AGV navigation method and system for aircraft maintenance site, medium and equipment

Abstract

The invention discloses an AGV (Automatic Guided Vehicle) navigation method and system for an aircraft maintenance site, a medium and equipment, and the method comprises the steps: collecting original local point cloud data of the maintenance site through an AGV platform, removing environment interference point cloud data, and carrying out the processing through employing a self-adaptive point cloud clustering algorithm based on density and spatial distribution characteristics, obtaining target local point cloud data of the target aircraft; the target local point cloud data and a complete machine three-dimensional point cloud database are registered through an NDT algorithm, a rigid body transformation matrix of a target complete machine point cloud model corresponding to the target aircraft relative to an AGV body coordinate system is obtained, the space attitude of an AGV platform under the complete machine point cloud coordinate system is obtained, and relative positioning is achieved; by taking the current positioning position of the AGV platform as a starting point and the maintenance operation point as a target point, automatically generating an optimal navigation path and controlling the AGV platform to travel along the optimal navigation path in an autonomous navigation manner; therefore, the AGV positioning precision can be improved, and the correctness of an autonomous navigation path is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aircraft maintenance site positioning, and in particular to an AGV navigation method and system for an aircraft maintenance site, a computer readable storage medium and a terminal device. BACKGROUND

[0002] With the continuous development of industrial automation and intelligent technology, AGV (Automated Guided Vehicle) has been widely used in the fields of aviation, logistics, manufacturing, etc. due to its autonomous navigation, environmental perception and path planning capabilities. In a regular environment, AGV usually relies on SLAM (Simultaneous Localization and Mapping) technology to achieve efficient operation, and is particularly suitable for scenarios with rich geometric features and clear structures, such as hangars or automated warehouse environments, etc. SLAM constructs an environment map in real time to provide support for path planning and obstacle avoidance, and exhibits high stability and reliability.

[0003] Point cloud data generated by laser radar is the basis of SLAM technology. According to the point cloud data, an environment map can be constructed in real time to achieve centimeter-level positioning and dynamic path planning. For example, in the aircraft maintenance scene, point cloud data can accurately identify the structure features or damage positions of the aircraft body, guiding AGV to perform tool handling tasks. However, in the complex structure target operation scene of aircraft maintenance, the existing positioning and path planning schemes based on point cloud data usually have the problem of over-segmentation of point cloud clustering of complex geometric targets at multiple angles, which will affect the recognition accuracy when identifying targets through point cloud clustering, thereby affecting the positioning accuracy of AGV. In addition, there is also a low overlap registration problem between local point cloud and global point cloud, which will affect the registration accuracy of local point cloud and global point cloud, and also affect the positioning accuracy of AGV, thereby causing path deviation problems. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide an AGV navigation method and system for an aircraft maintenance site, a computer readable storage medium and a terminal device, which can solve the problems of over-segmentation of point cloud clustering of complex geometric targets at multiple angles and low overlap registration between local point cloud and global point cloud, improve the recognition accuracy of the target aircraft and the point cloud registration accuracy, thereby improving the positioning accuracy of AGV and ensuring the correctness of the autonomous navigation path of AGV.

[0005] To achieve the above purpose, the embodiments of the present application provide an AGV navigation method for an aircraft maintenance site, comprising:

[0006] In the maintenance operation site of the target aircraft, integrated point cloud data covering the surrounding environment is obtained by using an AGV platform;

[0007] Collect original local point cloud data of the maintenance work site by using the AGV platform, remove environmental interference point cloud data from the integrated point cloud data, and process the local point cloud data after removing the interference by using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics to obtain target local point cloud data corresponding to the target aircraft.

[0008] The target local point cloud data is registered with a preset whole machine three-dimensional point cloud database by using an NDT algorithm to obtain a target whole machine point cloud model corresponding to the target aircraft and a rigid transformation matrix of the target whole machine point cloud model relative to the AGV body coordinate system, and the spatial pose of the AGV platform in the whole machine point cloud coordinate system is obtained according to the rigid transformation matrix to realize the relative positioning of the AGV platform; wherein the whole machine three-dimensional point cloud database includes whole machine point cloud models corresponding to multiple typical aircraft models, which are generated by point clouds of key structural components of the aircraft and are in a unified whole machine point cloud coordinate system.

[0009] The navigation map template corresponding to the target whole machine point cloud model is called, and the path node coordinates in the navigation map template are converted according to the rigid transformation matrix.

[0010] Based on the converted navigation map template, the optimal navigation path is automatically generated with the current positioning position of the AGV platform as the starting point and the maintenance work point as the target point, and the AGV platform is controlled to autonomously navigate and travel along the optimal navigation path.

[0011] To achieve the above purpose, the embodiment of the application also provides an AGV navigation system for an aircraft maintenance site, comprising:

[0012] An environmental point cloud data acquisition module is configured to acquire integrated point cloud data covering the surrounding environment by using an AGV platform at a maintenance work site of a target aircraft.

[0013] A local point cloud data acquisition module is configured to collect original local point cloud data of the maintenance work site by using the AGV platform, remove environmental interference point cloud data from the integrated point cloud data, and process the local point cloud data after removing the interference by using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics to obtain target local point cloud data corresponding to the target aircraft.

[0014] The local point cloud data registration and positioning module is configured to register the target local point cloud data with a preset whole machine three-dimensional point cloud database by using an NDT algorithm, to obtain a target whole machine point cloud model corresponding to the target aircraft and a rigid body transformation matrix of the target whole machine point cloud model relative to an AGV body coordinate system, and to obtain a spatial pose of the AGV platform in a whole machine point cloud coordinate system according to the rigid body transformation matrix, so as to realize relative positioning of the AGV platform.

[0015] The navigation map coordinate conversion module is configured to call a navigation map template corresponding to the target whole machine point cloud model, and to perform coordinate conversion on path node coordinates in the navigation map template according to the rigid body transformation matrix.

[0016] The autonomous navigation path planning module is configured to generate an optimal navigation path based on the converted navigation map template, with a current positioning position of the AGV platform as a starting point and a maintenance work point as a target point, and to control the AGV platform to autonomously navigate and travel along the optimal navigation path.

[0017] The embodiment of the application further provides a computer readable storage medium including a stored computer program, which, when executed, controls a device in which the computer readable storage medium is located to perform the AGV navigation method for an aircraft maintenance site.

[0018] The embodiment of the application further provides a terminal device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the AGV navigation method for an aircraft maintenance site.

[0019] Compared with the prior art, the embodiment of the present application provides an AGV navigation method, system, computer readable storage medium and terminal device for an aircraft maintenance site. First, in the target aircraft maintenance operation site, the integrated point cloud data covering the surrounding environment is obtained by using the AGV platform. Then, the original local point cloud data of the maintenance operation site is collected by using the AGV platform, the environmental interference point cloud data is removed from the integrated point cloud data, and the adaptive point cloud clustering algorithm based on the density and spatial distribution characteristics is used to process the local point cloud data after removing the interference, so as to obtain the target local point cloud data corresponding to the target aircraft. Then, the NDT algorithm is used to register the target local point cloud data with the preset whole machine three-dimensional point cloud database, so as to obtain the target whole machine point cloud model corresponding to the target aircraft and the rigid transformation matrix of the target whole machine point cloud model relative to the AGV body coordinate system, and the spatial pose of the AGV platform in the whole machine point cloud coordinate system is obtained according to the rigid transformation matrix, so as to realize the relative positioning of the AGV platform. Then, the navigation map template corresponding to the target whole machine point cloud model is called, and the path node coordinates in the navigation map template are converted according to the rigid transformation matrix. Finally, based on the converted navigation map template, the current positioning position of the AGV platform is taken as the starting point, and the maintenance operation point is taken as the target point, so as to automatically generate the optimal navigation path and control the AGV platform to autonomously navigate and travel along the optimal navigation path. The embodiment of the present application can solve the over-segmentation problem of point cloud clustering of a complex geometric target under multiple angles and the low overlap registration problem of local point cloud and global point cloud, improve the recognition accuracy of the target and the point cloud registration accuracy, thereby improving the positioning accuracy and ensuring the correctness of the AGV autonomous navigation path. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flowchart of an AGV navigation method for an aircraft maintenance site provided by an embodiment of the present application;

[0021] Figure 2 is a structural schematic diagram of original local point cloud data collected by a laser radar provided by an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a whole machine point cloud coordinate system provided by an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of path planning and obstacle avoidance provided by an embodiment of the present application;

[0024] Figure 5 is a structural block diagram of an AGV navigation system for an aircraft maintenance site provided by an embodiment of the present application;

[0025] Figure 6 is a structural block diagram of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0027] It should be noted that based on the SLAM technology, there are many mature mobile robot products on the market, such as sweeping robots, hotel delivery robots, etc. The sweeping robot is mainly used for ground cleaning in family or commercial places, and can realize functions such as automatic mapping, whole house coverage cleaning, regional cleaning, breakpoint continuous cleaning, etc. In order to complete the above tasks, the sweeping robot usually carries a laser radar or a camera, uses the SLAM technology to construct a map and realize real-time positioning, and in the path planning part, determines the optimal cleaning route according to the built map, usually uses a grid search and coverage optimization algorithm, and its obstacle avoidance algorithm can also ensure the safe operation of the sweeping robot in complex areas such as furniture, corners or step edges. Unlike the sweeping robot, the goal of the hotel delivery robot is not to traverse the entire area, but to deliver goods to the designated room. Its navigation system usually relies on laser radar to build a building map, and cooperates with the SLAM technology to realize autonomous positioning and path planning, and mostly uses the combination of A* or Dijkstra algorithm global path planning and DWA (Dynamic Window Approach) or VFH (Vector Field Histogram) algorithm local path planning method to cope with dynamic changes such as pedestrians and obstacles.

[0028] In addition to the autonomous mapping method based on SLAM technology, GNSS (Global Navigation Satellite System) positioning, beacon positioning and map matching positioning are also common robot positioning technologies. GNSS positioning mainly relies on GPS (Global Positioning System) or Beidou satellite signals, and can provide real-time position service in a global range, and is widely used in unmanned vehicles, agricultural robots and outdoor surveying robots. Beacon positioning locates beacons by deploying ZigBee, RFID (Radio Frequency Identification), UWB (Ultra Wide Band) or ultrasonic wave technology in the environment, and the robot determines its position by measuring the distance or angle with the beacon, which has high precision and practicality in indoor navigation, warehouse transportation and factory logistics. Map matching positioning uses laser radar or visual sensor to collect local environment features and matches them with the existing global map to estimate the pose information, which is commonly used in subway inspection robots and industrial guide vehicles.

[0029] Although the above-mentioned positioning technologies have been maturely applied in different scenarios, in the complex structure target operation scene of aircraft maintenance, these positioning technologies all have the problem of insufficient adaptability, which will be analyzed as follows:

[0030] Firstly, although the service mobile robot based on SLAM technology has the ability of autonomous positioning and navigation, it highly depends on stable environment mapping. This kind of mobile robot needs to build a complete map before operation, and mainly uses environmental features as the reference for positioning and path planning, which is suitable for closed spaces with regular structure and stable obstacle distribution. However, in the aircraft maintenance operation scene, the target object maintenance station is not fixed, the position is random, and there are a large number of temporary facilities and dynamic interference around, which makes the updating and maintenance of the map frequent and difficult. In addition, the positioning of this kind of mobile robot focuses more on the absolute position of the mobile robot in the scene, while in the aircraft maintenance operation scene, the relative position between the mobile robot and the target object should be focused on. Therefore, the applicability of this kind of mobile robot in the aircraft maintenance operation scene is not strong.

[0031] Secondly, although GNSS positioning technology can provide wide-area position information in outdoor environment, its signal is easily blocked in indoor or semi-closed maintenance plant, which leads to significant decrease of positioning accuracy. At the same time, the position information provided by GNSS is relative to the geographic coordinate system, which cannot meet the relative positioning demand of taking the target object as the reference in the aircraft maintenance operation scene, and cannot adapt to the production demand of aircraft maintenance.

[0032] Finally, beacon positioning relies on pre-set ZigBee, RFID, UWB and other fixed positioning base stations, and achieves local high-precision positioning through ranging or angle measurement. Although it has been successfully applied in indoor scenarios such as warehouse and logistics, it is not flexible enough to adapt to different maintenance tasks in the complex scenario of aircraft maintenance operation due to the complexity of beacon layout, high cost and poor stability in the operation area with frequent changes and high mobility of personnel and equipment.

[0033] Since the above-mentioned positioning technologies cannot meet the requirements of precise positioning and dynamic navigation of complex geometric targets in the aircraft maintenance scenario, the core of this scenario is that the AGV needs to maintain relative position perception with the aircraft, that is, it must take the aircraft itself as the positioning reference and cannot rely on the environment or external signals. To achieve such relative positioning based on the characteristics of the aircraft itself, a technical means is needed to accurately capture the geometric shape of the aircraft. Devices such as laser radar can directly collect three-dimensional point cloud data of the aircraft surface. These point clouds can fully reflect the inherent characteristics of the aircraft, such as contour and component structure, and are the core basis for AGV to identify the aircraft and determine the relative position. Therefore, collecting point clouds by laser radar and processing them become a better choice for realizing positioning in this scenario.

[0034] However, as described in the background art, when actually processing point cloud data, two key problems will naturally arise due to the complexity of the aircraft itself and the characteristics of AGV operation, as follows:

[0035] (1) Over-segmentation problem of point cloud clustering of complex geometric targets under multiple angles

[0036] An aircraft is a complex geometric target, and its components (such as wings, fuselage, tail, engines and landing gear) have significant differences in geometric characteristics, making it impossible to successfully cluster each part of the aircraft based on a specific geometric characteristic. Moreover, when AGV collects point cloud data around the aircraft, the point cloud shape collected from different angles may be completely different. For example, when collecting from the front, the point cloud of the aircraft engine may appear as a spatial circle, while when collecting from the side, it may appear as a semi-circular cone. In addition to geometric characteristics, the density of point cloud collected by laser radar is also affected by target surface shape, distance from the radar, and other factors, further exacerbating the unevenness and complexity of the data.

[0037] Under such complex conditions, if the traditional point cloud clustering method is based on density or geometric rules for segmentation, it will often lead to over-segmentation problem, making it difficult to aggregate the point clouds of multiple components of the aircraft into a whole large point cloud cluster, resulting in the AGV's inability to identify that the target is an aircraft, thereby affecting the identification accuracy when identifying the target aircraft through point cloud clustering, and making it impossible to use it as a reference for precise positioning, thereby affecting the positioning accuracy of the AGV.

[0038] Therefore, how to effectively remove the environmental point cloud and cluster the multiple components of the aircraft into a single cluster without losing local geometric features is one of the key problems that must be overcome by the embodiments of the present application.

[0039] (2) Low overlap registration problem of local point cloud and global point cloud

[0040] The local point cloud is usually composed of a partial region of the aircraft, and the geometric coverage is significantly lower than that of the global point cloud. For example, in a certain frame of local point cloud, the nose and the engine point cloud around it, or the rear fuselage and the tail point cloud, these local point clouds only account for 10% to 30% of the total range of the global point cloud, the number of effective matching point pairs is insufficient, resulting in insufficient global geometric constraint, easy to fall into local optimal solution, and difficult to stably construct a rigid transformation matrix.

[0041] In addition, due to the use of different laser radars when collecting the global point cloud and the local point cloud, the local point cloud is sparse and there is a significant difference in density with the global point cloud. The global point cloud structure is dense and can fully reflect the geometric profile of the aircraft, while the local point cloud is sparse and the feature expression ability is limited, which leads to the fact that the traditional registration algorithm is prone to mismatching under the influence of point cloud sparsity and density difference, which will affect the registration accuracy of the local point cloud and the global point cloud.

[0042] Further, due to the significant non-uniformity of the sampling distribution of the local point cloud, the geometric information density of different regions is greatly different, for example, the point cloud is dense at the nose and wing, and the point cloud is sparse in the tail region. This non-uniform distribution leads to unstable feature extraction and further weakens the information in the overlapping region. Sparsity will reduce the effective feature density in the overlapping region. Even if a certain region is geometrically covered, it may not have enough key points or feature points for the registration algorithm to use due to sparse sampling, so the sparsity problem can also be attributed to the special performance of low overlap.

[0043] Under the above conditions, the AGV positioning accuracy is limited due to error accumulation in the registration process of the local point cloud and the global point cloud, thereby causing path deviation problems.

[0044] Therefore, how to accurately register the point cloud under the condition of low overlap is the second key problem that must be overcome by the embodiments of the present application.

[0045] In order to solve the above key problems, the embodiments of the present application provide an AGV navigation method for an aircraft maintenance site, as shown in Figure 1 The method comprises steps S11 to S15:

[0046] Step S11, in the maintenance operation site of the target aircraft, use the AGV platform to obtain integrated point cloud data covering the surrounding environment.

[0047] In one of the optional embodiments, the integrated point cloud data covering the surrounding environment is obtained by using the AGV platform, specifically comprising:

[0048] The multiple laser radars carried by the AGV platform collect point cloud data of the surrounding environment to obtain multi-source point cloud data;

[0049] For the field of view overlap area of the multiple laser radars, the key point cloud data pairs of the same object from different laser radars in the field of view overlap area are selected, the radar scanning error is obtained according to the key point cloud data pairs, and the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system is corrected according to the radar scanning error;

[0050] According to the corrected external parameter transformation matrix, the multi-source point cloud data is converted in coordinates to convert the multi-source point cloud data into the AGV body coordinate system;

[0051] The multi-source point cloud data after coordinate conversion is collected together to form integrated point cloud data covering the surrounding environment.

[0052] In the specific implementation of the embodiment, the multiple laser radars carried by the AGV platform can be used to collect point cloud data of the surrounding environment of the AGV platform in the target aircraft maintenance operation site, and the multi-source point cloud data collected by the multiple laser radars is obtained accordingly. After that, since the installation position and attitude relationship of each laser radar has been determined in the calibration stage, the multi-source point cloud data collected by each laser radar can be uniformly transformed into the AGV body coordinate system through the known external parameters of each laser radar, that is, the multi-source point cloud data is registered and fused by using the external parameter transformation matrix of each laser radar relative to the AGV body coordinate system, and integrated point cloud data covering the surrounding environment of the AGV platform with high density and strong continuity is generated.

[0053] In order to further improve the registration accuracy, especially for the coincident object boundary (such as large target contour line) in the field of view overlap area of multiple laser radars, the embodiment of the application introduces a fine tuning mechanism based on local feature points, that is, before the coordinate transformation of the multi-source point cloud data, the fine tuning mechanism based on local feature points is used to select the key point cloud data pairs of the same object in the field of view overlap area collected by different laser radars to calculate the radar scanning error of the laser radar with error, and the external parameter transformation matrix of the laser radar with error relative to the AGV body coordinate system is corrected according to the calculated radar scanning error. After the correction is completed, the multi-source point cloud data is converted to the AGV body coordinate system according to the corrected external parameter transformation matrix, and then the multi-source point cloud data after coordinate conversion is collected together to form integrated point cloud data with high density and strong continuity covering the environment around the AGV platform.

[0054] For example, the AGV platform carries four laser radars, which are installed at the four corners of the AGV platform to realize omnidirectional point cloud perception of the environment around the AGV platform. Each laser radar can independently collect three-dimensional point cloud data in its field of view, covering the front, rear and left and right side areas of the AGV platform, and correspondingly forming four groups of point cloud segments with spatial complementarity (i.e. multi-source point cloud data).

[0055] For example, the parameters of the laser radar are as follows: 32-line mechanical laser radar, horizontal scanning angle 360 degrees, vertical scanning angle 89.55 degrees, ranging accuracy ±3cm.

[0056] In one of the optional embodiments, the radar scanning error is obtained according to the key point cloud data pair, and the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system is corrected according to the radar scanning error, specifically including:

[0057] According to the formula Δ=[Δx, Δy, Δz] T The radar scanning error is calculated; wherein Δx, Δy, Δz represent the coordinate deviations of the key point cloud data pair in the x-axis, y-axis and z-axis of the AGV body coordinate system.

[0058] According to the formula t'=t+Δ, the translation vector of the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system is corrected; wherein t represents the translation vector of the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system, and t' represents the corrected translation vector.

[0059] In combination with the above embodiments, in the specific implementation of the present embodiment, firstly, the coordinate deviations of the key point cloud data collected from different laser radars in the field of view overlap region of the same object (at least including the key point cloud data collected from the laser radar with errors) in the x-axis, y-axis and z-axis of the AGV body coordinate system are calculated, and the coordinate deviation of the x-axis is Δx, the coordinate deviation of the y-axis is Δy, and the coordinate deviation of the z-axis is Δz, and then the radar scanning error Δ of the laser radar with errors is calculated according to the formula Δ=[Δx, Δy, Δz] T , and then the translation vector t of the extrinsic transformation matrix of the laser radar with errors relative to the AGV body coordinate system is corrected according to the formula t'=t+Δ, and the corrected translation vector t' is obtained.

[0060] For example, the AGV platform carries four laser radars, the AGV body coordinate system is set as the reference coordinate system for unified alignment of point clouds, and in the calibration stage of the laser radars, the installation pose parameters of each laser radar relative to the AGV body coordinate system are known, including the rotation relationship and the spatial displacement, and the installation pose parameters can be represented as a set of extrinsic transformation matrices T∈R 4×4 , wherein the extrinsic transformation matrix of the i-th laser radar relative to the AGV body coordinate system is: Corresponding to the four laser radars, R i represents the rotation matrix from the radar coordinate system of the i-th laser radar to the AGV body coordinate system, t i represents the corresponding translation vector.

[0061] Suppose that the i-th laser radar has no error, then the multi-frame point cloud data collected by the i-th laser radar can be represented as: P i ={p1, p2,...,p n}, the j-th (j=1, 2,..., n) frame of point cloud data p i in P j is expanded into homogeneous form Then, the extrinsic transformation matrix T i of the i-th laser radar relative to the AGV body coordinate system is used for transformation: That is, the j-th frame of point cloud data p j is projected to the AGV body coordinate system to obtain its expression in the AGV body coordinate system as p j_AGV , and correspondingly, each frame of point cloud data in P i is subjected to coordinate transformation, and the expression of P i in the AGV body coordinate system is obtained as P i_AGV, thereby achieving the initial fusion of each point cloud data in the unified AGV body coordinate system; accordingly, after completing the coordinate unification of the four sets of point cloud data collected by the four lidars, these four sets of point cloud data can be combined into a set of integrated point cloud data: And integrate point cloud data P fused It covers the area around the AGV platform and has high spatial density and geometric continuity.

[0062] Furthermore, assuming that the i-th laser radar has an error, then, as described above, the radar scanning error Δ = [Δx, Δy, Δz] can be calculated T , then, for the external parameter transformation matrix T of the i-th laser radar relative to the AGV body coordinate system i , can be calculated based on the radar scanning error Δ=[Δx,Δy,Δz] T T i The translation vector t in i Make corrections: i '=t i +Δ, thereby compensating for the errors of the i-th laser radar caused by actual installation errors, scanning offsets or inconsistent calibration, and improving the coincidence and continuity of the spliced ​​point cloud. Correspondingly, the corrected external parameter transformation matrix of the i-th laser radar is For the j-th frame point cloud data p j The expression for coordinate transformation is:

[0063] Step S12: Using the AGV platform to collect the original local point cloud data of the maintenance operation site, eliminating the environmental interference point cloud data according to the integrated point cloud data, and using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics to process the local point cloud data after interference elimination to obtain the target local point cloud data corresponding to the target aircraft.

[0064] It can be understood that after using multiple laser radars carried by the AGV platform to collect the original local point cloud data of the maintenance operation site of the target aircraft, since there will be some environmental interference point cloud data in the original local point cloud data, that is, in the integrated point cloud data obtained in step S11, it is necessary to eliminate the environmental interference point cloud data in the original local point cloud data based on the obtained integrated point cloud data, and obtain the local point cloud data after eliminating the interference accordingly. After that, subsequent clustering processing is performed to effectively extract the target local point cloud data corresponding to the target aircraft based on the local point cloud data after eliminating the interference.

[0065] It should be noted that even if the environmental interference point cloud data in the original local point cloud data is not considered, due to the wide scanning range of the laser radar during operation, the obtained original local point cloud data may also include other aircraft, maintenance equipment, personnel and obstacles around the target aircraft and other stray targets in addition to the target aircraft itself, as shown in Figure 2 The structure diagram of the original local point cloud data collected by the laser radar provided by an embodiment of the present application is shown, which represents the typical scene information collected by the laser radar, according to Figure 2 It can be seen that the original local point cloud data not only includes the target aircraft, but also includes two surrounding aircrafts and obstacles in the field, therefore, in order to avoid non-target point cloud interference in subsequent calculation, an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics is needed to process the original local point cloud data, so as to automatically separate multiple point cloud clusters in the scene and select the main point cloud cluster to which the target aircraft belongs.

[0066] It should be noted that before the adaptive point cloud clustering algorithm based on density and spatial distribution characteristics is used for clustering processing, related processing of ground point cloud elimination and interested region selection can also be performed, wherein the ground point cloud elimination can be: fitting the dominant plane in the point cloud by using an algorithm, and determining the point cloud on the plane as the ground and performing elimination processing; the interested region selection can be: by setting the threshold range of the x-axis, y-axis and z-axis, limiting a space region in which the target aircraft may exist, eliminating the point cloud outside the space region, and only retaining the point cloud within the space region for subsequent clustering processing; or, the ground point cloud elimination and the interested region selection can also be implemented in other ways, both of which belong to the conventional operation in point cloud processing, and the embodiments of the present application are not limited in detail.

[0067] In one of the optional embodiments, the adaptive point cloud clustering algorithm based on density and spatial distribution characteristics is used to process the local point cloud data after the interference is eliminated, to obtain the target local point cloud data corresponding to the target aircraft, which specifically includes:

[0068] The mean shift algorithm is used to dynamically super-voxelize according to the local density and geometric information of the local point cloud data after the interference is eliminated, to generate a voxel space with density and geometric information; wherein the attributes of each voxel in the voxel space include point density, geometric center and normal vector;

[0069] A point cloud similarity graph is constructed according to the voxel space, and the normal vector is introduced as a new similarity measure in the point cloud similarity graph; wherein the definition of the new similarity measure is: s(x i ,x j ) represents the voxel x i and the voxel x ja new similarity measure between voxels i and j, i and j represent voxel indices, d xy , d z represents the horizontal distance, vertical distance between voxels x i and x j , θ represents the normal vector angle between voxels x i and x j , σ xy , σ z , σ θ represents the scale parameter corresponding to the horizontal distance, vertical distance, normal vector angle;

[0070] An adaptive scale parameter setting method is adopted to dynamically adjust the scale parameter through the local standard deviation of the voxel neighborhood, and an adjusted adaptive scale parameter is obtained.

[0071] According to the point cloud similarity graph and the adjusted adaptive scale parameter, a sparse adjacency graph is constructed to obtain a sparse similarity matrix, and a normalized Laplacian matrix is generated according to the sparse similarity matrix; wherein, L represents the normalized Laplacian matrix, D represents the degree matrix, and W represents the sparse similarity matrix.

[0072] Eigenvalue decomposition is performed on the normalized Laplacian matrix to obtain K1 eigenvectors corresponding to the smallest K1 eigenvalues, and after mapping the K1 eigenvectors to a low-dimensional feature space, a K-Means algorithm is used for clustering to obtain an initial clustering result; wherein, K1 is a positive integer.

[0073] The initial clustering result is dynamically post-processed by small cluster merging, large cluster decomposition, and environment point cloud removal to obtain a processed clustering result, and a main point cloud cluster to which the target aircraft belongs is selected from the processed clustering result to obtain target local point cloud data corresponding to the target aircraft.

[0074] In one of the optional embodiments, the adaptive scale parameter setting method is adopted to dynamically adjust the scale parameter through the local standard deviation of the voxel neighborhood, and an adjusted adaptive scale parameter is obtained, which specifically includes:

[0075] The adjusted adaptive scale parameter is calculated according to the formula σ k =std({||x i -x j |||x j ∈KNN(x i )}); wherein, σ k represents the adjusted adaptive scale parameter, k is the scale parameter index, std() represents the standard deviation of the distance between the voxel neighborhood points, ||x i -x j|| indicates the calculation of voxel x i With voxel x j The distance between them, KNN(x i ) represents the calculation of voxel x i The K2 nearest neighbor point set, x j ∈KNN(x i ) represents voxel x j is the voxel x i A domain point in the set of K2 nearest neighboring points, where K2 is a positive integer.

[0076] In combination with the above embodiments, in the specific implementation of this embodiment, first, in order to reduce the computational complexity of point cloud data and retain local geometric features, the point cloud data can be supervoxelized. That is, the Mean Shift algorithm can be used to dynamically supervoxelize the local point cloud data according to the local density and geometric information after eliminating interference, and a voxel space with density and geometric information is generated accordingly. The attributes of each voxel in the voxel space include point density, geometric center, and normal vector, and PCA (Principal Component Analysis) can be used to calculate the normal vector of the voxel as a key attribute describing its local geometric characteristics. Then, a point cloud similarity graph can be constructed based on the generated voxel space, and when constructing the point cloud similarity graph, the normal vector of the voxel is introduced as a new similarity measure. Based on the traditional Gaussian similarity function, a new similarity measure (i.e., an improved multi-attribute similarity measure) is defined as: s(x i ,x j ) represents voxel x i With voxel x j A new similarity measure between, i and j represent voxel indices, d xy Represents voxel x i With voxel x j The horizontal distance between z Represents voxel x i With voxel x j The vertical distance between the voxels, θ represents the voxel x i With voxel x j The angle between the normal vectors (i.e. the angle between the normal vectors of the two), σ xy , σ z , σ θ, which is a scale parameter corresponding to the horizontal distance, the vertical distance, and the normal vector angle, and the similarity function comprehensively considers the influence of spatial position, geometric direction, and local density, which is helpful to accurately describe the point cloud distribution of complex geometric targets; then, in order to solve the problem that the traditional spectral clustering depends on the globally fixed scale parameter, an adaptive scale parameter setting method can be used, specifically, the scale parameter of each voxel is dynamically adjusted by the local standard deviation of the voxel neighborhood, and the adjusted adaptive scale parameter is obtained accordingly, wherein the adjusted adaptive scale parameter σ k is: k σ i =std({||x j -x j |||x i ∈KNN(x i )}),k is a scale parameter index, std() represents the standard deviation of the distance between the voxel neighborhood points, ||x j -x i || represents the distance between the voxel x j and the voxel x i , KNN(x j ) represents the K2 nearest neighborhood point set of the voxel x i , x j ∈KNN(x i ) represents a field point in the K2 nearest neighborhood point set of the voxel x T , and K2 is a positive integer, which can automatically adjust the clustering scale according to the change of point cloud density and improve the adaptability of spectral clustering to complex point clouds; then, in the specific implementation of spectral clustering, the sparse adjacency graph can be constructed by using the obtained point cloud similarity graph and the adjusted adaptive scale parameter, and the sparse similarity matrix can be generated according to the constructed sparse similarity matrix, and the normalized Laplacian matrix can be generated according to the constructed sparse similarity matrix, wherein the expression of the normalized Laplacian matrix L is: D represents the degree matrix, and W represents the sparse similarity matrix; then, the generated normalized Laplacian matrix is subjected to eigenvalue decomposition, through the eigenvalue decomposition, the K1 eigenvectors corresponding to the K1 smallest eigenvalues are obtained (i.e. K1 eigenvectors, and K1 is a positive integer), and after the K1 eigenvectors are mapped to a low-dimensional feature space, the K-Means algorithm is used for clustering to complete the voxel grouping, and the initial clustering result is obtained accordingly; finally, the initial clustering result is dynamically post-processed, the dynamic post-processing includes small cluster merging, large cluster decomposition, and environment point cloud removal, to further optimize the clustering and segmentation effect, and the processed clustering result is obtained accordingly, and the final achievement of point cloud clustering is a complete local point cloud cluster, and then the main point cloud cluster to which the target aircraft belongs can be selected from the processed clustering result, and the target local point cloud data corresponding to the target aircraft is obtained accordingly.

[0077] As Figure 2 shown, in the case of including target aircraft, two surrounding aircraft and obstacle point cloud data in the original local point cloud data, the embodiment of the application can combine multiple geometric features such as spatial density distribution, field size, and normal vector consistency for clustering recognition, and automatically filter out the point cloud cluster with continuous spatial distribution, regular morphological characteristics, and consistent local normal vector as the target aircraft. The scattered obstacles are easily excluded in clustering due to small shape, low density, or unstable structure.

[0078] Step S13, using the NDT algorithm to register the target local point cloud data with the preset whole machine three-dimensional point cloud database, obtaining the target whole machine point cloud model corresponding to the target aircraft and the rigid transformation matrix of the target aircraft relative to the AGV body coordinate system, and obtaining the spatial pose of the AGV platform in the whole machine point cloud coordinate system according to the rigid transformation matrix, to realize the relative positioning of the AGV platform; wherein the whole machine three-dimensional point cloud database includes the whole machine point cloud model corresponding to multiple typical aircraft models, which is generated by the point cloud of the key structural components of the aircraft and is in a unified whole machine point cloud coordinate system.

[0079] In one of the optional embodiments, the whole machine three-dimensional point cloud database is pre-established by the following steps:

[0080] Using multiple laser radars carried by the AGV platform to collect multi-view point cloud data of key structural components of multiple typical aircraft models; wherein the key structural components of the aircraft at least include the fuselage, the wing, the tail and the landing gear;

[0081] For each typical aircraft model, the collected multi-view multi-frame point cloud data is denoised and spliced to generate a whole machine point cloud model corresponding to each typical aircraft model;

[0082] A unified whole machine point cloud coordinate system is established for the whole machine point cloud model corresponding to each typical aircraft model; wherein the whole machine point cloud coordinate system takes the preset reference point of the aircraft body as the origin, takes the direction along the fuselage pointing to the tail as the x-axis, takes the direction perpendicular to the fuselage and pointing to the left wing as the y-axis, and takes the direction perpendicular to the fuselage and pointing to the upper as the z-axis.

[0083] In combination with the above embodiments, in the specific implementation, a three-dimensional point cloud database of the whole machine can be established in advance for a plurality of typical machine types (for example, A320 series and B737 series, etc.) involved in the project, to serve as basic data support (generally completed before the project implementation, and in subsequent navigation positioning tasks, the point cloud model of the whole machine of the corresponding machine type can be directly called as needed for registration), and the specific establishment process is as follows: first, a plurality of laser radars carried by the AGV platform are used to collect point cloud data of the key structural components of the aircraft of a plurality of typical machine types from multiple perspectives, so that the three-dimensional point cloud database of the whole machine is composed of the point cloud data collected from multiple perspectives for each typical machine type, and covers the fuselage, wings, tail and landing gear and other key structural components of the aircraft of each typical machine type; then, the plurality of frames of point cloud data collected from multiple perspectives for each typical machine type are subjected to denoising and splicing processing, and a unified and complete whole machine point cloud model corresponding to each typical machine type is correspondingly generated; finally, in order to facilitate subsequent coordinate conversion and pose estimation, a coordinate system can be established on the aircraft body for all whole machine point cloud models, as a whole machine point cloud coordinate system, as shown in Figure 3 Fig. 1 is a schematic diagram of a whole machine point cloud coordinate system provided by an embodiment of the present application, and the whole machine point cloud coordinate system takes a preset reference point of the aircraft body as a coordinate origin, for example, the preset reference point can be the aircraft center of gravity or the landing gear at the aircraft nose, etc. Figure 3 Fig. 2 shows that the coordinate origin is set at the aircraft center of gravity, the x-axis points to the direction of the fuselage towards the tail, the y-axis is perpendicular to the fuselage and points to the direction of the left wing, and the z-axis is perpendicular to the fuselage and vertically points to the upward direction, and by establishing a unified whole machine point cloud coordinate system for all whole machine point cloud models, it can be ensured that the model has a clear and stable spatial reference base.

[0084] In one of the optional embodiments, the NDT algorithm is used to register the target local point cloud data with the preset three-dimensional point cloud database of the whole machine, to obtain a target whole machine point cloud model corresponding to the target aircraft and a rigid transformation matrix of the target whole machine point cloud model relative to the AGV body coordinate system, and to obtain the spatial pose of the AGV platform in the whole machine point cloud coordinate system according to the rigid transformation matrix, to realize the relative positioning of the AGV platform, and specifically includes:

[0085] The NDT algorithm is used to register and compare the target local point cloud data with each whole machine point cloud model in the preset three-dimensional point cloud database of the whole machine in sequence, and the whole machine point cloud model with the smallest fitting error is selected as the target whole machine point cloud model corresponding to the target aircraft;

[0086] The target local point cloud data is registered in space with the target whole machine point cloud model, a geometric constraint relationship between the target aircraft and the target whole machine point cloud model is constructed in the three-dimensional space through a probability distribution model, and a rigid transformation matrix of the target whole machine point cloud model relative to the AGV body coordinate system is output.

[0087] Invert the rigid transformation matrix to convert the known pose of the AGV platform in the AGV body coordinate system to the whole-machine point cloud coordinate system, to obtain the spatial pose of the AGV platform in the whole-machine point cloud coordinate system, thereby obtaining the current positioning position of the AGV platform relative to the target aircraft in the maintenance work site.

[0088] In combination with the above embodiment, in specific implementation, the NDT (Normal Distributions Transform) algorithm can be used first to sequentially register and compare the target local point cloud data corresponding to the target aircraft with each whole-machine point cloud model in the preset whole-machine three-dimensional point cloud database, and select the whole-machine point cloud model with the smallest fitting error as the target whole-machine point cloud model corresponding to the target aircraft; then, the target local point cloud data corresponding to the target aircraft is spatially registered with the selected target whole-machine point cloud model, the NDT algorithm can construct the geometric constraint relationship between the target aircraft and the target whole-machine point cloud model in the three-dimensional space through the probability distribution model, and finally output the rigid transformation matrix (including rotation and translation information) of the target whole-machine point cloud model relative to the AGV body coordinate system; since the target local point cloud data corresponding to the target aircraft is collected with the laser radar as a reference, based on the rigid transformation matrix obtained through registration, the known pose of the AGV platform in the AGV body coordinate system can be converted to the whole-machine point cloud coordinate system through inverse operation on the rigid transformation matrix, and the spatial pose of the AGV platform in the whole-machine point cloud coordinate system is correspondingly obtained, thereby realizing the relative positioning of the AGV platform in the maintenance work site, that is, obtaining the current positioning position of the AGV platform relative to the target aircraft in the maintenance work site.

[0089] Step S14: calling a navigation map template corresponding to the target whole-machine point cloud model, and performing coordinate conversion on the path node coordinates in the navigation map template according to the rigid transformation matrix.

[0090] In combination with the above embodiment, in specific implementation, the navigation map template corresponding to the selected target whole-machine point cloud model can be called, and the coordinate conversion is performed on each path node coordinate in the navigation map template according to the rigid transformation matrix obtained through registration, so that the pose of the target aircraft in the actual maintenance work site is kept consistent, and the path nodes are accurately aligned along the actual contour of the target aircraft, which is executable.

[0091] It should be noted that the navigation map template can be pre-set according to the specific maintenance task requirements, and is usually based on the maintenance work points of each part of the aircraft, combined with the actual situation of the maintenance site to formulate each path node, and a navigation trajectory is generated through an artificial path planning mode; wherein each typical machine type is provided with a corresponding navigation map template, and because the typical machine type and the whole machine point cloud model are one-to-one corresponding, the navigation map template and the whole machine point cloud model are one-to-one corresponding, and the navigation map template is used to align the path with the actual attitude of the aircraft after registration is completed.

[0092] In step S15, based on the navigation map template after coordinate conversion, the optimal navigation path is automatically generated from the current positioning position of the AGV platform as the starting point and the maintenance work point as the target point, and the AGV platform is controlled to autonomously navigate and travel along the optimal navigation path.

[0093] In combination with the above embodiment, after the alignment of the navigation map template and the current attitude of the target aircraft is completed, the set starting point and target point can be selected from the current positioning position of the AGV platform according to the navigation map template after coordinate conversion, as shown in Figure 4 The starting point is the current positioning position of the AGV platform (i.e. the initial pose of the AGV platform), and the target point is the maintenance work point that the AGV platform needs to reach, and then an optimal navigation path is automatically generated according to the selected starting point and target point in combination with the existence of obstacles near the target aircraft, for example, Figure 4 As shown in the figure, there is an obstacle near the nose of the target aircraft, so local obstacle avoidance is needed when planning the path; further, the generated optimal navigation path can be loaded into the chassis control module of the AGV platform to control the AGV platform to autonomously navigate and travel according to the optimal navigation path, realizing autonomous movement from the current positioning position of the AGV platform to the maintenance work point.

[0094] It should be noted that the path planning can be based on the path node information in the navigation map template after coordinate conversion, combined with the spatial position relationship for judgment and scheduling, to finally form a continuous and executable path trajectory, and the path planning can be directly implemented using existing technologies, and the embodiments of the present application are not limited in detail.

[0095] In summary, the AGV navigation method for aircraft maintenance sites provided by the embodiments of the present application has the following beneficial effects:

[0096] (1) No need to rely on environment mapping, suitable for complex scenes

[0097] In the aircraft maintenance site, the situation is complex and changeable, the maintenance station of each aircraft is not fixed every time, and the maintenance equipment in the environment also moves frequently, so that the traditional map maintenance based on the SLAM technology suffers great challenges; the embodiment of the application focuses on the invariable quantity in the changing scene, that is, the shape of the aircraft is invariable, by directly extracting the point cloud of the target aircraft and registering with the whole machine point cloud model, without pre-building the environment map, it can stably run in the disordered and frequently changing scene, which is different from the requirement of building a global map in the traditional SLAM technology in a static or semi-static environment, and has stronger environmental adaptability;

[0098] (2) With the target object as the reference, the self-positioning is more robust and flexible

[0099] Compared with the GNSS positioning limited by satellite signal shielding or the beacon positioning dependent on environmental layout and susceptible to interference, the embodiment of the application takes the target aircraft as the positioning reference, obtains the relative pose information through point cloud registration, does not need external signal support, and does not depend on environmental features or auxiliary marker equipment, and still can maintain high positioning accuracy and stability in the semi-closed and dynamically changing complex scene such as the hangar, has stronger environmental robustness, at the same time, does not need manual control and external positioning facilities before use, has high autonomy, and is convenient to deploy, and is especially suitable for the industrial site environment with large space limitation and frequent structural changes;

[0100] (3) The self-adaptive point cloud clustering has high recognition accuracy

[0101] In a multi-target scene (such as the presence of maintenance tools, other aircraft and the like near the aircraft), the traditional point cloud clustering is easy to misclassify non-target objects into the map or navigation reference, leading to recognition deviation and affecting the recognition accuracy, the embodiment of the application introduces the self-adaptive point cloud clustering algorithm, can effectively eliminate the environmental interference points, extracts the target aircraft point cloud, realizes more accurate and clean input data, and provides a basis for subsequent point cloud processing, thereby improving the positioning accuracy of the subsequent AGV and ensuring the correctness of the autonomous navigation path of the AGV;

[0102] (4) Support stable registration of local point cloud under low overlap and high sparsity conditions

[0103] Compared with the problem that the ICP (Iterative Closest Point) algorithm and the like are easy to fall into local optimization in low overlap point cloud, the embodiment of the application performs probabilistic registration on the local point cloud and the global model by using the NDT algorithm, even if only 10% to 30% of the target aircraft is covered, high-precision pose estimation can be completed, thereby improving the positioning accuracy of the AGV and ensuring the correctness of the autonomous navigation path of the AGV, and at the same time, the robustness under the condition of limited view angle or sparse perception is also significantly improved.

[0104] The embodiment of the present application also provides an AGV navigation system for an aircraft maintenance site, which is used for realizing the AGV navigation method for an aircraft maintenance site in any of the above-mentioned embodiments. Figure 5 As shown in FIG. 1, which is a structural block diagram of an AGV navigation system for an aircraft maintenance site according to an embodiment of the present application, the system comprises:

[0105] An environmental point cloud data acquisition module 11 is configured to acquire integrated point cloud data covering the surrounding environment by using an AGV platform at a target aircraft maintenance site;

[0106] A local point cloud data acquisition module 12 is configured to acquire original local point cloud data of the target aircraft maintenance site by using the AGV platform, remove environmental interference point cloud data from the original local point cloud data according to the integrated point cloud data, and process the local point cloud data after removing the interference by using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics, so as to obtain target local point cloud data corresponding to the target aircraft.

[0107] A local point cloud data registration and positioning module 13 is configured to register the target local point cloud data with a preset whole-machine three-dimensional point cloud database by using an NDT algorithm, acquire a target whole-machine point cloud model corresponding to the target aircraft and a rigid transformation matrix of the target whole-machine point cloud model relative to an AGV body coordinate system, and acquire a spatial pose of the AGV platform in the whole-machine point cloud coordinate system according to the rigid transformation matrix, so as to realize relative positioning of the AGV platform; wherein the whole-machine three-dimensional point cloud database comprises whole-machine point cloud models corresponding to multiple typical aircraft models, which are generated by point clouds of key structural components of the aircraft and are in a unified whole-machine point cloud coordinate system.

[0108] A navigation map coordinate conversion module 14 is configured to call a navigation map template corresponding to the target whole-machine point cloud model, and perform coordinate conversion on path node coordinates in the navigation map template according to the rigid transformation matrix.

[0109] An autonomous navigation path planning module 15 is configured to automatically generate an optimal navigation path based on the navigation map template after the coordinate conversion, take a current positioning position of the AGV platform as a starting point, and take a maintenance work point as a target point, and control the AGV platform to autonomously navigate and travel along the optimal navigation path.

[0110] Preferably, the environmental point cloud data acquisition module 11 specifically comprises:

[0111] An environmental point cloud data acquisition unit is configured to acquire multi-source point cloud data by using multiple laser radars carried by the AGV platform to perform point cloud data acquisition on the surrounding environment.

[0112] An error correction unit is configured to, for a field of view overlap area of multiple laser radars, select a key point cloud data pair of a same object from different laser radars in the field of view overlap area, acquire a radar scanning error according to the key point cloud data pair, and correct an external parameter transformation matrix of the laser radar relative to an AGV body coordinate system according to the radar scanning error.

[0113] An environmental point cloud coordinate conversion unit is configured to perform coordinate conversion on the multi-source point cloud data according to the corrected external parameter transformation matrix, so as to convert the multi-source point cloud data to the AGV body coordinate system.

[0114] An integrated point cloud data forming unit is configured to collect the coordinate-converted multi-source point cloud data together to form integrated point cloud data covering the surrounding environment.

[0115] Preferably, the error correction unit acquires a radar scanning error according to the key point cloud data pair, and corrects an external parameter transformation matrix of the laser radar relative to the AGV body coordinate system according to the radar scanning error, and specifically includes:

[0116] According to a formula Δ=[Δx, Δy, Δz] T a radar scanning error is calculated and obtained; wherein Δx, Δy, Δz represent coordinate deviations of the key point cloud data pair in the x-axis, y-axis and z-axis of the AGV body coordinate system.

[0117] According to a formula t'=t+Δ, a translation vector of the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system is corrected; wherein t represents the translation vector of the external parameter transformation matrix of the laser radar relative to the AGV body coordinate system, and t' represents the corrected translation vector.

[0118] Preferably, the local point cloud data acquisition module 12 specifically includes:

[0119] A super voxel processing unit is configured to perform dynamic super voxelization on the local density and geometric information of the interference-removed local point cloud data by using a mean shift algorithm, to generate a voxel space with density and geometric information; wherein the attributes of each voxel in the voxel space include point density, geometric center and normal vector.

[0120] A point cloud similarity graph construction unit is configured to construct a point cloud similarity graph according to the voxel space, and introduce the normal vector as a new similarity measure in the point cloud similarity graph; wherein the new similarity measure is defined as: s(x i ,x j ) represents a new similarity measure between voxel x i and voxel x j , i and j represent voxel indexes, and dxy , d z denotes the horizontal distance, the vertical distance, and the normal vector angle between the voxel x i and the voxel x j , and θ denotes the normal vector angle between the voxel x i and the voxel x j , and σ xy , σ z , σ θ denotes the scale parameter corresponding to the horizontal distance, the vertical distance, and the normal vector angle;

[0121] The scale parameter adjustment unit is configured to adopt an adaptive scale parameter setting method, dynamically adjust the scale parameter through a local standard deviation of a voxel neighborhood, and obtain an adjusted adaptive scale parameter.

[0122] The matrix generation unit is configured to adopt a sparse adjacency graph to construct a sparse similarity matrix according to the point cloud similarity graph and the adjusted adaptive scale parameter, and generate a normalized Laplacian matrix according to the sparse similarity matrix; wherein, L denotes the normalized Laplacian matrix, D denotes a degree matrix, and W denotes the sparse similarity matrix.

[0123] The point cloud clustering unit is configured to perform eigenvalue decomposition on the normalized Laplacian matrix, obtain K1 eigenvectors corresponding to the smallest K1 eigenvalues, and then perform clustering on the K1 eigenvectors after mapping the K1 eigenvectors to a low-dimensional feature space, to obtain an initial clustering result; wherein K1 is a positive integer.

[0124] The target local point cloud acquisition unit is configured to perform dynamic post-processing such as small cluster merging, large cluster decomposition, and environment point cloud removal on the initial clustering result, to obtain a processed clustering result, and filter out a main point cloud cluster to which the target aircraft belongs from the processed clustering result, to obtain target local point cloud data corresponding to the target aircraft.

[0125] Preferably, the scale parameter adjustment unit adopts an adaptive scale parameter setting method, dynamically adjusts the scale parameter through a local standard deviation of a voxel neighborhood, and obtains an adjusted adaptive scale parameter, and specifically includes the following steps:

[0126] The adjusted adaptive scale parameter is obtained according to the formula σ k = std({||x i -x j |||x j ∈KNN(x i )}) ; wherein σ k denotes the adjusted adaptive scale parameter, k is a scale parameter index, std() denotes a standard deviation of a distance between voxel neighborhood points, and ||x i-x j || indicates the calculation of voxel x i With voxel x j The distance between them, KNN(x i ) represents the calculation of voxel x i The K2 nearest neighbor point set, x j ∈KNN(x i ) represents voxel x j is the voxel x i A domain point in the set of K2 nearest neighboring points, where K2 is a positive integer.

[0127] Preferably, the system further includes a whole-machine point cloud database establishment module, which is used to pre-establish the whole-machine three-dimensional point cloud database through the following steps:

[0128] Utilize multiple laser radars on the AGV platform to collect multi-perspective point cloud data of key structural components of various typical aircraft models; wherein the key structural components of the aircraft include at least the fuselage, wings, tail and landing gear;

[0129] For each typical aircraft model, the collected multi-view multi-frame point cloud data is denoised and spliced ​​to generate a corresponding whole-machine point cloud model for each typical aircraft model;

[0130] A unified whole-aircraft point cloud coordinate system is established for the whole-aircraft point cloud model corresponding to each typical aircraft model; wherein, the whole-aircraft point cloud coordinate system takes the preset reference point of the aircraft body as the origin, the direction along the fuselage pointing to the tail as the x-axis, the direction perpendicular to the fuselage and pointing to the left wing as the y-axis, and the direction perpendicular to the fuselage and pointing upward as the z-axis.

[0131] Preferably, the local point cloud data registration and positioning module 13 specifically includes:

[0132] a local point cloud data registration unit, configured to sequentially register and compare the target local point cloud data with each whole aircraft point cloud model in a preset whole aircraft three-dimensional point cloud database using an NDT algorithm, and select the whole aircraft point cloud model with the smallest fitting error as the target whole aircraft point cloud model corresponding to the target aircraft;

[0133] A rigid body transformation matrix acquisition unit is used to spatially register the target local point cloud data with the target whole aircraft point cloud model, construct a geometric constraint relationship between the target aircraft and the target whole aircraft point cloud model in three-dimensional space through a probability distribution model, and output a rigid body transformation matrix of the target whole aircraft point cloud model relative to the AGV body coordinate system;

[0134] The AGV platform positioning unit is used for inverse operation on the rigid body transformation matrix, so as to convert the known posture of the AGV platform in the AGV body coordinate system to the whole machine point cloud coordinate system, obtain the spatial posture of the AGV platform in the whole machine point cloud coordinate system, and thus obtain the current positioning position of the AGV platform in the maintenance work site relative to the target aircraft.

[0135] It should be noted that the AGV navigation system for the aircraft maintenance site provided in the embodiments of the present application can realize all processes of the AGV navigation method for the aircraft maintenance site described in any of the above embodiments, and the functions and technical effects of each module and unit in the system are the same as those of the AGV navigation method for the aircraft maintenance site described in the above embodiments, which will not be repeated here.

[0136] The embodiments of the present application also provide a computer readable storage medium, including a stored computer program, which controls the device where the computer readable storage medium is located to execute the AGV navigation method for the aircraft maintenance site described in any of the above embodiments when running.

[0137] The embodiments of the present application also provide a terminal device, as shown in Figure 6 The terminal device includes a processor 10, a memory 20, and a computer program stored in the memory 20 and configured to be executed by the processor 10, and the processor 10 realizes the AGV navigation method for the aircraft maintenance site described in any of the above embodiments when executing the computer program.

[0138] Preferably, the computer program can be divided into one or more modules / units (such as computer program 1, computer program 2, …), which are stored in the memory 20 and executed by the processor 10 to complete the present application. The one or more modules / units can be a series of computer program instruction segments that can complete a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0139] The processor 10 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 10 can also be any conventional processor. The processor 10 is a control center of the terminal device, and connects various parts of the terminal device through various interfaces and lines.

[0140] The memory 20 mainly includes a program storage area and a data storage area. The program storage area can store an operating system, at least one application required by a function, etc., and the data storage area can store related data, etc. In addition, the memory 20 can be a high-speed random access memory, and can also be a non-volatile memory such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., or the memory 20 can also be other volatile solid-state storage devices.

[0141] It should be noted that the terminal device described above can include, but is not limited to, a processor and a memory, and those skilled in the art can understand that, Figure 6 The structural block diagram shown is only an example of the structure of the terminal device described above, and does not constitute a limitation on the structure of the terminal device described above. The terminal device described above can include more or fewer components than those shown, or combine certain components, or different components.

[0142] The above is only a preferred embodiment of the present application, and it should be noted that those skilled in the art can make several improvements and modifications without departing from the technical principles of the present application. These improvements and modifications should also be considered within the scope of protection of the present application.

Claims

1. An AGV navigation method for aircraft maintenance sites, characterized in that: include: At the target aircraft maintenance site, the AGV platform is used to obtain integrated point cloud data covering the surrounding environment; The AGV platform is used to collect original local point cloud data of the maintenance operation site, and the environmental interference point cloud data is eliminated according to the integrated point cloud data. The local point cloud data after interference is eliminated is processed using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics to obtain target local point cloud data corresponding to the target aircraft; The NDT algorithm is used to align the target local point cloud data with a preset whole-aircraft 3D point cloud database to obtain the target whole-aircraft point cloud model corresponding to the target aircraft and its rigid body transformation matrix relative to the AGV body coordinate system. The spatial posture of the AGV platform in the whole-aircraft point cloud coordinate system is obtained based on the rigid body transformation matrix to achieve relative positioning of the AGV platform. The whole-aircraft 3D point cloud database includes whole-aircraft point cloud models corresponding to various typical aircraft models, generated from point clouds of key aircraft structural components, and located in a unified whole-aircraft point cloud coordinate system. Calling a navigation map template corresponding to the target whole-machine point cloud model, and performing coordinate transformation on the path node coordinates in the navigation map template according to the rigid body transformation matrix; Based on the navigation map template after coordinate conversion, with the current positioning position of the AGV platform as the starting point and the maintenance operation point as the target point, the optimal navigation path is automatically generated and the AGV platform is controlled to autonomously navigate along the optimal navigation path.

2. The AGV navigation method for aircraft maintenance sites according to claim 1, characterized in that: The use of the AGV platform to obtain integrated point cloud data covering the surrounding environment specifically includes: Use multiple laser radars on the AGV platform to collect point cloud data of the surrounding environment and obtain multi-source point cloud data; For the overlapping areas of the fields of view of multiple lidars, select key point cloud data pairs from different lidars of the same object in the overlapping areas of the fields of view, obtain the radar scanning error based on the key point cloud data pairs, and correct the extrinsic parameter transformation matrix of the lidar relative to the AGV body coordinate system based on the radar scanning error; Performing coordinate transformation on the multi-source point cloud data according to the corrected extrinsic parameter transformation matrix to transform the multi-source point cloud data into the AGV body coordinate system; The multi-source point cloud data after coordinate conversion are gathered together to form integrated point cloud data covering the surrounding environment.

3. The AGV navigation method for aircraft maintenance sites according to claim 2, characterized in that: The step of obtaining a radar scanning error based on the key point cloud data and correcting an external parameter transformation matrix of the laser radar relative to the AGV body coordinate system based on the radar scanning error specifically includes: According to the formula Δ=[Δx,Δy,Δz] T Calculate and obtain the radar scanning error; wherein Δx, Δy, and Δz represent the coordinate deviations of the key point cloud data with respect to the x-axis, y-axis, and z-axis in the AGV body coordinate system; The translation vector of the extrinsic transformation matrix of the laser radar relative to the AGV body coordinate system is corrected according to the formula t'=t+Δ; where t represents the translation vector of the extrinsic transformation matrix of the laser radar relative to the AGV body coordinate system, and t' represents the corrected translation vector.

4. The AGV navigation method for aircraft maintenance sites according to claim 1, wherein: The adaptive point cloud clustering algorithm based on density and spatial distribution characteristics is used to process the local point cloud data after interference removal to obtain the target local point cloud data corresponding to the target aircraft, specifically including: Using a mean shift algorithm, dynamic supervoxelization is performed based on the local density and geometric information of the local point cloud data after removing interference, generating a voxel space with density and geometric information; wherein the attributes of each voxel in the voxel space include point density, geometric center, and normal vector; A point cloud similarity graph is constructed according to the voxel space, and the normal vector is introduced into the point cloud similarity graph as a new similarity metric; wherein the new similarity metric is defined as: s(x i ,x j ) represents voxel x i With voxel x j A new similarity measure between, i and j represent voxel indices, d xy d z Represents voxel x i With voxel x j The horizontal and vertical distances between them, θ represents the voxel x i With voxel x j The normal vector angle between xy , σ z , σ θ Indicates the scale parameters corresponding to horizontal distance, vertical distance, and normal vector angle; Adopting an adaptive scale parameter setting method, dynamically adjusting the scale parameter according to the local standard deviation of the voxel neighborhood to obtain an adjusted adaptive scale parameter; According to the point cloud similarity graph and the adjusted adaptive scale parameter, a sparse similarity matrix is ​​constructed using a sparse adjacency graph, and a normalized Laplacian matrix is ​​generated according to the sparse similarity matrix; wherein, L represents the normalized Laplace matrix, D represents the degree matrix, and W represents the sparse similarity matrix; Performing eigenvalue decomposition on the normalized Laplace matrix to obtain eigenvectors corresponding to the first K1 eigenvalues ​​with the smallest eigenvalues, and mapping the K1 eigenvectors to a low-dimensional feature space, and then performing clustering using the K-Means algorithm to obtain an initial clustering result; wherein K1 is a positive integer; The initial clustering results are subjected to dynamic post-processing of merging small clusters, decomposing large clusters, and removing environmental point clouds to obtain processed clustering results, and the main point cloud cluster to which the target aircraft belongs is screened from the processed clustering results to obtain target local point cloud data corresponding to the target aircraft.

5. The AGV navigation method for aircraft maintenance sites according to claim 4, characterized in that: The adaptive scale parameter setting method is used to dynamically adjust the scale parameter by using the local standard deviation of the voxel neighborhood to obtain the adjusted adaptive scale parameter, specifically including: According to the formula σ k =std({||x i -x j |||x j ∈KNN(x i )})Calculate the adjusted adaptive scale parameter; where σ k represents the adjusted adaptive scale parameter, k is the scale parameter index, std() represents the standard deviation of the calculated voxel neighborhood distance, ||x i -x j || indicates the calculation of voxel x i With voxel x j The distance between them, KNN(x i ) represents the calculation of voxel x i The K2 nearest neighbor point set, x j ∈KNN(x i ) represents voxel x j is voxel x i A domain point in the set of K2 nearest neighboring points, where K2 is a positive integer.

6. The AGV navigation method for aircraft maintenance sites according to claim 1, characterized in that: The whole machine three-dimensional point cloud database is pre-established by the following steps: Utilize multiple laser radars on the AGV platform to collect multi-perspective point cloud data of key structural components of various typical aircraft models; wherein the key structural components of the aircraft include at least the fuselage, wings, tail and landing gear; For each typical aircraft model, the collected multi-view multi-frame point cloud data is denoised and spliced ​​to generate a corresponding whole-machine point cloud model for each typical aircraft model; A unified whole-aircraft point cloud coordinate system is established for the whole-aircraft point cloud model corresponding to each typical aircraft model; wherein, the whole-aircraft point cloud coordinate system takes the preset reference point of the aircraft body as the origin, the direction along the fuselage pointing to the tail as the x-axis, the direction perpendicular to the fuselage and pointing to the left wing as the y-axis, and the direction perpendicular to the fuselage and pointing upward as the z-axis.

7. The AGV navigation method for aircraft maintenance sites according to claim 1, characterized in that: The NDT algorithm is used to register the target local point cloud data with the preset whole-machine 3D point cloud database, obtain the target whole-machine point cloud model corresponding to the target aircraft and its rigid body transformation matrix relative to the AGV body coordinate system, and obtain the spatial posture of the AGV platform in the whole-machine point cloud coordinate system according to the rigid body transformation matrix to achieve relative positioning of the AGV platform, specifically including: The NDT algorithm is used to sequentially align and compare the target local point cloud data with each whole aircraft point cloud model in a preset whole aircraft three-dimensional point cloud database, and the whole aircraft point cloud model with the smallest fitting error is selected as the target whole aircraft point cloud model corresponding to the target aircraft; Perform spatial registration on the target local point cloud data and the target whole-machine point cloud model, construct a geometric constraint relationship between the target aircraft and the target whole-machine point cloud model in three-dimensional space through a probability distribution model, and output a rigid body transformation matrix of the target whole-machine point cloud model relative to the AGV body coordinate system; An inverse operation is performed on the rigid body transformation matrix to convert the known posture of the AGV platform in the AGV body coordinate system to the whole aircraft point cloud coordinate system, thereby obtaining the spatial posture of the AGV platform in the whole aircraft point cloud coordinate system, thereby obtaining the current positioning position of the AGV platform relative to the target aircraft at the maintenance work site.

8. An AGV navigation system for aircraft maintenance sites, characterized in that: include: The environmental point cloud data acquisition module is used to acquire integrated point cloud data covering the surrounding environment at the maintenance site of the target aircraft using the AGV platform; a local point cloud data acquisition module, configured to use the AGV platform to collect original local point cloud data of the maintenance operation site, eliminate environmental interference point cloud data from the integrated point cloud data, and process the local point cloud data after interference elimination using an adaptive point cloud clustering algorithm based on density and spatial distribution characteristics to obtain target local point cloud data corresponding to the target aircraft; A local point cloud data registration and positioning module is used to register the target local point cloud data with a preset whole-aircraft 3D point cloud database using an NDT algorithm, obtain a target whole-aircraft point cloud model corresponding to the target aircraft and its rigid body transformation matrix relative to the AGV body coordinate system, and obtain the spatial posture of the AGV platform in the whole-aircraft point cloud coordinate system based on the rigid body transformation matrix to achieve relative positioning of the AGV platform; wherein the whole-aircraft 3D point cloud database includes whole-aircraft point cloud models corresponding to multiple typical aircraft models, generated from point clouds of key structural components of the aircraft, and located in a unified whole-aircraft point cloud coordinate system; A navigation map coordinate conversion module is used to call a navigation map template corresponding to the target whole machine point cloud model, and perform coordinate conversion on the path node coordinates in the navigation map template according to the rigid body transformation matrix; The autonomous navigation path planning module is used to automatically generate the optimal navigation path based on the navigation map template after coordinate conversion, with the current positioning position of the AGV platform as the starting point and the maintenance operation point as the target point, and control the AGV platform to autonomously navigate along the optimal navigation path.

9. A computer-readable storage medium, characterized in that The invention comprises a stored computer program, which controls the device where the computer-readable storage medium is located to execute the AGV navigation method for aircraft maintenance sites according to any one of claims 1 to 7 when the computer program is run.

10. A terminal device, characterized in that: The system comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the AGV navigation method for an aircraft maintenance site according to any one of claims 1 to 7 is implemented.

Images