A high dynamic tailing star image exposure center time plane positioning method

Abstract

The application discloses a high-dynamic smearing star image exposure center time image plane positioning method, and belongs to the field of telescope optical detection image processing. The method comprises the following steps: firstly, filtering a high-dynamic star map through an anisotropic Gaussian kernel to obtain an enhanced image and a direction image, and then accurately extracting the exposure center position of a star image based on the enhanced image and the direction image; performing linear fitting on each extracted star image by using a least square method to obtain a motion trajectory of the star image; and calculating an integral midpoint along the motion trajectory of the star image as the exposure center position of the star image by using the gray level accumulation sum of the normal direction of each point on the trajectory. The application realizes accurate pixel position determination of a smearing star image under a high-dynamic background, can effectively deal with the problem that the star image center position is difficult to accurately obtain under complex smearing conditions caused by high-dynamic star image smearing and platform turntable jitter, significantly increases the star image exposure center time image plane positioning precision compared with a traditional centroid algorithm, and finally improves the performance of a star tracker.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of telescope optical detection image processing, and particularly relates to a high-dynamic tailing star image exposure center time instant image plane positioning method. BACKGROUND

[0002] In the application scenarios of star sensors, star trackers and space debris tracking and positioning, star image detection is a core link of photoelectric telescopes, and the positioning accuracy of the pixel position of the centroid directly restricts the attitude determination accuracy and the astronomical positioning index of the system. The current technical challenge is that: the high dynamic motion causes the tailing effect of the target star image or the star and the arc effect caused by image rotation, so that the traditional centroid algorithm is difficult to accurately position the star image exposure center position on the image plane, such as the extracted position not on the target track, the calculation error caused by jitter, etc. In the high dynamic scene, the star imaging is disturbed by noise factors such as long integration time, large carrier angular motion and platform turntable jitter, and a variety of shapes of tailing instead of ideal point features are formed on the detector surface. The traditional centroid calculation method based on gray-scale image is no longer applicable to the star image position calculation under the condition of non-uniform tailing. In a strict sense, the system needs to determine the pixel position of the moving star image on the image plane at the integration center time instead of the centroid. Therefore, the core task at this time is to accurately position the star image position at the integration center time from the non-uniform energy distribution area. However, the classical centroid positioning algorithm is mainly aimed at static ideal point star images, and is not suitable for high dynamic star tailing in complex conditions. Therefore, it is of important theoretical significance and engineering application value to study the high dynamic star image exposure center time position determination method, which helps to improve the dynamic performance of the star tracker.

[0003] The current high dynamic star centroid positioning technology mainly realizes performance improvement through hardware optimization and algorithm compensation. The hardware enhancement scheme focuses on the improvement of the imaging system, such as the time delay integration technology (TDI) proposed by Roelof team, which reduces the vertical direction tailing effect by optimizing the imaging circuit timing, but this scheme increases the complexity of the imaging circuit and reduces the effective photosensitive area by more than 30%. Pasetti et al. introduced a multi-pixel cooperative compensation mechanism based on the TDI architecture, but the motion parameter estimation has significant deviation, and the dynamic compensation accuracy is limited.

[0004] Algorithm compensation mainly relies on image degradation model reconstruction. Zhang W N et al. uses adaptive wavelet threshold method for noise suppression, and then realizes motion blur restoration through wiener filter. However, the scheme depends on the accurate prior knowledge of degradation function and noise power spectrum, and the engineering applicability is limited. Sun T team constructs a kinematic model based on gyro angular velocity, and combines Richardson-Lucy (RL) algorithm for iterative restoration. Although the dynamic model accuracy can be improved, the RL algorithm is sensitive to noise interference, and multiple iterations are required for optimization, resulting in processing delay. In view of the defects of RL algorithm, Ma L H et al. uses multi-sub-region growth strategy for pretreatment, but the fixed size template leads to energy loss of star image; Sun T et al. introduce global morphological filter to improve signal-to-noise ratio, but the processing delay exceeds the real-time threshold. Jiang J et al. fuse the denoising and degradation function estimation of improved Radon transform, and accelerate the RL iteration through second-order vector extrapolation, but the whole frame restoration mode restricts the processing efficiency, and the dynamic scene adaptability is insufficient.

[0005] The above algorithm is mainly based on the assumption that the star image moves at a uniform linear speed in the imaging plane. This model ignores the potential influence of image rotation and high-frequency vibration on the shape of the star image point spread function. In addition, existing research generally lacks in-depth quantitative analysis of the correlation between key imaging parameters, such as different exposure times and star image motion speed, and the algorithm restoration effect. Considering the above factors, the existing image restoration algorithm has inherent limitations in dealing with dynamic degradation problems. This directly leads to the fact that the accuracy of the traditional centroid positioning algorithm is significantly restricted when processing star image data after dynamic blur restoration. SUMMARY

[0006] The technical problem solved by the present application is to solve the problem of low star image centroid positioning accuracy of the traditional centroid positioning method when facing high dynamic star image trailing gray distribution unevenness and trailing trajectory complexity, and to provide a high dynamic trailing star image exposure center time plane positioning method. By fusing the high dynamic star image exposure center time plane positioning algorithm of star image trailing trajectory fitting and integral midpoint calculation, the accuracy of star image exposure center time plane positioning in high dynamic environment is significantly improved. The method improves the positioning accuracy of star image center position in high dynamic conditions.

[0007] The technical solution of the present application is as follows:

[0008] A high dynamic trailing star image exposure center time plane positioning method, comprising:

[0009] Step S1: calculate a discrete anisotropic Gaussian kernel with different angles and variances as a filter, filter the image, obtain an enhanced image L and a direction image D from the anisotropic Gaussian kernel information that produces the maximum response, and then use a non-maximum suppression algorithm to refine the direction image D and the enhanced image L to obtain a result image;

[0010] Step S2: a single star image tail in the result map is proposed, and a pixel set thereof is fitted into a polynomial function of a motion trajectory according to a least square principle;

[0011] Step S3: a symmetric one-dimensional integration interval is defined along a local normal direction of each pixel point of the motion trajectory, and a local cumulative sum is calculated;

[0012] Step S4: based on the local cumulative sum in the normal direction, a point making the integration sums on both sides equal is searched as an integration midpoint along the motion trajectory of the star image.

[0013] Step S5: steps S2-S4 are repeated until the integration midpoint, i.e., the star image position at the exposure center moment, is calculated for all extracted star image tails.

[0014] A computing device comprises at least one processor and a memory storing program instructions; when the program instructions are read and executed by the processor, the computing device performs a high-dynamic tail star image exposure center moment image plane positioning method.

[0015] A readable storage medium storing program instructions, when the program instructions are read and executed by a computing device, the computing device performs a high-dynamic tail star image exposure center moment image plane positioning method.

[0016] The present application has the following beneficial effects:

[0017] The present application improves the star image exposure center moment image plane positioning accuracy, realizes high-precision positioning of tail star images in images under high-dynamic background by fusing star image tail trajectory fitting and integration midpoint calculation, significantly improves the positioning accuracy of star images under complex motion scenes, effectively overcomes the problem of uneven gray distribution caused by non-uniform motion of targets on the image plane under dynamic imaging conditions, and the feature distortion phenomenon caused by tail shape distortion of star images, and further improves the correctness and robustness of star image exposure center moment image plane positioning under complex working conditions.

[0018] The present application can improve the dynamic performance of star trackers, accurately position the star image exposure center moment image plane by calculating the exposure center using the star image motion trajectory, and is suitable for various tail star images under complex conditions, and improves the dynamic performance of all-day tracking. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the high-dynamic tail star image exposure center moment image plane positioning method according to the present application;

[0020] Figure 2 is an original star map in a specific embodiment;

[0021] Figure 3is a star trajectory fitting result graph in the specific embodiment;

[0022] Figure 4 is a star image plane positioning result graph at the high dynamic star image integration center moment in the specific embodiment;

[0023] Figure 5 is a result graph of star image plane positioning at the simulation image exposure center moment. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0025] Figure 1 is a flow chart of a high dynamic star image exposure center moment image plane positioning method according to the present application. As shown in Figure 1 , the method comprises:

[0026] Step S1: calculate a discrete anisotropic Gaussian kernel with different angles and variances as a filter to filter an image, obtain an enhanced image L and a direction image D from anisotropic Gaussian kernel information generating a maximum response, and then use a non-maximum suppression algorithm (NMS) to refine the direction image D and the enhanced image L to obtain a result image.

[0027] Step S2: propose a single star image trail in the result image, and fit a pixel point set thereof into a polynomial function of a motion trajectory according to a least square method.

[0028] Step S3: define a symmetric one-dimensional integration interval along a local normal direction of each pixel point of the motion trajectory to calculate a local cumulative sum. To calculate a local energy aggregation value of any point on the trajectory along its normal direction, a one-dimensional integration path needs to be defined and a line integration operation needs to be performed. Step S3 can comprise: determining a local normal direction of each pixel point of the motion trajectory; taking a symmetric space along the normal direction; and calculating a local cumulative sum of pixels in the interval.

[0029] Step S4: based on the local cumulative sum in the normal direction, find a point making the integration sums on both sides equal as an integration midpoint along the star motion trajectory.

[0030] Step S5: repeat steps S2-S4 until the integration midpoint, i.e. the star position at the exposure center moment, is calculated for all extracted star image trails.

[0031] In S1, the enhanced image L and the orientation map D are obtained from the anisotropic Gaussian kernel information that produces the maximum response. This includes: using a maximum value aggregation algorithm, for each pixel, retaining the largest value among all anisotropic Gaussian kernel responses as the value of the enhanced image L(m); similarly, the orientation map D is obtained from the anisotropic Gaussian kernel orientation information that produces the maximum response. The result image is refined using NMS technology, including: for each pixel, if the straightness value L(m) at the current pixel position m is the largest relative to other pixel values ​​in the direction of D(m), then that value is retained; otherwise, it is invalid. Finally, the image after extracting the star pattern is obtained.

[0032] Step S1 includes: calculating discrete anisotropic Gaussian kernels with different angles and variances as filters to filter the image, where m represents the image coordinates in two-dimensional integer coordinates. In it, its second-order anisotropic Gaussian kernel Defined as:

[0033] ,

[0034] ,

[0035] Among them, anisotropic factors and rotation matrix Defined as:

[0036] ,

[0037] ,

[0038] These represent scale and direction, respectively. Represents two-digit integer coordinates. and Let represent the scale set and the direction set, respectively, and g be the filtering formula for a first-order anisotropic Gaussian kernel. The robustness control scale is represented by its value; the larger the value, the more pronounced the stretching effect of the Gaussian kernel. The result of filtering image I using a discrete second-order anisotropic Gaussian kernel is as follows:

[0039] ,

[0040] in, Indicates the region of interest.

[0041] The maximum value aggregation algorithm is used to retain the maximum value among all anisotropic Gaussian kernel responses for each pixel as the value of the enhancement map L(m):

[0042] ,

[0043] The trajectory pattern also needs to be calculated in subsequent processing. As mentioned earlier, anisotropic Gaussian kernels with the same direction as the trajectory produce a peak response, and the direction information is represented by the anisotropic Gaussian kernel that produces the maximum response. That is, the trajectory pattern D(m) is obtained by the following formula:

[0044] ,

[0045] Non-maximum suppression (NMS) techniques are applied to refine the two filtered images. Specifically, for each pixel position m in the enhanced image L, it is determined whether pixel m is a local maximum point within its local spatial neighborhood (e.g., a 3×3 or 5×5 window), i.e., whether L(m) is greater than the intensity values ​​of all its spatially neighboring pixels. Simultaneously, based on the gradient direction information D(m) provided by the orientation map D at the corresponding position, it is examined whether pixel m also has the same gradient in its local one-dimensional neighborhood along that gradient direction.

[0046] Pixel m is retained and considered a potential candidate for a valid star center point only if both of the above conditions are met simultaneously; otherwise, the pixel value will be suppressed (usually set to zero). After this NMS processing flow, the final output is a binarized image in which the star target has been effectively extracted and refined.

[0047] Further, step S2 includes extracting a single celestial image, the pixel dataset of which is... The target function is a fourth-order polynomial:

[0048] ,

[0049] in, These are undetermined coefficients. According to the principle of least squares, we need to minimize the weighted sum of squared residuals between the observed and fitted values:

[0050] ,in, This is the actual observational data for this celestial phenomenon.

[0051] The objective function simplifies to:

[0052] ,

[0053] Among them, matrix The elements are ,vector coefficient vector .

[0054] Furthermore, in step S3, a symmetrical one-dimensional integration interval is defined along the local normal direction of each pixel on the motion trajectory, and the local cumulative sum is calculated, including:

[0055] S3-1, for each discrete point on the trajectory, calculate its corresponding local normal direction vector. The direction pattern D(m) typically represents the principal direction of the image gradient at that point, and its normal direction is perpendicular to this principal direction. This local normal direction is defined as:

[0056] ,

[0057] in, It is the direction of the tangent at that point. It is the direction of the normal at that point. It is a point on the trajectory.

[0058] To calculate the trajectory of any point m along its normal direction To obtain the local energy aggregation value, a one-dimensional path needs to be defined and an accumulation operation needs to be performed.

[0059] S3-2, For a trajectory point m, define a symmetric one-dimensional integration interval along its local normal direction. This interval is centered at m and extends to a distance of... For all pixels within the interval along the normal direction. The formula for calculating the local cumulative sum is defined as follows:

[0060] ,

[0061] ,

[0062] in, It is a pixel point within the interval along the normal direction. It is the grayscale value of that point. It is a point The sum of all gray values ​​within the interval It is a point on the trajectory.

[0063] Furthermore, in step S4, based on the local summation along the normal direction, a point is found along the trajectory of the celestial bodies that makes the sums of the integrals on both sides equal, serving as the midpoint of the integral. , Let these represent the coordinates of the midpoint of the integral, and let them be the coordinates of the midpoint. The formula is:

[0064] ,

[0065] in, It is the total number of points on the trajectory.

[0066] The method of the present invention will be described in detail below through specific examples:

[0067] Combination Figure 1 As shown, for Figure 2 The original star map shown is processed as follows:

[0068] Step S1: filter the original star map by discrete anisotropic Gaussian kernel, specifically, the anisotropic Gaussian kernel size is 20x20, angle belongs to 0° to 180°, variance The trajectory map L and the direction map D are obtained from the anisotropic Gaussian kernel information that produces the maximum response, and then the NMS technique is used to refine the result map.

[0069] Step S2: according to the result map calculated in S1, extract a single star image in the result map, and fit the star image pixels to a quartic polynomial function according to the least square method, as shown in Figure 3 .

[0070] Step S3: calculate the integral midpoint of the trailing trajectory according to the quartic polynomial calculated in S2. First, calculate the normal direction of each point on the star image trajectory, and define the displacement interval along the normal direction of any point on the star image trajectory, and then sum all the points in the interval.

[0071] Step S4: calculate the integral midpoint so that the integral sum of the two sides along the trajectory is equal.

[0072] Step S5: repeat steps S2-S4 until the centroid of all extracted star images, i.e. the star image position at the exposure center time, Figure 4 is the exposure center time image plane positioning result of multiple star images. In order to verify the positioning accuracy of the proposed algorithm, a comparative experiment is carried out on the simulation star map data. As shown in Figure 5 , the true value of the exposure center time corresponding to the simulation star image is marked with "o", and the exposure center time image plane position calculated by the algorithm is marked with "x". By systematically comparing, the positioning accuracy of the algorithm can be visually verified.

[0073] In summary, the present application discloses a high dynamic star image exposure center time image plane positioning method along the trajectory, which comprises the following steps: first, filter the high dynamic star map by anisotropic Gaussian kernel to obtain an enhanced map and a direction map, and extract star images based thereon; then, linearly fit each extracted star image by least square method to obtain the motion trajectory of the star image; and calculate the integral midpoint of the normal direction of each point on the motion trajectory as the exposure center of the star image. The present application realizes accurate positioning of star images in high dynamic background, and can effectively deal with the complex phenomenon of high dynamic star image trailing. Compared with the traditional centroid positioning algorithm, the present application significantly increases the image plane positioning accuracy of star image exposure center time under high dynamic complex trailing star image, and finally improves the performance of star tracker.

[0074] While the application has been described with reference to particular embodiments thereof, it should be understood that these are merely illustrative of the principles and applications of the application. It will thus be appreciated that numerous modifications can be made to the illustrative embodiments and that other arrangements can be devised without departing from the spirit and scope of the present application as defined by the appended claims. It will be understood that the features described in connection with one embodiment can be used in conjunction with other embodiments.

Claims

1. A high dynamic tailing star image exposure center time image plane positioning method, characterized in that, Comprising: Step S1: calculating different angle and variance of discrete anisotropic Gaussian kernel as filter, filtering the image, getting enhanced image L and direction image D from anisotropic Gaussian kernel information which produces maximum response, then using non-maximum suppression algorithm to refine direction image D and enhanced image L to get result image; Step S2: proposing single star image tail in result image, fitting its pixel set as polynomial function of motion trajectory according to least square method principle; Step S3: defining a symmetric one-dimensional integration interval along the local normal direction of each pixel point along the motion trajectory, calculating local cumulative sum; Step S4: taking local cumulative sum in normal direction as basis, finding the point which makes the integral sum on both sides equal as integral midpoint along the motion trajectory of star image; Step S5: repeating steps S2-S4 until the integral midpoint of all extracted star image tails, i.e. star image position at exposure center time, is calculated; In step S3, a symmetric one-dimensional integration interval is defined along the local normal direction of each pixel point along the motion trajectory, and the local cumulative sum is calculated, comprising: S3-1, for each discrete point on the trajectory, the corresponding local normal direction vector is calculated, and the local normal direction is defined as: ， wherein, is the tangent direction of the point, is the normal direction of the point, is a certain point on the trajectory; S3-2, for a trajectory point m, define a symmetric one-dimensional integration interval along its local normal direction The interval is centered at m and extends to a distance ; compute the local accumulated sum for all pixels within the interval along the normal direction The formula is defined as: ， ， wherein, is a certain pixel point within the normal direction interval, is the gray value of the point, is the point the cumulative sum of all gray values within the interval, is a certain point on the trajectory; The integral midpoint is found by searching along the star motion trajectory for a point where the two side integrals are equal , The coordinates of the integral midpoint are given by ， wherein, is the total number of points on the track, is a certain point on the track.

2. The high dynamic smear star image exposure center time instant image plane positioning method according to claim 1, characterized in that, In S1, enhanced image L and direction image D are obtained from anisotropic Gaussian kernel information which produces maximum response, comprising: using maximum aggregation algorithm, for each pixel point, the maximum value of all anisotropic Gaussian kernel responses is retained as the value of enhanced image L(m), and the same, direction image D is obtained from the direction information of the anisotropic Gaussian kernel which produces the maximum response; using NMS technology to refine to get the result image, comprising: for each pixel, if the straightness value L(m) at the current pixel position m is maximum relative to the values of other pixels in the direction of D(m), the value is retained, otherwise it will be invalid, finally the image after extracting star image is obtained.

3. The high dynamic smear star image exposure center time instant image plane positioning method according to claim 1, characterized in that, Step S1 comprises calculating a discrete anisotropic Gaussian kernel of different angles and variances as a filter, filtering the image comprises: m denotes the image coordinates, in two-dimensional integer coordinates whose second-order anisotropic Gaussian kernel is defined as: ， ， where the anisotropy factor and the rotation matrix is defined as: ， ， respectively denote scale and direction, where denotes two-dimensional integer coordinates, and respectively denote scale set and direction set, g is a filtering formula of first-order anisotropic Gaussian kernel, denotes robust control scale, the greater the value, the more obvious the stretching effect of the Gaussian kernel; the result of filtering the image I with the discrete second-order anisotropic Gaussian kernel is: ， wherein, represents the region of interest; Using maximum aggregation algorithm, for each pixel point, the maximum value of all anisotropic Gaussian kernel responses is retained as the value of enhanced image L(m): ， Trajectory direction image D(m) is obtained by the following formula: 。 4. The high dynamic smear star image exposure center time instant image plane positioning method according to claim 3, characterized in that, S1 comprises: Using non-maximum suppression algorithm to refine direction image D and enhanced image L to get result image, comprising: for each pixel position m in enhanced image L, it is judged whether pixel m is a maximum value point in its local spatial neighborhood; at the same time, according to the gradient direction information D(m) provided by the corresponding position of direction image D, it is investigated whether pixel m is also in the same gradient in its local one-dimensional neighborhood in the gradient direction; Only when the above two conditions are met at the same time, pixel m is retained as a potential effective star image center point candidate; otherwise, the pixel value will be suppressed; After this non-maximum suppression processing process, the final output is the binary image of the star image target after effective extraction and refinement.

5. The high dynamic smear star exposure center time instant image plane positioning method according to claim 4, characterized in that, Step S2 includes extracting a single star image whose pixel data set is , and the fitting target is a fourth-order polynomial function: ， wherein are to be determined coefficients; According to the least square method principle, the weighted residual sum of squares of observation value and fitting value is minimized: wherein, is the real observation data of the star appearance; The objective function is simplified as: ， where the elements of the matrix are , the vector , and the coefficient vector .

6. The high dynamic smear star exposure center time instant image plane positioning method according to claim 5, characterized in that, S3 comprises: determining the local normal direction of each pixel point along the motion trajectory; taking symmetric space along the normal direction; calculating the local cumulative sum of pixels in the interval.

7. A computing device, comprising: Comprising: At least one processor and a memory having program instructions stored therein; When the program instructions are read and executed by the processor, the computing device is caused to perform the high dynamic trailing star image exposure center-of-time image plane positioning method as claimed in any one of claims 1-6.

8. A readable storage medium storing program instructions, characterized in that, When the program instructions are read and executed by the computing device, the computing device is caused to perform the high dynamic trailing star image exposure center-of-time image plane positioning method as claimed in any one of claims 1-6.

Images