A high dynamic star point target extraction and positioning method based on pixel combination

Abstract

The application provides a high dynamic star point target extraction and positioning method based on pixel combination, comprising: S1, star point target rough extraction. Obtain the motion parameters of star points; construct a star point rough extraction template; perform rough extraction of star point targets based on image entropy. S2, star point accurate positioning. Take all pixels of the star point target as a whole, establish a Gaussian mixture model; determine the center line of the star point target by using principal component analysis method; match the star point Gaussian mixture model with the actual star point along the center line direction, and determine the length and width boundary of the star point tail through maximum likelihood estimation; calculate the star point centroid position by using the gray centroid method. The method takes the star point target as a whole composed of N star point pixels adjacent to each other, proposes a star point extraction and positioning algorithm based on pixel combination, can realize reliable extraction of star point targets with a signal-to-noise ratio lower than 3 or even close to 1, and the positioning accuracy can be improved by about 3 times compared with the conventional threshold method.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of star point extraction, and in particular to a high-dynamic star point target extraction and positioning method based on pixel combination. BACKGROUND

[0002] A star sensor is a main high-precision attitude measurement instrument in spacecraft attitude measurement, has the advantages of high precision, strong anti-interference, and independent navigation without dependence on other systems, and is widely applied to various spacecraft attitude measurement systems. The star sensor takes stars as observation targets, images a starry sky region in the direction of an optical axis, and then performs attitude calculation of a spacecraft through subsequent star point extraction, identification and positioning. In the above working process, star point extraction is the first step of spacecraft attitude calculation, and the star point detection and centroid positioning precision affect the accuracy of the subsequent steps.

[0003] With the continuous development of space technology, rapid maneuvering of a spacecraft has become a major requirement of a space mission, and the dynamic performance and update rate of the star sensor are increasingly required. Under high-dynamic conditions such as initial orbiting, maneuvering and large-angle attitude adjustment of the spacecraft, the star point image spots captured on the imaging surface of the star sensor will present a long strip-shaped "smear". Under high-dynamic conditions, the star target imaging produces a smear, and even breaks, and the signal-to-noise ratio sharply decreases. When the traditional threshold method is used to extract the star target, either a large positioning error is generated, or the star point bright spot cannot be detected, resulting in extraction failure.

[0004] Star point extraction and centroid positioning technology under high dynamic condition is a hot research topic in recent years. There are two main categories of star point extraction and positioning under high dynamic condition. One category needs high-precision gyroscopes and other external devices to obtain accurate motion information, but it is not in line with the development direction of star sensor miniaturization due to its large size and high cost, and the method of fusing external devices is not universal in star sensors. For example, Liheng M et al. proposed a high dynamic star point extraction method based on template matching. The degradation function is obtained from the real-time motion parameters, thereby constructing the degradation template of dynamic star points. The best matching between the fuzzy star map and the star point template is achieved by using the correlation criterion. However, it needs three orthogonal gyroscopes to complete the real-time acquisition of motion parameters, and the signal-to-noise ratio of the star point target that can be detected is at a normal level of greater than 3. Lian Da et al. proposed a star point registration and compensation method based on correlation matching. The star point dynamic template is established to determine the star point position by correlation matching, and the star point broken part is compensated by using the template. However, the dynamic template is established on the basis of real-time iterative estimation of the star point static model by using Kalman filter. After establishing the star point static spot model, Pandi et al. generate a star point dynamic trailing template by using the Bresenham line generation algorithm, and perform correlation matching on the star point to determine the position. The establishment of the dynamic template also needs the angular velocity information provided by the gyroscope. The other category uses image enhancement methods to process the star map to improve the extraction and positioning ability of dark and weak star points, but the methods for estimating the direction of star points and determining the imaging area of star points are not perfect, so some star points with lower signal-to-noise ratio cannot be extracted. For example, Jinyan et al. proposed a star target extraction method with adaptive window selection. When the star is broken, the broken star is “actively grown” based on mathematical morphology, and the star spot is repaired. However, the selection of the structure element and the gray value used for compensation is simple, and the real gray distribution of the spot is not considered, which may cause spot distortion. Especially when the star spot presents as a straight line with a slope, noise blocks are easily misjudged as star image areas, and the extraction accuracy is not ideal. Zeng Fen et al. proposed a method based on frame window shift superposition gray enhancement for extracting dark and weak star points under high dynamic condition, which improves the signal-to-noise ratio of weak stars in the window and makes them extractable. However, this method is mainly used to improve the detection rate of weak stars, and the positioning accuracy is not studied and discussed. Wang Hou et al. proposed a method of integrating and enhancing the gray value of the light spot along the direction of the star point. The image is enhanced, and then the star light spot is detected. However, the integral image will distort the brightness distribution of the star point, causing inaccurate centroid estimation. SUMMARY

[0005] The purpose of the present application is to provide a high dynamic star target extraction method based on pixel joint, which can realize reliable extraction of star point targets with a signal-to-noise ratio of less than 3 or even close to 1, and the positioning accuracy is about 3 times higher than that of the conventional threshold method.

[0006] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

[0007] According to an aspect of the present application, there is provided a high dynamic star point target extraction and positioning method based on pixel combination, comprising the following steps:

[0008] S1, star point target rough extraction, comprising:

[0009] S11, obtaining motion parameters of star points;

[0010] S12, constructing a star point rough extraction template;

[0011] S13, performing rough extraction of star point targets based on image entropy;

[0012] S2, star point accurate positioning, comprising:

[0013] S21, regarding all pixels of star point targets as a whole, establishing a Gaussian mixture model;

[0014] S22, determining a center line of star point targets by using a principal component analysis method;

[0015] S23, along the direction of the center line, matching the star point Gaussian mixture model with actual star points, and determining the length and width boundaries of star point tails by maximum likelihood estimation;

[0016] S24, calculating the centroid position of star points by using a gray centroid method.

[0017] In an embodiment, the S11 in the method specifically comprises: selecting a brightest star in a star map, and calculating the motion direction and tail length of the star point by using Radon transformation.

[0018] In an embodiment, the S12 in the method specifically comprises: establishing an inner template consistent with the shape of the star point target, and obtaining an outer template by adding a part of starry sky background area outside the inner template.

[0019] In an embodiment, the S13 in the method specifically comprises:

[0020] S131, scanning the whole image by using a rough extraction sliding window composed of the inner template and the outer template;

[0021] S132, calculating the image entropy related value of the current position;

[0022] S133, judging whether there is a star tail target in the template in the current position.

[0023] In an embodiment, the S132 in the method comprises:

[0024] S1321, if the maximum value and the average value of the pixel gray scale in the region where the inner template is located reach the set threshold value in the current position, it is considered that the position is suspected to have a star point; if the set threshold value is not reached, the rough extraction sliding window is moved to the next position;

[0025] S1322, for the position suspected to have a star point, the rough extraction sliding window is calculated at the position:

[0026] the image entropy of the W2 region under the outer template mask, that is, Img_Entropy(W2) = H(W2);

[0027] W0 region inside the outer template and outside the inner template, that is, W0 = W2-W1;

[0028] the image entropy of the W0 region, that is, Img_Entropy(W0) = H(W0);

[0029] S1323, the image entropy difference value of the W2 region and the W0 region is calculated;

[0030] H(W0)-H(W2) = Img_Entropy(W0)-Img_Entropy(W2);

[0031] If the entropy difference value reaches the set threshold value, it is considered that the position has a star tail.

[0032] In an embodiment, the S21 in the method specifically comprises: by an entropy algorithm, detecting a star tail rectangular region in a star chart, all pixels in the star tail rectangular region are collected as {i|i = 1, 2,..., N}, the pixels can be divided into a label set as {j|j = Ω1, Ω2}, wherein Ω1 represents that the pixel belongs to the background, and Ω2 represents that the pixel belongs to the star strip, and a probability density function of the pixel is defined as the following formula:

[0033]

[0034] wherein, the parameters Θ of the Gaussian mixture model j ={μ j ,σ j}, μ and σ respectively represent the mean and the standard deviation of the Gaussian distribution.

[0035] In an embodiment, the S22 in the method specifically comprises: binarizing the pixel set selected by the outer template frame to obtain a star point region highlight pixel set, and obtaining a projection direction of a first principal component by principal component analysis, and further determining the position and direction of the star point target center line from the first principal component direction.

[0036] In an embodiment, the S23 in the method specifically comprises:

[0037] S231, according to the center line position determined by the principal component analysis method, taking the center line as the center, setting two parallel straight lines with a distance of 3 pixels as the width boundary of the star point tail, and selecting two points on the center line as the two end points of the star point target, thus determining a closed region, assuming that the two end points and the boundary constitute a star strip region Ω2, and the rest in the rectangular window is a background region Ω1;

[0038] S232, calculating the joint likelihood expectation function L(Θ) under the current pixel label division {j|j=Ω1,Ω2} in the closed region,

[0039]

[0040] Then the end point position is changed, and the corresponding assumed star strip region and background region also change until all end point combinations in the end point value range are traversed, and the joint likelihood expectation function under all conditions is obtained;

[0041] S233, according to the expectation maximization criterion, when the likelihood expectation function L(Θ) of the image takes the maximum value, the closest to the true situation is It can be considered that the region corresponding to the division is the optimal star point tail imaging region, and the model built is best matched with the actual target.

[0042] In an embodiment, the S24 in the method specifically comprises:

[0043] The gray centroid method is used to calculate the star point centroid position (x p , y p ), and the calculation is as follows:

[0044]

[0045] Wherein, x i and y i are coordinate values of the detected star point imaging region, and I i is the pixel gray value corresponding to (x i , y i ).

[0046] The beneficial effects of the embodiment of the application are:

[0047] 1. The method does not rely on auxiliary devices such as gyroscopes to provide motion information, and the motion direction and tail length of the star points can be obtained by Radon transform assisted by bright stars in a single star map, so as to complete the modeling of the star point target;

[0048] 2. In the process of extracting the star point target, the texture information of the star point target itself is not damaged. First, the coarse extraction of the star point target is realized based on Radon transform and image entropy, and then the accurate positioning of the star point target is realized based on Gaussian mixture model and maximum likelihood estimation, without image enhancement of the star map, so as to not cause distortion of the star point distribution;

[0049] 3. The method breaks through the conventional extraction idea of "from details to the whole", and regards the star point target as a whole composed of N adjacent star point pixels, "from the whole to the details", and proposes a star point extraction and positioning algorithm based on pixel combination. The method has significant advantages in extraction accuracy and positioning accuracy, can realize reliable extraction of the star point target with a signal-to-noise ratio lower than 3 or even close to 1, and can improve the positioning accuracy by about 3 times compared with the conventional threshold method. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0051] The above features and advantages of the present application can be better understood after reading the detailed description of the embodiments of the present application in combination with the following drawings. In the drawings, the components are not necessarily drawn to scale, and the components having similar related properties or features can have the same or similar reference numerals.

[0052] Figure 1 is a schematic diagram of the coarse extraction inner template (a), the coarse extraction outer template (b) and the combined coarse extraction sliding window (c) constructed in the method.

[0053] Figure 2 is a schematic diagram of the region in the image entropy calculation step;

[0054] Figure 3 is a flowchart of the embodiment of the method. DETAILED DESCRIPTION

[0055] The present application will be described in detail below in combination with the drawings and specific embodiments. It should be noted that the aspects described below in combination with the drawings and specific embodiments are only exemplary and should not be understood as any limitation on the scope of protection of the present application.

[0056] Referring to Figure 3 The embodiment provides a high dynamic star object extraction method based on pixel combination, comprising the following steps:

[0057] Step S1, star point target rough extraction:

[0058] S11, obtaining the motion parameters of the star point. A bright star with the maximum signal-to-noise ratio in the star map is selected, and Radon transformation is performed on the spectrum map of the region of the bright star. The projection angle at which the maximum value is obtained by Radon transformation is used to obtain the bright and dark stripe inclination angle d in the spectrum map and the central bright stripe width d. The star point motion angle β and the star point tail length L are calculated according to the following formula respectively:

[0059]

[0060]

[0061] In the formula, M and N are respectively the number of pixels in the length and width directions of the star map in the region of the bright star.

[0062] S12, constructing a star point rough extraction template. When the star sensor rotates around a non-optical axis direction, the motion directions of the star points at different positions in the star map are approximately the same, so that a star point rough extraction template for the star map can be established according to the motion parameters of the bright star. An inner template (as shown in a of Figure 1 ) is established, and an outer template (as shown in b of Figure 1 ) is obtained by adding a part of the star background region on the basis of the inner template. The combination of the inner and outer templates can detect the local gray scale distribution change, and form a star point target rough extraction window (as shown in c of Figure 1 );

[0063] S13, performing rough extraction of the star point target based on image entropy.

[0064] S131, scanning the full map using the rough extraction sliding window composed of the inner template and the outer template;

[0065] S132, calculating the image entropy related value at the current position; including:

[0066] S1321, if the maximum value and the average value of the pixel gray scale in the region where the inner template is located both reach the set threshold value at the current position, it is considered that the star point exists at the position; if the set threshold value is not reached, the rough extraction sliding window is moved to the next position;

[0067] S1322, for the position where the star point is suspected to exist, the rough extraction sliding window is calculated at the position respectively:

[0068] ​Image entropy of the W2 region under the outer template mask, i.e. Img_Entropy(W2) = H(W2);

[0069] W0 region inside the outer template and outside the inner template, i.e. W0 = W2 - W1;

[0070] Image entropy of the W0 region, i.e. Img_Entropy(W0) = H(W0).

[0071] The positional relationship of each region is shown in Figure 2 .

[0072] S1323, calculating the image entropy difference of the W2 region and the W0 region;

[0073] H(W0) - H(W2) = Img_Entropy(W0) - Img_Entropy(W2);

[0074] S133, judging whether there is a star trail target in the inner template at the current position.

[0075] If the inner template is located at a position completely framing the star trail at this time, the entropy value of the W0 region, i.e. the region excluding the W1 region of the inner template from the W2 region of the outer template, will have a large difference from the entropy value of the W2 region of the outer template. If the entropy difference reaches a set threshold value, it is considered that there is a star trail at the position.

[0076] At this point, the coarse extraction of the star target is completed based on the Radon transform and the image entropy described above, but the accuracy is far from enough for the star centroid positioning. Therefore, it is necessary to further refine the star imaging region to obtain more accurate and higher-precision star positioning.

[0077] Step S2, star precise positioning:

[0078] S21, regarding all pixels of the star target as a whole, a Gaussian mixture model is used to describe the star image. In a star trail rectangular region of the star image detected by the entropy algorithm, the set of all pixels is {i|i = 1, 2,..., N, and the label set of the pixels is {j|j = Ω1, Ω2}, wherein Ω1 represents that the pixel belongs to the background, and Ω2 represents that the pixel belongs to the star trail. The probability density function of the pixel is defined as the following formula:

[0079]

[0080] wherein the parameters Θ of the Gaussian mixture model j = {μ j , σ j}, μ and σ respectively represent the mean and standard deviation of the Gaussian distribution.

[0081] S22, the position and direction of the star point target center line are determined by using a principal component analysis (PCA) method. The pixel set selected by the outer template frame is binarized to obtain a star point region high brightness pixel set, and the projection direction of the first principal component is obtained by principal component analysis. The position and direction of the star point target center line are further determined from the first principal component direction.

[0082] S23, the star point Gaussian mixture model is matched with the actual star point along the center line direction, and the length and width boundaries of the star point tail are determined by maximum likelihood estimation. According to the center line position determined by the principal component analysis method, the center line is taken as the center, the width is referenced to the size of the static star point, two parallel straight lines with a distance of 3 pixels are set as the width boundaries of the star point tail, and two points on the center line are selected as the two end points of the star point target. Thus, a closed region is determined, which is assumed to be composed of the two end points and the boundary. The remaining part in the rectangular window is the background region Ω1. Then, the joint likelihood expectation function L(Θ) under the current pixel label division {j|j=Ω1,Ω2} in the closed region is calculated.

[0083]

[0084] Then, the end point position is changed, and the assumed star strip region and background region also change accordingly. Until all end point combinations in the end point value range are traversed, and the joint likelihood expectation function under all conditions is obtained.

[0085] According to the expectation maximization (EM) criterion: when the likelihood expectation function L(Θ) of the image takes the maximum value, it is closest to the true situation. That is It can be considered that the region corresponding to the division at this time is the optimal star point tail imaging region, and the model built is best matched with the actual target.

[0086] S24, the star point centroid position (x p , y p ) is calculated by using the gray centroid method, and the calculation is as follows:

[0087]

[0088] Where x i and y i are the coordinate values of the detected star point imaging region, and I i is the pixel gray value corresponding to (x i , y i ).

[0089] Compared with the prior art, the method has the following beneficial effects:

[0090] 1. The method does not rely on auxiliary devices such as gyroscopes to provide motion information. In a single star map, the motion direction and tail length of the star points can be obtained by Radon transform assisted by bright stars, and the modeling of the star point target is completed.

[0091] 2. In the process of extracting the star point target, the texture information of the star point target itself is not destroyed. First, the coarse extraction of the star point target is realized based on Radon transform and image entropy, and then the accurate positioning of the star point target is realized based on Gaussian mixture model and maximum likelihood estimation. There is no image enhancement for the star map, which will not cause distortion of the star point distribution.

[0092] 3. The method breaks through the conventional "from details to the whole" extraction idea, and regards the star point target as a whole composed of N adjacent star point pixels. A star point extraction and positioning algorithm based on pixel combination is proposed. The method has significant advantages in extraction accuracy and positioning accuracy, and can realize reliable extraction of star point targets with a signal-to-noise ratio of less than 3 or even close to 1. At the same time, the positioning accuracy can be improved by about 3 times compared with the conventional threshold method.

[0093] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the embodiments can be referred to each other.

[0094] Although the above methods are illustrated and described as a series of acts for simplicity of explanation, it should be understood and appreciated that the methods are not limited by the order of acts, as some acts can occur in different orders and / or concurrently with other acts according to one or more embodiments. Not all of the acts are required, however, for the practice or completion of the methods.

[0095] The foregoing description of the present disclosure has been presented for purposes of illustration and description. Various modifications to the present disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the examples described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0096] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

Claims

1. A high dynamic star point target extraction and positioning method based on pixel joint, characterized in that, The method comprises the following steps: S1, star point target coarse extraction, comprising: S11, obtaining the motion parameters of the star point; the S11 specifically comprises: selecting a brightest star in the star map, and calculating the motion direction and the tail length of the star point by using Radon transform; obtaining the light and dark stripe inclination in the spectrum diagram from the projection angle of the maximum value obtained by the Radon transform and the central bright stripe width wherein the calculation formula of the star point motion angle and the star point tail length is as follows: , M and N are respectively the pixel number in the length direction and the width direction of the star map of the bright star region. S12, constructing a star point coarse extraction template; the S12 specifically comprises: establishing an inner template consistent with the length and direction of the star point target, and obtaining an outer template by adding a part of the star sky background area on the basis of the inner template, wherein the inner and outer templates are combined to detect local gray scale distribution changes and form a star point target coarse extraction window; S13, performing coarse extraction of the star point target based on image entropy; the S13 specifically comprises: S131, scanning the entire image using a coarse extraction sliding window composed of the inner template and the outer template; S132, calculating an image entropy correlation value at the current position; S133, determining whether there is a star point tail target in the inner template at the current position; S2, star point accurate positioning, comprising: S21, regarding all pixels of the star point target as a whole, and establishing a Gaussian mixture model; S22, determining a star point center line using a principal component analysis method; S23, matching the star point Gaussian mixture model with the actual star point along the center line direction, and determining the length and width boundaries of the star point tail through maximum likelihood estimation; S24, calculating the star point centroid position using a gray centroid method.

2. The pixel joint-based high dynamic star point target extraction and positioning method according to claim 1, characterized in that, The S132 comprises: S1321, if the maximum value and the average value of the pixel gray scale of the region where the inner template is located at the current position both reach a set threshold value, it is considered that there is a star point at the position; if the set threshold value is not reached, the coarse extraction sliding window is moved to the next position; S1322, for the position where the star point is suspected to exist, the coarse extraction sliding window at the position is calculated respectively: under the outer mask image entropy of the region, i.e. ; inside the outer mold plate and outside the inner mold plate region, i.e. ; the image entropy of the region, i.e. ; S1323, calculating region and image entropy difference value of the region ； If the entropy difference value reaches a set threshold value, it is considered that there is a star point tail at the position.

3. The pixel joint-based high dynamic star point target extraction and positioning method according to claim 2, characterized in that, The S21 specifically includes: by the entropy algorithm detects one star point in the star map, all pixels in the tailing rectangular region are collected as The pixel can be divided into a label set Wherein The pixel belongs to the background, The pixel belongs to the star strip, and the probability density function of the pixel is defined as the following formula: ， where the parameters of the Gaussian mixture model , and represent the mean and standard deviation of the Gaussian distribution, respectively.

4. The pixel joint-based high dynamic star point target extraction and positioning method according to claim 3, characterized in that, The S22 specifically comprises: binarizing the pixel set framed by the outer template to obtain a star point region high-luminance pixel set, and obtaining the projection direction of the first principal component through principal component analysis to further determine the position and direction of the star point target center line.

5. The pixel joint-based high dynamic star point target extraction and positioning method according to claim 4, characterized in that, The S23 specifically comprises: S231, the center line position determined according to the principal component analysis method, taking the center line as the center, the width referring to the static star point size, setting two parallel straight lines with a distance of 3 pixels as the width boundary of the star point tail, selecting two points on the center line as the two end points of the star point target, thus determining a closed area, assuming that the two end points and the boundary constitute a star strip area , the rest in the rectangular window is the background area ; S232, compute the joint likelihood expectation function under the current pixel label partition S232, compute the joint likelihood expectation function under the current pixel label partition , ， Then the end point position is changed, and the corresponding assumed star strip region and background region also change until all end point combinations in the end point value range are traversed, and the joint likelihood expectation function under all conditions is obtained; S233, Expectation of likelihood function of image according to expectation maximization criterion At the time of taking maximum value, the closest to the real situation, that is It can be considered that the region corresponding to the division at this time is the optimal star point tail imaging region, and the model built is best matched with the actual target.

6. The pixel joint-based high dynamic star point target extraction and positioning method according to claim 5, characterized in that, The S24 specifically comprises: The centroid position of the star point is calculated using a grayscale centroid method The following formula is calculated: ， wherein, and is a coordinate value of a star point imaging region detected, is a corresponding pixel gray value.

Images