An aircraft wheel recognition method and system based on three-dimensional point cloud

Abstract

The application provides an airplane wheel recognition method and system based on a three-dimensional point cloud, which comprehensively uses a clustering algorithm, a least square method for fitting a circle, and a position feature discrimination method to recognize airplane wheels, further reduces the consumption of computing resources, can be applied to specific airplanes by modifying parameters according to the data of the specific airplanes, does not need to be trained, can be applied to specific airplanes more quickly, has strong robustness, and can realize unmanned and full automation in a subsequent task target recognition stage.

Description

Technical Field

[0001] This invention belongs to the technical field of aircraft wheel recognition, and particularly relates to an aircraft wheel recognition method and system based on three-dimensional point clouds. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the process of attaching pylons to the fuselage, identifying the spatial position and attitude of the target pylon is fundamental and a prerequisite for achieving the entire automated operation. Since the pylon is located in a specific position under the fuselage, locating it directly from outside the fuselage is quite difficult. Based on field research, the ideal starting point for the attachment operation is 5-10 meters away from the aircraft. Within this distance, the aircraft fuselage is too large, making complete identification difficult, and other parts of the fuselage are also much higher than the fuselage itself. Within the height range of the attachment target, the aircraft wheels are relatively independent of the fuselage and possess distinct characteristics.

[0004] In recent years, target recognition algorithms based on 3D point clouds have developed rapidly, especially in fields such as autonomous driving, robot navigation, environmental modeling, and drones. Early target recognition algorithms mainly relied on traditional geometric feature extraction and matching methods, including geometric model-based matching, point cloud segmentation, and feature description techniques. These methods achieve target recognition by matching objects in the point cloud with predefined geometric models (such as spheres, cubes, cylinders, etc.). These methods work well when the object shape is simple and regular.

[0005] With the rapid development of deep learning technology, especially the successful application of Convolutional Neural Networks (CNNs) in image processing, more and more research is introducing deep learning methods into 3D point cloud object recognition. The most representative algorithm is PointNet, which first proposed a method for directly processing unordered point clouds and overcame the limitations of point cloud disorder through global feature pooling, achieving good classification and segmentation results. However, it lacks in-depth mining of local geometric information, thus still having certain limitations when processing complex-shaped objects. PointNet++ introduces hierarchical feature learning and local feature aggregation mechanisms, which can effectively extract local features at different scales and gradually construct global features, thereby improving the accuracy and robustness of object recognition. With the continuous development of these deep learning algorithms, more point cloud-based object recognition methods have emerged, such as DGCNN (Dynamic Graph CNN), VoxelNet, and PointCNN. These methods further improve the performance of 3D point cloud object recognition by introducing innovative techniques such as graph convolution, voxelization, or dynamic graph structures.

[0006] In target recognition tasks based on 3D point clouds, especially in specific scenarios such as the identification of special aircraft wheels, deep learning algorithms, while providing powerful feature learning capabilities, may face challenges in pre-training the model for practical applications. This is because such methods typically require large amounts of labeled data for training, and data on special aircraft is not readily available; obtaining a 3D point cloud dataset of aircraft wheels is often difficult. In such cases, the applicability of deep learning methods is limited.

[0007] In summary, the current challenge is to achieve real-time identification of aircraft wheels, especially those of special-type aircraft, by extracting and analyzing the geometric features of point clouds in the absence of a large amount of labeled data, so as to automate the loading of cargo into the inner and outer compartments of the fuselage. Summary of the Invention

[0008] To overcome the shortcomings of the existing technology, this invention provides a method and system for aircraft wheel recognition based on three-dimensional point clouds. It integrates clustering algorithms, least squares method for circle fitting, and position feature discrimination to identify aircraft wheels, further reducing computational resource consumption and making the algorithm more applicable.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] In a first aspect, the present invention provides a method for aircraft wheel identification based on three-dimensional point clouds, comprising:

[0011] Acquire and preprocess the lidar point cloud data of the aircraft wheels;

[0012] The preprocessed lidar point cloud data is clustered using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters. The positional relationship of each point cloud cluster is determined based on the actual positional relationship of the aircraft wheels.

[0013] For point cloud clusters that conform to positional relationships, the distribution of points is fitted using the least squares method to find the best-fit circle, and the validity of the best-fit circle is determined.

[0014] Calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud clusters for the turbine wheel.

[0015] Secondly, the present invention provides an aircraft wheel recognition system based on three-dimensional point clouds, comprising:

[0016] The acquisition module is configured to acquire and preprocess the lidar point cloud data of the aircraft wheels.

[0017] The position determination module is configured to: cluster the preprocessed lidar point cloud data using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters; and determine the position relationship of each point cloud cluster based on the actual position relationship of the aircraft wheels.

[0018] The fitting module is configured to: fit the distribution of points in a point cloud cluster that conforms to the positional relationship using the least squares method to find the best fitting circle, and determine whether the best fitting circle is valid.

[0019] The identification module is configured to: calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud cluster of the wheel.

[0020] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0021] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0022] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0023] The above one or more technical solutions have the following beneficial effects:

[0024] This invention integrates clustering algorithms, least squares circle fitting, and position feature discrimination to identify aircraft wheels, further reducing computational resource consumption. It can be applied by modifying parameters based on the specific aircraft's own data without the need for learning and training, allowing for faster application to specific aircraft. It has strong robustness and can achieve unmanned and fully automated target identification in subsequent attach missions.

[0025] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of the aircraft wheel recognition method based on three-dimensional point cloud in Embodiment 1 of the present invention;

[0028] Figure 2 This is a schematic diagram illustrating the principle of region of interest division in Embodiment 1 of the present invention;

[0029] Figure 3 This is a schematic diagram of the radius filtering principle in Embodiment 1 of the present invention;

[0030] Figure 4 This is a schematic diagram of the DBSCAN clustering algorithm in Embodiment 1 of the present invention;

[0031] Figure 5(a) is a diagram showing the effect of fitting a circle to the point cloud of the machine wheel in Embodiment 1 of the present invention;

[0032] Figure 5(b) is a diagram showing the effect of fitting a circle to the point cloud of debris in Embodiment 1 of the present invention;

[0033] Figure 6 This is a schematic diagram illustrating the calculation of the final target position and pose in Embodiment 1 of the present invention. Detailed Implementation

[0034] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0035] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0036] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0037] Example 1

[0038] This embodiment discloses an aircraft wheel recognition method based on three-dimensional point clouds, including:

[0039] Acquire and preprocess the lidar point cloud data of the aircraft wheels;

[0040] The preprocessed lidar point cloud data is clustered using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters. The positional relationship of each point cloud cluster is determined based on the actual positional relationship of the aircraft wheels.

[0041] For point cloud clusters that conform to positional relationships, the distribution of points is fitted using the least squares method to find the best-fit circle, and the validity of the best-fit circle is determined.

[0042] Calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud clusters for the wheel.

[0043] Based on the determined target cloud cluster of the aircraft wheels, and the positional relationship between the aircraft belly and the aircraft wheels, the midpoint coordinates of the two rear wheels of the aircraft are calculated to determine the target position of the aircraft belly; and based on the positional relationship between the installation orientation of the lidar and the orientation of the aircraft, the target pose of the aircraft belly is determined.

[0044] This embodiment installs the lidar on the trailer and integrates algorithms such as clustering, least squares circle fitting, and position feature discrimination, further reducing computational resource consumption and making the algorithm more applicable. It can be applied by modifying parameters according to the specific aircraft's own data without the need for learning and training, and can be applied to specific aircraft more quickly, with strong robustness.

[0045] The following is combined with Figure 1 This embodiment provides a detailed description of an aircraft wheel recognition method based on 3D point clouds, specifically including:

[0046] Step 1: Acquire the LiDAR point cloud data of the aircraft and perform preprocessing.

[0047] Specifically, after capturing the LiDAR point cloud data, filtering is first performed. By setting the boundaries of the point cloud data coordinate values, the region of interest is divided, and Boolean indexes are used to filter points located within the region of interest from the original point cloud. These filtered points are then fed into the Adaptive Radius Filter (AROR) algorithm for noise reduction. The radius of the neighborhood is automatically adjusted according to the local density of each point, thereby providing a more accurate filtering effect in different density regions, thus cleaning up noise and irrelevant points. The remaining point cloud data is then sent to subsequent clustering and positional relationship determination.

[0048] Region of Interest (ROI) segmentation is an important technique in 3D point cloud processing. It aims to filter points within specific regions from large amounts of point cloud data, allowing focus on task-relevant target objects, reducing computational load and improving processing efficiency. For example... Figure 2 As shown, a set of boundary values ​​(e.g., minimum and maximum values ​​on the X, Y, and Z axes) are set based on prior knowledge or task requirements to define the region of interest. In the task of aircraft wheel identification, the space under the aircraft fuselage is pre-defined as the region of interest: the forward direction of the trailer coincides with the positive X-axis direction of the lidar, and the initial position of the trailer is generally between 5-15m from the target. Considering the positional relationship between the three wheels and the wheel height, the x-axis range (depth) can be limited to within 18m, the y-axis range (lateral) to within 5m, and the z-axis range (height) to within 1m. Therefore, the region of interest of the lidar is defined as follows:

[0049] 1≤x≤18

[0050] -5≤y≤5

[0051] 0 <z≤1

[0052] Based on the defined minimum and maximum ranges for each coordinate axis, all points that meet the criteria are selected, while points outside the range are removed. This method is simple and efficient, and can flexibly adapt to different application scenarios. Subsequent clustering, feature determination, and other processing steps can be concentrated in these key areas, significantly improving processing accuracy and speed.

[0053] Adaptive Radius Outlier Removal (AROR) is a dynamic radius filtering method based on local density. Compared to radius filtering (ROR), it adaptively adjusts the filtering radius according to the density of the point cloud, ensuring sufficient clustering neighborhood radius in sparse regions while avoiding excessively large neighborhood ranges in dense regions. This ensures better removal of irrelevant points in sparse regions and better preservation of details in locally dense regions. Figure 3 As shown, the local density is first calculated, a fixed radius range r0 is set (0.3 in this embodiment), and a point p is selected. i The number of points within the cluster neighborhood radius r0 is N(p i ):

[0054] N(p i )={p j |d(p i ,p j (≤r0}

[0055] Wherein d(p i ,p j () is point p i and point p j The Euclidean distance between them. The neighborhood is spherical with a volume of:

[0056]

[0057] Based on this, the local density v i Defined as the ratio of the number of neighboring points to the volume of the neighboring area, the local density is:

[0058]

[0059] Next, the process of dynamically adjusting the radius is based on local density. In regions with high local density, there are more points in the neighborhood, so a smaller radius can be chosen to avoid failing to remove irrelevant points at the edges, thus affecting the filtering effect. Conversely, in sparse regions, there are fewer neighborhood points, so the radius needs to be increased to ensure that enough neighborhood points participate in the filtering. Specifically, the radius can be dynamically adjusted using the following formula:

[0060]

[0061] Here, C is a constant that controls the maximum radius. This formula indicates that regions with higher local density have smaller radii, while regions with lower density have larger radii. To avoid the radius being too small, a minimum radius value r can be set. min :

[0062] r i ≥r min

[0063] In this way, the radius of dense regions will automatically decrease, while the radius of sparse regions will increase to ensure the filtering effect.

[0064] Then, during the filtering process, a threshold for the number of neighboring points is set to τ (typically τ = 5). For each point p... i The dynamically adjusted radius r can be used. i To find its neighboring points. If the number of neighboring points is less than a preset threshold, the point is considered an outlier and can be deleted or adjusted.

[0065] The advantages of adaptive radius filtering are that the algorithm is simple and efficient, and it is particularly suitable for sparse point clouds or scenarios with large local density changes, such as scanning open experimental sites or real-time processing tasks.

[0066] Step 2: Cluster the preprocessed lidar point cloud data using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters. Determine the positional relationship of each point cloud cluster based on the actual positional relationship of the aircraft wheels.

[0067] Specifically, in order to clearly segment the point cloud data initially, DBSCAN clustering is first performed on the point cloud data retained in step 1. By optimizing the clustering neighborhood radius and the minimum number of points, adjacent points are grouped into a cluster. Then, the number of clusters is counted, and the positional relationship of each cluster of point clouds is determined.

[0068] The improved DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm designed to group high-density regions into the same cluster, while treating low-density regions as noise. Unlike traditional partition-based clustering methods, DBSCAN does not require pre-specifying the number of clusters; instead, it automatically identifies the cluster structure based on density relationships in point cloud data.

[0069] The improvements in this embodiment mainly target the optimization of two parameters: the clustering neighborhood radius selection method based on k-distance graphs and the neighborhood minimum number of points selection method based on Gaussian kernel density estimation.

[0070] In the traditional DBSCAN algorithm, the parameters ∈ (cluster neighborhood radius) and MinPts (minimum number of points) are key parameters that determine the clustering results. The selection of these parameters has a significant impact on the clustering results, and they often need to be manually adjusted for different datasets, increasing the difficulty of using the algorithm. The purpose of optimizing parameter selection is to automatically adjust these parameters based on the data distribution and local density changes, thereby improving the clustering performance and adaptability of DBSCAN in different scenarios.

[0071] The k-distance graph-based clustering neighborhood radius selection method calculates local variations using a k-distance graph and automatically adjusts the clustering neighborhood radius for each point. The k-distance graph evaluates the local variations of a point by calculating the distance to its k-th nearest neighbor. Commonly used k values ​​are 4 or 5 (usually corresponding to MinPts, but not necessarily the same).

[0072] The specific steps are as follows: First, for each data point p i Calculate the distance d from it to its k nearest neighbors. i,k The k-nearest neighbor distances of all points are sorted in ascending order. Then, the distances of each point to its k-nearest neighbors are plotted on a graph to construct a k-distance graph. The horizontal axis of the k-distance graph represents the index of the point, and the vertical axis represents the distance from the point to its k nearest neighbors. In the k-distance graph, there is usually a clear inflection point or elbow (i.e., a position where the distance changes significantly), indicating that after this distance, the local density of data points begins to decrease sharply. The index number of the point in the k-distance graph corresponding to this elbow point is calculated, and the distance from the corresponding point to p is found based on the index number. i The distance is used as the radius of the new local clustering neighborhood ∈′. The elbow position of the k-distance map can be detected by finding the “inflection point” of the curve, i.e., the elbow, through second-order differencing (i.e., curvature):

[0073] Δ 2 d i =(d i+2,k -d i+1,k )-(d i+1,k -d i,k )

[0074] Where, Δd i Representative point p j The distance to the k-th nearest neighbor and point p i+1 The difference in distances to the k-th nearest neighbor. The elbow usually corresponds to the point where the second-order difference is maximized, i.e.:

[0075] i * =argmax(Δ 2 d i )

[0076] By finding the index i corresponding to the maximum second difference* We can find the elbow position in the k-distance graph, and the ∈′ value corresponding to this position is the optimal clustering neighborhood radius.

[0077] The neighborhood minimum point selection method based on Gaussian kernel density estimation optimizes the traditional DBSCAN method of selecting MinPts, as a fixed value may not effectively process data for regions with varying densities. Therefore, the Gaussian kernel density estimation-based method dynamically adjusts the neighborhood minimum point threshold MinPts based on the Gaussian kernel density of each point's region, thereby improving clustering quality. Gaussian kernel density estimation is performed by calculating the minimum point density (p) for each point. i The local density of a point is estimated by a weighted average of points within its neighborhood. The formula is as follows:

[0078]

[0079] Wherein, ρ(p i () is point p i Gaussian kernel density; ||p i -p j || is the Euclidean distance between the two points; σ is the standard deviation of the Gaussian kernel, which controls the rate of distance decay; |N(p i | represents the number of neighboring points. This formula uses a Gaussian kernel function to weight the points within the neighborhood, focusing on those closest to p. i Points closer to the center have a larger weight, while points farther away have a smaller weight. The weight of point p can be calculated by summing the weights of all points in its neighborhood. i The local density at point p i The degree of "crowding" in its local area. For each point p i The following formula can be used to adaptively adjust MinPts:

[0080]

[0081] Among them, MinPts i It is point p i The corresponding optimized MinPts′ uses α as a constant, typically chosen as 2 or 3, to adjust its value to better reflect the density characteristics of the data. According to the formula, regions with high local density will have smaller MinPts′ values ​​because these regions are already dense enough that fewer neighboring points are needed to identify them as core points. Conversely, low-density regions will have larger MinPts′ values ​​because these regions are sparse and may require more neighboring points to be identified as core points.

[0082] Based on the optimized selection method, suitable clustering neighborhood radii ∈′ and minimum number of points MinPts′ are obtained. For example... Figure 4As shown, in this embodiment, the algorithm processes each point as follows: First, the algorithm checks whether the ∈′ neighborhood of a point contains at least MinPts′ points; if this condition is met, the point is considered a core point, and the surrounding points are grouped into the same cluster. The core point and the points in its neighborhood constitute a cluster, and other points in the cluster can continue to expand until all points are clustered or labeled as noise; through clustering, if there are no adjacent objects within the ∈′ distance, it is considered a cluster of point clouds, and the wheel point cloud cluster is initially aggregated into a whole, which is beneficial for subsequent judgment.

[0083] Step 3: For point cloud clusters that meet the positional relationship, use the least squares method to fit the distribution of points to find the best fitting circle, and determine whether the best fitting circle is valid.

[0084] Specifically, based on the actual positional relationship of the aircraft wheels, the target wheels should form an isosceles triangle. The centroid coordinates of each point cloud cluster are extracted, a threshold for side length difference is set, and the Euclidean distance between the centroids of each cluster is calculated. Within the threshold's allowable range, point cloud clusters that meet the relative positional conditions are retained, and the cluster's center point is added to the candidate set for subsequent geometric feature determination.

[0085] Specifically, aircraft wheels exhibit distinct cylindrical characteristics and fixed geometric dimensions, making the direction of geometric feature determination relatively clear. First, based on the point cloud cluster's point coordinate boundaries, the height consistency of points within each cluster is checked. The z-coordinates of all points in the cluster are extracted, and the maximum and minimum values ​​are compared with a set range to ensure the upper and lower surface heights of the target object are reasonable. Point cloud clusters that do not meet the range requirements are removed from the target candidate set. For those that meet the requirements, the least squares method is used to fit the distribution of points within the cluster to find the optimal fitting circle. Based on this, the center and radius of each cluster are estimated, and the validity of the fitting circle is determined according to the set size.

[0086] Least squares circle fitting is a common method used in 3D point cloud data processing to determine the optimal circle fit by minimizing errors. Its basic principle is to find the optimal parameters (center coordinates and radius) of the circle by minimizing the sum of the squares of the distances between the data points and the fitted circle. The goal of least squares is to find a circle such that the distance from each point to the circle is as small as possible, thus making the fitted result as close as possible to the actual point cloud distribution.

[0087] In a two-dimensional plane, the equation of a circle is usually expressed as:

[0088] (xa) 2 +(yb) 2 =r 2

[0089] Where (a,b) are the coordinates of the center of the circle, and r is the radius of the circle.

[0090] The goal of least squares fitting is to determine the center (a, b) and radius r of a circle such that the sum of the squares of the distances from each point in the point set to the circle is minimized. That is, to minimize the following error function:

[0091]

[0092] Here, n is the number of data points, and xi and yi are the coordinates of the i-th point. By minimizing this error function, the optimal center coordinates and radius can be obtained.

[0093] To facilitate computation, the equations can be linearized. By expanding the equations and introducing auxiliary variables, the least squares method is typically solved using matrix operations.

[0094] First, transform the equation of the circle into:

[0095] x 2 +y 2 +Dx+Ey+F=0

[0096] Where D = -2a, E = -2b and F = a 2 +b 2 -r 2 These are the parameters to be determined. By performing a least-squares fit on all points, the parameters D, E, and F can be obtained, and then the values ​​of the center and radius of the circle can be calculated. This process can be achieved by solving a system of linear equations.

[0097] The effect of fitting the circle in the experiment is as follows: Figures 5(a)-5(b) As shown, the left side displays the effect of fitting a circle to the point cloud clusters of the machine wheels, while the right side displays the effect of fitting a circle to the point cloud clusters of the stacked objects. In a real-world scenario, projecting the machine wheel point cloud along the Y-axis onto the xz plane reveals a distinct arc structure. This arc structure is a key characteristic that distinguishes the machine wheels from other objects. Therefore, we first project all point cloud clusters in the scene along the Y-axis onto the xz plane, and then perform circle fitting. The closer the shape of the fitted circle is to the distribution of the point cloud, the higher the coverage value will be in subsequent calculations, which is beneficial for the final target identification.

[0098] Step 4: Calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud cluster for the wheel.

[0099] Specifically, the coverage of the best-fit circle is calculated, the degree of matching between the distribution of all points in each cluster of point clouds and the corresponding best-fit circle is measured, and it is determined whether the coverage exceeds a predetermined threshold. The point cloud clusters that meet the conditions are sorted, and the top three clusters are selected from largest to smallest coverage as the final target point cloud clusters for the aircraft wheel.

[0100] The coverage calculation of the best-fit circle for point cloud is an indicator that measures the degree of matching between the fitted circle and the actual distribution of point cloud data. The higher the coverage, the more the best-fit circle can represent the distribution of point cloud data.

[0101] First, for each point (x) in each point cloud cluster i y i Calculate the distance from the point to the best-fit circle. Assuming the center of the best-fit circle is (a, b) and its radius is r, then the distance d between the point and the best-fit circle is... i It can be calculated using the following formula:

[0102]

[0103] Where, d i This represents the distance from the i-th point to the best-fit circle.

[0104] The calculation method for coverage can be expressed as:

[0105]

[0106] The threshold is used to determine whether a point is within the effective range of the fitted circle. Only when the difference between the distance from the point to the center of the circle and the radius is less than or equal to the threshold is the point considered to have an impact on coverage. The coverage value is usually between 0 and 1; the larger the value, the better the coverage effect of the fitted circle. In the project, the coverage of the wheel target is between 0.85 and 0.93, and the coverage of other objects is between 0.1 and 0.8. Based on this judgment condition, the targets can be identified well.

[0107] Step 5: Based on the determined target cloud cluster of the aircraft wheels, and the positional relationship between the aircraft belly and the aircraft wheels, calculate the midpoint coordinates of the two rear wheels of the aircraft to determine the target position of the aircraft belly; and based on the positional relationship between the installation orientation of the lidar and the orientation of the aircraft, determine the target pose of the aircraft belly.

[0108] Specifically, after obtaining the target point cloud clusters of the wheels, the front and rear wheels are distinguished based on the coordinate attributes of the midpoints of each cloud cluster. Furthermore, the target position and pose can be estimated from the positional relationship of the isosceles triangles of the three wheels.

[0109] Based on the actual layout of the aircraft, the horizontal position of the midpoint of the fuselage coincides with the horizontal position of the midpoint of the two rear wheels. The internal pylons are located on both sides of the cabin. The relative positions of the three are fixed. The calculated coordinates of the midpoint of the two rear wheels are the final target points. The horizontal coordinates of the pylons can be obtained through appropriate data compensation.

[0110] Since the lidar is installed on the trailer, the positive direction of its radar coordinate system X-axis coincides with the forward direction of the vehicle body. At the same time, the front wheel is located on the vertical line of the midpoint of the line connecting the two rear wheels. The target pose can be determined by calculating the angle between the line connecting the final target point in the lidar coordinate system, the midpoint of the front wheel, and the radar X-axis.

[0111] like Figure 6 As shown, XOY is the lidar coordinate system. In the figure, a, b, and c are the midpoints of the point cloud clusters of the front wheel and the two rear wheels, respectively. z is the midpoint of the line segment bc, which is also the required final target position. Subsequently, the coordinates of the internal fuselage pylon can be easily obtained by using the coordinates of point z and the aircraft's own size data. The angle θ between az and the X-axis is the angle between the aircraft's forward direction in the radar coordinate system and the positive X-axis of the radar coordinate system. When the radar is installed, the positive X-axis is the same as the forward direction of the attachment vehicle. At the same time, during the attachment operation, the forward direction of the attachment vehicle is required to be the same as the forward direction of the aircraft. Therefore, θ is the attitude angle that the attachment vehicle needs to deflect, which is the target attitude.

[0112] Experimental verification shows that the algorithm model in this embodiment is only 18kb in size, with an accuracy rate of 93.7% in identifying wheel targets and a recognition time of 0.15s. The data shows that it meets the actual requirements of the mounting task, namely, high accuracy, lightweight model, and fast recognition speed. At the same time, it can be used on another special type of aircraft by only modifying the wheel and compensation parameters, which shows that it has high universality.

[0113] Example 2

[0114] The purpose of this embodiment is to provide an aircraft wheel recognition system based on three-dimensional point clouds, including:

[0115] The acquisition module is configured to acquire and preprocess the lidar point cloud data of the aircraft wheels.

[0116] The position determination module is configured to: cluster the preprocessed lidar point cloud data using the clustering domain radius and minimum number of points to obtain multiple point cloud clusters; and determine the position relationship of each point cloud cluster based on the actual position relationship of the aircraft wheels.

[0117] The fitting module is configured to: fit the distribution of points in a point cloud cluster that conforms to the positional relationship using the least squares method to find the best fitting circle, and determine whether the best fitting circle is valid.

[0118] The identification module is configured to: calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud cluster of the wheel.

[0119] In further embodiments, the following is also provided:

[0120] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0121] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0122] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0123] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0124] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0125] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0126] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0127] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0128] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0129] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for aircraft wheel recognition based on 3D point clouds, characterized in that, include: Acquire lidar point cloud data of aircraft wheels and perform preprocessing; The preprocessed lidar point cloud data is clustered using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters. The positional relationship of each point cloud cluster is determined based on the actual positional relationship of the aircraft wheels. For point cloud clusters that conform to positional relationships, the distribution of points is fitted using the least squares method to find the best-fit circle, and the validity of the best-fit circle is determined. Calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud clusters for the wheel. Specifically, the preprocessed lidar point cloud data is clustered using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters. The positional relationship of each point cloud cluster is then determined based on the actual positional relationship of the aircraft wheels. Based on the clustering neighborhood radius and the minimum number of points, adjacent points are clustered into a point cloud cluster; Based on the geometric positional relationship of the aircraft wheels, the centroid coordinates of each cloud cluster are extracted, and the Euclidean distance between the centroids of each cloud cluster is calculated. The positional relationship of the corresponding point cloud clusters is determined based on the calculated Euclidean distance; It also includes: based on the determined target point cluster of the aircraft wheels, and based on the positional relationship between the aircraft belly and the aircraft wheels, calculating the midpoint coordinates of the two rear wheels of the aircraft to determine the target position of the aircraft belly; and based on the positional relationship between the installation orientation of the lidar and the orientation of the aircraft to determine the target pose of the aircraft belly.

2. The aircraft wheel recognition method based on three-dimensional point clouds as described in claim 1, characterized in that, The acquired lidar point cloud data of the aircraft wheels is preprocessed, specifically including: By defining the boundaries of the point cloud data coordinate values, the region of interest is divided, and Boolean indexes are used to filter out the point cloud data located in the region of interest from the acquired LiDAR point cloud data. An adaptive radius filtering algorithm is used to denoise the selected point cloud data, and outliers are filtered out based on the set radius and neighborhood point number threshold.

3. The aircraft wheel recognition method based on three-dimensional point clouds as described in claim 1, characterized in that, For point cloud clusters that conform to positional relationships, the distribution of points is fitted using the least squares method to find the best-fit circle, and the validity of the best-fit circle is determined; specifically: Based on the geometric features of aircraft wheels, the consistency of point height within each point cloud cluster is checked according to the point coordinate boundaries of the point cloud cluster. For point cloud clusters that exhibit high consistency, the distribution of the point cloud clusters is fitted using the least squares method to find the best-fit circle. Based on the set size of the fitted circle, determine whether the best fitted circle corresponding to the point cloud cluster is valid.

4. The aircraft wheel recognition method based on three-dimensional point clouds as described in claim 1, characterized in that, Calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud cluster for the wheel; specifically: Calculate the distance between each point in the point cloud cluster and the center of the corresponding best-fit circle; By setting a distance threshold, the number of point clouds within the effective range of the best-fit circle is determined, and the coverage corresponding to the point cloud cluster is calculated based on the determined point cloud data. The corresponding point cloud clusters are sorted according to the calculated coverage to determine the target point cloud cluster for the aircraft wheel.

5. The aircraft wheel recognition method based on three-dimensional point cloud as described in claim 1, characterized in that, The cluster neighborhood radius is determined based on the k-distance plot, and the minimum number of points is determined based on Gaussian kernel density estimation.

6. An aircraft wheel recognition system based on three-dimensional point clouds, characterized in that, include: The acquisition module is configured to acquire and preprocess the lidar point cloud data of the aircraft wheels. The position determination module is configured to: cluster the preprocessed LiDAR point cloud data using the clustering neighborhood radius and minimum number of points to obtain multiple point cloud clusters; and determine the position relationship of each point cloud cluster based on the actual position relationship of the aircraft wheels; specifically: Based on the clustering neighborhood radius and the minimum number of points, adjacent points are clustered into a point cloud cluster; Based on the geometric positional relationship of the aircraft wheels, the centroid coordinates of each cloud cluster are extracted, and the Euclidean distance between the centroids of each cloud cluster is calculated. The positional relationship of the corresponding point cloud clusters is determined based on the calculated Euclidean distance; The fitting module is configured to: fit the distribution of points in a point cloud cluster that conforms to the positional relationship using the least squares method to find the best fitting circle, and determine whether the best fitting circle is valid. The identification module is configured to: calculate the coverage of the effective best-fit circle, sort the corresponding point cloud clusters according to the calculated coverage, and determine the target point cloud cluster of the wheel; The target pose module is configured to: determine the target position of the aircraft belly based on the determined target point cloud cluster of the aircraft wheels and the positional relationship between the aircraft belly and the aircraft wheels; and determine the target pose of the aircraft belly based on the positional relationship between the installation orientation of the lidar and the orientation of the aircraft.

7. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-5.

Images