Local Discrete Point Multidimensional Feature Fusion Matching Method and Experimental Platform Based on Image Data

Abstract

The present invention relates to a method and experimental platform for multi-dimensional feature fusion matching of local discrete points based on data, which includes the construction of a multi-dimensional feature standard library, single-feature local discrete point matching, and multi-dimensional feature fusion matching. The experimental platform is based on global spherical discrete point data, constructs multiple standard feature libraries from multi-dimensional feature information as matching labels, uses the sum of the Euclidean distances between the matching result discrete points and the local discrete points to be matched as the matching index, and finally uses ICP registration to register the matching result discrete points and the local discrete points to be matched, calculates its matching index, and performs cyclic replacement matching on each independent matching result to minimize the matching index, completing the backend fusion of the matching results based on multi-dimensional features, and finally realizing the position matching recognition of the local discrete point map in the global spherical discrete point map. It realizes the fusion of multi-dimensional features, avoids the interference of noise and pseudo-discrete points, and effectively improves the matching accuracy of local discrete points on the premise of reducing the storage capacity and improving the speed.

Description

Technical Field

[0001] The present invention relates to the technical field of aerospace, and particularly to a multi-dimensional feature fusion matching method and experimental platform for local discrete points based on image data. Background Art

[0002] In recent years, image matching has developed vigorously, which is one of the important development directions in the fields of computer vision and aerospace, and has been widely applied. Matching through point feature data in images is one of the important links in image matching, and among them, discrete point matching based on data is particularly important. In fields such as ground scene matching and starlight navigation, the matching process based on point features often involves the position recognition and matching of local discrete point maps in the global spherical discrete point map. Therefore, anti-disturbance and high-precision discrete point matching algorithms are the key technologies in this field. The key to this algorithm lies in the feature construction and fusion matching of local discrete point data.

[0003] Currently, local discrete point matching is mainly based on single angular distance feature matching. Among them, the triangle algorithm has been widely applied in engineering due to its low computational complexity and strong portability. However, the algorithm based on single feature matching has low feature dimensions, serious feature redundancy and mismatching phenomena, and ultimately leads to low accuracy and poor anti-disturbance ability of the matching algorithm. Currently, many multi-feature triangle matching algorithms based on discrete points have been proposed. All combinations of any three navigation stars in the figure need to be traversed and combined, and various feature information needs to be constructed. When the number of points in the field of view (i.e., in the figure) is n, the number of combinations that need to be traversed and searched is C n 3 , that is, if there are 100 stars in the figure, C 100 3 = 100×99×98 / (3×2×1)=161700 times of matching are required, which has the problem of high computational complexity. At the same time, due to the fixed triangle matching based only on the main discrete points, the problem of mismatching and the like cannot be fundamentally solved. The matching algorithm based on constructing features with multiple main discrete points needs to construct multiple reference triangle models and use the segmented fitting search method for recognition, and a huge multi-field-of-view feature library needs to be constructed. Although the redundancy problem of the matching algorithm is reduced, there are still problems such as a large feature library capacity.

[0004] Traditional discrete point matching algorithms are mostly applicable to the matching of local discrete point data with low noise or no noise, and lack adaptability to strong noise, position relationship disturbance and other situations that occur in the actual matching process. In view of this deficiency, there is currently no perfect matching experimental platform based on discrete point data, and the matching algorithm part thereof is difficult to achieve matching recognition with strong anti-disturbance ability, high precision and low computational complexity. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide an experimental platform for multi-dimensional feature fusion matching of local discrete points based on data.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0007] In the first aspect, the present invention provides a method for multi-dimensional feature fusion matching of local discrete points based on data. The process of this method is as follows:

[0008] Obtain the global spherical discrete point coordinate data and the local discrete point data to be matched, number each global spherical discrete point in the global spherical discrete point coordinate data, and set the matching parameters.

[0009] Construct a multi-dimensional feature standard library:

[0010] Taking the imported global spherical discrete point coordinate data as a reference, calculate the three-dimensional angular distance between any two global spherical discrete points, and obtain the sine value of the three-dimensional angular distance, that is, the sine angular distance. The numbers of any two determined global spherical discrete points are respectively the main discrete point number and the secondary discrete point number. A two-dimensional table is formed with the main discrete point number and the secondary discrete point number. The content of the two-dimensional table is the sine angular distance between the main discrete point and the secondary discrete point. Store the sine angular distances between all any two global spherical discrete points in the two-dimensional table to obtain the sine angular distance matching search table, which serves as the standard feature library for matching based on the sine angular distance between two discrete points.

[0011] Taking each global spherical discrete point as the center, select the two global spherical discrete points with the smallest three-dimensional angular distance from it to form an included angle. Taking the line connecting the current global spherical discrete point and the global spherical discrete point with the smallest angular distance as a side, calculate the size of the included angle as the included angle degree of the current global spherical discrete point. Construct a one-dimensional included angle matching search table with all global spherical discrete points and the corresponding included angle degrees, which serves as the standard feature library for three-discrete point matching.

[0012] Taking each global spherical discrete point as the center, select the three global spherical discrete points with the smallest three-dimensional angular distance from it to form a quadrilateral. Taking the three-dimensional angular distance values between adjacent two global spherical discrete points as the side lengths, calculate the sum of the four sides of the quadrilateral, that is, the sum of the side lengths, which is also the angular distance sum of these four global spherical discrete points. Construct a one-dimensional angular distance sum matching search table with all global spherical discrete points and the corresponding angular distance sums, which serves as the standard feature library for four-discrete point matching.

[0013] So far, the construction of the multi-dimensional feature standard library is completed.

[0014] Respectively obtain the sine angular distance, included angle degree, and angular distance sum feature values of the local discrete point data to be matched, and perform search matching in the above sine angular distance matching search table, one-dimensional included angle matching search table, and one-dimensional angular distance sum matching search table respectively to obtain the corresponding local discrete point maps.

[0015] The three obtained local discrete point maps are subjected to feature fusion by using the ICP point cloud registration method to obtain the final matching result.

[0016] In a second aspect, the present invention provides a method for multi-dimensional feature fusion matching of local discrete points based on data, which is used for matching local discrete points in ground scene matching or starlight navigation. The process of this method is as follows:

[0017] Obtain the global spherical discrete point coordinate data for ground scene matching or starlight navigation and the local discrete point data to be matched, number each global spherical discrete point in the global spherical discrete point coordinate data, and set the matching parameters;

[0018] Construct a multi-dimensional feature standard library:

[0019] Taking the imported global spherical discrete point coordinate data as a reference, calculate the three-dimensional angular distance between any two global spherical discrete points, and obtain the sine value of the three-dimensional angular distance, that is, the sine angular distance. Arbitrarily select the numbers of two global spherical discrete points as the main discrete point number and the secondary discrete point number respectively, form a two-dimensional table with the main discrete point number and the secondary discrete point number, and the content of the two-dimensional table is the sine angular distance between the main discrete point and the secondary discrete point. Store the sine angular distances between all arbitrary two global spherical discrete points in the two-dimensional table to obtain the sine angular distance matching search table, which is used as the standard feature library for matching based on the sine angular distance between two discrete points;

[0020] Taking each global spherical discrete point as the center, select the two global spherical discrete points with the smallest three-dimensional angular distance from it to form an included angle. Taking the line connecting the current global spherical discrete point and the global spherical discrete point with the smallest angular distance as a side, calculate the degree of the included angle formed. Construct a one-dimensional included angle matching search table with all global spherical discrete points and the corresponding included angle degrees, which is used as the standard feature library for three-discrete-point matching;

[0021] Taking each global spherical discrete point as the center, select the three global spherical discrete points with the smallest three-dimensional angular distance from it to form a quadrilateral. Taking the three-dimensional angular distance values between adjacent two global spherical discrete points as the side lengths, calculate the sum of the four sides of the quadrilateral, that is, the sum of the side lengths, which is also the sum of the angular distances of these four global spherical discrete points. Construct a one-dimensional angular distance sum matching search table with all global spherical discrete points and the corresponding angular distance sums, which is used as the standard feature library for four-discrete-point matching;

[0022] Thus, the construction of the multi-dimensional feature standard library for ground scene matching or starlight navigation is completed;

[0023] Respectively obtain the sine angular distance, included angle degree, and angular distance sum feature values of the local discrete point data to be matched, and perform search matching in the above-mentioned sine angular distance matching search table, one-dimensional included angle matching search table, and one-dimensional angular distance sum matching search table respectively to obtain the corresponding local discrete point maps;

[0024] Perform feature fusion on the three obtained local discrete point maps to obtain the final matching result, and achieve the matching of local discrete points in ground scene matching or starlight navigation.

[0025] In a third aspect, the present invention provides a data-based experimental platform for multi-dimensional feature fusion matching of local discrete points, characterized in that: the experimental platform includes a multi-dimensional feature standard library, a single-feature local discrete point matching module, and a multi-dimensional feature fusion matching module;

[0026] The multi-dimensional feature standard library includes a standard feature library with at least three independent features;

[0027] The single-feature local discrete point matching module is used to search and match the single feature value calculated from the local discrete point data to be matched with the standard feature library of the corresponding independent feature;

[0028] The multi-dimensional feature fusion matching module includes feature fusion and ICP discrete point registration. The ICP discrete point registration is used for discrete point group registration and quantifying the Euclidean distance sum between two discrete point groups; after the independent feature matching results come out, fusion is performed at the back end. Based on the matching result of one independent feature, the matching results obtained from the standard libraries of other independent features are cyclically replaced in turn. Each time a result replacement is completed, the ICP point cloud registration algorithm is used for the replaced matching result and the original input local discrete points to calculate the Euclidean distance between the corresponding two discrete points, and the Euclidean distance sum of all corresponding discrete points is obtained. The matching result with the minimum distance sum is retained as the final matching result.

[0029] Preferably, the multi-dimensional feature standard library includes a standard feature library for matching based on the sine angle distance between two discrete points constructed based on the sine angle of two discrete points, a standard feature library for three-discrete-point matching constructed based on the relative position relationship of three discrete points, and a standard feature library for four-discrete-point matching constructed based on the positions of four discrete points. The three standard feature libraries are respectively independently used for feature search and matching of local discrete point data;

[0030] The multi-dimensional feature fusion matching module is a replacement fusion process based on the matching results of three single-feature independent matches. The matching result based on the sine angle distance between two discrete points is successively replaced by the matching results based on the three-point included angle and the four-point angle distance sum to minimize the Euclidean distance sum between discrete point groups and generate the multi-dimensional feature fusion matching result.

[0031] The software interface of the experimental platform includes a parameter setting area, a data import area, a single feature matching area, and a multi-dimensional feature matching area. The parameter setting area includes four text boxes for focal length, field of view range, image size, and matching accuracy. The data import area includes a button for importing global spherical discrete point data and a button for importing local discrete point data. The single feature matching area includes a two-point sine angular distance matching button, a three-point included angle matching button, and a four-point angular distance sum matching button. The three matching results are respectively displayed in the coordinate areas below the corresponding buttons. The multi-dimensional feature matching area includes a multi-dimensional feature matching button and a coordinate area for the multi-dimensional feature fusion matching result.

[0032] In this application, the multi-dimensional feature standard library includes a standard feature library with at least three independent feature values. The three independent features refer to independent matching with each other, and the matching results do not overlap.

[0033] Fourthly, the present invention provides a computer-readable storage medium, on which a computer program is stored. The program is characterized in that when executed by a processor, it can implement the steps of the above-mentioned method for multi-dimensional feature fusion matching of local discrete points based on data.

[0034] The beneficial effects of the present invention are as follows:

[0035] (1) The present invention uses the ICP algorithm to complete the registration of the point group for the matching result and the initial local discrete point map, and uses the Euclidean distance sum between the discrete point group of the matching result and the local discrete point group to be matched to quantify the difference between the matching result and the original map under the condition of multi-feature fusion.

[0036] (2) The present invention utilizes the relative position relationships between various discrete points. The matching based on the sine angular distance belongs to single main discrete point matching, and the matching based on the included angle and the sum of angular distances belongs to multi-main discrete point matching based on a single field of view, and constructs various position feature standard libraries, which can provide flexible and diverse matching methods and provide a prerequisite for multi-dimensional feature fusion.

[0037] (3) The multi-dimensional feature fusion matching proposed by the present invention first performs independent matching based on individual features and then fuses the matching results at the back end. Therefore, it is not necessary to traverse all combinations. When there are 100 stars in the map, only 100 * 3 = 300 times of combined matching are required, which reduces the computational complexity. In addition, only a two-dimensional sine angular distance library and two one-dimensional feature libraries are needed, and the capacity of the feature library is greatly reduced. The present invention improves the problems of low accuracy and poor anti-disturbance ability of the original single-feature matching algorithm, and at the same time solves the problems of high feature library capacity and high complexity of the existing multi-feature matching algorithms, realizing high-precision matching of local discrete points under noisy conditions, and still having good matching accuracy under the influence of strong noise such as image distortion and noise and position relationship disturbance during the imaging process.

[0038] (4) The experimental platform of the present invention calculates the angular distance between discrete points based on the coordinate data of local discrete points by setting parameters such as focal length, field of view range, image size, and matching accuracy for the local discrete point data. The angular distance is used to characterize the distance between discrete points, and the sine angular distance, included angle, angular distance sum, and three position features are introduced to depict the relative position relationship between discrete points. Multiple independent discrete point position features are used for matching, so it has stronger adaptability to noise compared with those based on a single feature. Through the fusion matching based on three independent relative position relationship features, the position region matching of local discrete point data in the global spherical discrete point data is completed, achieving a high-precision requirement with a matching accuracy higher than 98% under noisy conditions, a lower storage capacity, and a fast calculation speed, reducing the complexity. Although there is noise perturbation in the local discrete point position data compared with the standard feature library constructed by the global spherical discrete points in this application, a matching accuracy higher than 98% can still be achieved, demonstrating the adaptability of the present invention to noise perturbation. Description of the Drawings

[0039] Figure 1 It is a schematic diagram of the process of the multi-dimensional feature fusion matching method for local discrete points based on data of the present invention;

[0040] Figure 2 It is a schematic diagram of the flow of the ICP point cloud registration algorithm of the present invention;

[0041] Figure 3 It is a polar coordinate diagram of global spherical discrete points and an image coordinate diagram of local discrete points of the present invention;

[0042] Figure 4 It is an interface diagram of the experimental platform of the present invention. Detailed Embodiment

[0043] The technical solutions of the present invention will be clearly and completely described and explained below in conjunction with the embodiments and the content of the drawings, but this is not used as a limitation of the protection scope of this application.

[0044] As Figure 1 shown, the multi-dimensional feature fusion matching method for local discrete points based on data of the present invention constructs a multi-dimensional standard feature library based on global spherical discrete points, including a standard feature library constructed based on the sine angular distance between two points, a standard feature library constructed based on the included angle between three points, and a standard feature library constructed based on the sum of angular distances of four points. The local discrete point data features are respectively matched with the corresponding standard libraries to obtain matching results respectively. By circularly replacing the matching results and using ICP registration to minimize the sum of distances between point groups, the positioning matching of local discrete points in the global spherical discrete point set is realized.

[0045] The experimental platform for multi-dimensional feature fusion matching of local discrete points based on data of the present invention includes a multi-dimensional feature standard library, a single-feature local discrete point matching module, a multi-dimensional feature fusion matching module, a display unit, and a data acquisition unit, and realizes high-precision matching of the positions of local discrete points with noise in global spherical discrete points on the spherical surface.

[0046] The data acquisition unit is used to acquire global spherical discrete point coordinate data and local discrete point data to be matched, and number each global spherical discrete point in the global spherical discrete point coordinate data;

[0047] The display unit is used to input parameter data of local discrete points, including focal length, field of view, image size, matching accuracy, etc., and display the output results in each operation;

[0048] The multi-dimensional feature standard library includes a sine angular distance matching search table, a one-dimensional included angle matching search table, and a one-dimensional angular distance sum matching search table.

[0049] Taking the imported global spherical discrete point coordinate data as a reference, calculate the three-dimensional angular distance between any two global spherical discrete points, and obtain the sine value of the three-dimensional angular distance, which is the sine angular distance. The numbers of any two determined global spherical discrete points are respectively the main discrete point number and the secondary discrete point number. A two-dimensional table is formed with the main discrete point number and the secondary discrete point number. The content of the two-dimensional table is the sine angular distance between the main discrete point and the secondary discrete point. Store the sine angular distances between all any two global spherical discrete points in the two-dimensional table to obtain the sine angular distance matching search table, which is used as the standard feature library for matching based on the sine angular distance between two discrete points;

[0050] Taking each global spherical discrete point as the center, select two global spherical discrete points with the smallest three-dimensional angular distance from it to form an included angle. Calculate the size of the included angle with the line connecting the current global spherical discrete point and the global spherical discrete point with the smallest angular distance as the side as the included angle degree of the current global spherical discrete point. Construct a one-dimensional included angle matching search table with all global spherical discrete points and the corresponding included angle degrees, which is used as the standard feature library for three-discrete point matching;

[0051] Taking each global spherical discrete point as the center, select three global spherical discrete points with the smallest three-dimensional angular distance from it to form a quadrilateral. Taking the three-dimensional angular distance values between adjacent two global spherical discrete points as the side lengths, calculate the sum of the four sides of the quadrilateral, which is the sum of side lengths, that is, the angular distance sum of these four global spherical discrete points. Construct a one-dimensional angular distance sum matching search table with all global spherical discrete points and the corresponding angular distance sums, which is used as the standard feature library for four-discrete point matching;

[0052] In the single-feature local discrete point matching module, based on the imported local discrete point image coordinate data and the input local discrete point matching parameters, the matching parameters of the local discrete points include focal length, field of view range, image size, and matching accuracy. Based on the matching parameters and local discrete point coordinates, the local discrete point closest to the center point of the imaging plane is used as the main matching discrete point, and the local discrete points other than the main matching discrete point are used as secondary matching discrete points. Calculate the two-dimensional angular distance between the main matching discrete point and the secondary matching discrete points, and then obtain the sine value of the two-dimensional angular distance. Use this sine value as the sine angular distance, combine with the matching parameters, and search and match in the sine angular distance matching search table with the sine angular distance obtained from the local discrete points. When the sine angular distance value is less than the set matching accuracy, the matching is successful, and record the corresponding numbers of the main matching discrete point and the secondary matching discrete points obtained from the sine angular distance matching search table; if the matching is not successful, record the numbers of the main matching discrete point and the secondary matching discrete points as the same number as the main discrete point in the sine angular distance matching search table corresponding to the sine angular distance.

[0053] At the same time, calculate the included angle degree, angular distance, and eigenvalue of each local discrete point, combine with the matching parameters, and traverse and search for matching in the one-dimensional included angle matching search table and the one-dimensional angular distance and matching search table respectively. When the eigenvalue meets the matching accuracy requirements, it is regarded as the completion of the matching based on this feature, and use the discrete point number in the corresponding search table as the number of the current local discrete point.

[0054] The multi-dimensional feature fusion matching module is used to perform cyclic replacement based on the matching results after all single-feature-based matchings are completed. After each result replacement, use the ICP point cloud registration algorithm for the replaced matching results and the original input local discrete points, calculate the Euclidean distance between the corresponding two discrete points, obtain the sum of the Euclidean distances of all corresponding discrete points (all discrete points form a discrete point group), and retain the matching result with the smallest distance sum as the final matching result.

[0055] Preferably, the sine angular distance matching search table is based on the spherical global spherical discrete point polar coordinate data, and the sine angular distance features between each global spherical discrete point and other global spherical discrete points are obtained based on the polar coordinates. When performing matching based on the sine angular distance, search in this table. A total of three standard feature libraries are constructed in this application for three independent matchings.

[0056] In this application, the local discrete points are the discrete points on the imaging plane, and the global spherical discrete points are all the discrete points on the sphere. The sphere is three-dimensional, and the angular distance between two discrete points on the sphere is a three-dimensional angular distance. The imaging plane is two-dimensional, and the angular distance between two discrete points on the plane is a two-dimensional angular distance. The sine value of the two-dimensional angular distance and the sine value of the three-dimensional angular distance are both sine angular distances.

[0057] The present invention defines the three-dimensional coordinates of global spherical discrete points using three-dimensional spherical coordinates. With the origin of the spherical coordinate system as the center of the sphere, a sphere with a radius of r is generated. The spherical coordinates of the global spherical discrete points on this sphere are where θ ∈ (0, 360), The present invention represents the position distance between two global spherical discrete points using three-dimensional angular distance and calculates the three-dimensional angular distance based on the three-dimensional spherical coordinates. Angular distance is the size of the angle between the straight lines pointing from the observation point to these two objects, and can represent the positional relationship between two discrete points on the sphere. The calculation of the three-dimensional angular distance from the three-dimensional spherical coordinates is described as follows:

[0058]

[0059] where acos is the arccosine function; d3 is the three-dimensional angular distance between two global spherical discrete points, which is actually the number of degrees of the angle; is the first global spherical discrete point on the sphere, is the second global spherical discrete point on the sphere. Import the coordinate data of the global spherical discrete points into the experimental platform and number each global spherical discrete point. Then traverse and calculate the sine value of the three-dimensional angular distance between every two global spherical discrete points. Set the sine value of the three-dimensional angular distance that exceeds the visual field range to 1, and set the sine value of the three-dimensional angular distance to 1 when the two global spherical discrete points are the same, indicating that it is meaningless. Arbitrarily set the numbers of two global spherical discrete points as the main discrete point number and the secondary discrete point number respectively. Form a two-dimensional table with the main discrete point number and the secondary discrete point number. The content of the two-dimensional table is the sine angular distance between the main discrete point and the secondary discrete point. Store the sine angular distances between all arbitrary two global spherical discrete points in the two-dimensional table to obtain the sine angular distance matching search table.

[0060] Taking each global spherical discrete point as the center, select the two global spherical discrete points with the smallest angular distance from it to form an included angle. Calculate the size of the included angle with the connection line between the current global spherical discrete point and the two global spherical discrete points with the smallest angular distance (referring to the two discrete points closest to the current discrete point, and the three points form an included angle) as the side to obtain the included angle degree of the current discrete point. Construct a one-dimensional included angle matching search table with all global spherical discrete points and the corresponding included angle degrees as the standard feature library for three-discrete-point matching.

[0061] Taking each global spherical discrete point as the center, select the three global spherical discrete points with the smallest angular distance from it to form a quadrilateral. Using the three-dimensional angular distance values between adjacent two global spherical discrete points as the side lengths, calculate the sum of the four sides of the quadrilateral, which is the sum of the side lengths, that is, the angular distance sum of these four global spherical discrete points. A discrete point and the three closest points form a quadrilateral, and the side lengths of the quadrilateral are not necessarily equal. Construct a one-dimensional angular distance sum matching search table with all global spherical discrete points and the corresponding angular distance sums as the standard feature library for four-discrete-point matching.

[0062] Import the global spherical discrete point coordinate data. After completing the construction of three standard feature libraries (the standard feature library based on the sine angular distance matching of two discrete points, the standard feature library of three discrete point matching, and the standard feature library of four discrete point matching), set the matching parameters in the experimental platform, including the focal length f, the field of view α, the image size and the matching accuracy β. Calculate the focal length f and the spherical radius r of the global spherical discrete points according to the following formula from the field of view α and the image size as follows:

[0063]

[0064]

[0065] After setting the matching parameters, import the local discrete point coordinate data into the experimental platform. Let the two local discrete points be M'(x m ', y m ') and N'(x n ', y n '), and calculate the two-dimensional angular distance between the two local discrete points as follows:

[0066]

[0067] where d2 is the two-dimensional angular distance between the two local discrete points, O is the center of the sphere, O' is the center point of the imaging plane, is the length of the line connecting point O and point M', is the length of the line connecting point O and point N', are the lengths of the lines connecting point O' and points M', N' respectively. Take the local discrete point closest to O' as the main discrete point for matching, calculate the sine angular distance between the remaining secondary local discrete points and the main discrete point for matching, and search for the match in the sine angular distance matching search table. When the sine angular distance value is less than the set matching accuracy, the match is successful, and the corresponding number of the successful match is obtained from the sine angular distance matching search table. Otherwise, record the numbers of the main discrete point for matching and the secondary discrete point for matching as the same number as the main discrete point in the sine angular distance matching search table corresponding to the sine angular distance. In this embodiment, the matching accuracy is the reciprocal of the focal length.

[0068] Take each local discrete point as the center, and form the included angle of this local discrete point with the line connecting the two local discrete points with the smallest two-dimensional angular distance to it. Let the local discrete point be P'(x p ', y p '), and take the discrete point as P'(x p ', y p ') as the center. The two local discrete points with the smallest angular distance to it are respectively denoted as M'(x m ', y m ') and N'(xn ’, y n ’), the included angle is calculated as follows:

[0069]

[0070] Traverse each local discrete point to calculate the included angle and search for the closest result in the one-dimensional included angle matching search table. The number of the corresponding global spherical discrete point with the closest included angle in the one-dimensional included angle matching search table is used as its matching result.

[0071] Taking each local discrete point as the center, the three local discrete points with the smallest two-dimensional angular distance from it and this local discrete point form a quadrilateral. Taking the two-dimensional angular distance between two local discrete points as the side length, calculate the sum of the side lengths of this quadrilateral, and search for the closest angular distance sum result in the one-dimensional angular distance sum matching search table. The number of the global spherical discrete point with the closest angular distance sum is used as the matching result of this local discrete point.

[0072] The sine angular distance between two discrete points is a position feature, the included angle of three discrete points is a position feature, and the angular distance sum of four discrete points is a position feature. Through the above-mentioned matching results based on three single position features, the corresponding local discrete point maps are obtained respectively and displayed in the three matching result display frames of the display part of the experimental platform. After the three single feature matching results are displayed, multi-dimensional feature fusion matching can be performed.

[0073] Based on the matching result of the standard feature library based on the sine angular distance between two discrete points, use the matching results of the standard feature library of three discrete point matching and the matching results of the standard feature library of four discrete point matching to cyclically replace each discrete point (first, use the matching result of two discrete points (i.e., the local discrete point map obtained by matching through the sine angular distance matching search table) as the basis, and each time select a discrete point from the matching result of two discrete points and use the number of the corresponding discrete point obtained by three point matching (i.e., the local discrete point map obtained by matching through the one-dimensional included angle matching search table) for replacement, and then use the number of the corresponding discrete point obtained by four point matching (i.e., the local discrete point map obtained by matching through the one-dimensional angular distance sum matching search table) for replacement. If the matching index (the sum of Euclidean distances) is better, the replacement result is retained; if the index becomes worse, the replacement result is not retained. This process is completed for all discrete points). After each discrete point replacement, perform ICP point cloud registration on the replaced matching result map and the original input local discrete point map, and use the least squares estimation of the optimal transformation matrix as the objective function as follows:

[0074]

[0075] where E is the objective function, R and t are the rotation and translation matrices respectively, and N p is the number of local discrete points, X = {x1, x2, …, x Np}, and P = {p1, p2, …, p Np}, which are the corresponding point sets in the point cloud set. P is the discrete point set of the matching result, X is the original discrete point set to be matched, and i takes an integer not greater than N p .

[0076] The registration process of the ICP algorithm is as follows Figure 2 . By obtaining the corresponding point pairs between the initial local discrete point group to be matched and the discrete point set of the matching result, constructing a rotation and translation matrix based on the corresponding point pairs, calculating the objective function between the two point groups after transformation as the matching error, and performing iterative operations until the error value is minimized. After the ICP registration is completed, calculate the minimum error value of the current matching result and retain the matching result with the minimum error. After the ICP registration is completed, the process in Figure 2 is completed, and the R and t matrices are no longer changed. After all discrete points are replaced once, the local discrete point matching result based on multi-dimensional feature fusion is obtained. If more than half of the discrete points are accurately matched, the spherical region where the corresponding discrete points are located can be determined through the matching result, and the position of the local discrete point set to be matched in the global sphere can be determined.

[0077] Figure 3 The sphere in the left figure in Figure 4 contains 10,280 randomly generated discrete points, which are the global spherical discrete point data. The right figure is a local discrete point set of 1000 randomly generated discrete point image coordinate data containing random pseudo-discrete point noise and random position offset noise with parameters of a viewing range of 20°, an image size of 1024 * 1024 px, and a focal length of 512 / tan(10°). The matching accuracy is set to 1 / f. Import the global spherical discrete point data containing 10,280 random discrete points into

[0078] the experimental platform in

[0079] The display unit can adopt a display, an industrial computer, a liquid crystal touch screen, etc., Figure 4It is the software interface of the experimental platform, that is, the content presented by the display unit, including a parameter setting area, a data import area, a single feature matching area, and a multi-dimensional feature matching area. After entering the matching parameters in the four text boxes of focal length, field of view, image size, and matching accuracy, click the two buttons of importing global spherical discrete point data and importing local discrete point data respectively to complete the data import. Click the two-point sine angular distance matching button, three-point included angle matching button, and four-point angular distance sum matching button in sequence, and the three matching results are respectively displayed in the coordinate area below the corresponding button. The two-point matching result is plotted in the coordinate area above the two-point sine angular distance matching result, the three-point matching result is plotted in the coordinate area above the three-point included angle matching result, and the four-point matching result is plotted in the coordinate area above the four-point angular distance sum matching result. After completing the matching based on a single feature, finally click the multi-dimensional feature matching button to perform back-end fusion and secondary matching based on the three single feature matching results, and the final matching result is plotted in the coordinate area on the left side of the multi-dimensional feature fusion matching result.

[0080] The experimental platform of the present invention is based on global spherical discrete point data, constructs multiple standard feature libraries from multi-dimensional feature information as matching labels, uses the Euclidean distance sum between the matching result discrete points and the local discrete points to be matched as the matching index, and finally uses the ICP method to register the matching result discrete points and the local discrete points to be matched, calculates its matching index, performs cyclic replacement matching on each independent matching result to minimize the matching index, completes the back-end fusion of the multi-dimensional feature matching result, and finally realizes the position matching recognition of the local discrete point map in the global spherical discrete point map. The matching of local discrete points is realized by using three discrete point position features constructed based on data, and it can achieve higher accuracy and anti-perturbation to noise than the matching based on a single feature. The results of three different matching methods are fused at the back end, the three matching results are combined in a cyclic replacement manner, and the matching error is quantified to complete the fusion matching of multi-dimensional features, greatly reducing the storage and search difficulty of feature information required in the matching process, and simplifying the feature combination and construction process, effectively improving the local discrete point matching accuracy on the premise of reducing the storage capacity and increasing the speed.

[0081] Matters not described in the present invention are applicable to the prior art.

Claims

1. A method for multi-dimensional feature fusion matching of local discrete points based on image data, characterized in that, In the matching of local discrete points in ground scene matching or starlight navigation, the process of this method is as follows: Obtain the global spherical discrete point coordinate data for ground scene matching or starlight navigation and the local discrete point data to be matched, number each global spherical discrete point in the global spherical discrete point coordinate data, and set the matching parameters; Construct a multi-dimensional feature standard library: Calculate the three-dimensional angular distance between any two global spherical discrete points based on the imported global spherical discrete point coordinate data, and obtain the sine value of the three-dimensional angular distance, which is the sine angular distance. Arbitrarily select the numbers of two global spherical discrete points as the main discrete point number and the secondary discrete point number respectively, form a two-dimensional table with the main discrete point number and the secondary discrete point number, and the content of the two-dimensional table is the sine angular distance between the main discrete point and the secondary discrete point. Store the sine angular distances between all arbitrary two global spherical discrete points in the two-dimensional table to obtain the sine angular distance matching search table, which serves as the standard feature library for matching based on the sine angular distance between two discrete points; Taking each global spherical discrete point as the center, select the two global spherical discrete points with the smallest three-dimensional angular distance from it to form an included angle, calculate the size of the included angle with the line connecting the current global spherical discrete point and the global spherical discrete point with the smallest angular distance as the side as the included angle degree of the current global spherical discrete point, and construct a one-dimensional included angle matching search table with all global spherical discrete points and the corresponding included angle degrees, which serves as the standard feature library for three-discrete-point matching; Taking each global spherical discrete point as the center, select the three global spherical discrete points with the smallest three-dimensional angular distance from it to form a quadrilateral, calculate the sum of the four sides of the quadrilateral with the three-dimensional angular distance values between adjacent two global spherical discrete points as the side lengths, which is the sum of the side lengths, that is, the angular distance sum of these four global spherical discrete points, and construct a one-dimensional angular distance sum matching search table with all global spherical discrete points and the corresponding angular distance sums, which serves as the standard feature library for four-discrete-point matching; Thus, the construction of the multi-dimensional feature standard library for ground scene matching or starlight navigation is completed; Respectively obtain the sine angular distance, included angle degree, and angular distance sum feature values of the local discrete point data to be matched, and conduct search matching in the above-mentioned sine angular distance matching search table, one-dimensional included angle matching search table, and one-dimensional angular distance sum matching search table respectively to obtain the corresponding local discrete point maps; Perform feature fusion on the three obtained local discrete point maps to obtain the final matching result, realizing the matching of local discrete points in ground scene matching or starlight navigation.

2. A local discrete point multi-dimensional feature fusion matching experimental platform based on image data, characterized in that: This experimental platform includes a multi-dimensional feature standard library, a single-feature local discrete point matching module, and a multi-dimensional feature fusion matching module; The multi-dimensional feature standard library includes standard feature libraries of at least three independent features; The single-feature local discrete point matching module is used for search matching of the single feature value calculated from the local discrete point data to be matched with the standard feature libraries of the corresponding independent features; The multi-dimensional feature fusion and matching module includes feature fusion and ICP discrete point registration. The ICP discrete point registration is used for discrete point group registration and quantifying the Euclidean distance sum between two discrete point groups. After the independent feature matching results are obtained, fusion is performed at the backend. Based on the matching result of one independent feature, the matching results obtained from the standard libraries of other independent features are sequentially used for cyclic replacement. After each result replacement, the ICP point cloud registration algorithm is used for the replaced matching result and the original input local discrete points to calculate the Euclidean distance between the corresponding two discrete points, and the Euclidean distance sum of all corresponding discrete points is obtained. The matching result with the minimum distance sum is retained as the final matching result. The multi-dimensional feature standard library includes a standard feature library for matching based on the sine angle distance between two discrete points constructed based on the sine angle between two discrete points, a standard feature library for three-discrete-point matching constructed based on the relative position relationship of three discrete points, and a standard feature library for four-discrete-point matching constructed based on the positions of four discrete points. The three standard feature libraries are respectively and independently used for feature search and matching of local discrete point data. The multi-dimensional feature fusion and matching module is a replacement fusion process based on the matching results of three single features independently. The matching result based on the sine angle distance between two discrete points is sequentially replaced by the matching results based on the included angle of three points and the sum of the angle distances of four points to minimize the Euclidean distance sum between discrete point groups and generate the multi-dimensional feature fusion matching result. The software interface of the experimental platform includes a parameter setting area, a data import area, a single feature matching area, and a multi-dimensional feature matching area. The parameter setting area includes four text boxes for focal length, field of view range, image size, and matching accuracy. The data import area includes a button for importing global spherical discrete point data and a button for importing local discrete point data. The single feature matching area includes a button for two-point sine angle distance matching, a button for three-point included angle matching, and a button for four-point angle distance sum matching. The three matching results are respectively displayed in the coordinate areas below the corresponding buttons. The multi-dimensional feature matching area includes a multi-dimensional feature matching button and a coordinate area for the multi-dimensional feature fusion matching result.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it can implement the steps of the method described in claim 1.

Images