Multi-vehicle positioning method and system based on vehicle networking communication and cooperative perception

Abstract

The application discloses a multi-vehicle positioning method and system based on vehicle networking communication and cooperative sensing, which comprises the following steps: using a target detection algorithm combining motion compensation and deep learning to correct and calculate the original point cloud data output by the laser radar installed on the vehicle to be positioned, so as to obtain the accurate relative distance between the vehicle to be positioned and each selected anchor vehicle within the predetermined communication range; the vehicle to be positioned and the anchor vehicle are networked vehicles or intelligent networked vehicles; based on the accurate relative distance between the vehicle to be positioned and each selected anchor vehicle, combined with the vehicle center position broadcast by each selected anchor vehicle, a nonlinear distance observation equation is constructed; the nonlinear distance observation equation is solved by using an iterative linearization weighted least square algorithm, so as to obtain the position increment of the vehicle to be positioned which changes in real time, and then the global absolute coordinate of the vehicle to be positioned is calculated. The application can improve the positioning accuracy, robustness and real-time performance in a complex traffic environment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and autonomous driving technology, specifically relating to a multi-vehicle positioning method and system based on vehicle network communication and collaborative perception. Background Technology

[0002] With the rapid evolution of Intelligent Transportation Systems (ITS) and autonomous driving technologies, vehicles are facing stringent requirements for high-precision positioning capabilities in all weather conditions and scenarios. Currently, the Global Navigation Satellite System (GNSS) is the primary means of obtaining absolute position, providing fundamental support for vehicle navigation. However, in complex environments such as urban canyons, tunnels, and underground parking lots, satellite signals are often severely attenuated or even interrupted due to obstruction by tall buildings or multipath effects, making it difficult to meet the centimeter-level positioning accuracy required for autonomous driving.

[0003] To fill the positioning gap left by GNSS failure, existing technologies are mainly divided into two categories: single-vehicle intelligent positioning and multi-vehicle cooperative positioning. Single-vehicle positioning primarily relies on Inertial Measurement Units (IMUs) or Simultaneous Localization and Mapping (SLAM) technologies. However, IMUs suffer from drift errors that accumulate over time, while SLAM technology is costly in terms of map building, maintenance, and adaptability to dynamic environments. Therefore, multi-vehicle cooperative positioning based on Vehicle-to-Everything (V2X) has become a research hotspot. Traditionally, research on cooperative positioning has often focused on data interaction at the communication layer or backend fusion algorithms, typically assuming that the front-end perceived data is ideal or static, while ignoring the error propagation effect between the perception layer and the solution layer.

[0004] In real-world dynamic traffic environments, there is a strong physical coupling between the high-speed movement of vehicles and localization calculations. The scanning mechanism of lidar inevitably introduces motion distortion during vehicle movement, leading to shifts in the observed relative distances and angles. Directly inputting this distorted sensing data into the localization model amplifies the error nonlinearly. Furthermore, cooperative localization depends not only on individual vehicle observations but also on the constraints of multi-vehicle geometric configurations within the network topology. Therefore, in-depth research into the collaborative relationship between front-end sensing motion compensation and back-end geometric constraint calculations is crucial for achieving accurate and robust localization calculations even in decentralized networks where some observations are limited.

[0005] Although some progress has been made in multi-vehicle cooperative localization in recent years, and while the fusion of multi-source information has been considered, most solutions have neglected the heterogeneity of real-world traffic flow and the dynamic noise at the sensing sources. On the one hand, existing solutions typically assume that all traffic participants are intelligent connected and autonomous vehicles (CAVs) with high-precision sensing and communication capabilities, ignoring the current mixed traffic flow where regular vehicles (RVs) and connected vehicles (CVs) coexist, resulting in poor system availability at low penetration rates. On the other hand, localization algorithms neglect the impact of point cloud distortion caused by high-speed motion on ranging accuracy, and traditional least-squares solutions are prone to getting trapped in local optima or diverging results due to noise when dealing with nonlinear geometric constraints, failing to meet the stability requirements of lane-level localization.

[0006] To address the technical challenges in these areas, researchers urgently need to investigate the collaborative relationship between front-end perception and back-end computation, starting from motion distortion compensation at the perception source and the real-world characteristics of mixed traffic flow. Furthermore, they should introduce an iterative optimization algorithm with geometric constraints to construct a multi-vehicle cooperative localization method that can adapt to mixed traffic scenarios and possesses high noise resistance and real-time performance. Summary of the Invention

[0007] To address the aforementioned issues, this invention proposes a multi-vehicle positioning method and system based on vehicle-to-everything (V2X) communication and collaborative perception, which can improve positioning accuracy, robustness, and real-time performance in complex traffic environments.

[0008] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0009] In a first aspect, the present invention provides a multi-vehicle localization method based on vehicle network communication and cooperative perception, applied to vehicles to be located, including:

[0010] Using a target detection algorithm that combines motion compensation and deep learning, the raw point cloud data output by the lidar installed on the vehicle to be located is corrected and calculated to obtain the precise relative distance between the vehicle to be located and each selected anchor point vehicle within a predetermined communication range; wherein, the vehicle to be located and the anchor point vehicles are connected vehicles or intelligent connected vehicles.

[0011] Based on the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a nonlinear distance observation equation is constructed by combining the vehicle center position broadcast by each selected anchor point vehicle.

[0012] The nonlinear distance observation equation is solved by using an iterative linearized weighted least squares algorithm to obtain the real-time position increment of the vehicle to be located;

[0013] Based on the real-time changes in the position increment of the vehicle to be located, the global absolute coordinates of the vehicle to be located are calculated.

[0014] In conjunction with the first aspect, optionally, the method for calculating the precise relative distance between the vehicle to be located and the selected anchor point vehicles includes:

[0015] Motion distortion compensation is performed on the original point cloud data using forward and backward propagation with Kalman filtering to obtain motion-compensated point cloud data;

[0016] The point cloud data is processed using convolutional layers in a pre-defined deep neural network to generate an input feature map;

[0017] Using the channel attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed on the input feature map to obtain pooling results. The pooling results are then input into a multilayer perceptron for feature transformation, and finally, a channel attention feature map is generated using the sigmoid activation function. To enhance the response to vehicle speed-related characteristics;

[0018] By utilizing the spatial attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed along the channel dimension on the input feature map. The two pooling results are concatenated, and then feature fusion is performed through convolution. Finally, a spatial attention feature map is generated using the sigmoid activation function. To achieve focus on the point cloud contour region of the vehicle;

[0019] Channel attention feature map Spatial attention feature map Perform feature fusion processing to generate a feature fusion map;

[0020] The feature fusion map is enhanced to generate an enhanced feature fusion map. The enhanced feature fusion map is then input into the SSD detection head to obtain the three-dimensional bounding box of the vehicle. This three-dimensional bounding box contains the vehicle center coordinates of the vehicle to be located. Based on the vehicle center coordinates of the vehicle to be located and the vehicle center positions broadcast by each selected anchor point vehicle, the precise relative distance between the vehicle to be located and each selected anchor point vehicle is calculated.

[0021] In conjunction with the first aspect, optionally, the method for generating the motion-compensated point cloud data includes:

[0022] Forward propagation is performed based on IMU data from the inertial measurement unit installed on the vehicle to be located to predict the pose change trajectory of the vehicle to be located within the lidar scanning cycle.

[0023] When the lidar installed on the vehicle to be located completes a single-frame scan, the pose change trajectory is corrected based on the original point cloud data output by the lidar to obtain the optimized state at the end of the scan.

[0024] Based on the optimized state, backpropagation is performed to calculate the precise IMU pose at each laser point sampling time.

[0025] Using the precise IMU pose, the original point cloud data at discrete sampling times is projected to the end of the scan using the following mapping relationship to obtain motion-compensated point cloud data:

[0026] ,

[0027] In the formula, These are the coordinates of the point in the lidar coordinate system after motion compensation; It is a constant value that represents the spatial transformation relationship from the lidar coordinate system to the IMU coordinate system; This is the IMU pose transformation matrix obtained through backpropagation. Let be the rotation matrix in the IMU coordinate system relative to the global world coordinate system. Let be the translation vector of the IMU coordinate system relative to the global world coordinate system. To compensate for the point coordinates in the IMU coordinate system before.

[0028] In conjunction with the first aspect, optionally, the method for constructing the nonlinear distance observation equation includes:

[0029] Based on the pose change trajectory of the vehicle to be located within the LiDAR scanning period, let the estimated coordinates of the vehicle to be located be... These represent the locations of the vehicles to be located. , , The estimated coordinate positions in three directions, the first The vehicle center coordinates of each anchor point vehicle are The location is obtained directly from the vehicle to be located through the vehicle-to-everything (V2X) communication network, which consists of the vehicle to be located and the anchor point vehicle.

[0030] Construct the vehicle to be located and the first The nonlinear observation function between the vehicles at each anchor point is expressed as follows: ,in, For vehicles to be determined and the first The relative observation distance of vehicles at each anchor point It is the Euclidean norm;

[0031] Based on the nonlinear observation function and the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a system is constructed to determine the distance between the vehicle to be located and the selected anchor point vehicles. The nonlinear distance observation equation between vehicles at each anchor point is expressed as follows:

[0032] ,

[0033] In the formula, Indicates the vehicle to be located and the first Distance residuals between vehicles at each anchor point Indicates the vehicle to be located and the first The precise relative distance between vehicles at each anchor point.

[0034] In conjunction with the first aspect, optionally, the step of using an iterative linearized weighted least squares algorithm to solve the nonlinear distance observation equation to obtain the real-time changing position increment of the vehicle to be located includes:

[0035] Based on the nonlinear distance observation equation between the vehicle to be located and each selected anchor point vehicle within the predetermined communication range, an observation residual vector is formed. , The number of vehicles at the selected anchor points;

[0036] For the observation residual vector At the current location of the vehicle to be located At the next iteration, a first-order Taylor series expansion is performed, and higher-order terms are ignored to construct a linearized error equation. The expression of the linearized error equation is as follows:

[0037] ,

[0038] In the formula, This represents the position increment of the vehicle to be located. For the first The new information vector of the next iteration, i.e., based on the iterative estimate The calculated residual value; For the observed noise vector; For the nonlinear observation function in the iterative estimate The Jacobian matrix at point A is calculated using the following formula:

[0039] ;

[0040] ;

[0041] Solving the linearized error equation is transformed into a weighted norm minimization problem, specifically including:

[0042] Constructing about location increments The weighted quadratic objective function:

[0043] ,

[0044] In the formula, This is the ranging weight matrix;

[0045] According to the least squares principle, let the weighted quadratic objective function be expressed with respect to the position increment. The derivative is zero, yielding the value of the position increment. Positive definite equation:

[0046] ,

[0047] Solving the positive definite equation by matrix inversion yields the optimal position increment for the current iteration step. Update the vehicle center position according to the following formula:

[0048] ,

[0049] In the formula, For the first In the next iteration, the position increment The optimal estimate; Indicates the first The estimated value in the next iteration step;

[0050] Repeat the above steps until the L2 norm of the position increment is reached. Less than the preset convergence threshold Output the final solved coordinates Used as the global absolute coordinates of the vehicle to be located. .

[0051] In conjunction with the first aspect, optionally, the ranging weight matrix An adaptive update strategy based on IGG-III robust estimation theory is adopted, specifically including:

[0052] Construct a diagonal weight matrix Among them, diagonal elements Based on the vehicle to be located and the first Standardized residuals of the precise relative distance between vehicles at each anchor point Dynamic calculation:

[0053] ,

[0054] In the formula, For the a priori standard deviation of distance measurement, and These are the robust rights preservation threshold and the elimination threshold, respectively.

[0055] Introducing the robust estimation theory of IGG-III, the ranging weight matrix is ​​dynamically adjusted. To suppress the impact of abnormal noise on the positioning results.

[0056] In conjunction with the first aspect, the multi-vehicle positioning method may optionally further include:

[0057] The position covariance matrix is ​​obtained by inverting the coefficient matrix of the positive definite equation. .

[0058] In conjunction with the first aspect, optionally, the status information of each vehicle to be located and the anchor point vehicle is defined and stored by a unified information set, which is represented as:

[0059] ,

[0060] in, This represents the state vector of the vehicle to be located. - This represents the state vector of the anchored vehicle. This indicates the number of vehicles anchored in the vehicle-to-everything (V2X) communication network. Let represent the position covariance matrix of the vehicle to be located. Indicates the first A temporary or permanent unique identifier for a vehicle within a vehicle-to-everything (V2X) communication network. Indicates the first The relative distance between the anchored vehicle and the vehicle to be located. Indicates the first The vehicle center coordinates of the anchored vehicle; This represents the latitude and longitude coordinates of the aircraft in the WGS-84 coordinate system acquired by GNSS. Representing the The speed of the vehicle at each anchor point.

[0061] In conjunction with the first aspect, optionally, the system receives a set of broadcast information from vehicles at each anchor point in the vehicle-to-everything (V2X) communication network within a predetermined communication range;

[0062] The trace of the position covariance matrix of vehicles at each anchor point is calculated as a reliability index.

[0063] Select the first with the smallest trace One vehicle is selected as the anchor point vehicle and used as the positioning solution node.

[0064] Secondly, optionally, a multi-vehicle positioning system based on vehicle network communication and collaborative perception includes a storage medium and a processor;

[0065] The storage medium is used to store instructions;

[0066] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspects.

[0067] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0068] This invention proposes a multi-vehicle localization method and system based on vehicle-to-everything (V2X) communication and collaborative perception. First, a target detection algorithm combining motion compensation and deep learning is used to perform motion compensation on the raw point cloud data output by the LiDAR installed on the vehicle to be localized. Then, a deep learning network (incorporating a dual attention mechanism) is used to extract the 3D bounding box of the vehicle to be localized, thereby obtaining the precise relative distance between the vehicle to be localized and each selected anchor point vehicle within a predetermined communication range. Second, a nonlinear distance observation equation is constructed, and an iterative linearized weighted least squares (IL-WLS) algorithm is used. The nonlinear problem is linearized through Taylor series expansion, and a positive definite equation is constructed using the Jacobian matrix for iterative solution to obtain the global absolute coordinates of the vehicle to be localized. This effectively solves the vehicle localization problem in complex dynamic scenarios and exhibits high robustness and high real-time performance. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0070] Figure 1 This is one of the overall flowcharts of a multi-vehicle localization method based on vehicle network communication and cooperative perception according to an embodiment of the present invention;

[0071] Figure 2 This is the second schematic diagram of the overall process of a multi-vehicle localization method based on vehicle network communication and cooperative perception according to an embodiment of the present invention;

[0072] Figure 3 This is a schematic flowchart illustrating a method for calculating the precise relative distance between a vehicle to be located and selected anchor point vehicles according to an embodiment of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0074] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0075] Example 1

[0076] This invention provides a multi-vehicle localization method based on vehicle-to-everything (V2X) communication and cooperative perception, applicable to vehicles to be located, such as... Figures 1-2 As shown, it includes the following steps:

[0077] (1) Using a target detection algorithm that combines motion compensation and deep learning, the raw point cloud data output by the lidar installed on the vehicle to be located is corrected and calculated to obtain the precise relative distance between the vehicle to be located and each selected anchor point vehicle within a predetermined communication range; wherein, the vehicle to be located and the anchor point vehicles are connected vehicles or intelligent connected vehicles; the vehicle to be located and all the anchor point vehicles together constitute a vehicle-to-everything (V2X) communication network, i.e. Figure 1 In the specific implementation of the heterogeneous hybrid traffic cooperative topology, the vehicle-to-everything (V2X) communication network client is the V2X communication network.

[0078] (2) Based on the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a nonlinear distance observation equation is constructed by combining the vehicle center position broadcast by each selected anchor point vehicle;

[0079] (3) The nonlinear distance observation equation is solved by using the iterative linearized weighted least squares algorithm to obtain the real-time position increment of the vehicle to be located;

[0080] (4) Calculate the global absolute coordinates of the vehicle to be located based on the real-time changes in the vehicle's position increment.

[0081] The above scheme first utilizes a target detection algorithm combining motion compensation and deep learning to perform motion compensation on the raw point cloud data output by the LiDAR installed on the vehicle to be located. Then, a deep learning network (integrating a dual attention mechanism) is used to extract the 3D bounding box of the vehicle to be located, thereby obtaining the precise relative distance between the vehicle to be located and each selected anchor point vehicle within a predetermined communication range. Second, a nonlinear distance observation equation is constructed, and the iterative linearized weighted least squares (IL-WLS) algorithm is adopted. The nonlinear problem is linearized through Taylor series expansion, and a positive definite equation is constructed using the Jacobian matrix for iterative solution to obtain the global absolute coordinates of the vehicle to be located. This effectively solves the vehicle localization problem in complex dynamic scenarios and has high robustness and high real-time performance.

[0082] In one specific embodiment of the present invention, such as Figure 3 As shown, the method for calculating the precise relative distance between the vehicle to be located and the selected anchor point vehicles includes:

[0083] The original point cloud data is subjected to motion distortion compensation using forward and backward propagation of Kalman filtering (IESKF) to obtain motion-compensated point cloud data. In the specific implementation process, in order to reduce data redundancy and improve subsequent processing efficiency, the random sampling consistency plane fitting algorithm is also used to preprocess the motion-compensated point cloud data, identify and remove point cloud data belonging to the road surface, and retain only non-ground point cloud data containing vehicles and obstacles as input to the detection network.

[0084] The point cloud data is processed using the convolutional layer in a preset deep neural network to generate an input feature map. In the specific implementation process, non-ground point cloud data containing only vehicles and obstacles is also retained and fed into an improved PointPillars network to divide the three-dimensional space into vertical cylinders and encode them into pseudo images. The pseudo images are used as input to the convolutional layer of the preset deep neural network.

[0085] Using the channel attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed on the input feature map to obtain pooling results. The pooling results are then input into a multilayer perceptron for feature transformation, and finally, a channel attention feature map is generated using the sigmoid activation function. To enhance the response to vehicle speed-related characteristics;

[0086] By utilizing the spatial attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed along the channel dimension on the input feature map. The two pooling results are concatenated, and then feature fusion is performed through convolution. Finally, a spatial attention feature map is generated using the sigmoid activation function. To achieve focus on the point cloud contour region of the vehicle;

[0087] Channel attention feature map Spatial attention feature map Perform feature fusion processing to generate a feature fusion map;

[0088] The feature fusion map is enhanced to generate an enhanced feature fusion map. The enhanced feature fusion map is then input into the SSD detection head to obtain the three-dimensional bounding box of the vehicle. This three-dimensional bounding box contains the vehicle center coordinates of the vehicle to be located. Based on the vehicle center coordinates of the vehicle to be located and the vehicle center positions broadcast by each selected anchor point vehicle, the precise relative distance between the vehicle to be located and each selected anchor point vehicle is calculated.

[0089] In the above scheme, to accurately capture dynamic vehicles in complex mixed traffic flows and improve the feature extraction capability of the target detection algorithm in complex environments, a dual attention module is adopted in the feature extraction stage: First, a speed-aware channel attention mechanism is introduced. This mechanism adaptively assigns higher weights to key channels based on the inherent correlation between channel features and vehicle speed, thereby effectively suppressing static background noise. Second, a dynamic point cloud spatial attention mechanism is introduced, which focuses on the high-response areas of vehicle edge contours and potential motion directions in the spatial dimension. The weighted feature maps are deeply abstracted through a 2D-CNN backbone network, and finally, the SSD detection head regresses the 3D bounding box (3D bounding box) of the vehicle to be located. Then, the geometric center coordinates of the 3D bounding box are extracted as the vehicle center coordinates of the vehicle to be located. Combined with the vehicle center coordinates of each anchor point vehicle received, the precise relative distance vector between the vehicle to be located and each anchor point vehicle is calculated.

[0090] In one specific embodiment of the present invention, the method for generating motion-compensated point cloud data includes:

[0091] Forward propagation is performed based on IMU data from the inertial measurement unit installed on the vehicle to be located to predict the pose change trajectory of the vehicle to be located within the lidar scanning cycle.

[0092] When the lidar installed on the vehicle to be located completes a single-frame scan, the pose change trajectory is corrected based on the original point cloud data output by the lidar to obtain the optimized state at the end of the scan.

[0093] Based on the optimized state, backpropagation is performed to calculate the precise IMU pose at each laser point sampling time.

[0094] Using the precise IMU pose, the original point cloud data at discrete sampling times is projected to the end of the scan using the following mapping relationship to obtain motion-compensated point cloud data:

[0095] ,

[0096] In the formula, These are the coordinates of the point in the lidar coordinate system after motion compensation; It is a constant value that represents the spatial transformation relationship from the lidar coordinate system to the IMU coordinate system; This is the IMU pose transformation matrix obtained through backpropagation. Let be the rotation matrix in the IMU coordinate system relative to the global world coordinate system. Let be the translation vector of the IMU coordinate system relative to the global world coordinate system. To compensate for the point coordinates in the IMU coordinate system before.

[0097] In the above scheme, high-frequency kinematic data is acquired through an onboard inertial measurement unit (IMU), and forward propagation is performed using Iterative Error State Kalman Filter (IESKF) to predict the pose change trajectory of the vehicle to be located within the LiDAR scanning cycle. Subsequently, backpropagation correction is performed to address motion distortion caused by the LiDAR scanning mechanism. Specifically, after a single frame scan, the precise sampling timestamp of each laser point is traced back based on the optimized state, and the corresponding precise IMU pose is calculated using interpolation. All discrete points are then uniformly projected onto the coordinate system at the end of the scan, thus eliminating the ghosting phenomenon from a physical perspective.

[0098] In one specific embodiment of the present invention, the method for constructing the nonlinear distance observation equation includes:

[0099] Based on the pose change trajectory of the vehicle to be located within the LiDAR scanning period, let the estimated coordinates of the vehicle to be located be... These represent the locations of the vehicles to be located. , , The estimated coordinate positions in three directions, the first The vehicle center coordinates of each anchor point vehicle are The location is obtained directly from the vehicle to be located through the vehicle-to-everything (V2X) communication network, which consists of the vehicle to be located and the anchor point vehicle.

[0100] Construct the vehicle to be located and the first The nonlinear observation function between the vehicles at each anchor point is expressed as follows: ,in, For vehicles to be determined and the first The relative observation distance of vehicles at each anchor point It is the Euclidean norm;

[0101] Based on the nonlinear observation function and the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a system is constructed to determine the distance between the vehicle to be located and the selected anchor point vehicles. The nonlinear distance observation equation between vehicles at each anchor point is expressed as follows:

[0102] ,

[0103] In the formula, Indicates the vehicle to be located and the first Distance residuals between vehicles at each anchor point Indicates the vehicle to be located and the first The precise relative distance between vehicles at each anchor point.

[0104] In one specific embodiment of the present invention, the step of using an iterative linearized weighted least squares algorithm to solve the nonlinear distance observation equation to obtain the real-time changing position increment of the vehicle to be located includes:

[0105] Based on the nonlinear distance observation equation between the vehicle to be located and each selected anchor point vehicle within the predetermined communication range, an observation residual vector is formed. , The number of vehicles at the selected anchor points;

[0106] For the observation residual vector At the current location of the vehicle to be located At the next iteration, a first-order Taylor series expansion is performed, and higher-order terms are ignored to construct a linearized error equation. The expression of the linearized error equation is as follows:

[0107] ,

[0108] In the formula, This represents the position increment of the vehicle to be located. For the first The new information vector of the next iteration, i.e., based on the iterative estimate The calculated residual value; For the observed noise vector; For the nonlinear observation function in the iterative estimate The Jacobian matrix at point A is calculated using the following formula:

[0109] ;

[0110] ;

[0111] Solving the linearized error equation is transformed into a weighted norm minimization problem, specifically including:

[0112] Constructing about location increments The weighted quadratic objective function:

[0113] ,

[0114] In the formula, This is the ranging weight matrix;

[0115] According to the least squares principle, let the weighted quadratic objective function be expressed with respect to the position increment. The derivative is zero, yielding the value of the position increment. Positive definite equation:

[0116] ,

[0117] Solving the positive definite equation by matrix inversion yields the optimal position increment for the current iteration step. Update the vehicle center position according to the following formula:

[0118] ,

[0119] In the formula, For the first In the next iteration, the position increment The optimal estimate, Indicates the first The estimated value in the next iteration step;

[0120] Repeat the above steps until the L2 norm of the position increment is reached. Less than the preset convergence threshold Output the final solved coordinates Used as the global absolute coordinates of the vehicle to be located. .

[0121] In the above scheme, a nonlinear ranging observation model is constructed, and the iterative linearized weighted least squares (IL-WLS) algorithm is adopted. The nonlinear problem is linearized through Taylor series expansion, and a positive definite equation is constructed using the Jacobian matrix for iterative solution. Finally, the distance between the vehicle to be located and the first... The precise relative distance between vehicles at each anchor point.

[0122] In one specific embodiment of the present invention, the ranging weight matrix An adaptive update strategy based on IGG-III robust estimation theory is adopted, specifically including:

[0123] Construct a diagonal weight matrix Among them, diagonal elements Based on the vehicle to be located and the first Standardized residuals of the precise relative distance between vehicles at each anchor point Dynamic calculation:

[0124] ,

[0125] In the formula, For the a priori standard deviation of distance measurement, and These are the robust weight preservation threshold and the elimination threshold, respectively. Through this piecewise function, the weight of normal observations is maintained, the weight of abnormal observations is reduced, and extreme outliers are eliminated, thereby enhancing the robustness of the algorithm in non-line-of-sight environments.

[0126] Introducing the robust estimation theory of IGG-III, the ranging weight matrix is ​​dynamically adjusted. This is to suppress the impact of abnormal noise on the positioning results, thereby providing anti-interference capability for the least squares solution and ensuring that the final output position covariance matrix can truly reflect the positioning reliability, providing a reliable basis for the selection of anchor points.

[0127] In one specific embodiment of the present invention, the multi-vehicle positioning method further includes:

[0128] The position covariance matrix is ​​obtained by inverting the coefficient matrix of the positive definite equation. .

[0129] In one specific embodiment of the present invention, the status information of each vehicle to be located and the anchor point vehicle is defined and stored by a unified information set, which is represented as follows:

[0130] ,

[0131] in, This represents the state vector of the vehicle to be located. - This represents the state vector of the anchored vehicle. This indicates the number of vehicles anchored in the vehicle-to-everything (V2X) communication network. Let represent the position covariance matrix of the vehicle to be located. Indicates the first A temporary or permanent unique identifier for a vehicle within a vehicle-to-everything (V2X) communication network. Indicates the first The relative distance between the anchored vehicle and the vehicle to be located. Indicates the first The vehicle center coordinates of the anchored vehicle; This represents the latitude and longitude coordinates of the aircraft in the WGS-84 coordinate system acquired by GNSS. Representing the The speed of the vehicle at each anchor point.

[0132] In one specific embodiment of the present invention, the multi-vehicle positioning method further includes a global synchronization update step for multi-vehicle status information, specifically including: defining a synchronization time slice based on the LiDAR scanning cycle; checking the timestamp of the received V2X data packet (which includes vehicle information set) before each positioning calculation; retaining only data whose timestamp falls within the current time slice for least squares calculation, and discarding expired data with excessive delay to ensure the time consistency of each geometric constraint in the observation equation. Furthermore, the data transmission and reception timing of the V2X communication is strictly synchronized with the scanning cycle of the vehicle-mounted LiDAR, and all positioning calculations are completed within the time interval between the end of a single LiDAR scan and the start of the next frame. In the specific implementation process, the trace of the updated position covariance matrix is ​​monitored in real time. If the trace is less than the preset high-quality anchor point threshold, the role label of this vehicle in the cooperative topology is updated from the vehicle requesting positioning (i.e., the cooperating vehicle) to the anchor point vehicle providing the reference. The status information containing the updated position covariance matrix and role label is broadcast through V2X communication, so that this vehicle can participate in the positioning calculation of other surrounding vehicles as a high-confidence anchor point, thereby achieving error suppression and global propagation of confidence.

[0133] In one specific embodiment of the present invention, the method for selecting the anchor point vehicle includes:

[0134] Receives broadcast information sets from vehicles at various anchor points within the predetermined communication range of the vehicle-to-everything (V2X) communication network;

[0135] The trace of the position covariance matrix of vehicles at each anchor point is calculated as a reliability index.

[0136] Select the first with the smallest trace One vehicle is selected as the anchor point vehicle and used as the positioning solution node.

[0137] Example 2

[0138] This invention provides a multi-vehicle positioning system based on vehicle network communication and collaborative perception, including a storage medium and a processor;

[0139] The storage medium is used to store instructions;

[0140] The processor is configured to operate according to the instructions to execute the method according to any one of Embodiment 1.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0143] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0144] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0145] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0146] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A multi-vehicle localization method based on vehicle-to-everything (V2X) communication and collaborative perception, applied to vehicles to be located, characterized in that, include: Using a target detection algorithm that combines motion compensation and deep learning, the raw point cloud data output by the lidar installed on the vehicle to be located is corrected and calculated to obtain the precise relative distance between the vehicle to be located and each selected anchor point vehicle within a predetermined communication range; wherein, the vehicle to be located and the anchor point vehicles are connected vehicles or intelligent connected vehicles. Based on the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a nonlinear distance observation equation is constructed by combining the vehicle center position broadcast by each selected anchor point vehicle. An iterative linearized weighted least squares algorithm is used to solve the nonlinear distance observation equation to obtain the real-time position increment of the vehicle to be located. Based on the real-time changes in the position increment of the vehicle to be located, the global absolute coordinates of the vehicle to be located are calculated.

2. The multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 1, characterized in that, The method for calculating the precise relative distance between the vehicle to be located and the vehicles at each selected anchor point includes: Motion distortion compensation is performed on the original point cloud data using forward and backward propagation with Kalman filtering to obtain motion-compensated point cloud data; The point cloud data is processed using convolutional layers in a pre-defined deep neural network to generate an input feature map; Using the channel attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed on the input feature map to obtain pooling results. The pooling results are then input into a multilayer perceptron for feature transformation, and finally, a channel attention feature map is generated using the sigmoid activation function. To enhance the response to vehicle speed-related characteristics; By utilizing the spatial attention mechanism in a pre-defined deep neural network, global max pooling and global average pooling operations are performed along the channel dimension on the input feature map. The two pooling results are concatenated, and then feature fusion is performed through convolution. Finally, a spatial attention feature map is generated using the sigmoid activation function. To achieve focus on the point cloud contour region of the vehicle; The channel attention feature map Spatial attention feature map Perform feature fusion processing to generate a feature fusion map; The feature fusion map is enhanced to generate an enhanced feature fusion map. The enhanced feature fusion map is then input into the SSD detection head to obtain the three-dimensional bounding box of the vehicle. This three-dimensional bounding box contains the vehicle center coordinates of the vehicle to be located. Based on the vehicle center coordinates of the vehicle to be located and the vehicle center positions broadcast by each selected anchor point vehicle, the precise relative distance between the vehicle to be located and each selected anchor point vehicle is calculated.

3. The multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 2, characterized in that: The method for generating the motion-compensated point cloud data includes: Forward propagation is performed based on IMU data from the inertial measurement unit installed on the vehicle to be located to predict the pose change trajectory of the vehicle to be located within the lidar scanning cycle. When the lidar installed on the vehicle to be located completes a single-frame scan, the pose change trajectory is corrected based on the original point cloud data output by the lidar to obtain the optimized state at the end of the scan. Based on the optimized state, backpropagation is performed to calculate the precise IMU pose at each laser point sampling time. Using the precise IMU pose, the original point cloud data at discrete sampling times is projected to the end of the scan using the following mapping relationship to obtain motion-compensated point cloud data: ， In the formula, These are the coordinates of the point in the lidar coordinate system after motion compensation; It is a constant value that represents the spatial transformation relationship from the lidar coordinate system to the IMU coordinate system; This is the IMU pose transformation matrix obtained through backpropagation. Let be the rotation matrix in the IMU coordinate system relative to the global world coordinate system. This is the translation vector of the IMU coordinate system relative to the global world coordinate system. To compensate for the point coordinates in the IMU coordinate system before.

4. The multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 2, characterized in that: The method for constructing the nonlinear distance observation equation includes: Based on the pose change trajectory of the vehicle to be located within the LiDAR scanning period, let the estimated coordinates of the vehicle to be located be... These represent the locations of the vehicles to be located. , , The estimated coordinate positions in three directions, the first The vehicle center coordinates of each anchor point vehicle are The location is obtained directly from the vehicle to be located through the vehicle-to-everything (V2X) communication network, which consists of the vehicle to be located and the anchor point vehicle. Construct the vehicle to be located and the first The nonlinear observation function between the vehicles at each anchor point is expressed as follows: ,in, For vehicles to be determined and the first The relative observation distance of vehicles at each anchor point It is the Euclidean norm; Based on the nonlinear observation function and the precise relative distance between the vehicle to be located and each selected anchor point vehicle, a system is constructed to determine the distance between the vehicle to be located and the selected anchor point vehicles. The nonlinear distance observation equation between vehicles at each anchor point is expressed as follows: ， In the formula, Indicates the vehicle to be located and the first Distance residuals between vehicles at each anchor point Indicates the vehicle to be located and the first The precise relative distance between vehicles at each anchor point.

5. A multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 4, characterized in that: The method employs an iterative linearized weighted least squares algorithm to solve the nonlinear distance observation equation, obtaining the real-time position increment of the vehicle to be located, including: Based on the nonlinear distance observation equation between the vehicle to be located and each selected anchor point vehicle within the predetermined communication range, an observation residual vector is formed. , The number of vehicles at the selected anchor points; For the observation residual vector At the current location of the vehicle to be located At the next iteration, a first-order Taylor series expansion is performed, and higher-order terms are ignored to construct a linearized error equation. The expression of the linearized error equation is as follows: ， In the formula, This represents the position increment of the vehicle to be located. For the first The information vector of the next iteration, i.e., based on the iterative estimate. The calculated residual value; For the observed noise vector; For the nonlinear observation function in the iterative estimate The Jacobian matrix at point A is calculated using the following formula: ； ； Solving the linearized error equation is transformed into a weighted norm minimization problem, specifically including: Constructing about location increments The weighted quadratic objective function: ， In the formula, This represents a weighted quadratic objective function. This is the ranging weight matrix; According to the least squares principle, let the weighted quadratic objective function be expressed with respect to the position increment. The derivative is zero, yielding the value of the position increment. Positive definite equation: ， Solving the positive definite equation by matrix inversion yields the optimal position increment for the current iteration step. Update the vehicle center position according to the following formula: ， In the formula, For the first In the next iteration, the position increment The optimal estimate; Indicates the first The estimated value in the next iteration step; Repeat the above steps until the L2 norm of the position increment is reached. Less than the preset convergence threshold Output the final solved coordinates Used as the global absolute coordinates of the vehicle to be located. .

6. A multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 5, characterized in that: The ranging weight matrix An adaptive update strategy based on IGG-III robust estimation theory is adopted, specifically including: Construct a diagonal weight matrix Among them, diagonal elements Based on the vehicle to be located and the first Standardized residuals of the precise relative distance between vehicles at each anchor point Dynamic calculation: ， In the formula, For the a priori standard deviation of distance measurement, and These are the robust rights preservation threshold and the elimination threshold, respectively. Introducing the robust estimation theory of IGG-III, the ranging weight matrix is ​​dynamically adjusted. To suppress the impact of abnormal noise on the positioning results.

7. A multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 5, characterized in that: The multi-vehicle positioning method also includes: The position covariance matrix is ​​obtained by inverting the coefficient matrix of the positive definite equation. .

8. A multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 7, characterized in that: The status information of each vehicle to be located and the vehicle at the anchor point is defined and stored by a unified information set, which is represented as follows: ， in, This represents the state vector of the vehicle to be located. - This represents the state vector of the anchored vehicle. This indicates the number of vehicles anchored in the vehicle-to-everything (V2X) communication network. Let represent the position covariance matrix of the vehicle to be located. Indicates the first A temporary or permanent unique identifier for a vehicle within a vehicle-to-everything (V2X) communication network. Indicates the first The relative distance between the anchored vehicle and the vehicle to be located. Indicates the first The vehicle center coordinates of the anchored vehicle; This represents the latitude and longitude coordinates of the aircraft in the WGS-84 coordinate system obtained by GNSS. Representing the The speed of the vehicle at each anchor point.

9. A multi-vehicle localization method based on vehicle network communication and cooperative perception according to claim 8, characterized in that: The methods for selecting anchor point vehicles include: Receives broadcast information sets from vehicles at various anchor points within the predetermined communication range of the vehicle-to-everything (V2X) communication network; The trace of the position covariance matrix of vehicles at each anchor point is calculated as a reliability index. Select the first with the smallest trace One vehicle is selected as the anchor point vehicle and used as the positioning solution node.

10. A multi-vehicle positioning system based on vehicle-to-everything (V2X) communication and cooperative perception, characterized in that, Including storage media and processor; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-9.

Images