Iterative registration optimization algorithm based on angle clustering

Abstract

The invention discloses an iterative registration optimization algorithm based on angle clustering, and the algorithm comprises the following steps: inputting an initial scene point cloud and an initial model point cloud, simplifying the initial scene point cloud and the initial model point cloud, and obtaining an initial corresponding point pair set; constructing a compatibility constraint, calculating a compatibility score of each pair of corresponding point pairs, and sequencing to construct a compatibility matrix; performing outer layer circulation: sequentially selecting the foremost point pair from the compatibility matrix as a first corresponding point pair; inner layer circulation: selecting a second corresponding point pair; clustering the corresponding point pairs based on the six rotational degrees of freedom, and selecting all significant clusters to generate a conversion hypothesis; and verifying and selecting an optimal conversion hypothesis as an output conversion matrix. The compatibility among all the corresponding point pairs is evaluated by constructing a compatibility matrix; a simple and effective clustering strategy is adopted, and all significant clusters are considered to generate a conversion hypothesis; the simplified point clouds and the key points are effectively combined through a hypothesis verification strategy, and the accuracy of alignment of the low-overlapping point clouds is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular is an iterative registration optimization algorithm based on angle clustering. Background Art

[0002] Three-dimensional point cloud registration technology is an important research direction in the field of computer vision. This task converts two point clouds into the same coordinate system by calculating the rigid transformation matrix through a transformation estimation algorithm. Currently, point cloud registration based on local features is a popular registration technology because it is robust to interference such as noise, occlusion, and outliers. However, in some challenging scenes (e.g., low overlap and noise), the initial corresponding point pairs generated by local features often contain a large number of erroneous corresponding point pairs (outliers). Achieving accurate and robust registration from these initial corresponding point pairs containing a large number of outliers is a challenging task.

[0003] Currently, point cloud registration methods based on local features are generally divided into two categories. One method first calculates the maximum consistent set from initial corresponding point pairs containing outliers using some geometric constraints (such as distance and angle constraints), and then uses this consistent set to calculate the transformation matrix. However, these geometric constraints can be ambiguous when the proportion of correct corresponding point pairs is low. The other method generally obtains the final transformation matrix by alternately generating and verifying transformation hypotheses. These methods can be further divided into three categories: local reference frame (LRF)-based methods, local reference axis (LRA)-based methods, and methods based on three corresponding point pairs. Among them, methods based on local LRF / A select one or two corresponding point pairs at a time and generate transformation hypotheses based on the LRF / A constructed on these corresponding point pairs. The LRF / A in these two methods is susceptible to various interferences, resulting in weak robustness. Methods based on three correspondences are not affected by the repeatability of LRF / A, but they require three different corresponding point pairs to generate transformation hypotheses each time, which leads to higher computational complexity. Some existing three-point based point cloud registration methods use some geometric constraints to preliminarily screen out some outliers to improve the efficiency and accuracy of the algorithm. However, when the outlier rate is high, obtaining an accurate transformation matrix is ​​still a challenge.

[0004] Therefore, it is crucial to design a point cloud registration algorithm that can accurately complete the registration task in complex and changing scenes. Summary of the Invention

[0005] To address the shortcomings of existing point cloud registration algorithms based on local features, a highly accurate and robust point cloud registration method—an iterative registration optimization algorithm based on angle clustering—is proposed. First, a new method for constructing a compatibility matrix is ​​designed to evaluate the compatibility between all corresponding point pairs. Then, an iterative technique for generating transformation hypotheses is designed. Within this iterative framework, a simple yet effective clustering strategy is proposed, and all significant clusters are considered to generate transformation hypotheses. Finally, a new hypothesis verification strategy is proposed that effectively combines simplified point clouds and key points to improve the accuracy of low-overlap point cloud alignment.

[0006] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0007] An iterative registration optimization algorithm based on angle clustering includes the following steps:

[0008] S10: Input the initial scene point cloud P and the initial model point cloud Q, and obtain the scene point cloud P after simplification. s And model point cloud Q s , obtain the scene point cloud P through feature matching s And model point cloud Q s The initial corresponding point pair set between p i ∈P s ,q i ∈Q s ;

[0009] S20, calculating the compatibility score of each pair of corresponding point pairs in the initial corresponding point pair set H by constructing compatibility constraints, and sorting the corresponding point pairs in descending order based on the compatibility scores to construct a compatibility matrix H′;

[0010] S30, from the compatibility matrix H′, according to the order of the corresponding point pairs in the matrix, select the first corresponding point pair (p i ,q i );

[0011] S40, based on the first corresponding point pair (p i ,q i ) compatibility, select the second corresponding point pair (p j ,q j );

[0012] S50, clustering corresponding point pairs that satisfy the three-point constraint based on the six rotational degrees of freedom, and selecting all significant clusters to generate transformation hypotheses;

[0013] S60, repeat steps S40 and S50 until the loop ends;

[0014] S70, repeat steps S30, S40, S50 and S60 until the loop ends;

[0015] S80, verifying all saved transformation hypotheses and selecting the best transformation hypothesis as the output transformation matrix;

[0016] S90. Output the conversion hypothesis corresponding to the maximum number of inliers.

[0017] Furthermore, in S20, the compatibility constraint is constructed by setting a compatibility score threshold to filter out corresponding point pairs that are considered unreliable using both absolute error and relative error calculation methods, and selecting the corresponding point pairs with the larger score from the remaining corresponding point pairs as the final compatibility score.

[0018] Furthermore, the specific steps for calculating the compatibility score of each pair of corresponding point pairs in the initial corresponding point pair set H are:

[0019] S201. Set compatibility score threshold t cmp ;

[0020] S202, respectively calculate the corresponding point pairs H i and H j The distance error compatibility score S between cmp1 and S cmp2 ;

[0021] S203, calculating a compatibility score;

[0022] The steps to construct the compatibility matrix H′ are:

[0023] S211, construct a first-order compatibility matrix M, whose elements M ij is the corresponding point pair H i and H j Compatibility scores between them;

[0024] S212, constructing a second-order compatibility matrix M′ based on the first-order compatibility matrix M;

[0025] S213. Assign a reliability score S(i) to each corresponding point pair, and obtain the reliability score by accumulating the compatibility scores of other corresponding point pairs compatible with the corresponding point pair in the matrix M′;

[0026] S214, simplifying the corresponding point scale to obtain the simplified number k of corresponding point pairs;

[0027] S215. Sort the initial corresponding point pair set H in descending order of scores, and select the first k corresponding point pairs from it to form a new set According to the positions of corresponding point pairs in H′, the compatibility submatrix M″ is extracted from the matrix M′.

[0028] Furthermore, in S40, in the i-th row of the compatibility submatrix M″, all elements greater than 0 from the i+1-th column to the end of the row are sorted in descending order according to their values, and the column indexes of the sorted elements in the matrix are stored in the vector v1; the first corresponding point pair (p i ,q i ), in the second loop, column index j is selected sequentially from vector v1 to determine the second corresponding point pair (p j ,q j ).

[0029] Furthermore, in step S50, the steps of generating a conversion hypothesis are:

[0030] S501, extract the corresponding relationship between the two selected i ,q i ) and (p j ,q j ) other compatible correspondences;

[0031] S502, alignment (p i ,q i ) and (p j ,q j ) to restrict 5 of the 6 degrees of freedom;

[0032] S503, constraining the last degree of freedom by clustering the rotation angles of other correspondences selected in step S501 and aligning the significant clusters by rotation;

[0033] S504: Generate a transformation hypothesis based on the above 6-DOF constraint.

[0034] Furthermore, the specific steps of step S501 are: multiplying the elements of the i-th row and the j-th row in the compatibility submatrix M″ one by one to obtain the vector V t , from vector V t Select the index corresponding to the value greater than 0 to form the index set I t , based on the index set I t , find all the i ,q i ) and (p j ,q j ) satisfies the corresponding relationship of the three-point constraint and is expressed as H r ;

[0035] The specific steps of step S502 are: Calculate and align two corresponding point pairs (p i ,q i ) and (p j ,q j ) of the rotation axis k, and then obtain the corresponding rotation matrix Rv and the translation vector t v , through the rotation matrix R v and the translation vector t v For H r All model points in the transform are converted to obtain a new corresponding point pair set H′ that satisfies the ternary ring constraint r And the rotation axis formed by aligning the two corresponding point pairs

[0036] The specific steps of step S503 are: calculate H′ r All corresponding point pairs in the rotation axis The rotation angle of H′ will be obtained r The rotation angles of all corresponding point pairs in θ are sorted in ascending order to form an angle set θ. The rotation angles in the set θ are clustered using the angle clustering method. The significant clusters are selected based on the significance of the number of clusters, and the angle center of each cluster is calculated in a weighted manner.

[0037] The specific steps of step S504 are as follows: after obtaining the angular centers of all significant clusters, calculate the rotation matrix corresponding to each angular center, and combine the rotation matrix R v and the translation vector t v , and obtain the transformation matrix that constrains the six degrees of freedom.

[0038] Furthermore, in step S80, the first transformation hypothesis is first verified with key points to determine whether the number of inliers reaches 4% of the total number of key points. If so, all subsequent transformation hypotheses are verified with key points, otherwise key points and simplified fusion verification are used.

[0039] An electronic device is also provided, comprising a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm as described above.

[0040] A storage medium is further provided, in which processor-executable instructions are stored. When the processor-executable instructions are executed by the processor, they are used to execute the algorithm described above.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] 1. The present invention utilizes a new method for establishing a compatibility matrix, combining the two thresholds of absolute error and relative error. That is, by setting a threshold to filter out corresponding point pairs that are considered unreliable by both the absolute error and relative error thresholds, and selecting the larger score from the remaining corresponding point pairs as the final compatibility score, the accuracy of the method for calculating the compatibility of corresponding point pairs can be effectively improved, thereby improving the accuracy of the algorithm.

[0043] 2. This paper uses an iterative angle clustering strategy to design a transformation hypothesis generation method based on angle clustering. Based on the consistency of the sixth rotational degree of freedom of the correct corresponding point pairs, the initial corresponding point pairs are clustered. First, two corresponding point pairs with high reliability are aligned to constrain the five degrees of freedom. Then, other corresponding rotation angles that are compatible with the two corresponding point pairs are clustered to achieve the distinction between correct corresponding point pairs and outliers, generate more accurate transformation hypotheses, and ensure the accuracy of the transformation hypotheses.

[0044] 3. The present invention utilizes a verification strategy that combines simplified point clouds with key points to achieve verification using key points in high-quality scenes to improve efficiency, and uses fusion verification of key points and simplified point clouds in low-quality scenes to ensure algorithm efficiency while improving accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the iterative registration optimization method based on angle clustering of the present invention.

[0046] Figure 2 1 is an example diagram of selecting the first and second corresponding point pairs, taking 10 corresponding point pairs as an example in the embodiment.

[0047] Figure 3 This is an example diagram of the six-degree-of-freedom registration process in the embodiment.

[0048] Figure 4 Schematic diagram of the ambiguity of three-point constraints in the embodiment.

[0049] Figure 5 1 is an example diagram of the effect of the distance from a point to the rotation axis on the rotation angle in the embodiment. DETAILED DESCRIPTION

[0050] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0051] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component. When a component is considered to be "fixed to" another component, it may be directly fixed to the other component or there may be a central component.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0053] See attached Figure 1 , is a flow chart of the iterative registration optimization method based on angle clustering of the present invention, comprising the following steps:

[0054] S10: Input the initial scene point cloud P and the initial model point cloud Q, and obtain the scene point cloud P after simplification. s And model point cloud Q s , obtain the scene point cloud P through feature matching s And model point cloud Q s The initial corresponding point pair set between p i ∈P s ,q i ∈Q s ;

[0055] S20, calculating the compatibility score of each pair of corresponding point pairs in the initial corresponding point pair set H by constructing compatibility constraints, and sorting the corresponding point pairs in descending order based on the compatibility scores to construct a compatibility matrix H′;

[0056] S30, from the compatibility matrix H′, according to the order of the corresponding point pairs in the matrix, select the first corresponding point pair (p i ,q i );

[0057] S40, based on the first corresponding point pair (p i ,q i ) compatibility, select the second corresponding point pair (p j ,q j );

[0058] S50, clustering corresponding point pairs that satisfy the three-point constraint based on the six rotational degrees of freedom, and selecting all significant clusters to generate transformation hypotheses;

[0059] S60, repeat steps S40 and S50 until the loop ends;

[0060] S70, repeat steps S30, S40, S50 and S60 until the loop ends;

[0061] S80, verifying all saved transformation hypotheses and selecting the best transformation hypothesis as the output transformation matrix;

[0062] S90. Output the conversion hypothesis corresponding to the maximum number of inliers.

[0063] Specifically, in step S20, the compatibility constraint is constructed by setting a compatibility score threshold to filter out corresponding point pairs that are considered unreliable using both absolute error and relative error calculation methods, and selecting the one with the larger score from the remaining corresponding point pairs as the final compatibility score, thereby improving the accuracy and robustness of the compatibility matrix. The specific steps are:

[0064] S201. Set compatibility score threshold t cmp ;

[0065] S202, respectively calculate the corresponding point pairs H i and H j The distance error compatibility score S between cmp1 and S cmp2 :

[0066]

[0067] Among them, H i and H j are the i-th and j-th corresponding point pairs in the initial corresponding point pair set H, d cmp is a distance parameter, S dist H i and H j The rigid distance difference between dist Calculated as:

[0068]

[0069] S cmp1 and S cmp2 There are some differences in the compatibility scores obtained by the two calculation methods. cmp1 Reflects the two corresponding points H i and H j The absolute distance error between cmp2 reflects the relative distance error between the two corresponding point pairs. i -p j ||or||q i -q j When the distance between || is large, (p i ,p j ) or (q i ,q j ) has little effect on the final registration transformation assumption. i ,q i ) and (p j ,q j), a larger distance error can be tolerated. In this case, S cmp2 By calculating min(||p i -p j ||,||q i -q j ||) and max(||p i -p j ||,||q i -q j ||) rather than the absolute distance difference, which can reduce the i -p j ||or||q i -q j ||The effect of a larger distance on the conversion hypothesis. On the contrary, when ||p i -p j ||or||q i -q j When the distance || is small, a small position error may have a greater impact on the transformation hypothesis of the final registration, so S cmp1 By setting a fixed threshold, outliers can be identified and excluded more accurately.

[0070] To this end, the present invention proposes a new compatibility score calculation method as follows:

[0071] S203. Calculate the compatibility score:

[0072]

[0073] Where η=max(S cmp1 (H i ,H j ),S cmp2 (H i ,H j )).

[0074] The compatibility score calculated by this method is calculated by setting the threshold t cmp Screen out S cmp1 and S cmp2 Both calculation methods consider unreliable corresponding point pairs and select the larger score from the remaining corresponding point pairs as the final compatibility score, thereby improving the accuracy and robustness of the compatibility matrix.

[0075] The steps to construct the compatibility matrix H′ are:

[0076] S211, constructing the first-order compatibility matrix M is to accurately represent the compatibility between any two corresponding relationships, and its elements M ij is the corresponding point pair H i and H j Compatibility score between them.

[0077] S212. To enable each compatibility score to reflect global compatibility, the present invention constructs a second-order compatibility matrix M′ based on the first-order compatibility matrix M:

[0078] M′=M·(M*M)(5)

[0079] where · represents the element-by-element product of the two matrices. In the second-order compatibility matrix M′, each element reflects the number of three-membered rings formed by two corresponding point pairs, where a three-membered ring is a ring consisting of three mutually compatible corresponding point pairs.

[0080] S213. Assign a reliability score S(i) to each corresponding point pair to indicate its reliability. The higher the score, the greater the credibility of the corresponding point pair as an inlier. The score is obtained by accumulating the compatibility scores of other corresponding point pairs compatible with the corresponding point pair in the matrix M′:

[0081]

[0082] Among them, S(i) represents the reliability score of the i-th corresponding point pair, M′ ij is the element of the second-order compatibility matrix M′;

[0083] S214. Streamlining the scale of corresponding points. Considering that the maximum value in the second-order compatibility matrix M′ can reflect the global compatibility level to a certain extent, the present invention proposes a method for streamlining the scale of corresponding points:

[0084] k=[-T*(S max / N)+T](7)

[0085] Among them, [] means rounding the value inside the symbol, k is the number of corresponding point pairs after simplification, T is the upper limit of k set to ensure efficiency, S max is the maximum value in the second-order compatibility matrix M′;

[0086] S215. Sort the initial corresponding point pair set H in descending order of scores, and select the first k corresponding point pairs from it to form a new set According to the positions of corresponding point pairs in H′, the compatibility submatrix M″ is extracted from the matrix M′.

[0087] In step S40, since the compatibility submatrix M″ is a symmetric matrix, the second corresponding point pair in the same row is before the first corresponding point pair, which will cause duplication with the two corresponding point pairs selected in the previous cycle, such as Figure 2To avoid repeated point selection, in the i-th row of the compatibility submatrix M″, all elements greater than 0 from the i+1th column to the end of the row are sorted in descending order according to their values. The column index of the sorted elements in the matrix is ​​stored in the vector v1; the first corresponding point pair (p i ,q i ), in the second loop, the column index j is selected from the vector v1 in sequence. According to this column index j, the second corresponding point pair (p j ,q j ).

[0088] In step S50, the conversion hypothesis is generated in the following four steps:

[0089] S501, extract the corresponding relationship between the two selected i ,q i ) and (p j ,q j ) other compatible correspondences;

[0090] S502, alignment (p i ,q i ) and (p j ,q j ) to limit 5 of the 6-DOF (degrees of freedom);

[0091] S503, constraining the last degree of freedom by clustering the rotation angles of other correspondences selected in step S501 and aligning the significant clusters by rotation;

[0092] S504: Generate a transformation hypothesis based on the above 6-DOF constraints.

[0093] Figure 3 The pipeline for generating transformation hypotheses is shown.

[0094] The specific steps of step S501 are as follows:

[0095] (1) Multiply the elements in the i-th row and the j-th row in the compatibility matrix M″ element by element, as follows:

[0096] V t =M″(i,:)·M″(j,:)(8)

[0097] Where M″(i,:) and M″(j,:) represent all elements in the i-th row and j-th row of the matrix M″ respectively, and · represents an element-by-element product operation.

[0098] From vector V t The index corresponding to the selected value greater than 0 is expressed as:

[0099] It =I(V t >0)(9)

[0100] Among them, I t is an index set, I is an indicator function, if the function is true, the value is 1, otherwise it is 0. Its purpose is to identify the i ,q i ) and (p j ,q j ) are compatible with other corresponding point pairs. These correspondences that satisfy the three-point constraint are expressed as where m is H r The number of corresponding point pairs in .

[0101] The specific steps of step S502 are as follows:

[0102] In order to align two corresponding point pairs (p i ,q i ) and (p j ,q j ), according to the Rodriguez formula, a rotation axis and a rotation angle are required. The rotation axis k is calculated as:

[0103]

[0104] in, The rotation angle is the vector and The angle between them.

[0105] Assuming the cosine of the angle is c, the rotation matrix R that aligns the two corresponding points is v and the translation vector t v It can be calculated as:

[0106]

[0107] Where [k] × is the skew-symmetric cross product matrix about the rotation axis k, expressed as:

[0108]

[0109] By the rotation matrix R v and the translation vector t v For H r Transform all model points in to obtain a new set of corresponding point pairs that satisfy the ternary ring constraint where q′ t =R v q t +t v .

[0110] The specific steps of step S503 are as follows:

[0111] Although the three-point ring constraint can effectively filter out some outliers, it also has ambiguity, such as Figure 4 For this purpose, the present invention aligns two corresponding point pairs (p i ,q i ) and (p j ,q j ), first calculate H′ r Then, we design a simple and effective clustering method to cluster the angles. Finally, we select all the significant clusters as much as possible to construct the transformation hypothesis. The specific steps are as follows:

[0112] (1) Calculate H′ r All corresponding point pairs in the rotation axis Rotation angle:

[0113] The calculation process is mainly divided into three steps. r The tth corresponding point pair (p t ,q′ t )’s rotation angle θ t For example, the three steps are as follows:

[0114] First, construct p t and q′ t Horizontal vector to the axis of rotation and The calculation is as follows:

[0115]

[0116] in, ||·|| represents the modulus of the orientation vector. and Unitize.

[0117] Then, perpendicular to the axis Establish a two-dimensional coordinate system on the plane and calculate and The angle with the X axis of the coordinate system. and The included angle from clockwise to X axis and It can be calculated as:

[0118]

[0119] Finally, the corresponding point pair (p t ,q′ t )’s rotation angle θ t Calculated as:

[0120]

[0121] θ t The range is [0°,360°).

[0122] (2) Angle clustering:

[0123] After obtaining H′ r After calculating the rotation angles of all corresponding point pairs in , they are sorted in ascending order to form an angle set θ, and a simple and effective angle clustering method is used to cluster the rotation angles in the set θ. The method consists of the following three steps.

[0124] First, perform offset subtraction on the angle set θ, which can be expressed as follows:

[0125] θ δ =θ 2:m -θ 1:m-1 (18)

[0126] Among them, θ 2:m Represents the second to last element in the angle set θ, θ 1:m-1 Represents the first to the second-to-last element in the angle set θ.

[0127] Then, from θ δ Select the angle greater than the threshold value τ θ The indexes of are used as classification interval points for classification. The selected index set can be expressed as:

[0128] θ outer =I(θ δ >τ θ )(19)

[0129] Finally, according to the index θ outer Get all cluster sets And the angles in these clusters are mapped to H′ r The index set of corresponding point pairs in Where n represents the index θ outer The number of clusters obtained.

[0130] (3) Select significant clusters:

[0131] Typically, the consistency of the inliers ensures that the corresponding point pairs contained in the largest cluster are inliers. However, when the inlier rate is very low, outliers may have a significant impact on the clustering results, resulting in the corresponding point pairs in the largest cluster not necessarily being inliers. To this end, the present invention designs a strategy for selecting appropriate clusters based on the significance of the number of clusters. The specific steps are as follows:

[0132] First, calculate the number of corresponding point pairs of all clusters to form a set U. Then, set a threshold τ and select clusters whose number is at least τ times the maximum number of clusters. The selected cluster index set I c It is expressed as follows:

[0133] I c =I(U>τ·max(U))(20)

[0134] Among them, max(U) means taking the largest element in the set U, and the range of τ is between 0 and 1. c In this way, even in the case of low inlier rate, reliable corresponding point pairs can be effectively screened out, thereby improving the robustness of the registration.

[0135] (4) Determine the clustering angle center:

[0136] After all significant clusters are determined, since all rotation angles within a cluster are consistent, the angular center of each cluster can be used as the consistent angle value of the cluster. Considering that under the same interference, the farther the point is from the rotation axis, the less its rotation angle is affected, such as Figure 5 As shown, the present invention uses a weighted approach to calculate the angular center of each cluster. Assume that the lth cluster The rotation angle in is θ g , whose weight w g It can be calculated as:

[0137]

[0138] in, Get clustering After the weights of all rotation angles in the cluster are calculated, the angular center of the cluster It can be calculated as:

[0139]

[0140] S504: Transformation hypothesis generation: After obtaining the angular centers of all significant clusters, first calculate the rotation matrix corresponding to each angular center according to formula (23). Then, combine the two corresponding point pairs (p i ,q i ) and (p j ,q j )’s rotation matrix and translation vector, we can get the transformation matrix that constrains the six degrees of freedom. Specifically, clustering For example, its angular center The corresponding rotation matrix is ​​recorded as The point cloud registration rotation matrix corresponding to the cluster and translation vectors It can be calculated as:

[0141]

[0142] Among them, |.| represents the number of elements in the set, R V To align two corresponding points (p i ,q i ) and (p j ,q j )’s rotation matrix, p d and q d For clustering A pair of corresponding relationships in .

[0143] In step S80, to increase the reliability of the verification, the present invention assigns a score to each transformation hypothesis to evaluate its reliability. This score is related to the reliability of the corresponding point pairs used to calculate the transformation hypothesis. Here, the score of the corresponding point pairs is used to calculate the score of the transformation hypothesis. Specifically, for the rotation matrix and translation vectors The corresponding score It can be calculated as:

[0144]

[0145] In step S80, the first transformation hypothesis is first verified using key points. A determination is then made as to whether the number of inliers reaches 4% of the total number of key points. If so, all subsequent transformation hypotheses are verified using key points. Otherwise, key points and simplified point clouds are fused for verification. Specifically, key point verification involves aligning the key points on two point clouds using the estimated transformation hypothesis to count the number of inliers. Verification using key point and simplified point cloud fusion involves aligning the key points and simplified point clouds on the two point clouds using the estimated transformation hypothesis, and then counting and summarizing the number of inliers. Finally, the transformation hypothesis corresponding to the largest number of inliers is selected as the output transformation matrix.

[0146] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm as described above.

[0147] An embodiment of the present invention further provides a storage medium storing processor-executable instructions, wherein the processor-executable instructions are used to execute the algorithm described above when executed by the processor.

[0148] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0149] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0150] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0152] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention's description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An iterative registration optimization algorithm based on angle clustering, characterized in that: The following steps are involved: S10: Input the initial scene point cloud P and the initial model point cloud Q, and obtain the scene point cloud P after simplification. s And model point cloud Q s , obtain the scene point cloud P through feature matching s And model point cloud Q s The initial corresponding point pair set between p i ∈P s ,q i ∈Q s ; S20, calculating the compatibility score of each pair of corresponding point pairs in the initial corresponding point pair set H by constructing compatibility constraints, and sorting the corresponding point pairs in descending order based on the compatibility scores to construct a compatibility matrix H′; S30, from the compatibility matrix H′, according to the order of the corresponding point pairs in the matrix, select the first corresponding point pair (p i ,q i ); S40, based on the first corresponding point pair (p i ,q i ) compatibility, select the second corresponding point pair (p j ,q j ); S50, clustering corresponding point pairs that satisfy the three-point constraint based on the six rotational degrees of freedom, and selecting all significant clusters to generate transformation hypotheses; S60, repeat steps S40 and S50 until the loop ends; S70, repeat steps S30, S40, S50 and S60 until the loop ends; S80, verifying all saved transformation hypotheses and selecting the best transformation hypothesis as the output transformation matrix; S90. Output the conversion hypothesis corresponding to the maximum number of inliers.

2. The iterative registration optimization algorithm based on angle clustering according to claim 1, characterized in that: In S20, the compatibility constraint is constructed by setting a compatibility score threshold to filter out corresponding point pairs that are considered unreliable using both absolute error and relative error calculation methods, and selecting the corresponding point pairs with the larger score from the remaining corresponding point pairs as the final compatibility score.

3. The iterative registration optimization algorithm based on angle clustering according to claim 1 or 2, characterized in that: The specific steps for calculating the compatibility score of each pair of corresponding points in the initial corresponding point pair set H are: S201. Set compatibility score threshold t cmp ; S202, respectively calculate the corresponding point pairs H i and H j The distance error compatibility score S between cmp1 and S cmp2 ; S203, calculating a compatibility score; The steps to construct the compatibility matrix H′ are: S211, construct a first-order compatibility matrix M, whose elements M ij is the corresponding point pair H i and H j Compatibility scores between; S212, constructing a second-order compatibility matrix M′ based on the first-order compatibility matrix M; S213. Assign a reliability score S(i) to each corresponding point pair, and obtain the reliability score by accumulating the compatibility scores of other corresponding point pairs compatible with the corresponding point pair in the matrix M′; S214, simplifying the corresponding point scale to obtain the simplified number k of corresponding point pairs; S215. Sort the initial corresponding point pair set H in descending order of scores, and select the first k corresponding point pairs from it to form a new set According to the positions of corresponding point pairs in H′, the compatibility submatrix M″ is extracted from the matrix M′.

4. The iterative registration optimization algorithm based on angle clustering according to claim 3, characterized in that: In S40, in the i-th row of the compatibility submatrix M″, all elements greater than 0 from the i+1th column to the end of the row are sorted in descending order according to their values, and the column indexes of the sorted elements in the matrix are stored in the vector v1; in the first layer, the first corresponding point pair (p i ,q i ), in the second loop, column index j is selected sequentially from vector v1 to determine the second corresponding point pair (p j ,q j ).

5. The iterative registration optimization algorithm based on angle clustering according to claim 3, characterized in that: In step S50, the steps of generating a conversion hypothesis are: S501, extract the corresponding relationship between the two selected i ,q i ) and (p j ,q j ) other compatible correspondences; S502, alignment (p i ,q i ) and (p j ,q j ) to restrict 5 of the 6 degrees of freedom; S503, constraining the last degree of freedom by clustering the rotation angles of other correspondences selected in step S501 and aligning the significant clusters by rotation; S504: Generate a transformation hypothesis based on the above 6-DOF constraint.

6. The iterative registration optimization algorithm based on angle clustering according to claim 5, characterized in that: The specific steps of step S501 are: multiply the elements of the i-th row and the j-th row in the compatibility sub-matrix M″ one by one to obtain the vector V t , from vector V t Select the index corresponding to the value greater than 0 to form the index set I t , based on the index set I t , find all the i ,q i ) and (p j ,q j ) satisfies the corresponding relationship of the three-point constraint and is expressed as H r ; The specific steps of step S502 are: Calculate and align two corresponding point pairs (p i ,q i ) and (p j ,q j ) of the rotation axis k, and then obtain the corresponding rotation matrix R v and the translation vector t v , through the rotation matrix R v and the translation vector t v For H r All model points in the transform are converted to obtain a new corresponding point pair set H′ that satisfies the ternary ring constraint r And the rotation axis formed by aligning the two corresponding point pairs The specific steps of step S503 are: calculate H′ r All corresponding point pairs in the rotation axis The rotation angle of H′ will be obtained r The rotation angles of all corresponding point pairs in θ are sorted in ascending order to form an angle set θ. The rotation angles in the set θ are clustered using the angle clustering method. The significant clusters are selected based on the significance of the number of clusters, and the angle center of each cluster is calculated in a weighted manner. The specific steps of step S504 are as follows: after obtaining the angular centers of all significant clusters, calculate the rotation matrix corresponding to each angular center, and combine the rotation matrix R v and the translation vector t v , and obtain the transformation matrix that constrains the six degrees of freedom.

7. The iterative registration optimization algorithm based on angle clustering according to claim 1, characterized in that: In step S80, the first transformation hypothesis is first verified with key points to determine whether the number of inliers reaches 4% of the total number of key points. If so, all subsequent transformation hypotheses are verified with key points, otherwise key points and simplified fusion verification are used.

8. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm according to any one of claims 1 to 7.

9. A storage medium storing instructions executable by a processor, characterized in that: The processor-executable instructions are used to perform the algorithm according to any one of claims 1 to 7 when executed by the processor.

Images