CN112815936B - Rapid all-sky-domain star map identification method and system for noise robustness - Google Patents
Rapid all-sky-domain star map identification method and system for noise robustness Download PDFInfo
- Publication number
- CN112815936B CN112815936B CN202011605130.XA CN202011605130A CN112815936B CN 112815936 B CN112815936 B CN 112815936B CN 202011605130 A CN202011605130 A CN 202011605130A CN 112815936 B CN112815936 B CN 112815936B
- Authority
- CN
- China
- Prior art keywords
- star
- distance
- mask
- template
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/02—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by astronomical means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Multimedia (AREA)
- Astronomy & Astrophysics (AREA)
- Automation & Control Theory (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a fast all-sky-domain star map identification method and a system for noise point robustness, wherein the star map identification method comprises the following steps: s1, selecting a main star to be identified in the star map; s2, constructing a neighbor candidate set; s3, acquiring a preprocessing star map; s4, acquiring a distance mask mode of the preprocessed star map; s5, calculating initial similarity, and screening out candidate templates for matching; s6, calculating the comprehensive similarity of the candidate templates; and S7, outputting the final recognition result. The star map recognition is carried out according to the method of the invention, so that the problem that a large number of star points in the field of view are converted to the outside of the field of view after the traditional method is decentralized can be solved, and the number of available fixed stars is increased; the problem of subsequent matching errors caused by high adjacent point selection error rate in the traditional method is solved; meanwhile, the robustness is very high under the condition that noise points such as missing stars or pseudo stars exist. In addition, the invention adopts non-iterative verification, and does not need to introduce additional star point identification for confirmation, thereby being more rapid and effective.
Description
Technical Field
The invention belongs to the technical field of star map identification, and particularly relates to a fast all-sky-domain star map identification method robust to noise points.
Background
The star sensor is a core component of autonomous navigation of the spacecraft, and realizes high-precision attitude measurement by observing fixed stars in the space. For astronomical navigation, a star can be regarded as an ideal point light source with certain spectral characteristics, which is located at infinity, is approximately static. If no prior attitude information is available, or the star sensor is restarted after a fault, the star sensor operates in a capture mode to perform all-day-domain star map identification. In the capturing mode, the star sensor firstly identifies the fixed star captured in the image and then calculates the attitude according to the position measurement value (stored in the fixed star catalogue) of the identified fixed star. The task of computing the pose can be done easily by the existing QUEST or triac methods, but the most challenging task is how to correctly identify stars in the captured image in the presence of noise such as missing stars, false stars, etc.
The existing star map identification method can be divided into two categories based on geometric matching and pattern identification. Geometric matching based methods utilize features such as angles and distances formed by a plurality of star points to construct and identify a star map database, either directly or indirectly by constructing polygons. The pattern recognition method is to match a specific pattern formed by a star point and other points around the star point with a pattern template pre-stored in template data to perform recognition, and a typical representative method is a grid method.
Compared with a method based on geometric matching, the method based on pattern recognition has the advantages that the template database occupies less memory, and is relatively insensitive to position noise, star point noise, and interference stars such as pseudo-stars and missing stars. However, the current methods based on pattern recognition also have some disadvantages: the selection of the main star to be identified enables a large number of star points in the visual field to be changed to the outside of the visual field, the number of available star points is reduced, the star points in the star map are not fully utilized, the mode information becomes sparse, and the subsequent mode matching is not facilitated; under the condition that noise exists, the adjacent point selection error rate is high, so that subsequent matching errors are caused; the extracted mode features cannot sufficiently reflect the space similarity of the star atlas, and the robustness to noise points is insufficient; in order to solve the problems of high error rate of selection of neighbor points and the like, the verification step needs to be completed by additionally introducing the identification of a plurality of star points in the star map, and the calculation complexity is high.
Disclosure of Invention
The invention mainly aims to provide a fast all-sky-domain star atlas identification method robust to noise points, and aims to solve the technical problem that the identification rate performance of the existing similar method is sensitive to the noise points.
In order to achieve the above object, the present invention provides a fast whole-sky-domain star atlas identification method robust to noise, which comprises the following steps:
step S1, selecting a main star to be identified in the star map: selecting a main star to be identified in the star map on the basis of star points extracted from the star map and under the condition of keeping the maximum number of star points in a view field;
step S2, constructing a neighbor candidate set: selecting a star point with the minimum distance to the main star as a first neighbor star from all star points which have the distance to the main star greater than a distance threshold and are smaller than the distance from the main star to the edge of a field of view, setting a distance error range, and adding all star points which have the distance to the main star within the distance error range as candidate neighbor stars into a neighbor star candidate set;
step S3, acquiring a preprocessing star map: translating and rotating the star map to enable the main star to translate to the center of a view field, and rotating to a 0-degree direction by taking the direction from the main star to the adjacent star as a direction reference to obtain a preprocessed star map;
step S4, obtaining a distance mask pattern of the preprocessed star map: performing distance mask feature extraction on the preprocessed star map to obtain a distance mask mode of the preprocessed star map;
step S5, calculating initial similarity, and screening out candidate templates for matching: performing initial similarity calculation on the distance mask pattern of the preprocessed star map and templates in a template database, screening N candidate templates with the highest similarity for subsequent matching, wherein N represents the number of the candidate templates, if the maximum initial similarity is higher than an initial threshold, directly outputting a corresponding template main star as an identification result, and rapidly ending the identification, otherwise, entering a step S6;
step S6, calculating the comprehensive similarity of the candidate templates: extracting logic mask patterns and distance mask patterns of the N candidate templates, firstly calculating logic similarity and distance similarity with the preprocessed star map, and then calculating comprehensive similarity;
step S7, outputting the final recognition result: and screening out a template with the maximum comprehensive similarity from the candidate templates, outputting the main star corresponding to the template with the maximum comprehensive similarity as an identification result if the corresponding comprehensive similarity is greater than a final selection threshold, and otherwise, performing recognition rejection on the current star map and finishing the identification.
Further, the detailed step of step S1 includes the following substeps S1.1 to S1.3:
s1.1, extracting all star points of a captured star map;
s1.2, selecting a point capable of reserving the most star points in a view field from the extracted star points as a main star to be identified;
and S1.3, calculating the distance from the main star to be identified to the edge of the field of view.
Further, the distance mask pattern in step S4 is a pattern extracted from the image and used for measuring local features of the distribution of the star points in the image, and the distance mask pattern of the preprocessed star map is obtained by: firstly, constructing a mask for a preprocessed star map, carrying out normalized shortest distance transformation on the preprocessed star map, and carrying out logic and operation on the mask and a normalized shortest distance transformation result; the mask is a binary image matrix consisting of 0 or 1; the normalized shortest distance transformation is to firstly carry out shortest distance transformation on the image and then carry out normalization processing.
Optionally, the mask construction method includes: setting all elements with a size k as a radius range as 1 by taking a pixel point where a star point is located as a center, and constructing a square matrix with dimensions of (2k +1) × (2k +1) and all elements of 1 as a mask; the value range of the mask size k is 20-80.
Further, the detailed step of step S5 includes the following substeps S5.1 to S5.3:
s5.1, performing initial similarity calculation on the distance mask mode of the preprocessed star map and a template in a template database;
the initial similarity is the similarity degree of the star point distribution of the preprocessed star map and the star point distribution of the template in the template database on the local features, and the calculation method comprises the following steps: firstly, carrying out logical AND operation on the distance mask mode and corresponding elements of the template to obtain an intermediate mode, and then carrying out summation operation on all elements of the intermediate mode;
s5.2, screening N candidate templates with the highest initial similarity, wherein N is the number of the candidate templates, and the value range of the number N of the candidate templates is 50-200;
s5.3, if the maximum initial similarity is higher than the initial threshold, directly outputting the corresponding template main star as an identification result, and finishing identification; the template master star refers to a master star corresponding to a template in the template database.
Further, the detailed step of step S6 includes the following substeps S6.1 to S6.5:
s6.1, calculating a logic mask mode of the candidate template;
the logical mask mode is a mode which is extracted from the image and used for measuring the global characteristics of the star point distribution in the image; the calculation method of the logic mask mode of the candidate template comprises the following steps: sliding the mask on the candidate template to enable the mask to cover star points of the candidate template, enabling the star points to be located at the center of the mask, and assigning the pixel points covered by the mask to be 1 and the pixel points not covered by the mask to be 0;
s6.2, calculating a distance mask mode of the candidate template;
the calculation method of the distance mask pattern of the candidate template comprises the following steps: firstly, constructing a mask for a candidate template, carrying out normalized shortest distance transformation on the candidate template, and carrying out logical AND operation on the mask and a normalized shortest distance transformation result;
s6.3, calculating the logic similarity between the logic mask mode of the candidate template and the preprocessed star map;
the method for calculating the logic similarity between the logic mask pattern of the candidate template and the preprocessed star map comprises the following steps: performing logical AND operation on the logical mask pattern of the candidate template and corresponding elements of the preprocessed star map to obtain an intermediate pattern, and then performing summation operation on all elements of the intermediate pattern;
s6.4, calculating the distance similarity between the distance mask mode of the candidate template and the preprocessed star map;
the method for calculating the distance similarity between the distance mask pattern of the candidate template and the preprocessed star map comprises the following steps: performing logical AND operation on the distance mask pattern of the candidate template and corresponding elements of the preprocessed star map to obtain an intermediate pattern, and then performing summation operation on all elements of the intermediate pattern;
s6.5, calculating the comprehensive similarity of the candidate template and the preprocessed star map by using the logic similarity and the distance similarity of the candidate template and the preprocessed star map; the comprehensive similarity is the comprehensive similarity of the distribution of the star points of the preprocessed star map and the distribution of the star points of the template in the template database on the global features and the local features.
Optionally, the distance error range in step S2 is [ dr min ,dr min +d]Wherein dr is min The distance between the first adjacent star and the main star, and d is a set distance error value.
Furthermore, the present invention also provides a noise robust fast all-sky-field star atlas identification system, comprising a computer device, the computer being programmed or configured to perform the steps of the noise robust fast all-sky-field star atlas identification method, or the computer device having stored in its memory a computer program programmed or configured to perform the noise robust fast all-sky-field star atlas identification method.
Furthermore, the invention also provides a satellite with the fast whole-sky-domain star map identification system robust to the noise point.
Furthermore, the present invention also provides a computer readable storage medium having stored thereon a computer program programmed or configured to perform the fast noise-robust all-sky-domain star atlas identification method.
Compared with the prior art, the technical scheme of the invention has the following beneficial technical effects:
1. by adopting the hierarchical identification structure of first candidate and then final selection and the comprehensive similarity matching method combining the logic similarity and the distance similarity, the robustness of the star map identification method to noise points such as missing stars, fake stars and the like is greatly improved, and the star map identification rate under severe conditions is improved.
2. By adopting the technical means of selecting the main star to be identified in the star map under the condition of keeping the maximum number of star points in the field of view, the defect of insufficient utilization of the star points in the star map by the conventional similar method is overcome, the utilization rate of the star points and the utilization rate of star map information are improved, and the subsequent star map identification process is facilitated.
3. By taking the star point which is out of the radius of the main star in a certain range, is closest to the main star and has a distance less than the distance from the main star to the edge as the first adjacent star, all the star points in the certain distance range of the first adjacent star are added into the adjacent star candidate set, the adjacent star selection error rate is greatly reduced, and a foundation is laid for the subsequent star map identification process.
In addition, the invention also embeds the independent verification step of the existing similar method into the recognition result output in the step S7, abandons the traditional low-efficiency iterative verification method, and further invents a non-iterative verification method, which has the advantages of rapidness and effectiveness without introducing additional star point recognition for confirmation.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
FIG. 1 is a schematic diagram of a star sensor capturing a star map with a false star and a missing star, wherein (a) is an ideal star map, and (b) is an actual star map with a false star and a missing star problem;
FIG. 2 is a problem presentation diagram of a conventional method for transforming a large number of star points in a field of view to the outside of the field of view after decentralization, thereby resulting in low utilization rate of the star points, wherein (a) is a star map of an original field of view, and (b) is a star map obtained by translating star points to be identified to the center of the field of view;
FIG. 3 is a schematic illustration of a neighbor not being in the field of view resulting in a neighbor selection error, wherein (a) is the case of a neighbor selection error and (b) is a representation that the correct neighbor is actually out of the field of view;
FIG. 4 is a schematic diagram of close star point distances and star point errors resulting in neighbor selection errors, wherein (a) is a case of neighbor selection errors and (b) is shown for a case where the distances from the centers of the correct neighbor and the incorrect neighbor are close;
FIG. 5 is a schematic diagram of a preprocessed star map obtained by decentralizing and rotation correcting original star points;
FIG. 6 is a diagram illustrating normalized shortest distance transform and mask coverage;
FIG. 7 is a schematic diagram of a distance mask pattern and different size distance masks, wherein (a) is the distance mask pattern and (b) is the different size distance masks;
FIG. 8 is a schematic diagram of the AND calculation in the similarity calculation;
FIG. 9 is a comparison graph of recognition performance tests using the method of the present invention and a conventional method;
FIG. 10 is a flow chart of the fast noise robust all-sky-domain star atlas identification method of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that all scalar and vector names in the embodiments of the present invention, such as "integrated similarity" value, are set for convenience of description, and some variables in the embodiments of the present invention should not be understood as indicating or implying their design tendencies. For clarity, the physical meanings of the symbols used in this specification are as shown in table 1 below.
TABLE 1 symbols and their meanings used in the present invention
In the present invention, unless otherwise explicitly specified and defined, "translation", "rotation", and the like terms used to describe relative spatial positional relationships are to be interpreted broadly. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
The fast whole-sky-domain star map identification method robust to noise points is different from the traditional star map identification method, and the core thought of the fast whole-sky-domain star map identification method comprises 7 steps shown in figure 10. The following describes in detail the steps and flows of the fast whole-sky-domain star map identification method robust to noise provided by the present invention and the beneficial effects obtained by using the method with reference to fig. 1 to 10.
Step S1, selecting a main star to be identified in the star map; specifically, the main star to be identified in the star map is selected on the basis of the star points extracted from the star map under the condition of keeping the maximum number of star points in the field of view.
Step S1 specifically includes the following substeps S1.1 to S1.3:
s1.1, extracting and capturing all star points I of the star map S through operations such as median filtering, binarization, morphological closed operation, 8-connected region and the like. These star points include problems of position error, false star, and missing star, please refer to fig. 1. In FIG. 1, (a) is an ideal star map, and (b) is an actual star map with the problems of pseudo-stars and missing stars;
s1.2, selecting a point capable of reserving the most star points in the field of view from the extracted star points I as a main star r to be identified.
In the conventional method for selecting a point to be identified, a point closest to the center of a field of view is selected as the point to be identified, but a large number of star points may be transformed to the outside of the field of view, resulting in a problem of low star point utilization rate, please refer to fig. 2; in fig. 2, (a) is the original view field star map, and (b) is the star map obtained by translating the star point to be identified to the center of the view field.
S1.3, calculating the distance bo from the main star r to be identified to the edge of the field of view.
See fig. 3, where (a) is the case of an error in selection for a neighboring star, and (b) is the presentation that the correct neighboring star is actually outside the field of view. If the distance dr of the neighboring star from the primary star is greater than the distance bo of the primary star r to the edge of the field of view, it may cause the neighboring star not to be in the field of view, resulting in a problem of a neighboring star selection error. Here the calculated distance bo of the dominant star r to the edge of the field of view. And screening the conditions for selection of the subsequent adjacent star.
Step S2, constructing a neighbor candidate set; selecting a star point with the minimum distance to the main star as a first neighbor star from all star points which have the distance to the main star greater than a distance threshold and are smaller than the distance from the main star to the edge of the field of view, setting a distance error range, and adding all star points which have the distance to the main star within the distance error range as candidate neighbor stars into a neighbor star candidate set.
The specific process of step S2 is as follows:
setting the distance threshold as radius b, selecting a star point with the minimum distance dr from the main star r as a first adjacent star from all star points of which the distance dr from the main star r is greater than radius b and less than the distance bo from the main star to the edge of the field of view, and recording the distance dr from the first adjacent star to the main star r as the distance dr min ；
Setting the range of the distance error as [ dr min ,dr min +d]Wherein d is a set distance error value, and all the star points of which the distance dr to the main star is within the distance error range satisfy dr e [ dr ∈ [) min ,dr min +d]All the star points of the relationship, as candidate neighbor stars, are added to the neighbor star candidate set nbs.
See fig. 4, where (a) is the case where an error is selected for a neighbor and (b) is the case where the distance from the center is close for the correct neighbor to the erroneous neighbor. The selection of the neighboring star is also wrong due to the fact that the distances of the star points are almost the same and the star point errors, so that possible options need to be considered when the neighboring star is selected, and the possible options need to be added into a set. And finally determining the adjacent star through subsequent matching.
Step S3, acquiring a preprocessing star map; translating and rotating the star map to enable the main star to translate to the center of a view field, and rotating to a 0-degree direction by taking the direction from the main star to the adjacent star as a direction reference to obtain a preprocessed star map;
the specific process of step S3 is as follows: taking a main star r as a translation reference and a vector from the main star r to an adjacent star nbs as a direction reference, carrying out translation and rotation correction on star points I of the star map, so that the main star r is positioned at the center of a field of view, and the vector from the main star r to the adjacent star nbs is positioned in the direction of 0 degree, obtaining corrected star points in the field of view, and forming a preprocessed star map I s 。
Referring to fig. 5, the original star points are decentered and rotation-corrected to obtain a schematic diagram of the preprocessed star map, and since the templates are generated according to the predetermined rule, the current star map needs to be converted according to the predetermined rule so as to perform similarity matching with the templates.
Let V denote the direction in which the principal star r points to the nearest star point nbs base Then, then
Wherein the content of the first and second substances,
is the coordinate of the nearest star point nbs,
is the coordinate of the primary star r.
The direction rotation angle can be calculated as
The rotation matrix can be calculated as
And taking the direction as a reference direction for capturing the direction rotation correction of the star map and the template. Then the star map is translated so that the main star r is located at the center of the image, and the vector V is translated shift Can be calculated as follows
Wherein w and h are the width and height of the star map respectively,
is the coordinate of the primary star r.
The coordinates of the translated star points are
Wherein the content of the first and second substances,
is the coordinates of the star point before translation,
are the coordinates of the translated star points.
The star map is subjected to rotation correction according to the reference direction, and the coordinates of the rotated star points can be obtained as
Wherein the content of the first and second substances,
is the coordinates of the star points after the translation,
is the coordinates of the rotated star point, and w and h are the width and height of the star map respectively.
The star points still located within the field of view of the star map are retained:
wherein the content of the first and second substances,
is the coordinates of the retained star point, j is 1,2, …, N r ，N r The number of star points retained.
The preprocessed star atlas I is obtained after the decentralization and the reference direction correction s 。
Step S4, obtaining a distance mask mode of the preprocessed star map; and performing distance mask feature extraction on the preprocessed star map to obtain a distance mask mode of the preprocessed star map. The distance mask pattern is a pattern extracted from the image for measuring the local feature of the star point distribution in the image, which contains the local distance information.
The specific process of step S4 is as follows:
computing preprocessed star map I s Distance mask pattern di 1. The method for acquiring the distance mask mode of the preprocessed star map comprises the following steps: firstly, a mask is constructed for the preprocessed star map, the preprocessed star map is subjected to normalized shortest distance transformation, and then the mask and the normalized shortest distance transformation result are subjected to logic and operation.
The simple grid can only use 1 or 0 to represent whether star points exist in the grid, but cannot represent the distance between the star points and the grid, and the local spatial similarity of the star map is ignored. To solve this problem, a mask of a certain size is constructed and the star map I is preprocessed s The shortest distance of (2) transforms the profile. The distance mask pattern construction method described below may be applied to the construction of the distance mask pattern of the preprocessed star map, or the distance mask pattern construction of the candidate template described in S6.2 of step S6.
The mask is a binary image matrix consisting of 0 or 1, and is constructed to represent the error range of the positions of the star points. In this embodiment, the specific construction method of the mask for defining the pixel point where each star point in the image is located and the neighborhood thereof is as follows: setting all elements taking a pixel point where a star point is located as a center and taking a size k as a radius range as 1, thereby obtaining a square matrix 1 with dimensions of (2k +1) × (2k +1) and all elements of 1 (2k+1)×(2k+1) And it is used as the mask of the present embodiment, where k is the mask size, as shown in fig. 6; fig. 6 shows a mask diagram when k is 2. In the present embodiment, the mask size k ranges from 20 to 80.
The mask construction method described above may be used to construct the logical mask patterns of the candidate templates described in S6.1 of step S6.
The normalized shortest distance transformation is to firstly carry out shortest distance transformation on the image and then carry out normalization processing. The shortest distance transform is a mathematical transform commonly used in the field of image processing, and is specifically defined in the present embodiment as: in all pixel point sets omega represented by an image, for any pixel point p contained in the image, the shortest distance of the pixel point p can be occupied by all star points in the imageIs a subset omega of the pixel points c The definition is as follows:
D(p)＝min{d(p,q)|q∈Ω c } (8)
wherein D (p) is the shortest distance of p points, p is a pixel point which we want to calculate the shortest distance, q represents a pixel point where a star point is located in the image, and belongs to a subset omega formed by the pixel points occupied by all star points in the image c . The distance is defined herein as the Euclidean distance, i.e. the Euclidean distance
d(p,q)＝(p x -q x ) 2 +(p y -q y ) 2 (9)
The benefit of this distance definition is that it is invariant to translation and rotation.
Then normalizing the shortest distance, specifically, normalizing the shortest distance conversion value of the pixel point closest to the pixel where the star point is located, namely the pixel where the star point is located, to 1 by using the dimension k of the mask, sequentially decreasing the shortest distance conversion values of the surrounding star points along with the increase of the distance, wherein the shortest distance conversion value is 0 when the distance reaches the radius k, and the shortest distance conversion value is less than 0 when the distance exceeds the radius k, so that the normalization formula can be written as
Wherein D is s (p) is the normalized shortest distance transform value of p points, D (p) is the shortest distance of p points, p is the pixel point we want to compute the shortest distance, and k is the mask size.
By performing normalized shortest distance transformation on the image, that is, calculating a normalized shortest distance transformation value of each pixel point in the image, a profile map of the shortest distance transformation can be obtained, as shown in fig. 6. It can be known that the normalized shortest distance transformation value measures the possibility that the current pixel point is the closest star point to the current pixel point, and if the p point is just at the pixel point of the star point, D is s (p) is 1, representing that the probability is maximized.
And sliding the mask with a certain size on the contour map with the shortest distance conversion to cover each star point of the contour map with the shortest distance conversion, wherein the pixel point where the star point is located is positioned at the center of the mask with the certain size. The covering is defined as logical and operation, and the value not covered to the pixel point becomes 0. Through the process, a distance mask pattern of the image is obtained.
As shown in fig. 7 (a), the image is subjected to normalized shortest distance transformation and mask coverage to construct a distance mask pattern capable of measuring the probability that a pixel point becomes a star point in a star point and its neighborhood. As shown in fig. 7 (b), a distance mask pattern having a center value of 1 represents the highest distance similarity, and surrounding pixels have values decreasing with increasing distance from the center, which also represents the decreasing distance similarity, and the similarity is 0 after exceeding the mask size k. The distance mask can thus be approximated as a function of distance, but it is more computationally efficient. In addition, different star point distance metric masks can be calculated according to the size of the star map.
Step S5, calculating initial similarity, and screening out candidate templates for matching: and (4) performing initial similarity calculation on the distance mask pattern of the preprocessed star atlas and templates in the template database, screening N candidate templates with the highest similarity, performing subsequent matching, wherein N represents the number of the candidate templates, directly outputting a corresponding template main star as a recognition result if the maximum initial similarity is higher than an initial threshold, and rapidly finishing recognition, otherwise, entering the step S6.
The initial similarity is the similarity degree of the star point distribution of the preprocessed star map and the star point distribution of the template in the template database on the local features. The template master star is a master star corresponding to the templates in the template database, and each different template corresponds to a different master star.
Step S5 includes the following substeps S5.1 to S5.3:
s5.1, performing initial similarity calculation on the distance mask pattern di1 of the preprocessed star map and the templates in the template library database SP.
Referring to fig. 8, the initial similarity calculation is performed by anding the distance mask pattern di1 of the preprocessed star map with the template. Specifically, the method for calculating the initial similarity includes: the distance mask pattern di1 is first logically anded with the corresponding elements of the template to obtain an intermediate pattern, and then all the elements of the intermediate pattern are summed.
Assume distance mask pattern is pat r The template database formed by all navigation stars in the navigation star table is SP ═ { pat ═ i And (4) matching the processed star map and the template based on AND, namely calculating the following formula:
wherein, match (pat) r ,pat i ) Preprocessing the star map distance mask pattern as pat r And template pat i The similarity of (c).
S5.2, screening out N candidate templates cds with highest initial similarity. N represents the number of candidate templates; the value range of N is 50-200; in this embodiment, 100N are used.
And taking 100 template reference stars with the maximum matching degree as candidate identification results of the main star to be identified for subsequent processing.
And S5.3, if the maximum initial similarity is higher than the initial threshold th1, directly outputting the corresponding template main star as a recognition result, and finishing recognition. The initial threshold th1 is an initial similarity threshold set for directly identifying whether the current star map can be identified, and in this embodiment, the initial threshold th1 is obtained by numerical calculation and optimization.
Step S6, calculating the comprehensive similarity of the candidate templates; extracting logic mask patterns and distance mask patterns of the N candidate templates, firstly calculating logic similarity and distance similarity with the preprocessed star map, and then calculating comprehensive similarity; the logical mask mode is a mode which is extracted from the image and used for measuring the global characteristics of the star point distribution in the image, and the logical mask mode contains the geometric occlusion information of the star point global distribution; the comprehensive similarity is the comprehensive similarity of the distribution of the star points of the preprocessed star map and the distribution of the star points of the template in the template database on the global features and the local features.
Step S6 includes the following substeps S6.1 to S6.5:
s6.1, the logical mask pattern di2 of the candidate template cds is calculated.
The logical mask patterns of the candidate templates are constructed using the sized mask construction method described in step S4. Specifically, the method for acquiring the logic mask pattern of the candidate template comprises: sliding a mask with a certain size on the candidate template, enabling the mask with the certain size to cover star points of the candidate template, enabling the star points to be located at the center of the mask, and assigning the pixel points covered by the mask to be 1 and the pixel points not covered by the mask to be 0; this results in the logical mask pattern di2 of the candidate template cds.
S6.2, the distance mask pattern di3 of the candidate template cds is calculated.
The method for acquiring the distance mask mode of the candidate template comprises the following steps: firstly, constructing a mask for a candidate template, carrying out normalized shortest distance transformation on the candidate template, and carrying out logical AND operation on the mask and a normalized shortest distance transformation result;
the construction of the distance mask patterns di3 for the candidate template cds is the same as the distance mask pattern construction method described in step S4, except that the object is no longer the preprocessed star map I s But rather the candidate template cds. The distance mask pattern di3 of the constructed candidate template cds is also similar to that of fig. 7.
S6.3, calculating the logical mask pattern di2 and the preprocessed Star map I of the candidate template cds s The logical similarity sc 1.
Logical mask pattern di2 and preprocessed star map I for candidate template cds s The calculation of the logical similarity sc1 is similar to the calculation method of the initial similarity described in step S5, and is performed by performing the logical mask pattern di2 of the candidate template cds and preprocessing the star atlas I s And is computationally implemented. Specifically, the logical mask pattern di2 of the candidate template cds is first combined with the preprocessed star map I s Performing logical AND operation on corresponding elements to obtain an intermediate mode, and performing logical AND operation on the intermediate modeAll elements are summed. Reference is made specifically to formula (11).
S6.4, calculating distance mask pattern di3 of candidate template cds and preprocessing star atlas I s Distance similarity sc 2.
Distance mask pattern di3 and preprocessed star map I for candidate template cds s The calculation of the logical similarity sc2 is similar to the calculation of the initial similarity described in step S5, and is performed by performing the distance mask pattern di3 of the candidate template cds and preprocessing the star map I s And is computationally implemented. Specifically, the distance mask pattern di3 of the candidate template cds is first combined with the preprocessed star map I s And performing logical AND operation on the corresponding elements to obtain an intermediate mode, and then performing summation operation on all elements of the intermediate mode. Reference may be made specifically to formula (11).
S6.5, calculating candidate template cds and preprocessing star atlas I s The overall similarity ts of (a).
Candidate templates cds and preprocessed star maps I s The overall similarity ts is defined by the logical similarity sc1 and the distance similarity sc2, and the priorities of the logical similarity sc1 and the distance similarity sc2 are taken into consideration. The logic similarity sc1 takes precedence over the distance similarity sc 2: if the logic similarity sc1 of the candidate template cds is larger, the comprehensive similarity ts of the candidate template cds is also larger; if the logic similarity sc1 of the candidate template cds is smaller, the comprehensive similarity ts is also smaller; if the logical similarity sc1 of the candidate templates cds is equal, the distance similarity sc2 of the candidate templates cds is compared, and if the distance similarity sc2 is larger, the comprehensive similarity ts is larger. Thus, candidate templates cds and preprocessed star maps I s Is defined as the overall similarity ts
ts＝1000×sc1+sc2 (12)
This ensures that the logical similarity sc1 takes precedence over the distance similarity sc 2. The comprehensive similarity ts is defined in order to change the 2 comparison operations for the logical similarity sc1 and the distance similarity sc2 into 1 comparison operation for the comprehensive similarity, thereby optimizing the calculation efficiency.
Step S7, outputting the final recognition result: and screening out a template with the maximum comprehensive similarity from the candidate templates, if the corresponding comprehensive similarity is greater than a final selection threshold value, outputting a main star corresponding to the template with the maximum comprehensive similarity as an identification result, otherwise, performing recognition rejection on the current star map, and finishing the identification.
The specific steps of step S7 are as follows:
firstly, screening out the candidate template main star id with the maximum comprehensive similarity ts, wherein the comprehensive similarity is sc.
And then, if the comprehensive similarity sc is greater than a final selection threshold th2, outputting the template main star id as an identification result, otherwise, performing recognition rejection on the current star map, and ending the identification. The final selection threshold th2 is a comprehensive similarity threshold set for determining whether the current star map can be successfully identified, and in this embodiment, the final selection threshold th2 is obtained by numerical calculation and optimization.
In order to verify the effectiveness of the method, the performance of the method of the invention and the performance of the traditional method are tested through simulation, a benchmark result is shown in fig. 9, an HDSC curve in fig. 9 is the result of the method of the invention, GMV represents a geometric voting algorithm, STOD represents an optimal database search tree algorithm, and Polarstar represents a polar star algorithm. It can be seen that the method provided by the invention still keeps the recognition rate of the star map with the noise points at a higher level, and has good robustness to the noise points.
By combining the above analysis, the flow chart of the noise robust fast all-sky-domain star map identification method of the present invention is shown in fig. 10. According to the method, by means of selection of the main star to be identified, most star points in the star map are reserved in the subsequent identification process, and the utilization rate of the star points in the star map is improved; through the construction of a neighbor point candidate set, the error rate of selection of the neighbor points of the star map is reduced; a distance mask is designed, so that the spatial relation among the grid points of the star map is fully utilized, and the robustness of star point position errors is improved; through the comparison of the comprehensive similarity of the logic mask mode and the similarity of the distance mask mode, the recognition result or the recognition rejection result can be directly output, and the recognition point is not required to be additionally introduced for verification, so that the method is quick and effective.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent expected posture designs made by using the contents of the present specification and the attached drawings, or other related technical fields directly/indirectly using the inventive concept are included in the scope of the present invention.
Claims (8)
1. A fast all-sky-domain star atlas identification method robust to noise points is characterized by comprising the following steps:
step S1, selecting a main star to be identified in the star map: selecting a main star to be identified in the star map on the basis of star points extracted from the star map and under the condition of keeping the maximum number of star points in a view field;
step S2, constructing a neighbor candidate set: selecting a star point with the minimum distance to the main star as a first neighbor star from all star points which have the distance to the main star greater than a distance threshold and are smaller than the distance from the main star to the edge of a field of view, setting a distance error range, and adding all star points which have the distance to the main star within the distance error range as candidate neighbor stars into a neighbor star candidate set;
step S3, acquiring a preprocessing star map: translating and rotating the star map to enable the main star to translate to the center of a view field, and rotating to a 0-degree direction by taking the direction from the main star to the adjacent star as a direction reference to obtain a preprocessed star map;
step S4, obtaining a distance mask pattern of the preprocessed star map: performing distance mask feature extraction on the preprocessed star map to obtain a distance mask mode of the preprocessed star map;
step S5, calculating initial similarity, and screening out candidate templates for matching: performing initial similarity calculation on the distance mask pattern of the preprocessed star map and templates in a template database, screening N candidate templates with the highest similarity for subsequent matching, wherein N represents the number of the candidate templates, if the maximum initial similarity is higher than an initial threshold, directly outputting a corresponding template main star as an identification result, and rapidly ending the identification, otherwise, entering a step S6;
step S6, calculating the comprehensive similarity of the candidate templates: extracting logic mask patterns and distance mask patterns of the N candidate templates, firstly calculating logic similarity and distance similarity with the preprocessed star map, and then calculating comprehensive similarity;
step S7, outputting the final recognition result: screening out a template with the maximum comprehensive similarity from the candidate templates, if the corresponding comprehensive similarity is larger than a final selection threshold value, outputting a main star corresponding to the template with the maximum comprehensive similarity as an identification result, otherwise, performing recognition rejection on the current star map, and finishing the identification;
the distance mask pattern in step S4 is a pattern extracted from the image and used for measuring local features of the distribution of the star points in the image, and the distance mask pattern of the preprocessed star map is obtained by: firstly, constructing a mask for a preprocessed star map, carrying out normalized shortest distance transformation on the preprocessed star map, and carrying out logic and operation on the mask and a normalized shortest distance transformation result; the mask is a binary image matrix consisting of 0 or 1; the normalized shortest distance transformation is to firstly carry out shortest distance transformation on the image and then carry out normalized processing;
the detailed step of step S6 includes the following substeps S6.1 to S6.5:
s6.1, calculating a logic mask mode of the candidate template;
the logical mask mode is a mode which is extracted from the image and used for measuring the global characteristics of the star point distribution in the image; the calculation method of the logic mask mode of the candidate template comprises the following steps: sliding the mask on the candidate template to enable the mask to cover star points of the candidate template, enabling the star points to be located at the center of the mask, and assigning the pixel points covered by the mask to be 1 and the pixel points not covered by the mask to be 0;
s6.2, calculating a distance mask mode of the candidate template;
the calculation method of the distance mask pattern of the candidate template comprises the following steps: firstly, constructing a mask for a candidate template, carrying out normalized shortest distance transformation on the candidate template, and carrying out logical AND operation on the mask and a normalized shortest distance transformation result;
s6.3, calculating the logic similarity between the logic mask mode of the candidate template and the preprocessed star map;
the method for calculating the logic similarity between the logic mask pattern of the candidate template and the preprocessed star map comprises the following steps: performing logical AND operation on the logical mask pattern of the candidate template and corresponding elements of the preprocessed star map to obtain an intermediate pattern, and then performing summation operation on all elements of the intermediate pattern;
s6.4, calculating the distance similarity between the distance mask mode of the candidate template and the preprocessed star map;
the method for calculating the distance similarity between the distance mask pattern of the candidate template and the preprocessed star map comprises the following steps: performing logical AND operation on the distance mask pattern of the candidate template and corresponding elements of the preprocessed star map to obtain an intermediate pattern, and then performing summation operation on all elements of the intermediate pattern;
s6.5, calculating the comprehensive similarity of the candidate template and the preprocessed star map by using the logic similarity and the distance similarity of the candidate template and the preprocessed star map; the comprehensive similarity is the comprehensive similarity of the distribution of the star points of the preprocessed star map and the distribution of the star points of the template in the template database on the global features and the local features.
2. The method for fast full-sky-range star atlas identification robust to noise point as claimed in claim 1, wherein the detailed step of step S1 includes the following substeps S1.1-S1.3:
s1.1, extracting all star points of a captured star map;
s1.2, selecting a point capable of reserving the most star points in a view field from the extracted star points as a main star to be identified;
and S1.3, calculating the distance from the main star to be identified to the edge of the field of view.
3. The method for fast whole-day-domain star atlas identification robust to noise according to claim 1, wherein the mask is constructed by: setting all elements with a size k as a radius range as 1 by taking a pixel point where a star point is located as a center, and constructing a square matrix with dimensions of (2k +1) × (2k +1) and all elements of 1 as a mask; the value range of the mask size k is 20-80.
4. The method for fast whole-sky-range star atlas identification robust to noise point as claimed in claim 1, wherein the detailed step of step S5 includes the following substeps S5.1-S5.3:
s5.1, performing initial similarity calculation on the distance mask mode of the preprocessed star map and a template in a template database;
the initial similarity is the similarity degree of the star point distribution of the preprocessed star map and the star point distribution of the template in the template database on the local features, and the calculation method comprises the following steps: firstly, carrying out logical AND operation on the distance mask mode and corresponding elements of the template to obtain an intermediate mode, and then carrying out summation operation on all elements of the intermediate mode;
s5.2, screening N candidate templates with the highest initial similarity, wherein N is the number of the candidate templates, and the value range of the number N of the candidate templates is 50-200;
s5.3, if the maximum initial similarity is higher than the initial threshold, directly outputting the corresponding template main star as an identification result, and finishing identification; the template master star refers to a master star corresponding to a template in the template database.
5. The method for fast whole-day-domain star atlas identification robust to noise point of claim 1, wherein the range of distance error in step S2 is [ dr min ,dr min +d]Wherein dr is min The distance between the first adjacent star and the main star, and d is a set distance error value.
6. A noise robust fast whole-sky-field star atlas identification system comprising a computer device programmed or configured to perform the steps of the noise robust fast whole-sky-field star atlas identification method of any one of claims 1-5, or a computer program stored in a memory of the computer device and programmed or configured to perform the noise robust fast whole-sky-field star atlas identification method of any one of claims 1-5.
7. A satellite with a fast whole-sky-field star map identification system robust against noise as claimed in claim 6.
8. A computer readable storage medium having stored thereon a computer program programmed or configured to perform the noise point robust fast all-sky-domain star map identification method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011605130.XA CN112815936B (en) | 2020-12-29 | 2020-12-29 | Rapid all-sky-domain star map identification method and system for noise robustness |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011605130.XA CN112815936B (en) | 2020-12-29 | 2020-12-29 | Rapid all-sky-domain star map identification method and system for noise robustness |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112815936A CN112815936A (en) | 2021-05-18 |
| CN112815936B true CN112815936B (en) | 2022-08-02 |
Family
ID=75854652
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011605130.XA Active CN112815936B (en) | 2020-12-29 | 2020-12-29 | Rapid all-sky-domain star map identification method and system for noise robustness |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112815936B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113566815B (en) * | 2021-08-16 | 2022-07-01 | 北京航空航天大学 | Construction method and device of star map recognition navigation triangle library |
| CN114359696A (en) * | 2022-01-07 | 2022-04-15 | 博微(宁波)新技术有限公司 | Weather map feature type identification and similarity matching system and method |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100504300C (en) * | 2007-01-22 | 2009-06-24 | 北京航空航天大学 | Method for recognizing non-gauged star map |
| US8131074B2 (en) * | 2008-09-11 | 2012-03-06 | Raytheon Company | Specific emitter identification using histogram of oriented gradient features |
| CN106289240A (en) * | 2016-08-18 | 2017-01-04 | 中国人民解放军国防科学技术大学 | A kind of two step coupling method for recognising star map based on primary |
| CN107816987B (en) * | 2017-09-12 | 2020-07-31 | 北京航空航天大学 | Star pattern recognition method based on spider web pattern and convolutional neural network |
-
2020
- 2020-12-29 CN CN202011605130.XA patent/CN112815936B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112815936A (en) | 2021-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021072879A1 (en) | Method and apparatus for extracting target text in certificate, device, and readable storage medium | |
| CN110032998B (en) | Method, system, device and storage medium for detecting characters of natural scene picture | |
| US9396583B2 (en) | Method of modelling buildings on the basis of a georeferenced image | |
| US8285056B2 (en) | Method and apparatus for computing degree of matching | |
| JP5468332B2 (en) | Image feature point extraction method | |
| CN112815936B (en) | Rapid all-sky-domain star map identification method and system for noise robustness | |
| US20130089260A1 (en) | Systems, Methods, and Software Implementing Affine-Invariant Feature Detection Implementing Iterative Searching of an Affine Space | |
| CN112348836B (en) | Method and device for automatically extracting building outline | |
| JPWO2009060975A1 (en) | Feature point arrangement collation apparatus, image collation apparatus, method and program thereof | |
| CN111696118A (en) | Visual loopback detection method based on semantic segmentation and image restoration in dynamic scene | |
| CN110544202B (en) | Parallax image splicing method and system based on template matching and feature clustering | |
| CN110738216A (en) | Medicine identification method based on improved SURF algorithm | |
| JP7451373B2 (en) | Drawing structuring system and drawing structuring method | |
| CN111353325A (en) | Key point detection model training method and device | |
| Shi et al. | Robust image registration using structure features | |
| JP2017151797A (en) | Geometry verification device, program and method | |
| Castellani et al. | Statistical 3D shape analysis by local generative descriptors | |
| CN113159103B (en) | Image matching method, device, electronic equipment and storage medium | |
| JP2011107878A (en) | Position detection apparatus and position detection method | |
| CN114626483A (en) | Landmark image generation method and device | |
| JP5414631B2 (en) | Character string search method, character string search device, and recording medium | |
| Nam et al. | Precise pattern alignment for die-to-database inspection based on the generative adversarial network | |
| CN111915645B (en) | Image matching method and device, computer equipment and computer readable storage medium | |
| Phogat et al. | Different image registration methods—an overview | |
| CN112215032A (en) | Method, device, electronic device and program medium for identifying target in image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |