Laser radar odometer method and system based on edge feature pair and circular structure constraint and storage medium

Abstract

The invention provides a laser radar odometer method and system based on edge feature pair and circular structure constraint, and a storage medium, and relates to the technical field of laser radar positioning and mapping measurement, and the method comprises the steps: obtaining a laser radar original point cloud, projecting the laser radar original point cloud to a two-dimensional image, calculating the local roughness, and screening candidate edge points; edge feature pairs are screened according to the roughness relation between the category edge points and the category internal points with the maximum roughness; calculating a normal vector of a fusion plane based on a vector relationship between the associated edge feature pairs, projecting the associated points onto the plane and converting the associated points into two-dimensional coordinates; respectively fitting two circles by using the projection point of the current frame and the correlation point of the previous frame, and constructing a weighted residual error based on the circle center position difference and the radius difference; and solving the pose estimation through nonlinear optimization. According to the method, the feature stability is improved, the robustness in non-structured environments with rich columnar structures such as forests is enhanced, and the pose estimation precision is improved.

Description

Technical Field

[0001] This invention relates to the field of lidar positioning and mapping measurement technology, specifically to lidar odometry methods, systems, and storage media based on edge feature pairs and circular structure constraints. Background Technology

[0002] LiDAR odometry is one of the mainstream technologies for achieving Simultaneous Localization and Mapping (SLAM), and it is widely used in fields such as autonomous driving and mobile robotics. Among them, the LOAM (Lidar Odometry And Mapping) algorithm, as a feature-matching-based LiDAR odometry, has become a classic representative due to its strong real-time performance and clear structure. Its basic process includes: selecting edge and surface feature points based on the curvature of the point cloud; and estimating inter-frame pose transformation by minimizing the geometric distances from edge points to lines and surface points to surfaces. Many subsequent LOAM-based variants also largely follow this process, namely, feature point extraction based on roughness combined with a distance residual minimization matching strategy.

[0003] However, this type of method relies on objects in the scene having distinct edges for edge matching and cannot handle objects with indistinct edges well. Especially in scenes like forests with abundant columnar features such as tree trunks and lacking clear edges, LOAM-based laser odometry still faces significant challenges. The edge points on the surface of such columnar objects are geometrically ambiguous, easily introducing matching errors and leading to a decrease in pose estimation accuracy.

[0004] To overcome the aforementioned problems, some studies have proposed improvement strategies. The 2022 study, "PLC-LiSLAM: LiDAR SLAM With Planes, Lines, and Cylinders," extracted cylinders for matching in addition to planes and lines. However, this method does not analyze the relationships between features and heavily relies on computationally intensive operations such as clustering, resulting in high computational time and poor robustness to weakly structured environments.

[0005] Furthermore, existing methods only select feature points based on roughness without further evaluating the quality of these feature points. This results in a large number of low-quality edge points participating in the matching process, significantly impacting the accuracy of pose estimation. Edge points are often distributed on the surfaces of various objects and are easily affected by structural type and sampling noise, making them particularly sensitive to the matching process. For example, edge points on grass or cylindrical surfaces are prone to introducing matching errors due to their ambiguous geometric representation. Edge points of different qualities contribute significantly to the residuals, with low-quality points even producing negative interference.

[0006] Chinese patent document CN115235506A discloses a lidar odometry method based on ICP, which discloses a technical solution that adopts a multi-scale feature extraction strategy. On a large scale, smooth points and rough points are divided according to curvature and six degrees of freedom motion is solved by ICP. On a small scale, noise points are eliminated by tangent vector, straightness and rate of change, and edge points and corner points are retained. This has the technical effect of reducing the influence of noise and improving the robustness of motion estimation. However, there are still problems such as lack of targeted geometric constraints for weak edge features such as columnar objects and failure to make full use of the pairing relationship between edge points.

[0007] Chinese patent document CN120762000A discloses a point-to-line weighted matching method for lidar based on edge point quality assessment. It discloses a technical solution that evaluates edge point quality and assigns weights based on category label consistency and orientation standard deviation. It filters effective edge points by position judgment and orientation consistency detection, and uses an exponential decay function to dynamically weight and optimize the residuals. This achieves the technical effect of suppressing the influence of low-quality points and improving the accuracy of pose estimation. However, it still has the problems of lacking structured constraints based on single-point quality assessment and difficulty in effectively calculating orientation consistency in scenes of columnar objects with blurred edges. Summary of the Invention

[0008] The purpose of this invention is to provide a lidar odometry method, system, and storage medium based on edge feature pairs and circular structure constraints, enabling the extraction and matching of scene features based on low computational information, establishing geometric structure constraints for columnar objects, improving feature stability, enhancing robustness in unstructured environments rich in columnar structures such as forests, and improving pose estimation accuracy.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A lidar odometry method based on edge feature pairs and circular structure constraints includes the following steps: S1: Obtain the original point cloud of the LiDAR, project the original point cloud onto a 2D image, perform target segmentation on the projected point cloud and assign a category label to each point, calculate the local roughness of each point and filter candidate edge points; S2: Within the same scan line of each point cloud cluster, extract the leftmost and rightmost points as category edge points, and extract the remaining points as category interior points. Calculate the local roughness of the category edge points and category interior points respectively. S3: For each point cloud cluster in each scan row, edge feature pairs are selected based on the roughness relationship between the leftmost category edge point, the rightmost category edge point, and the internal points of the category with the highest roughness. S4: Project the edge feature pairs of the current frame onto the point cloud coordinate system of the previous frame. Search for the first edge feature pair that is closest to the projected point pair in the edge feature pair set of the previous frame. Then search for the second edge feature pair that is closest to the first edge feature pair. Establish the correspondence between the edge feature pairs of the current frame and the two associated edge feature pairs of the previous frame. S5: Calculate the normal vector of the fusion plane based on the vector relationship between the two associated edge feature pairs in the previous frame, and project the projection points of the current frame edge feature pairs and the four points of the two associated edge feature pairs in the previous frame onto the plane passing through the origin and convert them into two-dimensional coordinates; S6: Using the edge features of the current frame, fit a first circle to one of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Use the edge features of the current frame to fit a second circle to another of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Construct a weighted residual based on the difference in the center position and radius of the two circles. S7: Optimize the residuals using a nonlinear optimization method to solve for the pose estimation of the current frame.

[0010] Furthermore, in S1, the calculation method for the projected point cloud I is as follows:

[0011] in, r α, c β These represent the row and column numbers of the laser scan, respectively, satisfying... r α ∈[0,N-1], c β ∈[0,M-1], where N represents the number of laser radar beams and M represents the number of times a certain beam is transmitted or received in one frame scan. (·) indicates a projection operation. p k It is a point cloud P One of the points.

[0012] Furthermore: In step S1, the target segmentation employs a fast segmentation algorithm, and the output category label representation for each point is as follows:

[0013] Where L(·) represents the category label to which the laser point belongs, and W represents the total number of labels. This represents one label in the overall labeling; multiple laser points with the same category label constitute a point cloud cluster.

[0014] Furthermore, in S2, the local roughness is calculated as follows:

[0015] in, Indicates the current point p k The local roughness, where S represents the set of nearest points used for calculation. This represents the number of points in set S. and These represent the distance values ​​between a point in set S and the current point.

[0016] Further: In S3, the selection criteria for edge feature pairs are:

[0017]

[0018] in, and These are the leftmost and rightmost category edge points, respectively; This indicates the calculation of local roughness. It is a fixed coefficient; The point inside the category with the highest roughness; Set of points inside the category The internal point of the i-th category.

[0019] Further: In S4, the current frame edge feature pair The projection relationship onto the coordinate system of the point cloud in the previous frame is as follows:

[0020]

[0021] in, This is the pose transformation matrix from the current frame k to the next frame k-1; Set of valid edge feature pairs in the previous frame The method for searching the first edge feature pair is as follows:

[0022] in, This represents the edge feature pair from the previous frame that is closest to the projection point pair; Set of valid edge feature pairs in the previous frame The method for searching for the second edge feature pair is as follows:

[0023] in, Indicates and The nearest edge feature pair from the previous frame.

[0024] Furthermore: In S5, the calculation method for the fused plane normal vector is as follows:

[0025]

[0026]

[0027]

[0028]

[0029] in, and , where are the vectors between two related edge feature pairs; n is the fusion normal vector. The homogenization coefficient, The merged unit plane normal vector; The method of projecting a point onto a plane passing through the origin is as follows:

[0030] in, This is the vector after projection. The vector before projection; Represent the projection points using plane basis vectors. Two-dimensional coordinates:

[0031] in, These are the two-dimensional coordinates after projection. , These are the selected basis vectors.

[0032] Furthermore: In S6, the fitting equation for the first circle is:

[0033] in, Let the coordinates of the center of the first circle be... Let the radius be the radius of the first circle. , , These are the two-dimensional coordinates of the projection point on the left side of the current frame, the two-dimensional coordinates of the first edge feature relative to the left side of the previous frame, and the two-dimensional coordinates of the first edge feature relative to the right side of the previous frame, respectively. The fitting equation for the second circle is:

[0034] in, Let the coordinates of the center of the second circle be... The radius of the second circle, These are the two-dimensional coordinates of the projection point on the right side of the current frame; The residual of the constructed circular structure is:

[0035] Where E is the residual, The coefficient of the central term, This is the coefficient for the radius term.

[0036] A lidar localization and mapping system, employing the lidar odometry method based on edge feature pairs and circular structure constraints described above for inter-frame pose estimation, includes: a lidar sensor for acquiring environmental point cloud data; The data processing unit is used to execute a lidar odometry method based on edge feature pairs and circular structure constraints, including projection of point cloud data, target segmentation, edge feature pair extraction, correlation matching, circular structure constraint construction, and pose optimization. The storage unit is used to store point cloud data, edge feature pairs, and pose estimation results.

[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lidar odometry method based on edge feature pairs and circular structure constraints as described above.

[0038] Compared with the prior art, the present invention has the following advantages: This invention extracts edge feature pairs and uses a dual screening mechanism based on category labels and roughness, which has stronger pairing stability and noise resistance compared to single-point features, effectively filtering out the interference of low-quality edge point pairs in pose estimation.

[0039] This invention utilizes the geometric properties of columnar objects to construct circular structural constraints based on edge features. By establishing dual residual constraints through differences in the center position and radius, it enables columnar objects such as tree trunks with indistinct edges to provide stable and persistent matching constraints, significantly enhancing the robustness of the system in unstructured environments such as forests.

[0040] This invention replaces the complex clustering operation in traditional three-dimensional space by fitting a circular structure with a two-dimensional projection plane. While ensuring the strength of the constraint, it reduces the computational complexity, avoids the real-time problems caused by high computational load, and improves the pose estimation accuracy and computational efficiency of the lidar SLAM system. Attached Figure Description

[0041] Figure 1 A flowchart of a lidar odometry method based on edge feature pairs and circular structure constraints provided by the present invention; Figure 2 This is a schematic diagram showing the boundary points and interior points of a category. Figure 3 This is a schematic diagram of the edge feature pairs from the previous frame. Figure 4 A schematic diagram of the circular structure residual construction; Figure 5 This is a schematic diagram of the structure of a lidar positioning and mapping system provided by the present invention. Detailed Implementation

[0042] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example 1 like Figure 1 As shown, the present invention provides a lidar odometry method based on edge feature pairs and circular structure constraints, comprising the following steps; S1: Obtain the original point cloud of the LiDAR, project the original point cloud onto a 2D image, perform target segmentation on the projected point cloud and assign a category label to each point, calculate the local roughness of each point and filter candidate edge points; S2: Within the same scan line of each point cloud cluster, extract the leftmost and rightmost points as category edge points, and extract the remaining points as category interior points. Calculate the local roughness of the category edge points and category interior points respectively. S3: For each point cloud cluster in each scan row, edge feature pairs are selected based on the roughness relationship between the leftmost category edge point, the rightmost category edge point, and the internal points of the category with the highest roughness. S4: Project the edge feature pairs of the current frame onto the point cloud coordinate system of the previous frame. Search for the first edge feature pair that is closest to the projected point pair in the edge feature pair set of the previous frame. Then search for the second edge feature pair that is closest to the first edge feature pair. Establish the correspondence between the edge feature pairs of the current frame and the two associated edge feature pairs of the previous frame. S5: Calculate the normal vector of the fusion plane based on the vector relationship between the two associated edge feature pairs in the previous frame, and project the projection points of the current frame edge feature pairs and the four points of the two associated edge feature pairs in the previous frame onto the plane passing through the origin and convert them into two-dimensional coordinates; S6: Using the edge features of the current frame, fit a first circle to one of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Use the edge features of the current frame to fit a second circle to another of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Construct a weighted residual based on the difference in the center position and radius of the two circles. S7: Optimize the residuals using a nonlinear optimization method to solve for the pose estimation of the current frame.

[0044] Example 2 like Figures 1-4 As shown, this invention proposes a lidar odometry method based on edge feature pairs and circular structure constraints. The specific implementation steps are as follows: Original point cloud projection: Let P be the original point cloud obtained at a certain moment. The calculation method for its projected point cloud I is as follows:

[0045] Where, r α , c β Let r represent the row number and column number of the laser scan, respectively, satisfying r α ∈[0,N-1], c β ∈[0,M-1], where N represents the number of laser radar beams and M represents the number of times a certain beam is transmitted or received in a single frame scan. (·) indicates a projection operation. p k It is a point in the point cloud P.

[0046] Target segmentation and category label assignment: Although there are already many mature target segmentation methods, in order to meet the real-time requirements of target segmentation, this paper adopts the fast segmentation method proposed in the 2016 paper "Fast Range Image-Based Segmentation of Sparse 3D LaserScans for Online Operation".

[0047] After inputting the projected point cloud into the segmentation algorithm, each laser point can be assigned a corresponding category label, as shown below:

[0048] Where L(·) represents the category label of the laser point, and W represents the total number of labels. Multiple laser points with the same category label constitute a point cloud cluster.

[0049] Category edge point selection and category interior point selection: like Figure 2 As shown, for each point cloud cluster, the leftmost and rightmost points within the same scan row are selected as category edge points, and the remaining points are selected as category interior points.

[0050] Roughness calculation for category edge points and category interior points: For each category edge point and category interior point in the projected point cloud, find the nearest points within a certain range within the same scan line and calculate their local roughness:

[0051] in, Indicates the current point p k The local roughness is given by S, where S represents the set of nearest points used for calculation, and |S| represents the number of points in set S. and These represent the distance values ​​between a point in set S and the current point.

[0052] Edge feature pair selection: Let the leftmost and rightmost category edge points of each point cloud cluster in each row be respectively... and The set of points within the category is .

[0053] For sets The points in the array are sorted in descending order of their roughness values, and the point with the highest roughness is selected. Category edge points and The comparison will determine the category edge points that satisfy the following formula. Defined as edge feature pairs:

[0054] in, This indicates the calculation of local roughness. It is a fixed coefficient. Edge features are associated with: The obtained edge feature pairs Projecting these points onto the coordinate system of the previous frame's point cloud yields the corresponding projection point pairs. Its projection relationship can be expressed as:

[0055] in, This is the pose transformation matrix from the current frame k to the next frame k-1, and its initial value is determined by the uniform motion model.

[0056] Point pairs obtained by projection In the set of effective edge feature pairs in the previous frame Search for the nearest edge feature pair in the middle, denoted as

[0057] in, This represents the edge feature pair from the previous frame that is closest to the projection point pair.

[0058] For the nearest edge feature pair from the previous frame In the set of effective edge feature pairs in the previous frame Search for the nearest edge feature pair in the middle, denoted as

[0059] in, Indicates and The nearest edge feature pair from the previous frame.

[0060] and for The associated edge feature pairs in the previous frame.

[0061] Circular structure residual construction: Calculate separately point to , point to 3D vector:

[0062] in, and These are vectors between two pairs of related edge features.

[0063] Calculation based on , normal to the fusion plane:

[0064]

[0065]

[0066] in, The homogenization coefficient, This is the normal vector of the merged unit plane.

[0067] Will Project them onto the plane passing through the origin, using the following method:

[0068] in, This is the vector after projection. This is the vector before projection.

[0069] Represent the projection points using plane basis vectors. Two-dimensional coordinates:

[0070] in, These are the two-dimensional coordinates after projection. , These are the selected basis vectors.

[0071] Following the steps above, we obtain The corresponding two-dimensional coordinates are respectively .

[0072] Given three points in a two-dimensional plane, a circle can be fitted using these points. The center of the circle can be obtained by solving a quadratic system of equations.

[0073] in, Let the coordinates be the center of the circle. Let be the radius.

[0074] The formula for calculating the radius is:

[0075] Similarly, according to A circle can also be fitted, where the center is... , radius is . The residual of the circular structure based on the center and radius is constructed as follows:

[0076] Where E represents the residual formed by the relationship between position and size. The coefficient of the central term, This is the coefficient for the radius term.

[0077] The residuals are optimized globally using nonlinear optimization methods (such as least squares) to obtain the optimal pose estimate for the current frame.

[0078] Example 3 like Figure 5 As shown, this embodiment provides a lidar localization and mapping system that uses the aforementioned lidar odometry method based on edge feature pairs and circular structure constraints for inter-frame pose estimation. The system includes a lidar sensor, a data processing unit, and a storage unit. The lidar sensor collects environmental point cloud data. The data processing unit performs steps such as point cloud projection, target segmentation, edge feature pair extraction, association matching, circular structure constraint construction, and pose optimization. The storage unit stores the point cloud data, edge feature pair sets, and pose estimation results. The data processing unit establishes the association between the current frame and historical frames by calling the edge feature pair set of the previous frame from the storage unit, achieving continuous pose estimation. This system can operate stably in scenes containing numerous columnar structures, such as forests and urban roads, and is suitable for applications such as mobile robot navigation and autonomous vehicle localization.

[0079] Example 4 This embodiment provides a computer-readable storage medium on which a computer program is stored. When executed by a processor, the computer program implements the various steps of the aforementioned lidar odometry method based on edge feature pairs and circular structure constraints. Specifically, during program execution, it completes the acquisition and projection of the original lidar point cloud, target segmentation and category label assignment, local roughness calculation and candidate edge point selection, extraction of category edge points and intra-category points, selection of edge feature pairs, projection and association matching of edge feature pairs, calculation of fusion plane normal vectors and coordinate transformation, construction of circular structure constraints, and optimization solution of pose estimation. The computer-readable storage medium can be a non-volatile storage medium such as read-only memory, random access memory, disk, or optical disk, which can be deployed and run on different computing platforms, providing a flexible software implementation scheme for lidar SLAM systems.

[0080] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A laser radar odometry method based on edge features with circular structure constraints, characterized by: Includes the following steps: S1: Obtain the original point cloud of the LiDAR, project the original point cloud onto a 2D image, perform target segmentation on the projected point cloud and assign a category label to each point, calculate the local roughness of each point and filter candidate edge points; S2: Within the same scan line of each point cloud cluster, extract the leftmost and rightmost points as category edge points, and extract the remaining points as category interior points. Calculate the local roughness of the category edge points and category interior points respectively. S3: For each point cloud cluster in each scan row, edge feature pairs are selected based on the roughness relationship between the leftmost category edge point, the rightmost category edge point, and the internal points of the category with the highest roughness. S4: Project the edge feature pairs of the current frame onto the point cloud coordinate system of the previous frame. Search for the first edge feature pair that is closest to the projected point pair in the edge feature pair set of the previous frame. Then search for the second edge feature pair that is closest to the first edge feature pair. Establish the correspondence between the edge feature pairs of the current frame and the two associated edge feature pairs of the previous frame. S5: Calculate the normal vector of the fusion plane based on the vector relationship between the two associated edge feature pairs in the previous frame, and project the projection points of the current frame edge feature pairs and the four points of the two associated edge feature pairs in the previous frame onto the plane passing through the origin and convert them into two-dimensional coordinates; S6: Using the edge features of the current frame, fit a first circle to one of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Use the edge features of the current frame to fit a second circle to another of the projection points and the corresponding points of the first and second edge feature pairs of the previous frame. Construct a weighted residual based on the difference in the center position and radius of the two circles. S7: Optimize the residuals using a nonlinear optimization method to solve for the pose estimation of the current frame.

2. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In step S1, the projected point cloud I is calculated as follows: Where, r α , c β Let r represent the row number and column number of the laser scan, respectively, satisfying r α ∈[0,N-1], c β ∈[0,M-1], where N represents the number of laser radar beams and M represents the number of beams in a single frame scan. The number of times a certain wire harness transmits or receives data. (·) indicates a projection operation, p k It is a point in the point cloud P.

3. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In step S1, a fast segmentation algorithm is used for target segmentation, and the class label representation of each point is as follows: Where L(·) represents the category label to which the laser point belongs, and W represents the total number of labels. This represents one label in the overall labeling; multiple laser points with the same category label constitute a point cloud cluster.

4. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In S2, the local roughness is calculated as follows: in, Indicates the current point p k The local roughness, where S represents the set of nearest points used for calculation. This represents the number of points in set S. and These represent the distance values ​​between a point in set S and the current point.

5. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In S3, the selection criteria for edge feature pairs are: in, and These are the leftmost and rightmost category edge points, respectively; This indicates the calculation of local roughness. It is a fixed coefficient; The point inside the category with the highest roughness; Set of points inside the category The internal point of the i-th category.

6. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In S4, the current frame edge feature pair The projection relationship onto the coordinate system of the point cloud in the previous frame is as follows: in, This is the pose transformation matrix from the current frame k to the next frame k-1; Set of valid edge feature pairs in the previous frame The method for searching the first edge feature pair is as follows: in, This represents the edge feature pair from the previous frame that is closest to the projection point pair; Set of valid edge feature pairs in the previous frame The method for searching for the second edge feature pair is as follows: in, Indicates and The nearest edge feature pair from the previous frame.

7. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In S5, the calculation method for the fusion plane normal vector is as follows: in, and , where are the vectors between two related edge feature pairs; n is the fusion normal vector. The homogenization coefficient, The merged unit plane normal vector; The method of projecting a point onto a plane passing through the origin is as follows: in, This is the vector after projection. The vector before projection; Represent the projection points using plane basis vectors. Two-dimensional coordinates: in, These are the two-dimensional coordinates after projection. , These are the selected basis vectors.

8. The lidar odometry method based on edge feature pairs and circular structure constraints according to claim 1, characterized in that: In S6, the fitting equation for the first circle is: in, Let the coordinates of the center of the first circle be... Let the radius be the radius of the first circle. , , These are the two-dimensional coordinates of the projection point on the left side of the current frame, the two-dimensional coordinates of the first edge feature relative to the left side of the previous frame, and the two-dimensional coordinates of the first edge feature relative to the right side of the previous frame, respectively. The fitting equation for the second circle is: in, Let the coordinates of the center of the second circle be... The radius of the second circle, These are the two-dimensional coordinates of the projection point on the right side of the current frame; The residual of the constructed circular structure is: Where E is the residual, The coefficient of the central term, This is the coefficient for the radius term.

9. A lidar positioning and mapping system, characterized in that, The method of lidar odometry based on edge feature pairs and circular structure constraints according to any one of claims 1-8 is used to perform inter-frame pose estimation, comprising: a lidar sensor for acquiring environmental point cloud data; The data processing unit is used to execute a lidar odometry method based on edge feature pairs and circular structure constraints, including projection of point cloud data, target segmentation, edge feature pair extraction, correlation matching, circular structure constraint construction, and pose optimization. The storage unit is used to store point cloud data, edge feature pairs, and pose estimation results.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the lidar odometry method based on edge feature pairs and circular structure constraints as described in any one of claims 1-8.

Images