An unmanned cluster coordinate system automatic alignment method and system based on point cloud registration

Abstract

The application discloses an unmanned cluster coordinate system automatic alignment method and system based on point cloud registration, and the method comprises the following steps: acquiring initial pose information of an unmanned cluster system and constructing point cloud data of the unmanned cluster; constructing a global coordinate system based on the initial pose information of the unmanned cluster system, and combining the point cloud data of the unmanned cluster to perform point cloud registration calculation, so as to obtain a fine registration transformation matrix of source point cloud features to the global coordinate system; and based on the fine registration transformation matrix of the source point cloud features to the global coordinate system, performing coordinate fusion of all unmanned systems, so as to obtain an unmanned cluster coordinate system automatic alignment result. The application can reduce control errors among multiple unmanned systems due to inconsistent coordinates, thereby improving the overall collaborative efficiency of the system. The application can be widely applied to the technical field of space coordinate alignment as an unmanned cluster coordinate system automatic alignment method and system based on point cloud registration.

Description

Technical Field

[0001] This invention relates to the field of spatial coordinate alignment technology, and in particular to an automatic alignment method and system for unmanned cluster coordinate systems based on point cloud registration. Background Technology

[0002] In unmanned swarm collaborative tasks, each unmanned system typically relies on onboard sensors (such as visual cameras or lidar) to complete environmental perception and pose estimation in its own local coordinate system. Due to the influence of initialization conditions, sensor installation, and positioning errors, the coordinate systems of different unmanned systems often have unknown rigid body transformation relationships, making it difficult to directly fuse multi-source data.

[0003] Currently, research on this issue mainly follows three technical approaches: One method is a coordinate unification method based on an external positioning system: In engineering practice, a common approach is to use an external high-precision positioning system (such as RTK-GNSS or optical motion capture systems) to provide a unified global coordinate reference for all unmanned systems. In this method, an external device determines a reference coordinate system, and each unmanned system outputs its pose information directly within the same coordinate system, thus avoiding coordinate alignment issues. This method is simple to implement and highly accurate, but its drawbacks are also obvious: firstly, it is highly dependent on external infrastructure and difficult to deploy outdoors or in complex environments; secondly, its applicability and flexibility are limited when the system scales up or the operating area is restricted.

[0004] Second, a vision-based multi-robot alignment method: These methods, based on visual odometry (VO), visual inertial odometry (VIO), or simultaneous localization and mapping (SLAM) technologies, achieve coordinate alignment through information sharing among multiple unmanned systems. The basic idea is that each unmanned system constructs its own local map or trajectory, and then spatial alignment between the systems is achieved through methods such as loop closure detection or graph optimization. While these methods do not rely on external devices, they are sensitive to initial pose and data association. Furthermore, these methods typically focus on alignment at the map or trajectory level, lacking direct support for building a unified coordinate reference at the system level to serve real-time control.

[0005] Thirdly, spatial alignment methods based on point cloud registration: This type of method estimates the spatial transformation relationship between point cloud data collected from different perspectives by matching them. Typical methods include a combination of feature-matching-based global registration and optimization-based local registration. This approach can directly estimate spatial transformation relationships using geometric information without the need for an external positioning system, showing good application potential. However, most existing point cloud registration methods are designed for registration problems between a single device or two frames of point clouds, and the results are usually relative transformation relationships. In unmanned swarm applications, there is still room for improvement in how to construct a unified reference coordinate system based on this method and consistently fuse multiple registration results. Furthermore, existing research focuses primarily on the registration algorithm itself, with less attention paid to the organization, management, and real-time application of registration results in unmanned swarms. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide an automatic alignment method and system for unmanned cluster coordinate systems based on point cloud registration, which can reduce control errors caused by coordinate inconsistencies between multiple unmanned systems, thereby improving the overall collaborative efficiency of the system.

[0007] The first technical solution adopted in this invention is: an automatic alignment method for the coordinate system of an unmanned cluster based on point cloud registration, comprising the following steps: Acquire the initial pose information of the unmanned swarm system and construct the unmanned swarm point cloud data; A global coordinate system is constructed based on the initial pose information of the unmanned swarm system, and point cloud registration calculation is performed in combination with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. Based on the fine registration transformation matrix from source point cloud features to the global coordinate system, coordinate fusion is performed on all unmanned systems to obtain the automatic alignment result of the unmanned cluster coordinate system.

[0008] Furthermore, the step of acquiring the initial pose information of the unmanned swarm system and constructing the unmanned swarm point cloud data specifically includes: The unmanned cluster is set to include several unmanned systems. The initial pose information of each unmanned system is determined, and one of the unmanned systems is selected as the reference unmanned system. Based on each unmanned system, environmental point cloud data is collected to construct unmanned cluster point cloud data.

[0009] Furthermore, the step of constructing a global coordinate system based on the initial pose information of the unmanned swarm system and performing point cloud registration calculations using the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system specifically includes: Using the selected reference unmanned system as a benchmark, a homogeneous transformation matrix is ​​constructed based on the initial pose information of each unmanned system. The local coordinate system of the reference unmanned system is then mapped to the global coordinate system to build the global coordinate system. The unmanned cluster point cloud data is preprocessed and feature extracted sequentially to obtain a feature descriptor with local geometric invariance. The feature descriptor includes target point cloud features and source point cloud features. Based on the global coordinate system, global coarse registration is performed on the target point cloud features and the source point cloud features to obtain the initial transformation matrix of coarse registration from the source point cloud features to the target point cloud features.

[0010] Based on the initial transformation matrix of coarse registration from source point cloud features to target point cloud features, local fine registration is performed to obtain the fine registration transformation matrix from source point cloud features to the global coordinate system.

[0011] Furthermore, the step of performing global coarse registration of the target point cloud features and the source point cloud features based on the global coordinate system to obtain the initial transformation matrix of coarse registration from the source point cloud features to the target point cloud features specifically includes: Based on the global coordinate system, the target point cloud features are matched with the source point cloud features to construct an initial correspondence between the target point cloud features and the source point cloud features. The correspondence between the initial target point cloud features and the source point cloud features is filtered by a random sampling consistency strategy, and inlier matching pairs that satisfy geometric constraints are selected. Based on the selected inlier matching pairs, the coarse registration initial transformation matrix of the target point cloud features is obtained by solving the problem with the goal of maximizing the consistency of inlier matching.

[0012] Furthermore, the step of performing local fine registration based on the initial coarse registration transformation matrix from source point cloud features to target point cloud features to obtain the fine registration transformation matrix from source point cloud features to the global coordinate system specifically includes: The initial transformation matrix of the coarse registration of the target point cloud features is used as the initial pose, and the correspondence between the nearest neighbor points of the source point cloud features and the target point cloud features is established iteratively.

[0013] Based on the nearest neighbor correspondence between the source point cloud features and the target point cloud features, the distance error between corresponding point pairs is minimized as the objective for iterative optimization until the preset requirements are met, thus obtaining the fine registration transformation matrix from the source point cloud features to the global coordinate system.

[0014] Furthermore, the step of performing coordinate fusion of all unmanned systems based on the fine registration transformation matrix from source point cloud features to the global coordinate system to obtain the automatic alignment result of the unmanned cluster coordinate system specifically includes: The source point cloud features are combined with the fine registration transformation matrix to the global coordinate system, and matrix operations are performed on the global coordinate system. Rotation fusion and translation fusion are performed respectively to finally generate the complete transformation matrix of the current unmanned system in the global coordinate system. Based on the complete transformation matrix of the current unmanned system in the global coordinate system, global coarse registration, local fine registration and matrix operation are performed on all unmanned systems in sequence to construct a unified transformation matrix for the unmanned cluster. The effectiveness of the registration results of the unified transformation matrix of the unmanned swarm is evaluated by the matching rate and root mean square error, and an effective unified transformation matrix of the unmanned swarm is obtained. The effective unmanned swarm unified transformation matrix is ​​converted into a standard pose expression form to obtain the automatic alignment result of the unmanned swarm coordinate system. The standard pose expression form includes position vector and attitude information.

[0015] Furthermore, it also includes setting registration termination conditions, scheduling the registration process by periodic triggering or on-demand triggering, stopping the calculation when the termination conditions are met, and re-triggering registration in a dynamic environment.

[0016] The second technical solution adopted in this invention is: an automatic alignment system for an unmanned cluster coordinate system based on point cloud registration, comprising: The first module is used to obtain the initial pose information of the unmanned swarm system and to construct the unmanned swarm point cloud data. The second module is used to construct a global coordinate system based on the initial pose information of the unmanned swarm system, and to perform point cloud registration calculations in combination with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. The third module is used to perform coordinate fusion of all unmanned systems based on the fine registration transformation matrix from source point cloud features to the global coordinate system, and obtain the automatic alignment result of the unmanned cluster coordinate system.

[0017] The beneficial effects of the method and system of the present invention are as follows: The present invention obtains the initial pose information of the unmanned swarm system and constructs the unmanned swarm point cloud data, and realizes coordinate system alignment by using the environmental point cloud data autonomously collected by the unmanned swarm system, so as to complete the spatial relationship estimation between multiple unmanned systems without relying on external positioning equipment. By constructing a reference unmanned system coordinate system and performing unified mapping based on point cloud registration results, a unified coordinate system can be established in unknown environments. Furthermore, a global coordinate system is constructed based on the initial pose information of the unmanned swarm system, and point cloud registration calculations are performed using the unmanned swarm point cloud data to obtain a fine registration transformation matrix from the source point cloud features to the global coordinate system. A two-stage processing approach of "global coarse registration + local fine registration" is adopted. Global registration is used to obtain the initial spatial transformation under conditions of large pose deviations, while local fine registration further optimizes the registration results. Introducing the initial pose information of the unmanned swarm system as an auxiliary constraint during the registration process helps improve the registration convergence stability, thereby enhancing the system's adaptability in complex environments. Finally, based on the fine registration transformation matrix from the source point cloud features to the global coordinate system, coordinate fusion is performed on all unmanned systems to obtain an automatic alignment result of the unmanned swarm coordinate system. A global coordinate benchmark is constructed using the reference unmanned system, and the registration results of each unmanned system relative to the reference unmanned system are uniformly mapped to the same coordinate system using a coordinate transformation fusion method, achieving a consistent expression of the unmanned swarm pose in a unified coordinate system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the steps of an automatic alignment method for an unmanned cluster coordinate system based on point cloud registration according to the present invention. Figure 2 This is a structural block diagram of an automatic alignment system for an unmanned cluster coordinate system based on point cloud registration, according to the present invention. Figure 3 This is a schematic diagram of the overall process for automatic alignment of the coordinate system of an unmanned cluster provided in a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the registration calculation process using a drone swarm as an example, provided in a specific embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the placement and display of drones according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram of the initial coordinate system of a UAV in Rviz provided in a specific embodiment of the present invention; Figure 7 This is a schematic diagram of point cloud data display in Rviz provided in a specific embodiment of the present invention; Figure 8 This is a schematic diagram of point cloud data display of a reference drone in Rviz provided in a specific embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0020] First, it should be noted that the purpose of the embodiments of the present invention is: 1) Reduce reliance on external positioning devices. The aim is to provide a coordinate alignment method that does not require external infrastructure such as RTK or optical capture, enabling unmanned swarms to achieve coordinate unification even indoors or in complex environments, thereby improving the flexibility and applicability of system deployment.

[0021] 2) Improve the autonomy and robustness of coordinate alignment in unmanned clusters. By introducing point cloud registration and multi-stage optimization mechanisms, the system can achieve stable alignment even with initial errors or environmental differences, reducing reliance on manual intervention and precise initial conditions.

[0022] 3) Construct a pose representation method for multiple unmanned systems under a unified global coordinate system. This aims to solve the problem that existing methods can only obtain relative poses. By establishing a unified reference coordinate system, the poses of multiple unmanned systems can be expressed under the same coordinate framework, thereby supporting collaborative perception and control.

[0023] 4) Implement a complete coordinate alignment process for engineering applications. Provide a complete methodology from data acquisition and registration calculation to result publication, enabling coordinate alignment results to directly serve the operation of actual systems and improving the engineering feasibility and practical value of the method.

[0024] Based on this, such as Figure 3 As shown, in the data acquisition stage of this embodiment, each unmanned system acquires 3D point cloud data of the environment through airborne sensors, and combines it with the initial pose information provided during system initialization as auxiliary input. In the registration calculation stage, taking the reference unmanned system as the benchmark, its point cloud is used as the target point cloud, and the point clouds of the other intelligent agents are used as the source point clouds. The correspondence between the point clouds is established through feature extraction and matching, and global registration and local fine registration are performed sequentially to obtain the spatial transformation relationship of each unmanned system relative to the reference intelligent agent. In the coordinate fusion stage, the above relative transformations are combined with the pose information of the reference unmanned system for unified mapping to construct the pose expression of multiple unmanned systems in a unified coordinate system. In the result output stage, the validity of the registration results is evaluated, and the pose information in the final unified coordinate system is output for subsequent system applications.

[0025] Reference Figure 1 This invention provides an automatic alignment method for the coordinate system of an unmanned cluster based on point cloud registration. The method includes the following steps: S100: Obtain the initial pose information of the unmanned swarm system and construct the unmanned swarm point cloud data; S110. Set the unmanned cluster to include several unmanned systems, determine the initial pose information of each unmanned system, and select one of the unmanned systems as the reference unmanned system; In this embodiment, system initialization and reference unmanned system determination are performed. Assume the unmanned cluster contains N unmanned systems, and the initial pose of each unmanned system is represented as: in: We select one of the unmanned systems as the reference unmanned system, denoted as the i.e., the nth unmanned system. Each unmanned system has an initial pose, which can be derived from the onboard state estimation module or a preset configuration, to provide a rough spatial relationship between the unmanned systems. The selection of the reference unmanned system can be fixed or dynamically determined based on point cloud quality, environmental observability, or motion stability, thereby improving the reliability of subsequent registration. The main purpose of this step is to establish a reference benchmark for a unified coordinate system.

[0026] S120: Based on each unmanned system, collect environmental point cloud data and construct unmanned cluster point cloud data.

[0027] In this embodiment, unmanned cluster point cloud data acquisition is performed. Each unmanned system collects environmental point cloud data, the first... The point cloud set of an unmanned system is denoted as: In practical implementation, point cloud data can be acquired through airborne depth sensors and can be cached or time-aligned as needed to reduce the impact of time differences between multi-source data on the registration results. Furthermore, when the system has high real-time requirements, asynchronous point clouds can be used directly for subsequent calculations. This step provides the necessary spatial geometric information foundation for the entire method.

[0028] S200. Construct a global coordinate system based on the initial pose information of the unmanned swarm system, and perform point cloud registration calculations in conjunction with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. S210. Using the selected reference unmanned system as a benchmark, construct a homogeneous transformation matrix based on the initial pose information of each unmanned system, map the local coordinate system of the reference unmanned system to the global coordinate system, and construct the global coordinate system. In this embodiment, a reference coordinate system is constructed. A global coordinate system is constructed based on the reference unmanned system, and its initial transformation is as follows: in, This is a preset rotation. In practical applications, this preset rotation can be used to unify coordinate systems of different sensors or to meet mission requirements, thereby improving system consistency. Through this step, the local coordinate system of the reference unmanned system can be mapped to the global coordinate system, providing a unified reference for coordinate transformations of all subsequent unmanned systems.

[0029] S220. Perform data preprocessing and feature extraction processing on the unmanned cluster point cloud data in sequence to obtain a feature descriptor with local geometric invariance. The feature descriptor includes target point cloud features and source point cloud features. In this embodiment, point cloud preprocessing and feature extraction are performed. The point cloud is downsampled and its normals are estimated, and features are extracted: in The feature extraction operator is represented. Specifically, the preprocessing process may include operations such as noise point removal, uniform downsampling, and normal direction estimation to reduce the point cloud size and improve data quality. Based on this, feature descriptors with local geometric invariance are extracted for subsequent point cloud matching. The form of the features is not limited, thus ensuring that the method has good adaptability and scalability.

[0030] S230. Based on the global coordinate system, perform global coarse registration between the target point cloud features and the source point cloud features to obtain the initial transformation matrix of coarse registration from the source point cloud features to the target point cloud features. Specifically, based on the global coordinate system, the target point cloud features are matched with the source point cloud features to construct an initial correspondence between the target point cloud features and the source point cloud features. The initial correspondence between the target point cloud features and the source point cloud features is filtered through a random sampling consistency strategy to select interior point matching pairs that satisfy geometric constraints. Based on the filtered interior point matching pairs, the initial transformation matrix of the coarse registration of the target point cloud features is obtained by solving the problem with the goal of maximizing the consistency of interior point matching.

[0031] In this embodiment, global coarse registration is performed. Based on feature matching, an initial transformation from the source point cloud to the target point cloud is estimated. This transformation is obtained by maximizing feature matching consistency. By establishing a set of feature correspondences and employing a random sampling consistency strategy to select matching pairs that satisfy geometric consistency, an initial rigid body transformation is estimated. This step enables coarse alignment of the point cloud even with large initial pose errors, providing reasonable initial values ​​for subsequent fine registration and thus avoiding getting trapped in local optima.

[0032] S240. Based on the initial transformation matrix of coarse registration from source point cloud features to target point cloud features, perform local fine registration to obtain the fine registration transformation matrix from source point cloud features to the global coordinate system.

[0033] Specifically, the initial transformation matrix of the coarse registration of the target point cloud features is used as the initial pose, and the correspondence between the nearest neighbors of the source point cloud features and the target point cloud features is established iteratively. Based on the correspondence between the nearest neighbors of the source point cloud features and the target point cloud features, the distance error of the corresponding point pairs is minimized as the target for iterative optimization until the preset requirements are met, and the fine registration transformation matrix from the source point cloud features to the global coordinate system is obtained.

[0034] In this embodiment, local fine registration is performed. Based on the coarse registration, the transformation parameters are further optimized: By continuously establishing nearest neighbor correspondences and iteratively updating transformation parameters, the error function gradually converges. Compared to coarse registration, this step significantly improves registration accuracy, enabling more precise alignment of the two point clouds in their local structures.

[0035] S300: Based on the source point cloud features to the fine registration transformation matrix of the global coordinate system, coordinate fusion of all unmanned systems is performed to obtain the automatic alignment result of the unmanned cluster coordinate system.

[0036] S310. Combine the source point cloud features with the fine registration transformation matrix of the global coordinate system and perform matrix operations with the global coordinate system. Perform rotation fusion and translation fusion respectively to finally generate the complete transformation matrix of the current unmanned system in the global coordinate system. In this embodiment, coordinate transformation fusion is performed. The relative transformation is mapped to the global coordinate system, and the rotation fusion can be expressed as: Similarly, translation fusion is represented as: The final transformation is: This step integrates the relative pose obtained from point cloud registration with the global pose of the reference unmanned system, achieving coordinate system transformation and unification. Through the combined rotation and translation operations described above, the local coordinates of each unmanned system can be mapped to a unified global coordinate system, which is a crucial step in realizing unmanned swarm collaboration.

[0037] S320. Based on the complete transformation matrix of the current unmanned system in the global coordinate system, perform global coarse registration, local fine registration and matrix operation processing on all unmanned systems in sequence to construct a unified transformation matrix for the unmanned cluster. In this embodiment, a unified coordinate system for the unmanned cluster is constructed. For all... The unmanned system repeatedly performs the above registration and fusion process. During this process, the pose of the unmanned system is uniformly expressed in the same coordinate system, thereby realizing the consistent representation of multi-node spatial information and providing a foundation for subsequent collaborative tasks.

[0038] S330. The effectiveness of the registration result of the unified transformation matrix of the unmanned swarm is evaluated by the matching rate and root mean square error, and an effective unified transformation matrix of the unmanned swarm is obtained. In this embodiment, the validity of the registration results is evaluated. This is based on the matching rate: Root mean square error: This is used to evaluate the registration quality and determine whether the registration result is reliable. When the registration quality does not meet the requirements, the result can be marked as invalid, triggering re-registration or the use of a spare pose, thereby avoiding error propagation and improving the overall robustness of the system.

[0039] S340. Convert the effective unmanned swarm unified transformation matrix into a standard pose expression form to obtain the automatic alignment result of the unmanned swarm coordinate system. The standard pose expression form includes position vector and attitude information.

[0040] In this embodiment, pose output is performed. The final transformation is then performed. The data is converted into a standard pose representation, including position vectors and attitude information (such as rotation matrices or quaternions), and then output to an external system. This output can be used in applications such as unmanned swarm cooperative control, path planning, or environmental mapping, and serves as a direct input for realizing system functions.

[0041] Furthermore, it should be noted that the embodiments of the present invention also include setting registration termination conditions, scheduling the registration process in a periodic or on-demand manner, stopping the calculation when the termination conditions are met, and re-triggering registration in a dynamic environment.

[0042] In this embodiment, an execution control and termination mechanism is implemented. Registration execution conditions are defined: In actual operation, the registration process is scheduled through a control module, and can be executed using periodic or on-demand triggering. When the termination condition is met, the current registration process stops to avoid unnecessary computational overhead. This mechanism also supports re-triggering registration in dynamic environments to maintain coordinate consistency.

[0043] In summary, the embodiments of the present invention have the following distinguishing technical features compared with the prior art: 1) By selecting a reference unmanned system and constructing a global reference coordinate system based on its initial pose, the pose relationships between multiple unmanned systems, which were originally only relative, can be uniformly mapped to the same coordinate frame. This mechanism avoids the problem that traditional point cloud registration only outputs relative transformations and is difficult to use directly for system-level applications.

[0044] 2) To address the characteristic that point cloud registration typically outputs relative poses, this embodiment of the invention designs a transformation fusion method that combines relative transformations with a reference coordinate system to obtain the absolute poses of each unmanned system in a unified coordinate system. This method ensures the consistency of multi-source registration results, enabling them to be directly used for collaborative control and information fusion.

[0045] 3) An initial transformation is obtained using a feature-matching-based global registration method, followed by fine-tuning through iterative optimization. This approach balances robustness under large-scale errors with accuracy in local alignment. Compared to single registration methods, this strategy exhibits more stable performance in complex environments and under conditions of large initial errors.

[0046] 4) This invention not only focuses on the registration algorithm itself, but also constructs a complete process from point cloud acquisition, registration calculation, result evaluation to coordinate publication, realizing a closed-loop engineering process for coordinate alignment of unmanned clusters. This method can be directly embedded into actual systems and has good real-time performance and scalability.

[0047] Therefore, the embodiments of the present invention have the following advantages compared with the prior art: 1) Reduce reliance on external positioning infrastructure and improve system deployment flexibility: Existing methods for unifying coordinates in unmanned clusters typically rely on external positioning devices such as RTK, motion capture systems, or fixed base stations to achieve unified positioning. While these methods can achieve high positioning accuracy, they require additional infrastructure deployment, resulting in high system setup costs, and their applicability is limited in indoor, underground, or obstructed environments.

[0048] This invention utilizes environmental point cloud data autonomously collected by unmanned systems to achieve coordinate system alignment, enabling spatial relationship estimation between multiple unmanned systems without relying on external positioning equipment. By constructing a reference unmanned system coordinate system and combining it with point cloud registration results for unified mapping, a unified coordinate system can be established in unknown environments. Therefore, this invention improves the system's applicability in complex environments, reduces the need for external equipment deployment, lowers the overall system cost, and increases the flexibility of engineering deployment.

[0049] 2) Achieve unified coordinate representation for unmanned clusters to improve the consistency of collaborative tasks: Existing point cloud registration methods mostly only obtain the relative transformation relationship between two sets of point clouds, making it difficult to directly achieve a consistent representation of unmanned swarms in a unified coordinate system. As the number of unmanned systems increases, the problem of difficulty in unifying multiple sets of relative coordinates easily arises.

[0050] This invention constructs a global coordinate reference by referencing an unmanned system, and utilizes a coordinate transformation and fusion method to uniformly map the registration results of each unmanned system relative to the reference unmanned system to the same coordinate system. The coordinate fusion relationship is as follows: The above method enables a consistent representation of the pose of unmanned swarms in a unified coordinate system. Compared to traditional methods that only output local relative poses, this invention is more suitable for application scenarios requiring global consistency, such as formation movement, collaborative perception, and multi-machine map fusion.

[0051] 3) Improve the robustness and accuracy of point cloud registration: Existing local point cloud registration methods typically rely on a good initial pose. When the initial error is large, they are prone to getting trapped in local optima, which can lead to registration failure.

[0052] This invention employs a two-stage processing approach: global coarse registration and local fine registration. Global registration is used to obtain the initial spatial transformation under conditions of large pose deviations, while local fine registration further optimizes the registration results. The local optimization objective can be expressed as: The above method can reduce the impact of initial errors on the registration results and improve the alignment accuracy between point clouds. Furthermore, this invention introduces the initial pose information of the unmanned system as an auxiliary constraint during the registration process, which helps improve the registration convergence stability and thus enhances the system's adaptability in complex environments.

[0053] 4) Improve system scalability and engineering adaptability: Existing unmanned swarms often rely on fixed software frameworks, sensor types, or specific registration algorithms, so the system structure needs to be redesigned when changing hardware platforms or application scenarios.

[0054] This invention does not limit the form of point cloud features, the global registration method, or the local optimization method, and can be replaced or extended according to actual needs. For example, the feature extraction part can use different types of local geometric descriptors, and the global registration part can also use different matching strategies. Therefore, this invention can be adapted to different types of unmanned platforms and sensor systems, and has good versatility and engineering scalability.

[0055] 5) Facilitates the implementation of real-time collaborative tasks in unmanned clusters: Because this invention employs a modular processing flow and supports periodic or on-demand coordinate update mechanisms, it can continuously update the spatial relationships between multiple unmanned systems in a dynamic environment.

[0056] Compared to methods applicable only to offline mapping or static registration, this invention is more suitable for tasks such as real-time collaborative control, dynamic formation movement, collaborative path planning, and environmental perception. By establishing a unified coordinate system, control errors caused by coordinate inconsistencies between multiple unmanned systems can be reduced, thereby improving the overall collaborative efficiency of the system.

[0057] Finally, the embodiments of the present invention will be explained and described in conjunction with the accompanying drawings: This embodiment uses an unmanned swarm system consisting of three UAVs as an example to illustrate the automatic coordinate system alignment method for unmanned swarms proposed in this invention. Each UAV is equipped with radar to collect three-dimensional point cloud data of the environment, and the point cloud registration and coordinate fusion calculation are completed through a ground station.

[0058] like Figure 5 As shown, three drones were placed at different locations within the same experimental environment, exhibiting significant differences in position and attitude among them. Since each drone uses its own body coordinate system as a local reference coordinate system, spatial information cannot be directly and uniformly expressed between the different drones. In this embodiment, each drone independently completes point cloud acquisition and pose initialization at the beginning of the experiment, and then sends the relevant data to the ground station processing module via a communication link.

[0059] like Figure 6 As shown, during the system initialization phase, each UAV establishes its own local coordinate system. Due to the different initial orientations of the UAVs, their coordinate axis directions also differ, with each of the three UAVs corresponding to a local coordinate system. In the RViz visualization interface, it can be observed that the coordinate axis directions of the different UAVs are not consistent; therefore, their point cloud data are all located under different local reference systems, making it impossible to directly complete map fusion and collaborative localization.

[0060] like Figure 7 As shown, each UAV collects point clouds of the experimental environment and generates corresponding environmental point clouds in its own local coordinate system. Due to spatial position deviations and coordinate direction differences between different UAVs, the multiple point clouds are significantly dispersed in space, making it impossible to form a unified environmental representation.

[0061] Select the drone located in the middle as the reference drone, and use its local coordinate system as the global reference coordinate system. The point cloud data of the reference drone is as follows: Figure 8As shown. All other UAVs establish spatial transformation relationships with the reference UAV through point cloud registration. The ground station performs registration calculations on the point clouds collected by different UAVs, such as... Figure 4 As shown, the global pose of the reference UAV is combined with the relative transformation obtained from point cloud registration to obtain the final pose of each UAV in a unified coordinate system. The ground station terminal outputs the spatial position and rotation matrix results of the other UAVs relative to the reference UAV.

[0062] The results above show that this embodiment can effectively calculate the spatial transformation relationship between different UAVs and realize a unified coordinate expression for multiple UAVs.

[0063] As can be seen from this embodiment, the present invention can achieve automatic coordinate system alignment using point cloud data autonomously collected by multiple UAVs without the need for external positioning infrastructure, thereby obtaining a consistent expression of the spatial information of multiple UAVs, and thus providing a unified spatial foundation for subsequent tasks such as collaborative mapping, formation control and path planning.

[0064] Reference Figure 2 An automatic alignment system for unmanned cluster coordinate systems based on point cloud registration, comprising: The first module 201 is used to acquire the initial pose information of the unmanned swarm system and construct the unmanned swarm point cloud data. The second module 202 is used to construct a global coordinate system based on the initial pose information of the unmanned swarm system, and to perform point cloud registration calculations in combination with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. The third module 203 is used to perform coordinate fusion of all unmanned cluster systems based on the fine registration transformation matrix from source point cloud features to the global coordinate system, and obtain the automatic alignment result of the unmanned cluster coordinate system.

[0065] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0066] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this is not intended to limit the scope of the embodiments of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A method for automatic alignment of coordinate systems in unmanned clusters based on point cloud registration, characterized in that, Includes the following steps: Acquire the initial pose information of the unmanned swarm system and construct the unmanned swarm point cloud data; A global coordinate system is constructed based on the initial pose information of the unmanned swarm system, and point cloud registration calculation is performed in combination with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. Based on the fine registration transformation matrix from source point cloud features to the global coordinate system, coordinate fusion is performed on all unmanned swarm systems to obtain the automatic alignment result of the unmanned swarm coordinate system.

2. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 1, characterized in that, The steps of obtaining the initial pose information of the unmanned swarm system and constructing the unmanned swarm point cloud data specifically include: The unmanned cluster system is set to include several unmanned systems. The initial pose information of each unmanned system is determined, and one of the unmanned systems is selected as the reference unmanned system. Based on each unmanned cluster system, environmental point cloud data is collected to construct unmanned cluster point cloud data.

3. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 2, characterized in that, The step of constructing a global coordinate system based on the initial pose information of the unmanned swarm system and performing point cloud registration calculations using the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system specifically includes: Using the selected reference unmanned swarm system as a benchmark, a homogeneous transformation matrix is ​​constructed based on the initial pose information of each unmanned swarm system. The local coordinate system of the reference unmanned system is then mapped to the global coordinate system, thus constructing the global coordinate system. The unmanned cluster point cloud data is preprocessed and feature extracted sequentially to obtain a feature descriptor with local geometric invariance. The feature descriptor includes target point cloud features and source point cloud features. Based on the global coordinate system, global coarse registration is performed on the target point cloud features and the source point cloud features to obtain the initial transformation matrix of coarse registration from the source point cloud features to the target point cloud features. Based on the initial coarse registration transformation matrix from source point cloud features to target point cloud features, local fine registration is performed to obtain the fine registration transformation matrix from source point cloud features to the global coordinate system.

4. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 3, characterized in that, The step of performing global coarse registration of the target point cloud features and the source point cloud features based on the global coordinate system to obtain the initial transformation matrix of coarse registration from the source point cloud features to the target point cloud features specifically includes: Based on the global coordinate system, the target point cloud features are matched with the source point cloud features to construct an initial correspondence between the target point cloud features and the source point cloud features. The correspondence between the initial target point cloud features and the source point cloud features is filtered by a random sampling consistency strategy, and inlier matching pairs that satisfy geometric constraints are selected. Based on the selected inlier matching pairs, the coarse registration initial transformation matrix of the target point cloud features is obtained by solving the problem with the goal of maximizing the consistency of inlier matching.

5. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 4, characterized in that, The step of performing local fine registration based on the initial coarse registration transformation matrix from source point cloud features to target point cloud features to obtain the fine registration transformation matrix from source point cloud features to the global coordinate system specifically includes: The initial transformation matrix of coarse registration of the target point cloud features is used as the initial pose, and the correspondence between the nearest neighbor points of the source point cloud features and the target point cloud features is established iteratively. Based on the nearest neighbor correspondence between the source point cloud features and the target point cloud features, the distance error between corresponding point pairs is minimized as the objective for iterative optimization until the preset requirements are met, thus obtaining the fine registration transformation matrix from the source point cloud features to the global coordinate system.

6. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 5, characterized in that, The step of performing coordinate fusion of all unmanned systems based on the fine registration transformation matrix from source point cloud features to the global coordinate system to obtain the automatic alignment result of the unmanned cluster coordinate system specifically includes: The source point cloud features are combined with the fine registration transformation matrix to the global coordinate system, and matrix operations are performed on the global coordinate system. Rotation fusion and translation fusion are performed respectively to finally generate the complete transformation matrix of the current unmanned system in the global coordinate system. Based on the complete transformation matrix of the current unmanned system in the global coordinate system, global coarse registration, local fine registration and matrix operation are performed on all unmanned systems in sequence to construct a unified transformation matrix for the unmanned cluster. The effectiveness of the registration results of the unified transformation matrix of the unmanned swarm is evaluated by the matching rate and root mean square error, and an effective unified transformation matrix of the unmanned swarm is obtained. The effective unmanned swarm unified transformation matrix is ​​converted into a standard pose expression form to obtain the automatic alignment result of the unmanned swarm coordinate system. The standard pose expression form includes position vector and attitude information.

7. The automatic alignment method for unmanned cluster coordinate systems based on point cloud registration according to claim 6, characterized in that, It also includes setting registration termination conditions, scheduling the registration process by periodic triggering or on-demand triggering, stopping calculation when the termination conditions are met, and re-triggering registration in a dynamic environment.

8. An automatic alignment system for an unmanned cluster coordinate system based on point cloud registration, characterized in that, Includes the following modules: The first module is used to obtain the initial pose information of the unmanned swarm system and to construct the unmanned swarm point cloud data. The second module is used to construct a global coordinate system based on the initial pose information of the unmanned swarm system, and to perform point cloud registration calculations in combination with the unmanned swarm point cloud data to obtain the fine registration transformation matrix from the source point cloud features to the global coordinate system. The third module is used to perform coordinate fusion of all unmanned swarm systems based on the fine registration transformation matrix from source point cloud features to the global coordinate system, and obtain the automatic alignment result of the unmanned swarm coordinate system.

Images