Method and system for fast extraction of aerial triangulation connection points based on hierarchical clustering

Abstract

The present application belongs to the field of three-dimensional geographic information processing, and relates to a method for extracting connection points of images rapidly by adopting hierarchical clustering to cluster feature points of all images. Firstly, feature points of all images in a survey area are merged into a set; then, clustering operation is performed on the set, and recursive clustering is performed on each clustering result; finally, similarity of feature points in each cluster is judged, and feature points with sufficient similarity are connection points of the same group. Through the method, the problem of feature point matching can be converted into a clustering problem, so that mature clustering methods are used to extract connection points, the efficiency of extraction of connection points in aerial triangulation is improved, and large-scale aerial triangulation of images can be performed.

Description

Technical Field

[0001] This invention belongs to the field of 3D geographic information processing and relates to a method for rapidly extracting image connectivity points by clustering feature points of all images using hierarchical clustering. The key technologies of this scheme are utilizing hierarchical clustering instead of matching and the termination condition for hierarchical clustering. Background Technology

[0002] Aerial triangulation is a crucial step in photogrammetry, used to reconstruct 3D models from 2D images, acquire digital elevation models and digital orthophotos, and thus perform 3D landscape simulation and achieve digital twins. Extracting tie points is an essential part of aerial triangulation, and the speed of tie point extraction significantly impacts the operating speed of the aerial triangulation system. In the era of big data, aerial triangulation images are characterized by their large quantity, varying quality, and significant spatiotemporal heterogeneity, placing high demands on the operating speed of aerial triangulation systems. Therefore, accelerating tie point extraction has become increasingly important.

[0003] The commonly used method for extracting connectivity points involves sequentially measuring the similarity of feature points on one image with other images to determine the connectivity points of that point in other images and grouping them together. This process is then repeated for the next feature point. While simple to implement, this method is inefficient and time-consuming due to the comparison of similarity between feature points across different images, making it unsuitable for aerial triangulation tasks involving a large number of images. Summary of the Invention

[0004] To address the shortcomings of current methods for extracting connector points in terms of computational efficiency and time, this invention proposes a fast connector point extraction method based on hierarchical clustering in aerial triangulation. The essence of connector point extraction is to find a set of feature points from various images that meet certain similarity requirements. Therefore, this invention merges the feature points from each image into a set, and then performs hierarchical clustering on this set until the feature points in each class are sufficiently similar. The feature points in each class are then considered as a group of connector points, thus completing the connector point extraction process. This method transforms the feature point matching problem into a clustering problem, utilizing clustering methods to achieve connector point extraction and improving the efficiency of connector point extraction.

[0005] The technical solution of this invention is a method for rapid extraction of aerial triangulation connection points based on hierarchical clustering. First, the feature points of all images in the survey area are merged into a set; then, clustering is performed on the set, and recursive clustering is performed on each clustering result; finally, the similarity of feature points in each cluster is judged, and feature points that are sufficiently similar are the same group of connection points.

[0006] The implementation process includes the following steps:

[0007] Step 1: Use a feature extraction algorithm to extract and describe feature points in all images of the survey area;

[0008] Step 2: Merge all feature points into a set, which serves as the root node of the clustering tree, and record the location information of each feature point on the image.

[0009] Step 3: Cluster the feature points within a node, and use the clustering results as child nodes of that node, and repeat this clustering process continuously.

[0010] Step 4: Check whether the similarity of feature points in child nodes reaches the threshold. If the threshold is reached, the feature points in child nodes will no longer be clustered.

[0011] Step 5: Take the feature points of each leaf node of the clustering tree as a set of connection points, and use the feature point information recorded in Step 2 to restore the connection points between images, thus completing the rapid extraction of connection points.

[0012] Furthermore, step 3 includes the following sub-steps:

[0013] Step 3.1: Estimate the number of clusters used in each clustering level;

[0014] Step 3.2: Use a clustering algorithm to cluster the feature points to be clustered to obtain child nodes;

[0015] Step 3.3: Repeat step 3.2 to recursively cluster the child nodes until the child nodes meet the termination condition. When all the child node results no longer need to be clustered, the hierarchical clustering is completed, and a clustering tree is obtained.

[0016] Furthermore, the formula used to calculate the cluster number in step 3.1 is as follows:

[0017]

[0018] Where N is the total number of feature points in all images, n is the average number of connection points in a set of connection points, and H is the number of layers required for hierarchical clustering.

[0019] Furthermore, in step 3.1, the number of clusters is directly taken as an empirical value.

[0020] Furthermore, step 3.2 specifically includes:

[0021] Step 3.2.1: Randomly select a feature point from the feature points as the cluster centroid. ci The clusters corresponding to each centroid are: i ;

[0022] Step 3.2.2, set each feature point jAssigned to the nearest ci Clustering i middle;

[0023] Step 3.2.3: Recalculate the cluster centroids based on the newly assigned clusters. ci :

[0024]

[0025] Where, x m For cluster X i Feature points in n i For cluster X i The cardinality;

[0026] Step 3.2.4: Repeat steps 3.2.2 and 3.2.3 until the required number of iterations is reached or the change in the centroid is less than the threshold ε.

[0027] ||Δx ci ||<ε

[0028] Where, Δx ci represents the change in the cluster centroids between the two iterations, and ‖·‖ represents the norm.

[0029] Furthermore, step 4 includes the following sub-steps:

[0030] Step 4.1: Calculate the maximum distance d between each feature point in the clustering results. max Second largest value 2ndmax and the third largest value 3rdmax ;

[0031] Step 4.2: Determine whether the endpoints of the three longest distances have a common feature point. If so, it indicates that the point may be an outlier or the clustering results are significantly heterogeneous, and further clustering is required. Proceed to step 4.4; otherwise, proceed to step 4.3.

[0032] Step 4.3: Calculate whether the similarity of the clustering results meets the requirements. The similarity r is based on the maximum distance d between each feature point. max Second largest value 2ndmax Perform the calculation. If the similarity is greater than a, the clustering results are considered to have consistent similarity, and the clustering ends. Otherwise, proceed to step 4.4.

[0033] Step 4.4, select the maximum value max The two corresponding feature points are used as the initial centroids for clustering, and the remaining centroids are selected randomly before clustering.

[0034] Furthermore, the formula for calculating the similarity r is as follows:

[0035]

[0036] This invention also provides a rapid extraction system for aerial triangulation connection points based on hierarchical clustering, comprising the following modules:

[0037] The feature point description module is used to extract and describe feature points for all images in the survey area using feature extraction algorithms.

[0038] The information recording module is used to merge all feature points into a set, which serves as the root node of the clustering tree, and records the location information of each feature point on the image.

[0039] The feature point clustering module is used to cluster the feature points within a node. The clustering result is used as the child node of that node, and this clustering process is repeated continuously.

[0040] The similarity judgment module is used to check whether the similarity of feature points in child nodes reaches a threshold. If the threshold is reached, the feature points in child nodes will no longer be clustered.

[0041] The connection point extraction module is used to take the feature points of each leaf node of the clustering tree as a group of connection points, and use the recorded feature point information to restore the connection points between images, thus completing the rapid extraction of connection points.

[0042] Furthermore, the system includes a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute the above-described method for rapid extraction of aerial triangulation connection points based on hierarchical clustering.

[0043] Furthermore, the system includes a readable storage medium storing a computer program, which, when executed, implements the method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described above.

[0044] Compared with existing technologies, the compression scheme of the present invention starts from the essence of tie point extraction, transforms the problem of feature point matching into a clustering problem, and thus uses mature clustering methods to achieve tie point extraction, improving the efficiency of tie point extraction in aerial triangulation and facilitating aerial triangulation of large-scale images. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the rapid extraction method for aerial triangulation connection points according to an embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of hierarchical clustering in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram illustrating outliers or cluster heterogeneity in an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram illustrating the clustering termination condition determination in an embodiment of the present invention. Detailed Implementation

[0049] This invention provides a fast method for extracting aerial triangulation tie points based on hierarchical clustering. The core of the method consists of three processes: first, extracting and describing image feature points and merging them into a set; then, performing hierarchical clustering on this set and detecting whether the clustering results meet similarity requirements; finally, mapping the clustering results back onto the image to obtain tie points.

[0050] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1 As shown, the method for rapid extraction of aerial triangulation connection points based on hierarchical clustering according to an embodiment of the present invention includes the following steps:

[0052] Step 1: Use feature extraction algorithms that can perform feature description and similarity measurement, such as SIFT artificial features or neural network learning feature extraction methods, to extract and describe feature points for all images in the survey area.

[0053] Step 2: All feature points are grouped into a set, which serves as the root node of the clustering tree, and the location information of each feature point on the image is recorded.

[0054] Step 3: Cluster the feature points within a node, and use the clustering results as the child nodes of that node. For example... Figure 2 As shown, the clustering results at each layer can be further divided into smaller sub-nodes through clustering. The specific steps of step 3 are as follows:

[0055] Step 3.1, estimate the number of clusters used in each clustering level (or directly use the empirical value k = 10). The calculation formula is as follows:

[0056]

[0057] Where N is the total number of feature points in all images, n is the average number of connection points in a set of connection points, and H is the number of layers required for hierarchical clustering.

[0058] Step 3.2: Using a clustering algorithm, cluster the feature points to be clustered to obtain child nodes. Specific steps include:

[0059] Step 3.2.1: Randomly select a feature point from the feature points as the cluster centroid. ci The clusters corresponding to each centroid are: i ;

[0060] Step 3.2.2, set each feature point jAssigned to the nearest ci Clustering i middle;

[0061] Step 3.2.3: Recalculate the cluster centroids based on the newly assigned clusters. ci :

[0062]

[0063] Where, x m For cluster X i Feature points in n i For cluster X i The cardinality (number of elements);

[0064] Step 3.2.4: Repeat steps 3.2.2 and 3.2.3 until the required number of iterations is reached or the change in the centroid is less than the threshold ε.

[0065] ||Δx ci ||<ε

[0066] Where, Δx ci represents the change in the cluster centroids between the two iterations, and ‖·‖ represents the norm.

[0067] Step 3.3: Repeat step 3.2 to recursively cluster the child nodes until the child nodes meet the termination condition in step 4. When all the child node results no longer need to be clustered, the hierarchical clustering is completed, and a clustering tree is obtained.

[0068] Step 4: Check whether the similarity of feature points in the child nodes meets the requirements. For example... Figure 3 Figure 4 As shown, if the requirements are met, clustering of feature points in child nodes will no longer be performed. Specific steps include:

[0069] Step 4.1: Calculate the maximum distance d between each feature point in the clustering results. max Second largest value 2ndmax and the third largest value 3rdmax ;

[0070] Step 4.2: Determine whether the endpoints of the three longest distances have a common feature point. If so, it indicates that the point may be an outlier or the clustering results are significantly heterogeneous, and further clustering is required. Proceed to step 4.4; otherwise, proceed to step 4.3.

[0071] Step 4.3: Calculate whether the similarity of the clustering results meets the requirements. The calculation formula is as follows:

[0072]

[0073] When the similarity is greater than 0.7, the clustering results are considered to have consistent similarity, and the clustering can be terminated; otherwise, proceed to step 4.4.

[0074] Step 4.4, select the maximum value max The two corresponding feature points are used as the initial centroids for clustering, and the remaining centroids are selected randomly before clustering.

[0075] Step 5: Take the feature points of each leaf node of the clustering tree as a set of connection points, and use the feature point information recorded in Step 2 to restore the connection points between images.

[0076] In specific implementation, the method proposed in the technical solution of this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the protection scope of this invention.

[0077] In some possible embodiments, a system for rapid extraction of aerial triangulation connection points based on hierarchical clustering is provided, comprising the following modules:

[0078] The feature point description module is used to extract and describe feature points for all images in the survey area using feature extraction algorithms.

[0079] The information recording module is used to merge all feature points into a set, which serves as the root node of the clustering tree, and records the location information of each feature point on the image.

[0080] The feature point clustering module is used to cluster the feature points within a node. The clustering result is used as the child node of that node, and this clustering process is repeated continuously.

[0081] The similarity judgment module is used to check whether the similarity of feature points in child nodes reaches a threshold. If the threshold is reached, the feature points in child nodes will no longer be clustered.

[0082] The connection point extraction module is used to take the feature points of each leaf node of the clustering tree as a group of connection points, and use the recorded feature point information to restore the connection points between images, thus completing the rapid extraction of connection points.

[0083] The specific implementation methods and steps of each module are the same, and will not be described in this invention.

[0084] In some possible embodiments, a system for rapid extraction of aerial triangulation connection points based on hierarchical clustering is provided, including a processor and a memory. The memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute the method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described above.

[0085] In some possible embodiments, a system for rapid extraction of aerial triangulation connection points based on hierarchical clustering is provided, including a readable storage medium on which a computer program is stored. When the computer program is executed, it implements the method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described above.

[0086] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A fast method for extracting aerial triangulation connection points based on hierarchical clustering, characterized in that, Includes the following steps: Step 1: Use a feature extraction algorithm to extract and describe feature points in all images of the survey area; Step 2: Merge all feature points into a set, which serves as the root node of the clustering tree, and record the location information of each feature point on the image. Step 3: Cluster the feature points within a node, and use the clustering results as child nodes of that node, and repeat this clustering process continuously. Step 3 includes the following sub-steps: Step 3.1, estimate the number of clusters used in each clustering level. ; Number of clusters in step 3.1 Directly using empirical values, or The following formula is used for calculation: in, The total number of feature points across all images. The average number of connection points in a set of connection points. The number of layers required for hierarchical clustering; Step 3.2: Use a clustering algorithm to cluster the feature points to be clustered to obtain child nodes; Step 3.2 specifically includes: Step 3.2.1, randomly select from feature points Each feature point is used as the cluster centroid. The clusters corresponding to each centroid are: ; Step 3.2.2, set each feature point Assigned to the nearest Clustering middle; Step 3.2.3: Recalculate the cluster centroids based on the newly assigned clusters. : in, For clustering Feature points in For clustering The cardinality; Step 3.2.4: Repeat steps 3.2.2 and 3.2.3 until the required number of iterations is reached or the change in the centroid is less than the threshold. : in, This represents the change in cluster centroids between two iterations. To take the norm; Step 3.3: Repeat step 3.2 to recursively cluster the child nodes until the child nodes meet the termination condition. When all the child node results no longer need to be clustered, the hierarchical clustering is completed, and a clustering tree is obtained. Step 4: Check whether the similarity of feature points in child nodes reaches the threshold. If the threshold is reached, the feature points in child nodes will no longer be clustered. Step 5: Take the feature points of each leaf node of the clustering tree as a set of connection points, and use the feature point information recorded in Step 2 to restore the connection points between images, thus completing the rapid extraction of connection points.

2. The method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described in claim 1, characterized in that: Step 4 includes the following sub-steps: Step 4.1: Calculate the maximum distance between each feature point in the clustering results. Second largest value and the third largest value ; Step 4.2: Determine whether the endpoints of the three longest distances have a common feature point. If they do, it means that the point is an outlier or the clustering result has obvious heterogeneity, and further clustering is needed. Proceed to step 4.4; otherwise, proceed to step 4.

3. Step 4.3: Calculate whether the similarity of the clustering results meets the requirements. Based on the maximum distance between each feature point Second largest value Perform calculations when If the clustering results are considered to have consistent similarity, the clustering process ends; otherwise, proceed to step 4.

4. Step 4.4, select the maximum value The two corresponding feature points are used as the initial centroids for clustering, and the remaining centroids are selected randomly before clustering.

3. The method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described in claim 2, characterized in that: Similarity The calculation formula is as follows: 。 4. A rapid aerial triangulation connection point extraction system based on hierarchical clustering, used to implement the rapid aerial triangulation connection point extraction method based on hierarchical clustering as described in any one of claims 1-3, characterized in that, Includes the following modules: The feature point description module is used to extract and describe feature points for all images in the survey area using feature extraction algorithms. The information recording module is used to merge all feature points into a set, which serves as the root node of the clustering tree, and records the location information of each feature point on the image. The feature point clustering module is used to cluster the feature points within a node. The clustering result is used as the child node of that node, and this clustering process is repeated continuously. The similarity judgment module is used to check whether the similarity of feature points in child nodes reaches a threshold. If the threshold is reached, the feature points in child nodes will no longer be clustered. The connector extraction module is used to take the feature points of each leaf node of the clustering tree as a group of connector points, and use the recorded feature point information to restore the connector points between images, thus completing the rapid extraction of connector points.

5. The rapid extraction system for aerial triangulation connection points based on hierarchical clustering as described in claim 4, characterized in that: It includes a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute the aerial triangulation connection point extraction method based on hierarchical clustering as described in any one of claims 1-3.

6. The aerial triangulation connection point rapid extraction system based on hierarchical clustering as described in claim 4, characterized in that: The method includes a readable storage medium on which a computer program is stored, and when the computer program is executed, it implements the method for rapid extraction of aerial triangulation connection points based on hierarchical clustering as described in any one of claims 1-3.

Images