A vector map based road network image registration method, device and aircraft
By using a road network image registration method based on vector graphics and utilizing feature points and topological similarity metrics, the problems of registration accuracy and speed in a wide area are solved, achieving high-precision and fast image registration.
Patent Information
- Application Number
- CN202310427905.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-20
- Publication Date
- 2025-11-21
- Estimated Expiration
- 2043-04-20
AI Technical Summary
Existing road network image registration methods have low registration accuracy and long computation time in a wide area, and are difficult to adapt to matching scenarios with unknown angles and heights.
The road network image registration method based on vector maps extracts feature points from the road network vector map, calculates vector rotation angles, establishes a feature database, and uses topological similarity metrics and spatial scale scaling relationships for image registration, including dividing sub-blocks, constructing candidate point sets and triangle sets for accurate matching.
It achieves high-precision and fast wide-area road network image registration, can accurately match under unknown information, is applicable to any scene and road network, reduces the amount of computation and improves the registration speed.
Smart Images
Figure CN116823900B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image matching technology, and more specifically, relates to a method, device and aircraft for road network image registration based on vector graphics. Background Technology
[0002] With the continuous development of aircraft technology, the autonomous and semi-autonomous navigation technologies supporting it are receiving increasing attention. Among the current basic navigation methods and their optimizations, celestial navigation, satellite navigation, and Doppler navigation all inevitably require receiving or transmitting electronic information, making the information acquisition process susceptible to interference. Inertial navigation, on the other hand, can be completed entirely within the aircraft, without relying on external information or transmitting it. However, due to integral errors, it requires supplementary navigation methods for periodic recalibration to eliminate these errors. These auxiliary methods all require storing relevant information about the planned flight path, such as topographic maps, remote sensing images, and geomagnetic vector maps. The inherent density of this data means that the stored data can only cover a relatively small area.
[0003] Road network image registration refers to determining the actual latitude and longitude coordinates of each point in a real-time captured remote sensing image based on a pre-stored road network map, thereby enabling positioning calibration, yaw angle calibration, and other functions during navigation. Existing road network image registration methods are mainly divided into vector map-based registration methods and point set-based registration methods.
[0004] Existing vector map matching methods, such as those calculating Hausdorff and Fréchet distances, are often designed for matching within small map areas and are unsuitable for matching scenarios with unknown angles or heights. Existing point-set-based matching methods, such as ICP and CPD, suffer from numerous anomalies and noise points during registration, resulting in significant computational errors and time consumption, making it difficult to obtain accurate results over wide areas. Summary of the Invention
[0005] To address the shortcomings and improvement needs of existing technologies, this invention provides a road network image registration method, device, and aircraft based on vector graphics. The aim is to propose a road network image registration method with high registration accuracy, fast calculation speed, and applicability to a wide area.
[0006] To achieve the above objectives, according to one aspect of the present invention, a road network image registration method based on vector graphics is provided, comprising:
[0007] Obtain the road network vector map of the target area, extract the feature points, calculate the vector angle with each feature point as a vertex, and establish a feature database using the vector angle as an index and the vertex number and vertex coordinates as data; the feature points include inflection points and intersection points;
[0008] Follow these steps to register the remote sensing images and road network vector maps for the target area:
[0009] (S1) Convert the remote sensing image into a vector image and extract the feature points in it as the points to be matched to obtain the set of points to be matched;
[0010] (S2) For each point to be matched, calculate the vector rotation angle with that point as the vertex, and search the feature database according to the vector rotation angle to obtain all candidate points of that point to be matched;
[0011] (S3) Divide the road network vector map into multiple sub-blocks. For each sub-block, count the number of candidate points that fall within the sub-block and use it as the initial screening score of the sub-block. Select the sub-blocks with the highest initial screening scores as candidate blocks. Each sub-block can completely surround the remote sensing image.
[0012] (S4) Construct a candidate point set based on the candidate block, and determine the candidate point set with the highest topological similarity to the point set to be matched as the best matching point set; the candidate points in each candidate point set correspond one-to-one with the points to be matched, and the candidate points in each candidate point set are located in the same candidate block.
[0013] (S5) Calculate the spatial scale scaling relationship and angle transformation relationship between the set of points to be matched and the best matching set, and calculate the latitude and longitude coordinates of each point in the remote sensing image to complete the registration.
[0014] Further, in step (S3), the road network vector map is divided into multiple sub-blocks, including:
[0015] The road network vector map is divided into multiple non-overlapping sub-blocks, each of which can completely surround the remote sensing image;
[0016] For each sub-block, three overlapping sub-blocks are created as follows: the sub-block is offset horizontally by half the size of the sub-block to obtain the first set of overlapping sub-blocks; the sub-block is offset vertically by half the size of the sub-block to obtain the second set of overlapping sub-blocks; the sub-block is offset horizontally and vertically by half the size of the sub-block to obtain the third set of overlapping sub-blocks.
[0017] Each sub-block and each overlapping sub-block are treated as a sub-block, thus completing the sub-block division.
[0018] In some optional embodiments, step (S4) calculates the topological similarity between the set of points to be matched and the set of candidate points in the following manner:
[0019] Establish an edge between every two points in the set of points to be matched. Using the length of each edge as a coordinate of one dimension, obtain the n*(n-1) / 2 dimensional coordinates C1 = (a12 ,a 13 …,a (n-1)n ); n represents the number of elements in the set of points to be matched and the set of candidate points; a ij Let represent the length of the edge between the i-th and j-th points to be matched, where i,j∈{1,2,…n};
[0020] Establish an edge between every two candidate points in the candidate point set, and use the length of each edge as a coordinate of one dimension to obtain the n*(n-1) / 2 dimensional coordinates C2 = (b 12 ,b 13 …,b (n-1)n );a ij This represents the length of the edge between the i-th candidate point and the j-th candidate point, where the i-th candidate point corresponds to the i-th point to be matched, and the j-th candidate point corresponds to the j-th point to be matched.
[0021] After normalizing coordinates C1 and C2, their similarity is calculated and used as the topological similarity between the set of points to be matched and the set of candidate points.
[0022] In some optional embodiments, in step (S4), the topological similarity between the set of points to be matched and the set of candidate points is calculated as follows:
[0023] Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2.
[0024] Find all corresponding triangle pairs in triangle set S2 and triangle set S1. Measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair.
[0025] The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points.
[0026] In some optional embodiments, in step (S4), the topological similarity between each candidate point set and the point set to be matched is calculated as follows:
[0027] Sort each candidate point set in ascending order of the number of elements;
[0028] For the first k candidate point sets, the topological similarity between the point set to be matched and the candidate point sets is calculated as follows:
[0029] Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2.
[0030] Find all corresponding triangle pairs in triangle set S2 and triangle set S1. Measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair.
[0031] The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points.
[0032] For the last Nk candidate point sets, the topological similarity between the point set to be matched and the candidate point sets is calculated as follows:
[0033] From the first k candidate point sets, select a candidate point set, and from this set, select m candidate points as auxiliary points;
[0034] For the current candidate point set, traverse the candidate points in it. For each candidate point traversed, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point to obtain the triangle set S3. Find all triangle pairs that correspond to the triangle set S3 and the triangle set S1. Perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than the preset threshold, the triangle pair is determined to be a similar triangle pair, and the topological similarity is incremented by 1.
[0035] After the traversal is complete, the topological similarity between the candidate point set and the point set to be matched is obtained;
[0036] Where N represents the number of candidate points; k is a positive integer, and k <N。
[0037] Furthermore, for the last Nk candidate point set, for the currently traversed candidate point, it also includes: counting the pairs of similar triangles containing the candidate point;
[0038] Furthermore, after obtaining the topological similarity between the candidate point set and the point set to be matched, it also includes:
[0039] Obtain the candidate point with the highest count in the candidate point set, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point, and obtain the triangle set S4; find all triangle pairs corresponding to triangle set S4 and triangle set S1, and perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair; otherwise, the triangle pair is determined to be a dissimilar triangle pair; delete the auxiliary points contained in the dissimilar similar triangle pairs from their respective candidate point sets.
[0040] Furthermore, the similarity between two triangles is measured as follows:
[0041] Map the two triangles to coordinate points in the triangular similarity space to obtain coordinates C1′ and C2′ respectively, and calculate the similarity between coordinates C1′ and C2′ as the similarity measure of the two triangles;
[0042] The coordinates of the triangle mapped to the triangular similarity space are: Among them, C min C mid and C max Let C represent the side length of the triangle, and C... min ≤C mid ≤C max .
[0043] Further, let A and R represent the set of points to be matched and the set of best matching points, respectively. Then step (S5) includes:
[0044] Calculate the centroid coordinates CenterA and CenterR of the point set to be matched A and the best matching point set R, respectively;
[0045] After subtracting the centroid coordinate CenterA from the coordinates of each element in the set of points to be matched A, transform the coordinates from the Cartesian coordinate system to the polar coordinate system; after subtracting the centroid coordinate CenterR from the coordinates of each element in the best matching set of points R, transform the coordinates from the Cartesian coordinate system to the polar coordinate system.
[0046] The spatial scale scaling relationship dρ and the angular transformation relationship dθ between the set of points to be matched and the best matching set are calculated using the following formulas:
[0047]
[0048] Calculate the rotation matrix M between the set of points to be matched A and the best matching set R according to the following formula:
[0049]
[0050] The coordinates (x′, y′) of each point in the remote sensing image are transformed into latitude and longitude coordinates (lng′, lat′) using the following formula:
[0051]
[0052] Where, (ρ ai ,θ ai (ρ) represents the polar coordinates of the i-th element in the set of points to be matched A. ri ,θ ri (x, y) represents the polar coordinates of the i-th element in the best-matching motor R; (x, y) represents the coordinates of the element in the set of points to be matched A; (lng, lat) represents the coordinates of the element in the best-matching set of points R; and E() represents the expected value.
[0053] According to another aspect of the present invention, a road network image registration device based on vector graphics is provided, comprising:
[0054] A computer-readable storage medium for storing computer programs;
[0055] And a processor for reading a program from a computer-readable storage medium and executing the road network image registration method based on vector graphics provided by the present invention.
[0056] According to another aspect of the present invention, an aircraft is provided which is equipped with a remote sensing image capturing device and a road network image registration device based on vector graphics provided by the present invention.
[0057] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:
[0058] (1) The road network image registration method based on vector graphics provided by this invention uses strong feature points such as inflection points and intersection points in the vector graphics as registration objects. It can effectively reduce the amount of computation and improve the registration speed while ensuring registration accuracy. It can also be applied to image registration of different road networks. In the registration process, firstly, all candidate points of each point to be matched are determined based on the matching results of feature points. Then, the most likely matching sub-block, i.e., candidate block, is determined based on the distribution of the candidate points of each feature point. This effectively narrows the matching range and further reduces the amount of computation, which is conducive to acceleration. Based on the determination of candidate blocks, the best matching point set is further determined based on the results of topological similarity measurement. This process does not depend on information such as angle change, scale change, rotation angle, and image resolution. It can also obtain accurate matching results when the relevant information is unknown. Since this invention has fast calculation speed and good real-time performance, it can be applied to image registration of any scene and any road network. Therefore, this invention can realize image registration of wide-area road networks.
[0059] Overall, the road network image registration method based on vector graphics provided by this invention has high registration accuracy, fast calculation speed, and is applicable to road network image registration over a wide area.
[0060] (2) The road network image registration method based on vector graphics provided by the present invention first divides the road network vector graphics into multiple sub-blocks in order to determine candidate blocks through preliminary screening and narrow down the subsequent matching range. In the preferred embodiment of the present invention, based on dividing the road network vector graphics into multiple sub-blocks, three sets of overlapping sub-blocks are constructed for each small block by offsetting half a sub-block. The resulting sub-block division result can avoid the satellite sub-image to be registered being divided into multiple sub-blocks, further improving the registration accuracy.
[0061] (3) The road network image registration method based on vector graphics provided by the present invention, in its preferred embodiment, constructs edges in the set of points to be matched and the set of candidate points, and constructs multi-dimensional coordinates corresponding to the topology of the two sets of points with the length of each edge as a coordinate of one dimension. Based on the similarity between the multi-dimensional coordinates, the topological similarity between the set of points to be matched and the set of candidate points can be measured quickly.
[0062] (4) The road network image registration method based on vector graphics provided by the present invention, in its preferred embodiment, constructs a triangle set using the set of points to be matched and the candidate set of points, and completes the measurement of the topological similarity between the set of points to be matched and the candidate set of points by a method of measuring the similarity of each pair of triangles. This measurement method makes the maximum use of the topological information of the two set of points and can further improve the measurement accuracy.
[0063] (5) The road network image registration method based on vector graphics provided by the present invention, in its preferred embodiment, adopts a grouping measurement method to measure the topological similarity between the set of points to be matched and the set of candidate points. Specifically, for a candidate point set with fewer elements (i.e., a simplified candidate set), a triangle structure is constructed, and the topological similarity between the set of points to be matched and the set of candidate points is measured by a pairwise triangle similarity measurement method. For a candidate point set with more elements (i.e., a confirmed candidate set), points in the simplified candidate set are used to assist in the measurement of topological similarity. This measurement method can reduce the number of triangle similarity measurements while ensuring measurement accuracy, thus accelerating the calculation of topological similarity measurement. In a further preferred embodiment, the best candidate point in the confirmed candidate set is found, and the best candidate point is used to identify erroneous candidate points among the auxiliary points selected from the simplified candidate set, further improving the subsequent measurement accuracy.
[0064] (6) The road network image registration method based on vector graphics provided by the present invention calculates the spatial scale scaling relationship dρ and the angle transformation relationship dθ between the two sets by converting the set of points to be matched and the candidate set of points to polar coordinates. This process has a small amount of computation and a fast calculation speed. Attached Figure Description
[0065] Figure 1 A flowchart of a road network image registration method based on vector graphics provided in an embodiment of the present invention;
[0066] Figure 2 This is an example of a road network vector diagram provided in an embodiment of the present invention;
[0067] Figure 3 The distribution results of candidate points in the feature database for a single vector rotation angle provided in the embodiments of the present invention;
[0068] Figure 4 This is a schematic diagram illustrating the candidate block determination method provided in an embodiment of the present invention.
[0069] Figure 5 This is a schematic diagram illustrating topological similarity measurement using a triangular space feature point matching method, as provided in an embodiment of the present invention.
[0070] Figure 6 This is a schematic diagram of the positioning results provided in an embodiment of the present invention;
[0071] Figure 7 This is a schematic diagram illustrating topological similarity measurement using an extended triangular space feature point matching method, as provided in another embodiment of the present invention. Detailed Implementation
[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0073] In this invention, the terms "first," "second," etc. (if present) in the invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0074] To address the technical problems of existing registration methods, such as large registration errors, long computation time, and inability to complete road network registration over wide areas, this invention provides a road network image registration method, device, and aircraft based on vector maps. The overall approach is as follows: based on vector maps, strong feature points such as inflection points and intersections are extracted as matching objects to reduce computational load while maintaining registration accuracy. During registration, candidate blocks are initially determined based on the matching results of feature points, further reducing computational load and accelerating registration. Among the candidate blocks, the best matching result is determined through topological similarity measurement to ensure final registration accuracy.
[0075] The following is an example.
[0076] Example 1:
[0077] A method for road network image registration based on vector graphics, the process of which is as follows: Figure 1 As shown.
[0078] To simultaneously improve registration accuracy and speed, this embodiment utilizes a road network vector map with higher storage density to provide road network information. Figure 2The image shows an example of a road network vector map. Road network vector maps can be downloaded from public databases. It should be noted that the original road network vector obtained often covers a larger area than the target area. In order to perform effective registration and reduce data storage and computation, a road network vector map that matches the target area can be obtained by cropping or other methods.
[0079] This embodiment will further utilize the road network vector map of the target area to construct a feature database: the specific process is as follows.
[0080] Feature points are extracted from the road network vector map. The vector rotation angle with each feature point as a vertex is calculated. The vector rotation angle is used as an index, and the vertex number and vertex coordinates are used as data to establish a feature database. The feature points specifically include strong feature points such as inflection points and intersection points.
[0081] The method for calculating vector rotation angles is as follows: Three consecutive points A, B, and C in a road network vector map can generate two vectors, V1(x1, y1) and V2(x2, y2). Converting these to polar coordinates (ρ1, θ1) and (ρ2, θ2) yields a vector rotation angle θ2-θ1 with point B as the vertex. It's easy to understand that for the same point, there may be multiple rotation angles with that point as the vertex in the road network vector map; correspondingly, multiple data entries will exist in the feature database.
[0082] It's easy to understand that inflection points represent points with large vector angles, and intersection points represent points where the same coordinates appear three or more times in the same vector map.
[0083] Based on the constructed feature database, the data distribution range of candidate points for specific vector corners can be quickly obtained by indexing vector corners, thus obtaining the spatial distribution of candidate points within the spatial domain of the road network in the remote sensing image. Optionally, in this embodiment, the feature database uses a B+ tree for data storage to facilitate subsequent searching. Since the construction process of the feature database is time-consuming and related to the size of the road network vector map, this embodiment completes the establishment of the feature database in advance before real-time image registration.
[0084] This embodiment also includes: completing the registration of remote sensing images and road network vector maps within the target area according to the following steps (S1) to (S5), the specific steps of which are as follows:
[0085] (S1) Convert the remote sensing image into a vector image and extract the feature points in it as the points to be matched to obtain the set of points to be matched;
[0086] The conversion of satellite images into vector graphics can be achieved through various road extraction algorithms, such as CoANet and VecRoad, as well as image skeletonization and vectorization methods, such as Zhang-Suen. No specific limitations are made here.
[0087] (S2) For each point to be matched, calculate the vector rotation angle with that point as the vertex, and search the feature database according to the vector rotation angle to obtain all candidate points of that point to be matched;
[0088] Figure 3 The figure shows the distribution of candidate points obtained after searching the feature database for a single vector corner. It is easy to understand that for a single point to be matched, there may be multiple corner vectors, and the search results corresponding to all these corner vectors together constitute the candidate points of the point to be matched.
[0089] (S3) Divide the road network vector map into multiple sub-blocks. For each sub-block, count the number of candidate points that fall within the sub-block and use it as the initial screening score of the sub-block. Select the sub-blocks with the highest initial screening scores as candidate blocks.
[0090] Each sub-block can completely surround the remote sensing image;
[0091] This embodiment effectively narrows the matching range by filtering out the sub-blocks that are most likely to be registered to the remote sensing image, i.e., candidate blocks, which further reduces the amount of computation and helps to speed up the process.
[0092] The method used in this embodiment to lock candidate blocks is similar to performing spatial distribution histogram statistics on candidate points. Specifically, all candidate points are recorded in the sub-block containing their latitude and longitude coordinates. Multiple candidate points corresponding to the same point to be matched are recorded only once in the same block. Figure 4 As shown; after completing the statistical counting, the sub-block whose counting result is consistent with the number of points to be matched, that is, the sub-block that contains candidate points of each point to be matched, is very likely the sub-block matched by the remote sensing image. Therefore, in this embodiment, such a sub-block is used as a candidate block. In order to avoid errors in the candidate points determined by the search database, in some other embodiments of the present invention, the self-selected block containing only some candidate points to be matched can also be used as a candidate block.
[0093] Considering that during the process of segmenting the road network vector map, the targets contained in the remote sensing image may be divided into multiple sub-blocks, in order to avoid this problem, as a preferred implementation, in step (S3) of this embodiment, the road network vector map is divided into multiple sub-blocks, such as... Figure 4 As shown, it includes:
[0094] The road network vector map is divided into multiple non-overlapping sub-blocks, each of which can completely surround the remote sensing image;
[0095] For each sub-block, three overlapping sub-blocks are created as follows: the sub-block is offset horizontally by half the size of the sub-block to obtain the first set of overlapping sub-blocks; the sub-block is offset vertically by half the size of the sub-block to obtain the second set of overlapping sub-blocks; the sub-block is offset horizontally and vertically by half the size of the sub-block to obtain the third set of overlapping sub-blocks.
[0096] Each sub-block and each overlapping sub-block are treated as a sub-block, thus completing the sub-block division.
[0097] This embodiment divides sub-blocks by constructing overlapping sub-blocks, which can ensure that the remote sensing image can fall completely into at least one sub-block, thereby avoiding the target being segmented into multiple sub-blocks and further improving the registration accuracy.
[0098] (S4) Construct a candidate point set based on the candidate blocks, and determine the candidate point set with the highest topological similarity to the point set to be matched as the best matching point set;
[0099] Each candidate point in a candidate point set corresponds one-to-one with the point to be matched, and the candidate points in each candidate point set are located in the same candidate block;
[0100] To accurately determine the optimal matching point set, this embodiment employs a triangular spatial feature point matching method to calculate the topological similarity between the point set to be matched and the candidate point set; see reference. Figure 5 The triangular space feature point matching method specifically includes:
[0101] Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2.
[0102] Find all corresponding triangle pairs in triangle set S2 and triangle set S1, and measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair. It is easy to understand that the corresponding triangle pair mentioned in this embodiment refers to two triangles from different sets, and the triangle vertices have a one-to-one correspondence.
[0103] The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points.
[0104] In this embodiment, the similarity between two triangles is measured as follows:
[0105] Map the two triangles to coordinate points in the triangular similarity space to obtain coordinates C1′ and C2′ respectively. Calculate the similarity between coordinates C1′ and C2′ as the similarity measure of the two triangles. The similarity between coordinates can be measured by Euclidean distance, vector dot product, etc.
[0106] The coordinates of the triangle mapped to the triangular similarity space are: Among them, C min C mid and C max Let C represent the side length of the triangle, and C... min ≤C mid ≤C max ;
[0107] This embodiment constructs a triangle set using the set of points to be matched and the set of candidate points, and measures the topological similarity between the set of points to be matched and the set of candidate points by a pairwise triangle similarity measurement method. This measurement method makes the most of the topological information of the two point sets and can further improve the measurement accuracy.
[0108] The threshold for determining whether two triangles are similar can be set according to the actual characteristics of the road network and the accuracy requirements. Optionally, in this embodiment, the threshold is set to 0.001.
[0109] (S5) Calculate the spatial scale scaling relationship and angle transformation relationship between the set of points to be matched and the best matching set, and calculate the latitude and longitude coordinates of each point in the remote sensing image to complete the registration.
[0110] See Figure 6 Step (S5) in this embodiment specifically includes:
[0111] Let A and R represent the set of points to be matched and the set of points to be matched, respectively. Calculate the centroid coordinates CenterA and CenterR of the set of points to be matched A and the set of points to be matched R, respectively. For any set of points, let (x... i y i Let ) represent an element, and n represent the number of elements. The formula for calculating the centroid (Center) is as follows:
[0112]
[0113] After subtracting the centroid coordinate CenterA from the coordinates of each element in the set of points to be matched A, transform the coordinates from the Cartesian coordinate system to the polar coordinate system; after subtracting the centroid coordinate CenterR from the coordinates of each element in the best matching set of points R, transform the coordinates from the Cartesian coordinate system to the polar coordinate system.
[0114] The spatial scale scaling relationship dρ and the angular transformation relationship dθ between the set of points to be matched and the best matching set are calculated using the following formulas:
[0115]
[0116] Let (X,Y) represent the coordinates of an element in the set of points to be matched A, and (lng,lat) represent the coordinates of an element in the set of best matching points R. Then, the calculation of dρ and dθ can be converted to:
[0117]
[0118] Calculate the rotation matrix M between the set of points to be matched A and the best matching set R according to the following formula:
[0119]
[0120] The coordinates (x′, y′) of each point in the remote sensing image are transformed into latitude and longitude coordinates (lng′, lat′) using the following formula:
[0121]
[0122] Where, (ρ ai ,θ ai (ρ) represents the polar coordinates of the i-th element in the set of points to be matched A. ri ,θ ri ) represents the polar coordinates of the i-th element in the best-matched motor R. E() represents the expected value. Through image registration, the latitude and longitude coordinates of each point in the remote sensing image can be determined, thus completing the positioning.
[0123] In practical applications, the center coordinates of the image can be converted into latitude and longitude coordinates for positioning calibration, and the rotation angle between the image and latitude and longitude coordinates can be used to calibrate the yaw angle.
[0124] In summary, this embodiment utilizes a vector corner feature database and region overlap method to obtain candidate points and approximate location areas, combined with triangular similarity spatial similarity measurement to achieve rapid vector map localization. Experiments show that in actual verification of matching image spatial range of 500m and road network vector map spatial range of 6500m, the average time of this embodiment is approximately 5ms.
[0125] Example 2:
[0126] A road network image registration method based on vector graphics is presented in this embodiment, which is similar to Embodiment 1 above. The difference lies in that, to further accelerate the registration process while maintaining accuracy, this embodiment employs a grouping acceleration method in step (S4) to measure the topological similarity between the set of points to be matched and all candidate point sets. The specific details of this grouping acceleration method are as follows:
[0127] Sort each candidate point set in ascending order of the number of elements;
[0128] For the first k candidate point sets, which are then used as a simplified candidate set, the topological similarity between the point set to be matched and the candidate point sets is calculated using the triangular space feature point matching method, i.e.:
[0129] Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2.
[0130] Find all corresponding triangle pairs in triangle set S2 and triangle set S1, and measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair. In this embodiment, the method of measuring triangle similarity and the threshold setting can be referred to the description in Embodiment 1 above.
[0131] The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points.
[0132] For the last Nk candidate point sets, which are used as confirmed candidate sets, the topological similarity between the point set to be matched and the candidate point sets is calculated as follows:
[0133] From the first k candidate point sets, select a candidate point set, and from this set, select m candidate points as auxiliary points;
[0134] For the current candidate point set, traverse the candidate points in it. For each candidate point traversed, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point to obtain the triangle set S3. Find all triangle pairs that correspond to the triangle set S3 and the triangle set S1. Perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than the preset threshold, the triangle pair is determined to be a similar triangle pair, and the topological similarity is incremented by 1.
[0135] After the traversal is complete, the topological similarity between the candidate point set and the point set to be matched is obtained;
[0136] In this embodiment, for candidate point sets with fewer elements (i.e., simplified candidate sets), a triangle structure is constructed, and the topological similarity between the point set to be matched and the candidate point set is measured by a pairwise triangle similarity measurement method. For candidate point sets with more elements (i.e. confirmed candidate sets), the points in the simplified candidate set are used to assist in the measurement of topological similarity. This measurement method can reduce the number of triangle similarity measurements while ensuring measurement accuracy, thus accelerating the calculation of topological similarity measurement.
[0137] To further improve the calculation accuracy of topological similarity, in this embodiment, for the last Nk candidate point set, i.e. the confirmed candidate set, during the traversal process, for the currently traversed candidate point, the method further includes: counting the similar triangle pairs containing the candidate point.
[0138] Furthermore, after obtaining the topological similarity between the candidate point set and the point set to be matched, it also includes:
[0139] Obtain the candidate point with the highest count in the candidate point set, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point, and obtain the triangle set S4; find all triangle pairs corresponding to triangle set S4 and triangle set S1, and perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair; otherwise, the triangle pair is determined to be a dissimilar triangle pair; delete the auxiliary points contained in the dissimilar similar triangle pairs from their respective candidate point sets.
[0140] This embodiment further improves the accuracy of subsequent measurements by finding the best candidate point in the confirmed candidate set and using the best candidate point to identify erroneous candidate points in the auxiliary points.
[0141] Example 3:
[0142] A road network image registration method based on vector graphics is described in this embodiment, which is similar to Embodiment 1 above. The difference lies in that, to further accelerate the registration speed while maintaining registration accuracy, step (S4) of this embodiment employs an extended triangular space feature point matching method to calculate the topological similarity between the set of points to be matched and the candidate set of points. (See also...) Figure 7 In this embodiment, the extended triangular space feature point matching method specifically includes:
[0143] Establish an edge between every two points in the set of points to be matched. Using the length of each edge as a coordinate of one dimension, obtain the n*(n-1) / 2 dimensional coordinates C1 = (a 12 ,a 13 …,a (n-1)n ); n represents the number of elements in the set of points to be matched and the set of candidate points; a ij Let represent the length of the edge between the i-th and j-th points to be matched, where i,j∈{1,2,…n};
[0144] Establish an edge between every two candidate points in the candidate point set, and use the length of each edge as a coordinate of one dimension to obtain the n*(n-1) / 2 dimensional coordinates C2 = (b 12 ,b 13 …,b (n-1)n );a ijThis represents the length of the edge between the i-th candidate point and the j-th candidate point, where the i-th candidate point corresponds to the i-th point to be matched, and the j-th candidate point corresponds to the j-th point to be matched.
[0145] After normalizing coordinates C1 and C2, their similarity is calculated as the topological similarity between the set of points to be matched and the set of candidate points. During normalization, the longest side or vector length can be selected from the multidimensional coordinates as the normalization metric. The similarity between coordinates can be measured by Euclidean distance or vector-point product, etc.
[0146] This embodiment constructs edges in the set of points to be matched and the set of candidate points, and uses the length of each edge as a coordinate of one dimension to construct multi-dimensional coordinates corresponding to the topology of the two sets of points. Based on the similarity between the multi-dimensional coordinates, the topological similarity between the set of points to be matched and the set of candidate points can be measured quickly.
[0147] Example 4:
[0148] A road network image registration device based on vector graphics, comprising:
[0149] A computer-readable storage medium for storing computer programs;
[0150] And a processor for reading a program from a computer-readable storage medium and executing the road network image registration method based on vector graphics provided in any one of embodiments 1 to 3 above.
[0151] Example 5:
[0152] An aircraft equipped with a remote sensing image capturing device and a road network image registration device based on vector graphics provided in Embodiment 4 above.
[0153] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for road network image registration based on vector graphics, characterized in that, include: Obtain the road network vector map of the target area, extract the feature points, calculate the vector turning angle with each feature point as a vertex, and establish a feature database using the vector turning angle as an index and the vertex number and vertex coordinates as data; the feature points include inflection points and intersection points; Complete the registration of the remote sensing image and the road network vector map within the target area according to the following steps: (S1) Convert the remote sensing image into a vector image and extract the feature points therein as matching points to obtain a set of matching points; (S2) For each point to be matched, calculate the vector rotation angle with that point as the vertex, and search the feature database according to the vector rotation angle to obtain all candidate points of the point to be matched; (S3) Divide the road network vector map into multiple sub-blocks. For each sub-block, count the number of candidate points that fall within the sub-block as the initial screening score of the sub-block. Select the sub-blocks with the highest initial screening scores as candidate blocks. Each sub-block can completely surround the remote sensing image. (S4) Construct a candidate point set based on the candidate block, and determine the candidate point set with the highest topological similarity to the point set to be matched as the best matching point set; the candidate points in each candidate point set correspond one-to-one with the points to be matched, and the candidate points in each candidate point set are located in the same candidate block; (S5) Calculate the spatial scale scaling relationship and angle transformation relationship between the set of points to be matched and the set of best matching points, and thereby calculate the latitude and longitude coordinates of each point in the remote sensing image to complete the registration.
2. The road network image registration method based on vector graphics as described in claim 1, characterized in that, In step (S3), the road network vector map is divided into multiple sub-blocks, including: The road network vector map is divided into multiple non-overlapping sub-blocks, each of which can completely surround the remote sensing image; For each sub-block, three overlapping sub-blocks are created as follows: the sub-block is offset horizontally by half the size of the sub-block to obtain the first set of overlapping sub-blocks; the sub-block is offset vertically by half the size of the sub-block to obtain the second set of overlapping sub-blocks; the sub-block is offset horizontally and vertically by half the size of the sub-block to obtain the third set of overlapping sub-blocks. Each sub-block and each overlapping sub-block are treated as a sub-block, thus completing the sub-block division.
3. The road network image registration method based on vector graphics as described in claim 1 or 2, characterized in that, Step (S4) calculates the topological similarity between the set of points to be matched and the set of candidate points in the following manner: An edge is established between every two points in the set of points to be matched. Using the length of each edge as a coordinate dimension, the n*(n-1) / 2 dimensional coordinates C1 = (a...) are obtained for the set of points to be matched. 12 ,a 13 …,a (n-1)n ); n represents the number of elements in the set of points to be matched and the set of candidate points; a ij Let represent the length of the edge between the i-th and j-th points to be matched, where i,j∈{1,2,…n}; An edge is established between every two candidate points in the candidate point set. Using the length of each edge as a coordinate dimension, the n*(n-1) / 2 dimensional coordinates C2 = (b...) of the candidate point set are obtained. 12 ,b 13 …,b (n-1)n );a ij This represents the length of the edge between the i-th candidate point and the j-th candidate point, where the i-th candidate point corresponds to the i-th point to be matched, and the j-th candidate point corresponds to the j-th point to be matched. After normalizing coordinates C1 and C2, their similarity is calculated and used as the topological similarity between the set of points to be matched and the set of candidate points.
4. The road network image registration method based on vector graphics as described in claim 1 or 2, characterized in that, In step (S4), the topological similarity between the set of points to be matched and the set of candidate points is calculated in the following manner: Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2. Find all corresponding triangle pairs in triangle set S2 and triangle set S1. Measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair. The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points.
5. The road network image registration method based on vector graphics as described in claim 1 or 2, characterized in that, In step (S4), the topological similarity between each candidate point set and the point set to be matched is calculated in the following manner: Sort each candidate point set in ascending order of the number of elements; For the first k candidate point sets, the topological similarity between the point set to be matched and the candidate point sets is calculated as follows: Obtain all triangles with vertices of the unmatched points in the unmatched point set, resulting in triangle set S1; obtain all triangles with vertices of the candidate points in the candidate point set, resulting in triangle set S2. Find all corresponding triangle pairs in triangle set S2 and triangle set S1. Measure the similarity between the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair. The number of similar triangle pairs is used as the topological similarity between the set of points to be matched and the set of candidate points. For the last Nk candidate point sets, the topological similarity between the point set to be matched and the candidate point sets is calculated as follows: From the first k candidate point sets, select a candidate point set, and from this set, select m candidate points as auxiliary points; For the current candidate point set, traverse the candidate points in it. For each candidate point traversed, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point to obtain the triangle set S3. Find all triangle pairs that correspond to the triangle set S3 and the triangle set S1. Perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than the preset threshold, the triangle pair is determined to be a similar triangle pair, and the topological similarity is incremented by 1. After the traversal is complete, the topological similarity between the candidate point set and the point set to be matched is obtained.
6. The road network image registration method based on vector graphics as described in claim 5, characterized in that, For the last Nk candidate point set, for the currently traversed candidate point, the process also includes: counting the pairs of similar triangles containing the candidate point; Furthermore, after obtaining the topological similarity between the candidate point set and the point set to be matched, it also includes: Obtain the candidate point with the highest count in the candidate point set, exhaustively enumerate the triangles formed by the auxiliary points and the candidate point, and obtain the triangle set S4; find all triangle pairs corresponding to triangle set S4 and triangle set S1, and perform similarity measurement on the two triangles in each triangle pair. If the measurement result is greater than a preset threshold, the triangle pair is determined to be a similar triangle pair; otherwise, the triangle pair is determined to be a dissimilar triangle pair; delete the auxiliary points contained in the dissimilar similar triangle pairs from their respective candidate point sets.
7. The road network image registration method based on vector graphics as described in any one of claims 4 to 6, characterized in that, The similarity between two triangles is measured as follows: Map the two triangles to coordinate points in the triangular similarity space to obtain coordinates C1′ and C2′ respectively, and calculate the similarity between coordinates C1′ and C2′ as the similarity measure of the two triangles; The coordinates of the triangle mapped to the triangular similarity space are: Among them, C min C mid and C max Let C represent the side length of the triangle, and C... min ≤C mid ≤C max .
8. The road network image registration method based on vector graphics as described in claim 1 or 2, characterized in that, Let A and R represent the set of points to be matched and the set of best matching points, respectively. Then, step (S5) includes: Calculate the centroid coordinates CenterA and CenterR of the point set to be matched A and the best matching point set R, respectively; After subtracting the centroid coordinate CenterA from the coordinates of each element in the set of points to be matched A, transform the coordinates from the Cartesian coordinate system to the polar coordinate system; after subtracting the centroid coordinate CenterR from the coordinates of each element in the best matching set of points R, transform the coordinates from the Cartesian coordinate system to the polar coordinate system. The spatial scale scaling relationship dρ and the angular transformation relationship dθ between the set of points to be matched and the best matching set are calculated using the following formulas: Calculate the rotation matrix M between the set of points to be matched A and the best matching set R according to the following formula: The coordinates (x′, y′) of each point in the remote sensing image are transformed into latitude and longitude coordinates (lng′, lat′) according to the following formula: Where, (ρ ai ,θ ai (ρ) represents the polar coordinates of the i-th element in the set of points to be matched A. ri ,θ ri (x, y) represents the polar coordinates of the i-th element in the best-matching motor R; (x, y) represents the coordinates of the element in the set of points to be matched A; (lng, lat) represents the coordinates of the element in the best-matching set of points R; and E() represents the expected value.
9. A road network image registration device based on vector graphics, characterized in that, include: A computer-readable storage medium for storing computer programs; And a processor for reading a program in the computer-readable storage medium and executing the road network image registration method based on vector graphics as described in any one of claims 1 to 8.
10. An aircraft, characterized in that, It is equipped with remote sensing image capturing equipment and the road network image registration equipment based on vector graphics as described in claim 9.