Visual slam feature matching method based on topological feature constraint

Abstract

The application relates to a visual SLAM feature matching method based on topological feature constraint, and belongs to the cross field of automation and computer technology. The method first extracts point features from an image and calculates corresponding descriptors, removes features not meeting requirements according to neighborhood density, response strength and descriptor redundancy, and obtains an optimized feature set; secondly, the features are organized into stable triangular topological units through triangulation, the geometric properties of each triangle are calculated, and unstable triangles are removed; thirdly, nearest neighbor matching is performed on the feature descriptors of adjacent image frames to generate candidate matching pairs; fourthly, topological consistency screening is performed to remove invalid matching; and finally, the matching pairs are weighted and accumulated in a time window of a set length, robust matching pairs are screened out, and the robust matching pairs are output. The application can provide reliable input for subsequent SLAM processing, reduce false matching, and improve the accuracy and stability of visual SLAM in a weak texture environment.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of automation and computer technology, and relates to a visual SLAM feature matching method based on topological feature constraints. Background Technology

[0002] Visual SLAM technology uses image sequences acquired by cameras to enable autonomous localization and environmental mapping for mobile platforms, and is widely used in fields such as intelligent vehicles, drone navigation, and robotics. Feature matching, as the core component of visual SLAM, provides support for pose estimation and mapping, and its matching accuracy directly determines the overall performance of the SLAM system.

[0003] The paper "Zhang Z, Zhu J, Wei J, Zhang G. A Fast Method for Point Feature Extracting and Matching[C] / / 2024 4th International Conference on Robotics, Automation and Intelligent Control 2024. IEEE." discloses a visual SLAM point feature extraction and matching scheme. It optimizes the feature point distribution by improving the Harris corner detection algorithm, constructs descriptors using circular neighborhood pixel gray-level histograms, and achieves feature matching based on Pop distance. In weakly repetitive texture scenes, because it relies on a single gray-level histogram descriptor to characterize features, the feature gray-level distribution of repetitive texture regions has high similarity, resulting in insufficient descriptor discriminability and easy mismatches where local features are similar but do not actually correspond.

[0004] The paper "Zhang Y, Dong M. Research on visual SLAM algorithm based on improved point-line feature fusion[C] / / 2023 International Conference on Pattern Recognition, Machine Vision and Intelligent Algorithms (PRMVIA).2023. IEEE." discloses a point-line feature fusion SLAM scheme based on the ORB-SLAM3 framework. It uses the GCNV2 self-supervised deep network to extract point features, combines threshold filtering and GMS algorithm to eliminate mismatches of point features, and improves the LSD algorithm to extract line features and optimizes line feature matching through MAD algorithm. In weakly repetitive texture scenes, the matching process of point features and line features is relatively independent, and no targeted constraints are designed for the local similarity of features in repetitive texture areas. When there are a large number of repetitive texture units in the scene, the accumulation of local matching deviations of point and line features can easily lead to overall matching errors.

[0005] The paper "Jiang X, Fu Y, Song S, Wang H, Cheng W, Zhu W. On visual SLAM algorithm based on point-line features in weak texture environment[C] / / 20238th International Conference on Intelligent Computing and Signal Processing(ICSP). 2023. IEEE." discloses a point-line feature SLAM scheme for weak texture environments. It extracts line features to supplement point features using the EDlines algorithm, improves the line feature reprojection error model, and constructs a point-line affine invariance constraint optimization backend. In weakly repeating texture scenes, the point-line affine invariance constraint is not good at distinguishing the similarity affine relationship of repeating texture units, making it difficult to effectively filter out incorrect matching pairs in repeating textures.

[0006] In summary, existing visual SLAM feature matching schemes suffer from high mismatch rates and insufficient robustness in environments with weak textures, repetitive textures, and complex environments. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a visual SLAM feature matching method based on topological feature constraints.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A visual SLAM feature matching method based on topological feature constraints, the method comprising: S1. Acquire single-frame images from the visual SLAM system, collect the monitoring point features of the single-frame images and calculate the corresponding descriptors, and optimize the feature point set according to neighborhood density, response intensity and descriptor redundancy rules. S2. Generate triangular topological unit structures on the optimized feature set through triangulation, and calculate the side length, interior angle, area, and normalized side ratio geometric properties of each triangular topological unit. S3. Perform nearest neighbor matching on the feature descriptors of adjacent image frames, and generate candidate matching pairs by combining Lowe ratio test, mutual inspection principle and preliminary geometric constraints including epipolar constraints. S4. Extract the relevant frames from each candidate matching pair. n For each neighborhood subgraph, the similarity indices of side length ratio consistency and interior angle consistency are calculated. The validity of the matching is determined by weighted synthesis of topological scores, and invalid matches are eliminated. S5. Within a set time window, perform a weighted cumulative score on the pass rate of the matching pairs, filter out robust matching pairs, and output the feature matching results.

[0009] Furthermore, step S1 includes at least the following steps: A keypoint detector is used to analyze a single frame of image. Perform feature point detection to obtain an initial set of feature points. ,in For feature points Two-dimensional coordinates, This represents the initial number of feature points; For feature set Each feature point in Extract the gradient features of its local neighborhood to generate floating-point descriptors. :

[0010] in Therefore The local neighborhood centered on, Operators are extracted from floating-point descriptors to obtain the final descriptor set. ; The feature set is filtered based on neighborhood density, response intensity, and descriptor redundancy. Feature points and their corresponding descriptors that do not meet the filtering rules are removed from the feature set. and descriptor subsets Remove from the list.

[0011] Furthermore, the neighborhood density, response intensity, and descriptor redundancy screening in step S1 are as follows: Neighborhood density filtering: Define neighborhood radius Calculate each feature point In radius Number of neighboring points: ,like ,Will and corresponding from and Remove from the middle; Response intensity filtering: A Gaussian difference pyramid is constructed on the original image, and several consecutive layers of Gaussian blurred images are grouped together. o The group index where the feature point belongs. The layer index of the feature point within the group; for the detected feature points First, its pixel coordinates in the original image are mapped to local coordinates within the corresponding group. Extract detector response value :

[0012] in, Indicates the first o DOG differential image data of the group Indicates taking the absolute value; sets the response value threshold. ,like Then the feature points and corresponding descriptors Remove from the set; Descriptor redundancy filtering: for any Descriptor similarity is calculated using the normalized similarity formula:

[0013] in, Calculate the Euclidean distance for the descriptors; set a similarity threshold. Spatial distance threshold If the descriptor similarity is satisfied And the spatial distance of feature points If the two feature points are deemed redundant and duplicated, the feature point with the larger response value is retained, and the feature point with the smaller response value is removed. Remove from the list.

[0014] Furthermore, in step S2, the set of triangular topological units is represented as:

[0015] in, Indicates by 、 、 The triangle formed , Point With point distance, Point With point distance, Point With point distance, , , Representing triangles vertex , , Corresponding interior angle size, Represents a triangle area, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio that constitutes the edge.

[0016] Furthermore, the process of constructing the triangular topological unit set in step S2 is as follows: S21. For each feature point Calculate the Euclidean distance between the point and all other feature points, and select the two closest points to form a triangle that satisfies the triangle inequality. , , and interior angle constraints; S22, Triangle Interior Point Detection: For triangular topological units composed of feature points... First, calculate the parameters of the circumcircle of the triangle, including the center. and radius Then, iterate through other feature points and find those that satisfy... point Perform subsequent interior point detection; for the points to be detected that are inside the circumcircle. Construct vector , , , , , ,definition:

[0017]

[0018]

[0019] like 、 、 If the signs are completely identical, that is, all signs are either positive or all signs are negative, then the judgment is... If it is located inside a triangle, it is otherwise considered a triangle. External points; if all points to be detected are determined to be external points, then No interior points, preserved Otherwise, return to step S21; S23. The triangle obtained in step S21 Calculations, including side length calculations , , Area calculation:

[0020]

[0021] in, It is half the perimeter. For area; S24. Stability Screening: If the area of ​​the triangle... ,or Then the triangle is removed, where This is the minimum threshold for the area of ​​the triangle. The threshold value for the side length ratio; S25. Calculation of geometric quantities of a triangle, including interior angle calculation. , , , and normalized edge ratio calculation , Then join in and mark the point After marking No longer participating in step S21; S26. Repeat steps S21-S25 until the remaining set of feature points can no longer form a valid triangle, finally obtaining the topological model. .

[0022] Furthermore, in step S3, candidate matching pairs are screened through descriptor matching and geometric constraints, including nearest neighbor matching, Lowe's ratio test, and preliminary geometric filtering, which are as follows: Proximity distance calculation: for the current time For adjacent frames and Triangular topological units composed of feature points and In clockwise order, and The floating-point descriptors of the three vertices are concatenated as [ ]| and[ ]| Using Euclidean distance measure, for[ ]| and[ ]| The Euclidean distance; For each triangular topological unit Calculate its relationship with all triangular topological units in adjacent frames. Obtain the triangle Most similar triangle The distance is denoted as the nearest neighbor distance. and second-similar triangles The distance is denoted as the second nearest neighbor distance. ; Lowe ratio test: If Then keep Nearest neighbor matching triangle As a candidate, The threshold is the ratio of the nearest distance to the second nearest distance; otherwise, the matched pair is discarded. ; Topological geometry filtering: Calculate matching triangle pairs Corresponding vertex pairs polar error ,in Based on the fundamental matrix, the epipolar constraint relationship between two frames is described by solving for the calibration parameters and matching point pairs of adjacent frames. To match the homogeneous coordinates corresponding to the vertex pixel coordinates, matrix operations are used to satisfy the epipolar constraint requirements; if the epipolar error... If so, then keep the matching pair. This is the epipolar error threshold. Otherwise, the matched pair is discarded. .

[0023] Furthermore, in step S4, through topological sub- Figure 1 Consistency verification filters out false matches, including The processes of extracting the neighborhood subgraph, calculating the similarity index, and determining the comprehensive score are as follows: Neighborhood subgraph extraction: for candidate matching triangle topological unit pairs , respectively in and Extracting from the triangular topology graph Rank neighborhood subgraph and ; Subgraph similarity index calculation: For subgraph pairs Calculate similarity metrics, including the consistency of side length ratios. , interior angle consistency ,in:

[0024]

[0025] Comprehensive Topology Score and Judgment: Weighted Composite Topology Score:

[0026] Set threshold ,like If, keep the matching pair; The matching pair is then removed. After comprehensive topology scoring, a preliminary set of reliable matching pairs is obtained. .

[0027] Furthermore, step S5 includes a time-window weighted accumulation and a robust matching and filtering process, wherein... Introducing a sliding time window to filter out instantaneous false matches by utilizing the temporal continuity of matched pairs: setting the time window length. The window covers the current frame. and before frame ,right Each cross-frame feature matching relationship in the data is denoted as . , indicating the current frame Features Compared to the past frame Features A matching association exists; define an indicator function. If in If a corresponding matching unit exists in the frame and passes topological verification, then ,otherwise 0; Calculate the weighted cumulative score:

[0028] in The weighting is time-based, with time nodes closer to the current frame having a higher weight. Set robustness threshold Filter matching pairs with temporal stability, where the robustness threshold is used. Represented as:

[0029] in The pass rate coefficient is the cumulative score of the matched pairs. If they are matched, they are marked as robust matching pairs and included in the final matching result set. Otherwise, they are discarded. Finally, the set of matching results is output. .

[0030] The beneficial effects of this invention are as follows: This invention effectively reduces the false matching rate in the feature matching process by constructing triangular topological units and introducing a topological consistency screening mechanism. This invention not only relies on the similarity of local descriptors but also comprehensively considers the spatial topological relationships between feature points, thus achieving accurate feature matching even in environments with repetitive or weak textures.

[0031] This invention introduces a time-series robust filtering mechanism that performs weighted cumulative scoring on matching pairs within a time window to filter out long-term stable matching pairs. This mechanism effectively filters out instantaneous false matches and enhances the system's robustness in complex scenarios such as dynamic environments or changes in lighting.

[0032] This invention improves the localization accuracy of visual SLAM systems. Accurate feature matching provides reliable input for pose estimation and mapping, enabling the system to achieve more precise autonomous localization in complex environments. By comprehensively utilizing topological feature constraints and temporal robust filtering, this invention enhances the overall stability of visual SLAM systems in complex environments, better addressing challenges such as scale changes, occlusion, and illumination interference, ensuring continuous and stable operation.

[0033] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall process of the visual SLAM feature matching method based on topological feature constraints according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the feature point selection process according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating an example of triangle topological feature formation according to an embodiment of the present invention. Detailed Implementation

[0035] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0036] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0037] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0038] Please see Figures 1-3 This is a visual SLAM feature matching method based on topological feature constraints.

[0039] Example This embodiment details the implementation process of a visual SLAM feature matching method based on topological feature constraints, such as... Figure 1 As shown, it includes at least the following steps: S1. Acquire single-frame images from the visual SLAM system, collect the monitoring point features of the single-frame images and calculate the corresponding descriptors, and optimize the feature point set according to neighborhood density, response intensity and descriptor redundancy rules. S2. Generate triangular topological unit structures on the optimized feature set through triangulation, and calculate the side length, interior angle, area, and normalized side ratio geometric properties of each triangular topological unit. S3. Perform nearest neighbor matching on the feature descriptors of adjacent image frames, and generate candidate matching pairs by combining Lowe ratio test, mutual inspection principle and preliminary geometric constraints including epipolar constraints. S4. Extract the relevant frames from each candidate matching pair. n For each neighborhood subgraph, the similarity indices of side length ratio consistency and interior angle consistency are calculated. The validity of the matching is determined by weighted synthesis of topological scores, and invalid matches are eliminated. S5. Within a set time window, perform a weighted cumulative score on the pass rate of the matching pairs, filter out robust matching pairs, and output the feature matching results.

[0040] Step S1 in this embodiment aims to filter feature points from the original image to provide a reliable foundation for subsequent topology construction. Specifically, it includes three sub-steps: feature detection, descriptor calculation, and feature quality assessment and removal. 1.1 Feature Detection A keypoint detector is used to analyze a single frame of image. Perform feature point detection to obtain an initial set of feature points. ,in For feature points Two-dimensional coordinates, This represents the initial number of feature points.

[0041] 1.2 Descriptor Computation For feature set Each feature point in Extract the gradient features of its local neighborhood to generate floating-point descriptors. :

[0042] in Therefore The local neighborhood centered on, Operators are extracted from floating-point descriptors to obtain the final descriptor set. .

[0043] 1.3 Feature Quality Assessment and Elimination The feature set is filtered based on neighborhood density, response intensity, and descriptor redundancy. Feature points and their corresponding descriptors that do not meet the filtering rules are removed from the feature set. and descriptor subsets Remove from the middle: 1.3.1 Neighborhood density screening Define neighborhood radius Calculate each feature point In radius Number of neighboring points: ,like ,Will and corresponding from and Remove from the list.

[0044] 1.3.2 Response Intensity Screening: A Difference of Gaussian (DoG) pyramid is constructed from the original image, dividing several consecutive layers of Gaussian blurred images into a group, where... o The group index where the feature point belongs. This is the layer index of the feature point within the group. For the detected feature points... First, its pixel coordinates in the original image are mapped to local coordinates within the corresponding group. Extract detector response value :

[0045] in, Indicates the first o DOG differential image data of the group Indicates taking the absolute value. Sets the response value threshold. ,like If the feature point has low contrast and is susceptible to noise interference, then the feature point is removed from the list. and corresponding descriptors Remove from the set.

[0046] 1.3.3 Descriptor Redundancy Filtering: For any Descriptor similarity is calculated using the normalized similarity formula:

[0047] in, Calculate the Euclidean distance between the descriptors. Set a similarity threshold. Spatial distance threshold If the descriptor similarity is satisfied And the spatial distance of feature points If the two feature points are found to be redundant and duplicated, then the feature point with the larger response value is retained, and the feature point with the smaller response value is removed. Remove from the list.

[0048] In this embodiment, step S2 organizes feature points into triangular topological units through triangulation, providing a structural basis for topology verification. Specifically, it includes two sub-steps: triangle generation and geometric quantity calculation. 2.1 Triangle Generation: Establishing a Triangle Set Model:

[0049] in, Indicates by 、 、 The triangle formed , Point With point distance, Point With point distance, Point With point distance, , , Representing triangles vertex , , Corresponding interior angle size, Represents a triangle area, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio constitutes the edge. The specific process is as follows: (1) For each feature point Calculate the Euclidean distance between the point and all other feature points, and select the two closest points to form a triangle that satisfies the triangle inequality. , , and interior angle constraints.

[0050] (2) Triangle interior point detection: For a triangular topological unit composed of feature points... First, calculate the parameters of the circumcircle of the triangle, including the center. and radius Then, iterate through other feature points and find those that satisfy... point Perform subsequent interior point checks.

[0051] For the point to be detected within the circumcircle Construct vector , , , , , ,definition:

[0052]

[0053]

[0054] like 、 、 If the signs are completely identical, that is, all signs are either positive or all signs are negative, then the judgment is... If it is located inside a triangle, it is otherwise considered a triangle. External points. If all points to be detected are determined to be... external points, then No interior points, preserved Otherwise, return to step 1.

[0055] (3) The triangle obtained in step (2) calculate: Side length calculation:

[0056] , And so on.

[0057] Area calculation:

[0058]

[0059] in It is half the perimeter. For area.

[0060] (4) Stability screening: If the area of ​​the triangle ,or If so, then the triangle is removed. This is the minimum threshold for the area of ​​the triangle. This is the threshold for the side length ratio.

[0061] (5) Calculation of geometric quantities of triangles, including interior angle calculation and normalized side ratio calculation.

[0062] Interior angle calculation, based on vertex For example:

[0063] in, It is a triangle Vertex in The corresponding interior angles, , Triangles vertex , The corresponding interior angles are calculated in the same way as above.

[0064] Normalized edge ratio calculation:

[0065] in, It is a triangle midpoint and The normalized edge ratio that constitutes the edge, , Similarly, the calculation method is the same as above. join in and mark the point After marking No longer participating in step 1.

[0066] (6) Repeat steps (1) to (5) until the remaining feature point set cannot form a valid triangle, and finally obtain the topological model. .

[0067] Step S3 in this embodiment filters candidate matching pairs through descriptor matching and geometric constraints, specifically including three sub-steps: nearest neighbor matching, Lowe's ratio test, and preliminary geometric filtering. 3.1 Calculation of nearest neighbor distance For the current moment For adjacent frames and Triangular topological units composed of feature points and In clockwise order, and The floating-point descriptors of the three vertices are concatenated as [ ]| and[ ]| Using Euclidean distance measure, for[ ]| and[ ]| The Euclidean distance.

[0068] For each triangular topological unit Calculate its relationship with all triangular topological units in adjacent frames. Obtain the triangle Most similar triangle The distance is denoted as the nearest neighbor distance. and second-similar triangles The distance is denoted as the second nearest neighbor distance. .

[0069] 3.2 Lowe's ratio test like Then keep Nearest neighbor matching triangle As a candidate, The threshold is the ratio of the nearest distance to the second nearest distance; otherwise, the matched pair is discarded. .

[0070] 3.3 Topological Geometry Filtering Calculate matching triangle pairs Corresponding vertex pairs polar error ,in Based on the fundamental matrix, the epipolar constraint relationship between two frames is described by solving for the calibration parameters and matching point pairs of adjacent frames. To match the homogeneous coordinates corresponding to the vertex pixel coordinates, matrix operations are used to satisfy the epipolar constraint requirements. If the epipolar error... If so, then keep the matching pair. This is the epipolar error threshold. Otherwise, the matched pair is discarded. .

[0071]

[0072] in This is the intrinsic parameter matrix of the camera. for arrive The rotation matrix, for arrive The translation vector.

[0073] Step S4 in this embodiment uses topological sub- Figure 1 Consistency verification filters out false matches, specifically including The three sub-steps are: extraction of the neighborhood subgraph, calculation of similarity index, and determination of comprehensive score. 4.1 Order neighborhood subgraph extraction For candidate matching triangle topological element pairs , respectively in and Extracting from the triangular topology graph Rank neighborhood subgraph and ; 4.2 Subgraph Similarity Index Calculation For subgraph pairs Calculate three similarity indicators: Consistency of side length ratio For the corresponding edge set calculate:

[0074] in This is the normalized edge ratio of the corresponding edge.

[0075] interior angle consistency For the corresponding vertex set calculate:

[0076] in , for , point , inside angle.

[0077] 4.3 Comprehensive Topology Score and Judgment Weighted composite topology score:

[0078] Set threshold ,like If, keep the matching pair; The matching pair is then removed. After comprehensive topology scoring, a preliminary set of reliable matching pairs is obtained. .

[0079] Step S5 in this embodiment filters out long-term stable matching pairs through time continuity verification, providing input for the SLAM backend module. Specifically, it includes two sub-steps: time window weighted accumulation and robust matching filtering. 5.1 Time Window Weighted Cumulative Introducing a sliding time window, the temporal continuity of matched pairs is used to filter out instantaneous false matches: Window parameter settings: Set the length of the time window The window covers the current frame. and before frame ; Weighted score calculation: for Each cross-frame feature matching relationship in the data is denoted as . , indicating the current frame Features Compared to the past frame Features A matching association exists. Define an indicator function. If in If a corresponding matching unit exists in the frame and passes topological verification, then ,otherwise 0.

[0080] Calculate the weighted cumulative score:

[0081] in The time weight is determined by the time period; the closer the time node is to the current frame, the higher the weight.

[0082] 5.2 Robust Matching Filtering Set robustness threshold We then filter for matching pairs that exhibit temporal stability. The threshold is calculated as follows:

[0083] in The pass rate coefficient is the cumulative score of the matched pairs. If they are matched, they are marked as robust matching pairs and included in the final matching result set. Otherwise, they are discarded. Finally, the set of matching results is output. .

[0084] like Figure 2 The diagram illustrates a visual SLAM feature point selection strategy based on topological feature constraints according to the present invention. Through a triple-rule selection process based on neighborhood density, response intensity, and descriptor redundancy, a feature set and corresponding descriptor sets are obtained. Black represents retained points, while points a, b, and c are discarded. The reason for this is that the feature points... In the neighborhood radius r The number of neighborhood points within a feature point is lower than a set threshold; detector response value Below a set threshold; for feature point c, calculate the similarity, point c If a feature point has a similarity to another point that exceeds a threshold, one of the feature points should be removed. The response values ​​of the two points should then be compared. c Points below similarity points are therefore removed. c.

[0085] like Figure 3 The image shows an example of triangle topological feature formation for visual SLAM feature matching based on topological feature constraints according to the present invention. Black triangles represent those successfully matched and retained. Feature descriptors of triangles from two adjacent frames are matched to calculate the frame... Mid-topological features With frames Middle descriptor and The distances are respectively = and = Obviously Then keep Nearest neighbor matching triangle As a candidate, This is the threshold for the ratio of the nearest neighbor distance to the second nearest neighbor distance.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A visual SLAM feature matching method based on topological feature constraints, characterized in that: The method includes: S1. Acquire single-frame images from the visual SLAM system, collect the monitoring point features of the single-frame images and calculate the corresponding descriptors, and optimize the feature point set according to neighborhood density, response intensity and descriptor redundancy rules. S2. Generate triangular topological unit structures on the optimized feature set through triangulation, and calculate the side length, interior angle, area, and normalized side ratio geometric properties of each triangular topological unit. S3. Perform nearest neighbor matching on the feature descriptors of adjacent image frames, and generate candidate matching pairs by combining Lowe ratio test, mutual inspection principle and preliminary geometric constraints including epipolar constraints. S4. Extract the relevant frames from each candidate matching pair. n For each neighborhood subgraph, the similarity indices of side length ratio consistency and interior angle consistency are calculated. The validity of the matching is determined by weighted synthesis of topological scores, and invalid matches are eliminated. S5. Within a set time window, perform a weighted cumulative score on the pass rate of the matching pairs, filter out robust matching pairs, and output the feature matching results.

2. The visual SLAM feature matching method based on topological feature constraints according to claim 1, characterized in that: Step S1 includes at least the following steps: A keypoint detector is used to analyze a single frame of image. Perform feature point detection to obtain an initial set of feature points. ,in For feature points Two-dimensional coordinates, This represents the initial number of feature points; For feature set Each feature point in Extract the gradient features of its local neighborhood to generate floating-point descriptors. : in Therefore The local neighborhood centered on, Operators are extracted from floating-point descriptors to obtain the final descriptor set. ; The feature set is filtered based on neighborhood density, response intensity, and descriptor redundancy. Feature points and their corresponding descriptors that do not meet the filtering rules are removed from the feature set. and descriptor subsets Remove from the list.

3. The visual SLAM feature matching method based on topological feature constraints according to claim 2, characterized in that: The neighborhood density, response intensity, and descriptor redundancy screening in step S1 are as follows: Neighborhood density filtering: Define neighborhood radius Calculate each feature point In radius Number of neighboring points: ,like ,Will and corresponding from and Remove from the middle; Response intensity filtering: A Gaussian difference pyramid is constructed on the original image, and several consecutive layers of Gaussian blurred images are grouped together. o The group index where the feature point belongs. The layer index of the feature point within the group; for the detected feature points First, its pixel coordinates in the original image are mapped to local coordinates within the corresponding group. Extract detector response values : in, Indicates the first o DOG differential image data of the group Indicates taking the absolute value; sets the response value threshold. ,like Then the feature points and corresponding descriptors Remove from the set; Descriptor redundancy filtering: for any Descriptor similarity is calculated using the normalized similarity formula: in, Calculate the Euclidean distance for the descriptors; set a similarity threshold. Spatial distance threshold If the descriptor similarity is satisfied And the spatial distance of feature points If the two feature points are deemed redundant and duplicated, the feature point with the larger response value is retained, and the feature point with the smaller response value is removed. Remove from the list.

4. The visual SLAM feature matching method based on topological feature constraints according to claim 2, characterized in that: In step S2, the set of triangular topological units is represented as: in, Indicates by 、 、 The triangle formed , Point With point distance, Point With point distance, Point With point distance, , , Representing triangles vertex , , Corresponding interior angle size, Represents a triangle area, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio that constitutes the edge, express midpoint With point The normalized edge ratio that constitutes the edge.

5. The visual SLAM feature matching method based on topological feature constraints according to claim 2, characterized in that: The process of constructing the triangular topological unit set in step S2 is as follows: S21. For each feature point Calculate the Euclidean distance between the point and all other feature points, and select the two closest points to form a triangle that satisfies the triangle inequality. , , and interior angle constraints; S22, Triangle Interior Point Detection: For triangular topological units composed of feature points... First, calculate the parameters of the circumcircle of the triangle, including the center. and radius Then, iterate through other feature points and find those that satisfy... point Perform subsequent interior point detection; for the points to be detected that are inside the circumcircle. Construct vector , , , , , ,definition: like 、 、 If the signs are completely identical, that is, all signs are either positive or all signs are negative, then the judgment is... If it is located inside a triangle, it is otherwise considered a triangle. External points; if all points to be detected are determined to be external points, then No interior points, preserved Otherwise, return to step S21; S23. The triangle obtained in step S21 Calculations, including side length calculations , , Area calculation: in, It is half the perimeter. For area; S24. Stability Screening: If the area of ​​the triangle... ,or Then the triangle is removed, where This is the minimum threshold for the area of ​​the triangle. The threshold value for the side length ratio; S25. Calculation of geometric quantities of a triangle, including interior angle calculation. , , , and normalized edge ratio calculation , Then join in and mark the point After marking No longer participating in step S21; S26. Repeat steps S21-S25 until the remaining set of feature points can no longer form a valid triangle, finally obtaining the topological model. .

6. The visual SLAM feature matching method based on topological feature constraints according to claim 4, characterized in that: In step S3, candidate matching pairs are screened through descriptor matching and geometric constraints, including nearest neighbor matching, Lowe's ratio test, and preliminary geometric filtering, which are as follows: Proximity distance calculation: for the current time For adjacent frames and Triangular topological units composed of feature points and In clockwise order and The floating-point descriptors of the three vertices are concatenated as [ ]| and[ ]| Using Euclidean distance measure, for[ ]| and[ ]| The Euclidean distance; For each triangular topological unit Calculate its relationship with all triangular topological units in adjacent frames. Obtain the triangle Most similar triangle The distance is denoted as the nearest neighbor distance. and second-similar triangles The distance is denoted as the second nearest neighbor distance. ; Lowe ratio test: If Then keep Nearest neighbor matching triangle As a candidate, The threshold is the ratio of the nearest distance to the second nearest distance; otherwise, the matched pair is discarded. ; Topological geometry filtering: Calculate matching triangle pairs Corresponding vertex pairs polar error ,in Based on the fundamental matrix, the epipolar constraint relationship between two frames is described by solving for the calibration parameters and matching point pairs of adjacent frames. To match the homogeneous coordinates corresponding to the vertex pixel coordinates, matrix operations are used to satisfy the epipolar constraint requirements; if the epipolar error... If so, then keep the matching pair. This is the epipolar error threshold. Otherwise, the matched pair is discarded. .

7. The visual SLAM feature matching method based on topological feature constraints according to claim 6, characterized in that: In step S4, false matches are filtered out through topological subgraph consistency verification, which includes... The processes of extracting the neighborhood subgraph, calculating the similarity index, and determining the comprehensive score are as follows: Neighborhood subgraph extraction: for candidate matching triangle topological unit pairs , respectively in and Extracting from the triangular topology graph Rank neighborhood subgraph and ; Subgraph similarity index calculation: For subgraph pairs Calculate similarity metrics, including the consistency of side length ratios. , interior angle consistency ,in: Comprehensive Topology Score and Judgment: Weighted Composite Topology Score: Set threshold ,like If, keep the matching pair; The matching pair is then removed. After comprehensive topology scoring, a preliminary set of reliable matching pairs is obtained. .

8. The visual SLAM feature matching method based on topological feature constraints according to claim 7, characterized in that: Step S5 includes a time window weighted accumulation and robust matching filtering process, wherein... Introducing a sliding time window to filter out instantaneous false matches by utilizing the temporal continuity of matched pairs: setting the time window length. The window covers the current frame. and before frame ,right Each cross-frame feature matching relationship in the data is denoted as . , indicating the current frame Features Compared to the past frame Features A matching association exists; define an indicator function. If in If a corresponding matching unit exists in the frame and passes topological verification, then ,otherwise 0; Calculate the weighted cumulative score: in The weighting is time-based, with time nodes closer to the current frame having a higher weight. Set robustness threshold Filter matching pairs with temporal stability, where the robustness threshold is used. Represented as: in The pass rate coefficient is the cumulative score of the matched pairs. If they are matched, they are marked as robust matching pairs and included in the final matching result set. Otherwise, they are discarded. Finally, the set of matching results is output. .

Images