velocity, t is the time difference between two adjacent frames, and Δφ x ,Δφ y ,Δφ z is a random variable with a small amplitude. According to the actual motion law of the booster, the correction problem of the attitude parameters can be simplified to the correction problem of a single attitude angle. According to equations (1), (9), (11) and (18), we can get φ x Two sets of solutions:

[0093]

[0094] Combined with the prior information of the booster motion parameters, the corrected attitude parameter φ x Expressed as:

[0095]

[0096] In formula (20), They are The prior probability of .

[0097] The specific implementation of step 4 is as follows:

[0098] According to the imaging characteristics of the target and the distribution law of pose measurement errors, the algorithm designs a real-time generation method of contour templates, constructs a similarity measurement function, combines the pose parameter correction values ​​and prior information to determine the posterior probability distribution of the pose parameters, optimizes the correction values, and achieves high-precision estimation of the pose parameters.

[0099] Step 4.1: Contour feature generation

[0100] The booster is primarily composed of three basic geometric shapes: cylinders, frustums, and Laval nozzles. Because the nozzle imaging features represent a relatively small portion of the overall contour and have poor imaging quality, they contain insufficient pose parameter information and are severely disturbed. Therefore, the nozzle imaging features are not considered in the similarity metric. The key to generating the target contour template lies in generating the imaging features of the cylinder and frustum, and combining these features at different poses.

[0101] Cylindrical imaging feature analysis

[0102] The pose parameters of the main part of the booster (cylinder) are defined as:

[0103] (1) Position parameter P1: expressed as the coordinates (X1, Y1, Z1) of the cylinder's centroid O1;

[0104] (2) Posture parameter V: V = (V x ,V y ,V z ), which is the same as the booster attitude parameter.

[0105] Cylindrical imaging model such as Figure 2 As shown in the figure, under different posture parameters, the types of cylindrical imaging features are different. The influence of posture parameters on imaging features is reflected in the camera optical center O C Distance to line L:

[0106] (1) If the camera optical center O C If the distance to line L is less than R, then there is no line passing through point O. C A plane tangent to the cylinder;

[0107] (2) If the camera optical center O C The distance to the line L is equal to R, then there is a point O C A plane tangent to the cylinder;

[0108] (3) If the camera optical center O C If the distance to the line L is greater than R, then there are two points passing through O. C A plane tangent to the cylinder.

[0109] In this case, the straight line L passes through the center of the cylinder, and the vector of the straight line is parallel to V.

[0110] Pass O C The number of planes tangent to the cylinder determines the type of imaged feature:

[0111] (1) When the number of sections is ≤1, the cylinder is imaged as a single ellipse (imaged from the bottom surface closer to the optical center);

[0112] (2) When the number of sections is 2, the characteristic contour of the cylindrical imaging consists of three parts: partial arc segments of two ellipses and a line segment (generator imaging, tangent to the ellipse).

[0113] A. Camera optical center O C Distance to line L ≤ R

[0114] When the camera optical center O C When the distance to the straight line L is ≤ R, the image of the cylinder is an ellipse. In this case, first calculate the center of the two bases of the cylinder and the optical center O C The distance between C The base circle C corresponding to the nearest center point is the original image of the imaging ellipse.

[0115] The edge point coordinates p of the bottom circle C i (X i ,Y i ,Z i ) can be obtained from formula (21):

[0116]

[0117] In formula (21), (X0, Y0, Z0) is the center coordinate of the bottom circle C (which can be converted from the cylinder posture parameters and cylinder height parameters), u(u x ,u y ,u z ),n(n x ,n y ,n z ) are two mutually perpendicular unit vectors on plane π, t represents O o P i The rotation angle relative to u.

[0118] According to formula (21), 5 points are evenly selected, and the coordinates of the imaging points are calculated according to the camera imaging model, as shown in formula (22):

[0119]

[0120] Solving the parametric equation of the imaging ellipse based on these five coordinate points is the target contour.

[0121] B. Camera optical center O C Distance to line L > R

[0122] When the camera optical center O C When the distance to the line L is greater than R, the imaging characteristics of the cylinder are a combination of a straight line and an elliptical curve. In this case, the key to obtaining the imaging characteristics is to obtain the tangent lines of the two imaging ellipses. The method for finding the tangent lines of the ellipses is as follows: Let the equation of the tangent plane passing through the optical center be AX + BY + CZ = 0. This tangent plane is tangent to the cylinder and must satisfy the following equation system:

[0123]

[0124] Find two sets of plane parameters (A i ,B i ,C i ), then calculate the point of tangency between the tangent plane and the cylinder base. The tangent point is the projection point of the center of the base circle onto the tangent plane, which can be obtained using the point-to-plane projection formula. Based on formula (22), the projection point of the tangent point is obtained, and connecting the projection points is the ellipse tangent.

[0125] Analysis of frustum imaging characteristics

[0126] The pose parameters of the booster head (cone) are expressed as:

[0127] (1) Position parameter P2: expressed as the coordinates (X2, Y2, Z2) of the center O2 of the bottom of the cone;

[0128] (2) Posture parameter V: the same as the normal vector of the cylinder axis.

[0129] The frustum imaging model is as follows Figure 3 As shown in Figure 2, under different posture parameters, the types of frustum imaging features are different. Let the vertex of the cone passing through the two bottom surfaces of the frustum be O p The influence of the posture parameters on the imaging characteristics is reflected in the straight line O C O p Angle θ with V i superior:

[0130] (1) If θ i <θ0, then there is no passing point O C A plane tangent to the frustum;

[0131] (2) If θ i =θ0, then there exists a point O C A plane tangent to the frustum;

[0132] (3) If θ i >θ0, then there are two passing points O C A plane tangent to the frustum.

[0133] Where θ0 represents the angle between the cone axis and the generatrix,

[0134] Same as the cylinder, passing through point O C The number of planes tangent to the frustum determines the type of imaging feature:

[0135] (1) When the number of sections is ≤1, the frustum is imaged as a single ellipse (imaged from the bottom surface with a large area);

[0136] (2) When the number of sections is 2, the characteristic contour of the frustum imaging consists of three parts: two ellipses and their tangents.

[0137] When θ i When θ ≤ θ0, the method for solving the imaging characteristics is the same as that for the cylinder. i When θ > θ0, the key to obtaining imaging features remains to determine the tangents of the two imaging ellipses. Similar to the method for finding tangents to a cylinder, first find the tangent plane of the frustum passing through the optical center, then find the point of tangency between the tangent plane and the frustum base. Assume that the equation of the tangent plane passing through the optical center is AX + BY + CZ = 0. Its tangency to the frustum must satisfy the following set of equations:

[0138]

[0139] Find two sets of plane parameters (A i ,B i ,C i ), then find the point of tangency between the bottom surface of the truncated cone and the plane, and the tangent line of the ellipse can be obtained based on the projection point of the point of tangency.

[0140] Combination feature analysis

[0141] When the booster body and head are combined, the combination of their imaging features needs to be considered. Different posture parameters correspond to different types of imaging features. The influence of posture parameters on imaging features is reflected in the line O C O p Angle θ with V i and O C The distance d to the straight line of the booster axis i On the basis of θ i and d i The types of booster imaging features can be divided into three categories:

[0142] (1) When d i When ≤R, the booster imaging feature is an ellipse;

[0143] (2) When d i >R and θ i When ≤θ0, the imaging features of the booster head are completely surrounded by the imaging features of the main body, and the combined features of the two imaging features are the same as the imaging features of the main body;

[0144] (3) When d i >R and θ i When θ > 0, the combined features of the body and head include three ellipses (partial elliptic curves) and their corresponding tangents.

[0145] The booster has a relatively large aspect ratio. Typically, the straight line features (generative imaging features) of the booster's contour are relatively complete and account for a large proportion of the overall contour. However, the curved features (base circle imaging features) account for a smaller proportion of the overall contour and are subject to noise and severe distortion. Furthermore, when generating imaging features based on target pose parameters, straight line features are less complex than curved features. Therefore, except in cases where the booster image approximates an ellipse, the algorithm retains only straight line features as components of the contour template.

[0146] Step 4.2: Similarity measurement function construction

[0147] The purpose of similarity measurement is to quantify the difference between the pose parameters corresponding to the template and the actual pose parameters by calculating the difference between the image eigenvalues ​​inside and outside the contour template and the difference between the contour eigenvalues ​​and the observed eigenvalues ​​(minimum envelope rectangle).

[0148] The booster surface has no obvious color, texture, or marking features, but there is a clear grayscale difference between the target and the background at the edge of the outline. Based on the imaging characteristics of the target, the algorithm uses the grayscale difference on both sides of the outline template as the basic indicator for measuring the similarity between the target and the template. The similarity measurement function is defined as follows:

[0149]

[0150] In formula (25), n is the number of pixels, p i,out ,p i,in is the template contour point p i The adjacent pixels are located on both sides of the target contour, and p i,out ,p i,in Located in p i On the perpendicular line of the tangent.

[0151] Although the similarity metric function shown in formula (25) can significantly distinguish between “small error” poses and “large error” poses, due to the lack of effective use of observation values, the distinction between “large error” poses is poor (F b,s Therefore, it is necessary to optimize formula (25) to enhance the ability of the similarity measurement function to distinguish different error samples.

[0152] As shown in equations (8) to (11), the eigenvalues ​​of the minimum envelope rectangle EnRec contain the target's pose information. Therefore, in order to better quantify the degree of deviation between the pose parameters corresponding to the template and the actual pose parameters, the algorithm introduces the eigenvalues ​​of EnRec into the similarity measurement function. The optimized similarity measurement function is expressed as:

[0153]

[0154] In formula (26), λ1, λ2 are correction coefficients, SEnRec Represents the area of ​​the extracted minimum envelope rectangle, S EnRec_tem Represents the area of ​​the minimum envelope rectangle corresponding to the template, S EnRec_new Represents the area of ​​the minimum envelope rectangle corresponding to the contour edge points in the template and the image.

[0155] The optimized similarity measurement function makes full use of the imaging features of the target and enhances the quantification ability of the pose error corresponding to the template, where:

[0156] (1) Correction coefficient λ1: The value of λ1 is related to |S EnRec -S EnRec_tem The value of | is negatively correlated, that is, the greater the size difference between the template and EnRec, the smaller the λ1 value, which is the penalty term for the template size difference;

[0157] (2) Correction coefficient λ2: The value of λ2 is less than k tem ,k r The value of > is negatively correlated, that is, the greater the angle difference between the template and EnRec, the smaller the λ2 value, which is the penalty term for the template angle difference.

[0158] Step 4.3: Parameter Optimization

[0159] The algorithm combines the pose parameter correction value, motion prior information and similarity measurement function to determine the posterior probability distribution of the pose parameters, optimizes the pose correction value and achieves accurate estimation of the pose parameters.

[0160] Assume that the pose measurement value of the nth frame image is P n , the motion parameter is ApInf n (The corresponding noise is v n , approximately obeys Gaussian distribution), then the posture distribution range determined based on prior information It can be expressed as:

[0161]

[0162] Assume that the pose parameter correction value of the n+1th frame image is because Correction based on observational information Therefore, you can use right Perform preliminary optimization to obtain the optimized distribution range like Figure 4 As shown:

[0163]

[0164] In order to ensure the robustness of the algorithm and ensure that the true value of the pose is included in the distribution range, this section combines the pose distribution before and after optimization for optimization. The distribution range after optimization Rn→n+1 Expressed as:

[0165]

[0166] In formula (29), |v n | for v n The maximum amplitude, generally speaking, v n Obeying Gaussian distribution, then |v n | Taking 3σ (σ is the Gaussian standard deviation) can meet the measurement requirements.

[0167] Determine the matching range R n→n+1 back:

[0168] (1) Sampling the pose parameters: Figure 5 As shown, in R n→n+1 Randomly collect N0 pose samples in the image, and then measure the similarity of the samples according to formula (26), assign weights to the samples, and obtain a set S with N0 elements (samples), as shown in Figure 6 As shown:

[0169] S={(P i ,ω i ),i=1~N0} (30)

[0170] In formula (30), P i is the pose parameter of the i-th sample, ω i is the weight of the i-th sample, the weight is non-negative, and the sum of all weights is 1. i According to the normalization of sample similarity, it can be expressed as:

[0171]

[0172] In formula (31), Indicates the similarity between the i-th sample and the image data.

[0173] (2) Resampling the pose parameters: Equation (30) reflects the distribution of sample similarity, which can approximate the posterior probability distribution of the pose parameters. Weight the elements in the sample set S to obtain the expected value of the pose parameter under this distribution.

[0174]

[0175] The sample variance is expressed as:

[0176]

[0177] In formula (33), express The jth parameter of Indicates P i The jth parameter of j Represents the standard deviation of the jth parameter. And σ value, establish the posterior distribution probability density function of the pose parameters, such as Figure 7 As shown:

[0178]

[0179] According to formula (34), The pose parameters within the range are resampled, and the number of sampling samples is N0. The samples are similarity measured and weighted, and the weighted pose parameter P is output:

[0180]

[0181] In formula (35), P i,2 ,ω i,2 They represent the pose parameters and weights corresponding to the sub-sampling samples respectively.

[0182] Compared with formula (29), formula (34) combines the similarity measure value, which can better reflect the posterior probability distribution of the pose parameters. On this basis, resampling, assignment, and weighting can further improve the accuracy of pose estimation.

[0183] Due to the adoption of the above-mentioned technical scheme, the beneficial effects of the present invention are as follows: the posture measurement method of a rotating aircraft based on contour features predicts the posture parameters according to the prior information of the target's motion, narrows the target range, and reduces the amount of data to be processed; based on the predicted values ​​of the posture parameters, the minimum envelope rectangle of the contour edge points is extracted, and the predicted values ​​of the posture parameters are corrected; according to the imaging characteristics of the rotating body and the camera imaging model, a real-time generation method of the target contour template is designed; according to the difference in the eigenvalues ​​of the images inside and outside the template and the difference between the eigenvalues ​​of the template and the observed values, a similarity measurement function is constructed, and a target posture parameter optimization method is established, thereby realizing high-precision measurement of the posture parameters of the rotating body under low signal-to-noise ratio conditions.

[0184] It should be recognized that the above description is only a specific embodiment of the present invention, and the present invention is not limited to the specific structure shown or described above. The claims will cover all variations within the spirit and scope of the present invention.

Claims

1. A method for measuring the position and posture of a rotating aircraft based on contour features, characterized in that: The steps include: Step 1, image preprocessing: input the image to be processed and perform image preprocessing, including distortion correction and Gaussian filtering; Step 2, pose prediction: predict pose parameters based on the prior information of target motion and the acquired measurement data; Step 3, pose correction: according to the observation information, i.e., contour features, the pose parameter prediction value is corrected; Step 4, pose optimization: Generate a contour template in real time based on the structural characteristics of the rotating body, and build a similarity measurement function based on this, establish a target pose parameter optimization method, and achieve high-precision measurement of pose parameters; The specific implementation of step 3 is as follows: Step 3.1: Combining the predicted values ​​of the target pose parameters, the target geometric parameters, and the camera imaging model, the target is located. The minimum envelope rectangle width w corresponding to the target imaging feature can be approximately expressed as: The value of length l is determined by the position parameter and the posture parameter, and can be approximately expressed as: The center coordinates of the rectangle (x r ,y r ) is expressed as: The slope k of the rectangle r Expressed as: Step 3.2: Set the adaptive gradient threshold according to the gradient difference between the target contour edge point and other edge points, extract the target contour edge point, and calculate the minimum envelope rectangle based on this; based on the distribution characteristics of the noise, the edge point number change rate R w ,R l Optimize the envelope rectangle for evaluation index, R w ,R l The definition is as follows: Step 3.3: Correct the predicted values ​​of the pose parameters based on the eigenvalues ​​of the rectangle: correct the position parameters based on the width and center of the rectangle; correct the pose parameters based on the length and slope of the rectangle and the target motion law.

2. The method for measuring the position and posture of a rotating aircraft based on contour features according to claim 1, wherein: The specific implementation method of step 2 is as follows: Step 2.1: Predict the pose parameters of the target in the current frame based on the measured values ​​of the target pose parameters in the previous frame image and the prior information of the motion parameters: Step 2.2: Update the motion parameters based on the prior information and the measurement information. The update criteria for the expected values ​​of the motion parameters are as follows:

3. The method for measuring the position and posture of a rotating aircraft based on contour features according to claim 1, wherein: The specific implementation method of step 4 is as follows: Step 4.1: Combine the target 3D model, target pose parameters, and camera imaging model to generate the target imaging contour features in real time; Step 4.2: Establish a similarity measurement function based on the difference between the image eigenvalues ​​inside and outside the contour template and the difference between the contour eigenvalues ​​and the observed eigenvalues ​​to quantify the difference between the pose parameters corresponding to the template and the actual pose parameters; Step 4.3: Determine the posterior probability distribution of the pose parameters by combining the pose parameter correction value, motion prior information, and similarity measurement function, optimize the pose correction value, and achieve accurate estimation of the pose parameters.

Images