A high dynamic stripe star image detection method and device based on a region growing algorithm

Abstract

The application discloses a high-dynamic streak star image detection method and device based on a region growing algorithm, and relates to the image processing technology in the fields of star sensors for astronomical navigation and space debris monitoring. The method comprises the following steps: S110, performing global threshold segmentation on an input high-dynamic star map to obtain a candidate seed region set; S120, traversing the candidate seed region, calculating the direction information of each region by using an anisotropic Gaussian filter, judging the possibility of the existence of a streak star image in the region, growing along the region direction if the possibility is greater than a preset threshold, and stopping growing if the possibility is less than the preset threshold; and S130, marking the pixels meeting the spatial distribution characteristics of a trailing streak star image during the growing process until all the candidate seed regions are traversed, and completing the streak star image detection. The application effectively realizes the high-dynamic streak star image detection under noise interference, and significantly enhances the suppression ability to various noise interferences.

Description

TECHNICAL FIELD

[0001] The present application relates to the image processing technology in the field of star sensor and space debris monitoring, and in particular to a high dynamic streak star image detection method and device based on a region growing algorithm. BACKGROUND

[0002] As the core measurement device for attitude measurement of a carrier such as a spacecraft, the performance of a star sensor directly determines the autonomous navigation accuracy and reliability of the spacecraft. The carrier on which the star sensor is installed, especially a high dynamic carrier, will produce rotation, movement, and jitter. The traditional star sensor improves the signal-to-noise ratio of star points by setting a longer exposure time, but in a high dynamic environment, the relative motion between the carrier and the star image during exposure will cause the streak star image on the image plane to produce a tail phenomenon, and the trajectory is determined by the motion direction and the exposure time. This tailing causes the energy of the streak star image to disperse to the surrounding pixels, resulting in a decrease in the gray value of the streak star image and the presence of noise. In actual applications, the carrier often presents a complex motion mode of combined linear motion, angular motion, and vibration, which further aggravates the non-linear characteristics of star image tailing and becomes a core technical bottleneck restricting the detection and positioning performance of the streak star image.

[0003] The existing solutions to the above problems mainly fall into two categories. The first category is to improve the imaging system to weaken the influence of high dynamic motion before the streak star image is imaged. For example, by increasing the motion compensation of optical and mechanical components, by optimizing the signal-to-noise ratio of star images through increasing the light flux, or by achieving a leap in photoelectric conversion efficiency through high quantum efficiency materials and low noise circuit design. This type of method has the disadvantages of high cost and increased load weight.

[0004] The second category is to detect the streak star image through image processing after the star image is imaged. This type of method mainly includes star image restoration algorithms and dynamic window optimization techniques to improve performance. Star image restoration algorithms include classical methods such as inverse filtering, Wiener filtering, and constrained least squares filtering, whose core principle is to inverse the original star image through a degradation model; there are also algorithms based on brightness adaptive screening mechanism, which preferentially identify high signal-to-noise ratio streak star images to establish templates and compensate for dark stars in reverse. Another method is the dynamic window optimization algorithm, and the dynamic window adaptive algorithm developed by the Beihang team analyzes the motion parameters of the streak star image in real time, dynamically matches the size and shape of the detection window, and reconstructs the broken streak star image trajectory combined with the morphological break repair technology.

[0005] The current streak star image detection algorithm still has the following shortcomings: the signal-to-noise ratio of the streak star image decreases due to the dispersed distribution of pixels in a high dynamic environment, the noise energy increases nonlinearly with the expansion of the dynamic range in complex working conditions, and at the same time, various factors cause the streak star image tailing to be complex, and the traditional algorithm is difficult to achieve reliable streak star image detection. SUMMARY

[0006] The technical solution of the present application: in view of the problems of high noise sensitivity, complex tailing track leading to feature degradation and inability to accurately obtain high-precision star image position in the existing star point detection technology, a high dynamic streak star image detection method and device based on region growing algorithm are proposed, which significantly improves the streak star image detection accuracy under high dynamic conditions while ensuring real-time through anisotropic Gaussian filtering and seed region growing based on the tailing direction, and improves the robustness and accuracy of high dynamic streak star image detection in a noisy environment.

[0007] The technical solution of the present application:

[0008] A high dynamic streak star image detection method based on region growing algorithm, comprising the following steps:

[0009] S110, global threshold segmentation is performed on the input high dynamic star map to obtain a candidate seed region set;

[0010] S120, the candidate seed region is traversed, the direction information of each region is calculated by anisotropic Gaussian filtering, and the possibility of the existence of streak star image in the region is judged, if the possibility is greater than a preset threshold, growth is performed along the region direction, if the possibility is less than the preset threshold, growth is stopped;

[0011] S130, in the growth process, pixels meeting the spatial distribution characteristics of the tailing streak star image are marked until all candidate seed regions are traversed, and streak star image detection is completed.

[0012] A high dynamic streak star image detection device based on region growing algorithm, comprising:

[0013] A threshold segmentation module is configured to perform global threshold segmentation on the input high dynamic star map to obtain a candidate seed region set;

[0014] A growth judgment module is configured to traverse the candidate seed region, calculate the direction information of each region by anisotropic Gaussian filtering, and judge the possibility of the existence of streak star image in the region, if the possibility is greater than a preset threshold, growth is performed along the region direction, if the possibility is less than the preset threshold, growth is stopped;

[0015] A traversal module is configured to mark pixels meeting the spatial distribution characteristics of the tailing streak star image in the growth process until all candidate seed regions are traversed, and streak star image detection is completed.

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

[0017] The method can effectively improve the detection accuracy, reliability and efficiency of the streak star image. The potential streak star image in the star map is marked by the anisotropic region self-growth algorithm, noise interference is effectively suppressed, and complex streak scenes of different causes can be coped with. Meanwhile, by reducing unnecessary region scanning, the resource consumption of redundant calculation in star map processing is effectively reduced. In addition, the direction-based region self-growth method further enhances the accuracy and robustness of streak star image detection under complex conditions.

[0018] The method can improve the dynamic performance of the star tracker. By reducing a large amount of noise interference in star map processing, the method is suitable for various streak star images in complex conditions, and improves the dynamic performance of tracking and positioning. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the high-dynamic streak star image detection method based on the region growth algorithm of the present application;

[0020] Figure 2 is the original star map and the streak star image detection result in the specific embodiment, wherein (a) is the original star map, (b) is the segmented star map, (c) is the three-dimensional histogram of the original star map, and (d) is the three-dimensional histogram of the segmented star map. DETAILED DESCRIPTION

[0021] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and the present disclosure should not be limited by the embodiments described herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0022] Figure 1 is a flowchart of the high-dynamic streak star image detection method based on the region growth algorithm of the present application. As shown in Figure 1 , the method comprises the following steps:

[0023] S110, performing global threshold segmentation on the input high-dynamic star map to obtain a candidate seed region set;

[0024] S110 can include performing background rough denoising processing on the high-dynamic star map obtained by the imaging end by using a global threshold segmentation algorithm to obtain a star map after background filtering and partial noise removal.

[0025] S120, traversing the candidate seed region, calculating the direction information of each region by using anisotropic Gaussian filtering, and judging the possibility of the existence of the streak star image in the region. If the possibility is greater than a preset threshold, the region is grown along the direction; if the possibility is less than the preset threshold, the growth is stopped.

[0026] S120 can include:

[0027] S120-1, filtering to obtain the direction map and straight line map of each region;

[0028] S120-2, using a maximum aggregation algorithm, for each pixel point, retaining the maximum value in all anisotropic Gaussian kernel responses as the value of the result map;

[0029] S120-3, calculating the direction trend of the region of interest, judging whether there is a possibility of a streaky star object in the region, and if so, taking the region as a seed point to grow along the direction information of the region of interest;

[0030] S120-4, calculating the ratio of the upper and lower triangular weights based on the direction of the region, and constraining the growth direction along the direction of the region to make the growing region of interest located inside the trailing streaky star object;

[0031] S120-5, repeating steps S120-1 to S120-4, stopping when the growth termination condition is reached, and traversing all seed regions. The growth termination condition is that there is no star point pixel that can grow in the region of interest.

[0032] Specifically, step S120 can further include:

[0033] Step S120-1: constructing an anisotropic Gaussian kernel filter set with different variances and angles . Taking the star map bright spots after threshold segmentation as the center of the seed region, circumscribing a fixed size region of interest ROI, and filtering the region of interest ROI with the anisotropic Gaussian kernel filter;

[0034] wherein the anisotropic Gaussian kernel filter set with different variances and angles is constructed, and the normalized formula of the filter is:

[0035] ,

[0036] ,

[0037] In the formula, x and y represent image coordinates, the rotation matrix , , , , respectively represent scale, direction and adaptive anisotropy factor, and the anisotropy factor , is a robustness control scale, and the larger the value, the more obvious the stretching effect of the Gaussian kernel, wherein represents a two-dimensional integer coordinate, and respectively denote a scale set and a direction set. is a second-order anisotropic Gaussian kernel, is an anisotropic Gaussian kernel. Therefore, the parameters of the anisotropic Gaussian kernel are scale, direction and anisotropy factor.

[0038] The starlight points segmented by the threshold value are taken as the seed region center, a fixed-size region of interest ROI is circled, a discrete anisotropic Gaussian kernel filter is used to filter the initial growth region ROI, including:

[0039] ,

[0040] wherein, is the filtering result of the discrete anisotropic Gaussian kernel filter and the region of interest ROI.

[0041] Step S120-2: A maximum aggregation strategy is adopted, for each pixel point, the maximum value in all anisotropic Gaussian kernel responses is selected as the pixel value of the corresponding position of the output image. The anisotropic Gaussian kernel consistent with the direction of the streaked star image tail generates a more significant peak response, and therefore the direction information of each pixel point can be directly represented by the anisotropic Gaussian kernel generating the maximum response;

[0042] wherein, a maximum aggregation strategy is adopted, for each pixel point, the maximum value in all anisotropic Gaussian kernel responses is selected as the pixel value of the corresponding position of the output image, including:

[0043] ,

[0044] The anisotropic Gaussian kernel consistent with the direction of the streaked star image tail generates a more significant peak response, and therefore the direction information of each pixel point can be directly represented by the anisotropic Gaussian kernel generating the maximum response, that is, the direction map D of the streaked star image tail in the region of interest ROI is obtained in the following way:

[0045] .

[0046] Step S120-3: The main direction information d of the center point P of the region of interest ROI is taken as the direction trend of the region of interest ROI, the number of pixels satisfying the direction consistency is counted. When the number of pixels exceeds the preset direction consistency threshold, it is considered that there is a possibility of the existence of the streaked star image in the region of interest, the potential streaked star candidate area is marked in the region of interest ROI, and the iterative segmentation algorithm based on region growing is started;

[0047] The main direction information d of the center point P of the region of interest ROI is taken as the direction trend of the region of interest ROI, the number of pixels satisfying the direction consistency is counted, and the calculation formula is as follows:

[0048] ,

[0049] in, It's a direction or trend. , represents the set of pixels in the region along direction d, where Let be the matrix radius. This indicates rounding to the nearest integer. These are the pixel coordinates within the region of interest (ROI), and they are positive integers. This indicates the direction information of point p within the direction pattern. This is a logical value indicating whether the directional information of a point p within the Region of Interest (ROI) belongs to a directional trend. If the number of pixels meeting the criteria is greater than a preset threshold, it is considered that there is a possibility of a trailing stellar pattern within the ROI and is marked, and the region begins to grow.

[0050] Step S120-4: Based on directional feature analysis, through statistical analysis of the upper triangular region... Regions and Lower Triangle The number of pixels within a region that conform to the dominant directional trend is used to construct the directional weight ratio parameter. This parameter is used to dynamically adjust the expansion boundary of the seed region along the normal direction, while constraining and adjusting the center position of the region in real time to ensure that the growth process is always confined within the morphological boundary of the trailing stripes; the dominant directional trend is the directional information d of the center point P of the region of interest (ROI) as the dominant directional trend of this ROI.

[0051] Based on directional feature analysis, calculate the upper triangle within the ROI region. Regions and Lower Triangle The number of pixels within a region that conform to the main directional trend determines whether to grow along the seed point towards the normal direction, as shown in the following formula:

[0052] ,

[0053] ,

[0054] ,

[0055] In the formula, U and L represent the upper and lower triangular weights formed based on the directional trend of the region, respectively. It is an upper triangular weighted logical matrix. It's a direction or trend. It is a lower triangular weighted logical matrix.

[0056] Step S120-5: Repeat steps S120-1 to S120-4 until the termination condition is met, determine whether the pixel of the region of interest passed meets the characteristics of the streaked star image, and mark whether the growth belongs to the streaked star image or noise interference, until all seed regions are traversed.

[0057] S130, during the growth process, mark the pixels that meet the spatial distribution characteristics of the streaked star image, until all candidate seed regions are traversed, and the streaked star image detection is completed.

[0058] The method of the present application will be described in detail below through specific embodiments: Specific embodiments:

[0060] For Figure 2 The original star map shown in (a) is processed:

[0061] S1, calculate the global threshold of the whole map , wherein mean and std represent the mean and variance of the gray scale of the original map, respectively;

[0062] S2, calculate the discrete second-order anisotropic Gaussian kernel with a size of 5x5, and the angle ranges from 0° to 180°, and every 10 degrees takes a value, and the bright spots of the star map after threshold segmentation are used as the center of the seed region, and the maximum value aggregation algorithm is used to filter to obtain the direction map and straight line map of the region;

[0063] S3, determine the direction trend of the region based on the region center point p , , determine whether the series of pixels in the direction belong to the direction trend, and the number of pixels meeting the requirements is greater than the threshold value, i.e. 2 times rds;

[0064] S4, if it is considered that there is a streaked star image in the region, grow from the seed region along the normal direction, and adjust the position of the seed region according to the weight ratio of the upper and lower triangles in the direction;

[0065] S5, traverse all seed regions to obtain the final streaked star image detection result map, such as Figure 2 (b), Figure 2 (c) and (d) in the three-dimensional histogram quantitatively compare the original noise distribution and the extracted streaked star image.

[0066] In summary, the application discloses a high dynamic streak star detection method based on anisotropic region self-growth, comprising the following steps: performing global threshold segmentation on an input high dynamic star map to obtain a candidate seed region set, then traversing the candidate regions, calculating the direction information of the regions by anisotropic Gaussian filtering and judging the possibility of the existence of streak star in the regions, growing along the region direction if the possibility is greater than a preset threshold, and stopping growing if the possibility is less than the preset threshold, marking the pixels meeting the spatial distribution characteristics of the trailing streak star in the growing process until all the candidate regions are traversed, and completing the streak star detection.

[0067] The application further provides a high dynamic streak star detection device based on a region growing algorithm, comprising:

[0068] A threshold segmentation module is configured to perform global threshold segmentation on an input high dynamic star map to obtain a candidate seed region set.

[0069] A growth judgment module is configured to traverse the candidate seed regions, calculate the direction information of the regions by anisotropic Gaussian filtering and judge the possibility of the existence of streak star in the regions, grow along the region direction if the possibility is greater than a preset threshold, and stop growing if the possibility is less than the preset threshold.

[0070] A traversal module is configured to mark the pixels meeting the spatial distribution characteristics of the trailing streak star in the growing process until all the candidate seed regions are traversed, and complete the streak star detection.

[0071] In the specification provided herein, a large number of specific details are explained. However, it can be understood that the embodiments of the application can be practiced without these specific details. In some examples, well-known methods, structures and techniques are not shown in detail in order not to obscure the understanding of the specification.

[0072] Although the application is described in terms of a limited number of embodiments, those skilled in the art, with the benefit of the above description, will appreciate that other embodiments are possible within the scope of the application described herein. Moreover, it should be noted that the language used in the specification has primarily been chosen for readability and instructional purposes and can not have been chosen to convey an exclusive or exhaustive explanation of the application.

Claims

1. A high dynamic streak star image detection method based on a region growing algorithm, characterized in that, The method comprises the following steps: S110, globally thresholding the input high-dynamic star map to obtain a candidate seed region set; S120, traversing the candidate seed region, calculating the direction information of each region by anisotropic Gaussian filtering, and judging the possibility of the existence of a streaked star in the region, if the possibility is greater than a preset threshold, growing along the region direction; if less than, stopping growing; S130, in the growing process, marking the pixels meeting the spatial distribution characteristics of the trailing streaked star until all the candidate seed regions are traversed, completing the streaked star detection; Step S120 comprises: Step S120-1: constructing an anisotropic Gaussian kernel filter group with different variances σ and angles θ; taking the star map bright spots after thresholding as the seed region center, circumscribing a fixed-size region of interest ROI, and filtering the region of interest ROI by the anisotropic Gaussian kernel filter; Step S120-2: adopting a maximum value aggregation strategy, for each pixel point, selecting the maximum value in all anisotropic Gaussian kernel responses as the pixel value of the corresponding position of the output image; the anisotropic Gaussian kernel consistent with the trailing direction of the streaked star will generate a more significant peak response, so the direction information of each pixel point is directly represented by the anisotropic Gaussian kernel generating the maximum response; Step S120-3: taking the main direction information d of the region of interest ROI center point P as the direction trend of the region of interest ROI, and counting the number of pixels meeting the direction consistency; when the number of pixels exceeds a preset direction consistency threshold, it is considered that there is a possibility of the existence of a trailing streaked star in the region of interest, a potential trailing star candidate area is marked in the region of interest ROI, and an iterative segmentation algorithm based on region growing is started; Step S120-4: Based on the direction feature analysis, the number of pixels in the upper triangle region and the lower triangle region that conform to the main direction trend is counted, and a direction weight ratio parameter is constructed, which is used to dynamically control the expansion boundary of the seed region along the normal direction, while constraining and adjusting the center position of the region in real time, to ensure that the growth process is always limited within the morphological boundary range of the trailing stripe. The direction weight ratio parameter is used to dynamically control the expansion boundary of the seed region along the normal direction, while constraining and adjusting the center position of the region in real time, to ensure that the growth process is always limited within the morphological boundary range of the trailing stripe.​ Step S120-5: repeating steps S120-1 to S120-4 until the termination condition is met, judging whether the traversed region of interest pixel meets the trailing streaked star characteristics, so as to mark whether the growth belongs to the trailing of the streaked star or noise interference, until all the candidate seed regions are traversed.

2. The high dynamic range streak star detection method based on region growing algorithm according to claim 1, characterized in that, S110 comprises: performing background coarse denoising processing on the high-dynamic star map obtained by the imaging end by a global thresholding algorithm to obtain a star map after filtering out part of the background noise.

3. The high dynamic streak star detection method based on region growing algorithm according to claim 1, characterized in that, In step S120-1, a set of anisotropic Gaussian kernel filters with different variances and angles are constructed, and the normalized formula of the filter is as follows: ， ， wherein denote image coordinates, a rotation matrix , , , denote scale, orientation and adaptive anisotropy factor, respectively, and the anisotropy factor , is a robustness control scale, the larger its value the more pronounced the stretching effect of the Gaussian kernel, wherein denote two-dimensional integer coordinates, and denote a set of scales and a set of orientations, respectively, is a second order anisotropic Gaussian kernel, is an anisotropic Gaussian kernel.

4. The high dynamic range streak artifact detection method based on region growing algorithm according to claim 3, characterized in that, In step S120-1, taking the star map bright spots after thresholding as the seed region center, circumscribing a fixed-size region of interest ROI, and filtering the initial growing region ROI by the discrete anisotropic Gaussian kernel filter, comprising: ， wherein, is the filtering result of the discrete anisotropic Gaussian kernel filter and the region of interest ROI.

5. The high dynamic range streak artifact detection method based on region growing algorithm according to claim 4, characterized in that, In step S120-2, a maximum value aggregation strategy is adopted, for each pixel point, the maximum value in all anisotropic Gaussian kernel responses is selected as the pixel value of the corresponding position of the output image, comprising: 。 6. The high dynamic range streak artifact detection method based on region growing algorithm according to claim 5, characterized in that, In step S120-2, the anisotropic Gaussian kernel consistent with the trailing direction of the streaked star will generate a more significant peak response, so the direction information of each pixel point is directly represented by the anisotropic Gaussian kernel generating the maximum response, that is, the direction map D in the region of interest ROI is obtained by the following way: 。 7. The high dynamic range streak artifact detection method based on region growing algorithm according to claim 6, characterized in that, In step S120-3, the main direction information d of the center point P of the region of interest ROI is taken as the direction trend of the region of interest ROI, the number of pixels satisfying the direction consistency is counted, and the calculation formula is as follows: ， wherein, is a direction trend, represents a set of pixel points in the region of direction d, wherein is a matrix radius, represents rounding to the nearest integer, is a pixel coordinate within a region of interest (ROI), is a positive integer, represents direction information of a point p within a direction map, is a logical value indicating whether the direction information of the point p within the region of interest (ROI) belongs to the direction trend.

8. The high dynamic range streak star detection method based on region growing algorithm according to claim 7, characterized in that, In step S120-4, based on directional feature analysis, the upper triangle within the ROI region is calculated. Regions and Lower Triangle The number of pixels within a region that conform to the main directional trend determines whether to grow along the seed point towards the normal direction, as shown in the following formula: ， ， ， where U and L represent upper and lower triangular weights based on the region direction tendency, respectively, is an upper triangular weight logical matrix, is a main direction tendency, is a lower triangular weight logical matrix.

9. A high dynamic range streak artifact detection device based on a region growing algorithm, characterized in that, The application discloses a region growing algorithm-based high-dynamic streak star detection method and device. The threshold segmentation module is configured to perform global threshold segmentation on the input high-dynamic star map to obtain a candidate seed region set. The growth judgment module is configured to traverse the candidate seed region, calculate direction information of each region by using an anisotropic Gaussian filter, and judge a possibility of existence of the streak star in the region. The traversal module is configured to mark pixels satisfying a spatial distribution feature of the trailing streak star during the growth process until all the candidate seed regions are traversed, and streak star detection is completed.

Images