Global position optimization method and system for automatic driving based on road network consistency inspection

Abstract

The application provides an automatic driving global position optimization method and system based on road network consistency verification, and relates to the technical field of vehicle positioning. The method comprises the following steps: determining a first candidate transformation matrix between a key non-static frame sequence of an odometer and at least one global candidate sub-track with the same form in a standard definition electronic map; performing consistency verification on the first candidate transformation matrix based on a KD tree and a road network in the standard definition electronic map, so as to obtain at least one feasible transformation matrix in the first candidate transformation matrix that satisfies the road network form consistency verification; and optimizing the automatic driving global position of a vehicle based on each feasible transformation matrix. The application can eliminate the difference information between the actual observation distance of an automatic driving vehicle and the prior position of a road node in an electronic map when positioning is performed by using a standard definition electronic map, and improve the positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of vehicle positioning technology, and in particular to a method and system for global location optimization of autonomous driving based on road network consistency verification. Background Technology

[0002] Continuous localization is crucial for navigation and planning of connected and autonomous vehicles (CAVs) in urban environments. Current mainstream global continuous localization systems, such as fusion systems of Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU), often suffer severe performance degradation due to signal interference, obstruction, and multipath effects. Visual or LiDAR-derived odometry can provide independent and continuous position estimates, enhancing the redundancy and robustness of global localization. However, suppressing odometry error accumulation typically relies on pre-built feature maps, which are costly and require frequent updates. Providing redundant global localization when these costly maps are missing or invalid poses a significant challenge.

[0003] Standard definition (SD) electronic maps, as low-cost and widely available online maps, provide static features of urban road networks and key environmental aspects. Due to their global consistency, SD electronic maps serve as a lightweight alternative to high-precision LiDAR maps, particularly suitable for situations where high-precision maps are unavailable or outdated. However, existing global localization methods using LiDAR-SD electronic map matching suffer from issues such as untimely updates, insufficient spatial accuracy, and limited semantic features in SD electronic maps, leading to discrepancies between the location estimates provided by SD electronic map road nodes and the actual observed locations. Summary of the Invention

[0004] This invention provides a global location optimization method and system for autonomous driving based on road network consistency verification. This addresses the shortcomings of existing technologies where standard-definition (SD) electronic maps suffer from untimely updates, insufficient spatial accuracy, and limited semantic features, leading to discrepancies between the location estimates provided by SD electronic map road nodes and the actual observed location. The method involves determining the first candidate transformation matrix between the odometer's key non-static frame sequence and the global candidate sub-trajectory, then using a KD-tree to find feasible transformation matrices from the first candidate transformation matrix. This optimizes the transformation matrix between the odometer and the global candidate sub-trajectory, and finally, based on the optimized transformation matrix, performs global positioning of the vehicle, thus reducing the discrepancy between the prior location provided by SD electronic map road nodes and the actual observed location.

[0005] This invention provides a global location optimization method for autonomous driving based on road network consistency verification, comprising the following steps:

[0006] Determine the first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in the standard definition electronic map;

[0007] Based on the KD tree and the road network in the standard definition electronic map, a consistency check is performed on each of the first candidate transformation matrices, and at least one feasible transformation matrix that satisfies the road network morphology consistency check is selected from each of the first candidate transformation matrices.

[0008] Based on the feasible transformation matrices described above, the global position of the vehicle for autonomous driving is optimized.

[0009] According to the present invention, an autonomous driving global location optimization method based on road network consistency verification includes the following steps: performing consistency verification on each first candidate transformation matrix based on a KD-tree and the road network in the standard definition electronic map; selecting at least one feasible transformation matrix from each first candidate transformation matrix that satisfies the road network morphology consistency verification; mapping each odometer point in the original odometer sequence to a global coordinate system based on the first candidate transformation matrix to obtain a global location point corresponding to each odometer point; finding the first nearest neighbor point corresponding to each global location point in the road network of the standard definition electronic map based on the KD-tree; determining the first average distance between all global location points and the first nearest neighbor point corresponding to each global location point; and performing consistency verification on each first candidate transformation matrix based on the first average distance, selecting at least one feasible transformation matrix from each first candidate transformation matrix that satisfies the road network morphology consistency verification.

[0010] According to the present invention, an autonomous driving global position optimization method based on road network consistency verification is provided. The optimization of the vehicle's autonomous driving global position based on each feasible transformation matrix includes: mapping the original odometer sequence to a global coordinate system based on each feasible transformation matrix to obtain a first global trajectory corresponding to each second transformation matrix; searching for a second nearest neighbor point corresponding to each position point in the first global trajectory in the road network of the standard definition electronic map based on the KD tree; determining a target transformation matrix based on the second nearest neighbor point; and optimizing the vehicle's autonomous driving global position based on the target transformation matrix.

[0011] According to the present invention, an autonomous driving global position optimization method based on road network consistency test is provided. The step of determining the target transformation matrix based on the second nearest neighbor point includes: determining a second candidate transformation matrix between the odometry key non-static frame sequence and at least one of the global candidate sub-trajectories based on minimizing the least squares estimation error; mapping the original odometry sequence to a global coordinate system based on each of the second candidate transformation matrices to obtain at least one second global trajectory; determining a second average distance between each position point in the second global trajectory and the second nearest neighbor point corresponding to each position point in each of the first global trajectories; and determining the second candidate transformation matrix corresponding to the minimum second average distance as the target transformation matrix.

[0012] According to the present invention, an autonomous driving global position optimization method based on road network consistency verification is provided, which determines the key non-static frame sequence of the odometer, including: acquiring the pose of each position point through a lidar odometer to obtain the original odometer sequence; and sampling from the original odometer sequence to obtain the key non-static frame sequence of the odometer.

[0013] According to the present invention, an autonomous driving global location optimization method based on road network consistency test is provided, which determines the global candidate sub-trajectory, including: determining multiple global candidate locations in the online open map; and determining at least one global candidate sub-trajectory from the multiple global candidate locations based on the continuity of each global candidate location.

[0014] This invention also provides an autonomous driving global location optimization system based on road network consistency verification, comprising the following modules:

[0015] The first determining module is used to determine the first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in the standard definition electronic map;

[0016] The second determining module is used to perform consistency checks on each of the first candidate transformation matrices based on the KD tree and the road network in the standard definition electronic map, and select at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices.

[0017] The optimization module is used to optimize the global position of the vehicle for autonomous driving based on the feasible transformation matrices.

[0018] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the global position optimization method for autonomous driving based on road network consistency verification as described above.

[0019] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the global position optimization method for autonomous driving based on road network consistency verification as described above.

[0020] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the global position optimization method for autonomous driving based on road network consistency verification as described above.

[0021] The present invention provides a method and system for optimizing the global position of autonomous driving based on road network consistency verification. This method determines a first candidate transformation matrix between a key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in a standard-definition (SD) electronic map. Based on a KD-tree and the road network in the SD electronic map, a consistency verification is performed on each of the first candidate transformation matrices, and at least one feasible transformation matrix that satisfies the road network shape consistency verification is selected from among the first candidate transformation matrices. Based on each feasible transformation matrix, the global position of the vehicle's autonomous driving is optimized. Thus, a KD-tree is constructed based on the vehicle's navigable road network nodes in the SD electronic map, and motion consistency verification is performed on the local laser odometer sequence of the first candidate transformation matrices based on this KD-tree. The first candidate transformation matrix that passes the consistency verification is used as a feasible transformation matrix, and the global position of the vehicle's autonomous driving is optimized based on the optimized transformation matrix, thus eliminating the difference between the actual observed distance of the autonomous vehicle and the prior position of the road nodes in the SD electronic map, and improving positioning accuracy. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the initial location estimation results provided by LiDAR-standard definition electronic map descriptor matching based on existing technology.

[0024] Figure 2 This is a flowchart illustrating the global location optimization method for autonomous driving based on road network consistency verification provided by the present invention.

[0025] Figure 3 This is a schematic diagram illustrating the inconsistency between the odometry pose sequence and the global position estimation sequence provided by this invention.

[0026] Figure 4This is a schematic diagram of a global position estimation sequence obtained by matching LiDAR-standard definition electronic map descriptors and performing motion continuity checks, as provided by the present invention.

[0027] Figure 5 This invention provides a global candidate trajectory available as a sub-trajectory. One of the schematic diagrams.

[0028] Figure 6 This invention provides a global candidate trajectory available as a sub-trajectory. The second diagram.

[0029] Figure 7 This is one of the projection results of the transformation matrix for estimating the correspondence between global candidate sub-trajectories and local keyframe odometry provided by this invention.

[0030] Figure 8 This is the second projection result of the transformation matrix for estimating the correspondence between global candidate sub-trajectories and local keyframe odometry provided by this invention.

[0031] Figure 9 This is a schematic diagram of the structure of the global position optimization system for autonomous driving based on road network consistency verification provided by the present invention.

[0032] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0034] Existing global positioning methods for LiDAR-standard definition electronic map data association can be divided into two categories: methods that rely on external initialization and methods that do not rely on external initialization.

[0035] Methods relying on external initialization: External positioning systems such as GNSS-IMU are susceptible to occlusion and multipath effects in complex urban environments. To address this, some methods utilize standard-definition (SD) electronic map road networks and building information as auxiliary observations to establish additional location estimation channels and reduce global positioning errors. Initialization methods such as integrating GPS, visual odometry, and SD electronic maps, using Transformer models to align LiDAR data with satellite maps, and combining visual structures with SD electronic map building data have improved urban positioning accuracy. However, these methods require weak GNSS to provide initial global position, making it difficult to achieve fully autonomous positioning solely based on vehicle-mounted perception.

[0036] Methods that do not rely on external initialization: Directly constructing the constraint relationship between vehicle-mounted perception and standard-definition electronic maps to achieve autonomous localization. For example, using deep learning descriptors and particle filtering to achieve autonomous localization between LiDAR perception and standard-definition electronic maps. Neural networks (OrientNet) achieve global pose regression, but their generalization is limited. To overcome the problem of relying on pre-training, combining Visual Simultaneous Localization and Mapping (SLAM) with Monte Carlo methods to initially achieve autonomous localization; using particle filters to compare real and simulated LiDAR data to achieve GPS-free localization. However, these methods have limited localization accuracy and may produce unreasonable estimates.

[0037] The following section first describes the global initial position estimation and sequence generation of LiDAR-standard definition electronic map descriptor matching. Then, it explains how to use the prior road network information provided by open street maps (standard definition electronic maps) to achieve continuous global positioning through odometer-assisted LiDAR-standard definition electronic map matching, and provides a formal description of this problem.

[0038] Define a prior descriptor database based on standard definition electronic maps as follows: The following formula (1):

[0039]

[0040] Among them, database Containing K candidate road nodes and their corresponding prior information, this database consists of three subsets: a set of road node coordinates. Virtual Plane Point Cloud and the set of structure boundary descriptors in, The expression is shown in the following formulas (2)-(4):

[0041]

[0042] Among them, P K Indicates a road node.

[0043]

[0044] in, This represents a virtual planar point cloud.

[0045]

[0046] in, This represents the first to the Kth structure boundary descriptors.

[0047] Each road node All coordinates are represented using the Universal Transverse Mercator Grid System (UTM), as shown in the following formula (5):

[0048]

[0049] Each virtual planar point cloud The LiDAR sensing range is generated by simulating the range of 360 two-dimensional points, as shown in formula (6):

[0050]

[0051] in, This represents the first two-dimensional point.

[0052] Each structure boundary descriptor It consists of 360 scalar values, as shown in the following formula (7):

[0053]

[0054] in, This represents the angle between the laser radar scanning direction and the local building boundary orientation in the θ-th direction, where θ represents any direction from 1 to 360 degrees.

[0055] The real-time input LiDAR scanning data is defined as the original 3D point cloud set, as shown in the following formula (8):

[0056]

[0057] Each point This represents the three-dimensional coordinates obtained from a single lidar scan. The length of j is determined by the input frame but is uncertain, and j is greater than or equal to 1.

[0058] The subset of LiDAR points belonging to architectural semantics is defined as architectural point cloud, as shown in formula (9):

[0059]

[0060] Among them, based on semantic segmentation from Extracted building lidar point set It only includes points identified as building surfaces. To obtain the online structural descriptor corresponding to the real-time building point cloud, the three-dimensional point set Lb is first projected onto a two-dimensional plane, and the projected points are grouped according to the scanning angle to form an equivalent planar building point cloud set, as shown in the following formula (10):

[0061]

[0062] Where θ represents the integer scan angle in the vehicle coordinate system, in degrees. Let θ be the equivalent two-dimensional projected building point at angle θ. To convert 3D semantic point clouds Convert to an angle-aligned equivalent planar point set The planar projection operator.

[0063] By calculating the angle between the building boundary orientation and the LiDAR scanning direction at each angle, an online boundary descriptor can be obtained. The real-time input descriptor is defined by the following formula (11):

[0064]

[0065] Where each m θ This indicates the scanning direction at angle θ and the corresponding projected building boundary point. The angle between azimuth and direction. For each input LiDAR frame, this can be determined by using the prior descriptor set. Perform a most similar descriptor search and obtain its representation in the road node set. The corresponding position index in the table determines the global initial position.

[0066] For example, real-time input LiDAR frames acquire the corresponding global candidate location P in a standard definition electronic map. * The following formula (12):

[0067]

[0068] Where, k * This is the optimal index with the lowest cost for descriptor similarity.

[0069] Furthermore, k * The calculation is as follows (13):

[0070]

[0071] in, Let represent the similarity measurement function between descriptors, arg represent the k corresponding to the minimum similarity, N is the threshold number of candidate descriptors, and min represents the minimum value.

[0072] Based on the initialization of the global candidate positions described above, a set of global candidate positions can be obtained for each input LiDAR frame. However, due to the temporal discontinuity and abrupt changes in the estimated positions, the output of this method is not yet sufficient to reliably support the continuous global localization of connected autonomous vehicles (CAVs).

[0073] The following is a problem definition for achieving autonomous continuous global positioning based on the output results, combined with lidar odometry constraints:

[0074] Given a lidar odometry system, the estimated pose of the lidar in the i-th frame is represented as Pio∈

[0075] SE(2). The complete lidar odometry is defined as follows: (14)

[0076]

[0077] Among them, l o The original odometer Po represents the length of the original odometer (i.e., the total number of input LiDAR frames), and the original odometer sequence. It is represented by a local coordinate system, the origin of which is fixed at the initial frame.

[0078] Each odometer pose All are associated with global candidate locations obtained through LiDAR-standard definition electronic map descriptor matching. The corresponding LiDAR odometry global candidate location sequence. This can be expressed as the following formula (15):

[0079]

[0080] in, For the l o Global candidate positions for frame lidar odometer.

[0081] Figure 1 This is a schematic diagram illustrating the initial location estimation results provided by LiDAR-standard definition electronic map descriptor matching based on existing technology. For example... Figure 1As shown, the horizontal axis represents the X-axis in the global coordinate system, the vertical axis represents the Y-axis, the green line represents the ground truth, the red line represents standard definition (OSM) matching, and the black line represents the road network. Many initial global positions obtained through LiDAR-OSM descriptor matching are scattered outside the actual trajectory, and there are also difficult-to-visualize temporal and spatial mismatches. It can be seen that LiDAR-OSM descriptor matching struggles to provide stable and continuous global position estimates corresponding to local odometry.

[0082] To address the challenges of continuous localization of CAVs based on LiDAR-standard definition electronic map matching, the problem can be decomposed into two key sub-tasks: cross-modal trajectory association and observation error elimination.

[0083] The cross-modal trajectory association problem requires selecting a globally consistent pose sequence from the candidate location set. Its core objective is to extract a global path with motion consistency from the LiDAR-standard definition electronic map matching results. This path is defined by the following formula (16):

[0084]

[0085] in, Represents the local odometer trajectory Global paths with similar spatiotemporal motion patterns To the candidate set The optimized pose selected from the options.

[0086] In determining candidate paths Subsequently, the observation error compensation problem can be formulated as: estimating a rigid body transformation T that maps the local coordinate system to the global UTM coordinate system, such that the average position deviation between the transformed odometry pose and the corresponding global reference position is minimized, and the optimal transformation matrix T is obtained. * The calculation is as follows (17):

[0087]

[0088] Where Dist(·) represents the Euclidean distance metric, i is the frame number of the input lidar sequence, and T is the local odometry sequence. The most similar candidate path projected onto the LiDAR-standard definition electronic map matching results The transformation matrix.

[0089] Descriptor matching location set obtained through LiDAR-standard definition electronic map matching In the process, a globally consistent candidate path is determined. Make the selected path match the odometer trajectory in motion mode. Maintain maximum similarity. Based on the candidate paths obtained through optimization. By solving the rigid body transformation T * To minimize the global registration error between the local odometry pose and the reference path, the final output is the globally aligned pose sequence of the connected autonomous vehicle in the UTM coordinate system, as shown in the following formula (18):

[0090]

[0091] in, This indicates that the globally aligned pose sequence is the global trajectory.

[0092] However, the global positioning method for matching LiDAR with standard definition electronic maps mentioned above suffers from problems such as untimely updates, insufficient spatial accuracy, and limited semantic features in standard definition electronic maps, resulting in discrepancies between the location estimates provided by road nodes in standard definition electronic maps and the actual observed locations.

[0093] To address the aforementioned issues, this invention provides a method for optimizing the global location of autonomous driving based on road network consistency verification. A K-Dimensional Tree (KD) is constructed based on the vehicle-accessible road network nodes within a static standard-definition electronic map. Motion consistency verification is then performed on the local laser odometry sequence of the first candidate transformation matrix based on this KD tree. The first candidate transformation matrix that passes the consistency verification is used as a feasible transformation matrix. The optimized transformation matrix is ​​then used to optimize the global location of the vehicle's autonomous driving system, eliminating the discrepancy between the actual observed distance of the autonomous vehicle and the prior location of road nodes on the standard-definition electronic map, thereby improving positioning accuracy.

[0094] The following is combined with Figures 2-8 The present invention describes an autonomous driving global location optimization method based on road network consistency verification. This method is applicable to any type of networked autonomous driving vehicle. The execution subject of this method can be an electronic device or an autonomous driving global location optimization method based on road network consistency verification installed in the electronic device. The autonomous driving global location optimization system based on road network consistency verification can be implemented by software, hardware, or a combination of both.

[0095] Figure 2 This is a flowchart illustrating the global location optimization method for autonomous driving based on road network consistency verification provided by the present invention, as shown below. Figure 2 As shown, the method includes the following:

[0096] Step 201: Determine the first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in the standard definition electronic map.

[0097] It should be noted that the odometer sequence is represented in a local coordinate system, while the global candidate sub-trajectory is represented in a global coordinate system. Under the action of the first candidate transformation matrix, the local coordinate system can be mapped to the global coordinate system. In other words, under the action of the first candidate transformation matrix, the position in the odometer sequence can be converted into a global reference position.

[0098] It should be noted that the position point of each frame in the key non-static frame sequence of the odometer is different.

[0099] Optionally, the first candidate transformation matrix can be obtained from the real-time acquired key non-static frame sequence of odometry and the global candidate sub-trajectory, or it can be pre-calculated and fixed. Here, all first candidate transformation matrices are candidate solutions.

[0100] Optionally, the first candidate transformation matrix can be obtained using the least squares error, or it can be obtained by further filtering after obtaining the initial transformation matrix using the least squares error.

[0101] For example, the key non-static frame sequence of the odometer can be represented by the following formula (19):

[0102]

[0103] in, i is the first pose in the original odometry sequence. M Indicates the last index of the keyframe sequence. For P o A subset of. The length of the sampled odometry sequence can be defined as l. so .

[0104] For example, the global candidate sub-trajectories can be represented by the following formula (20):

[0105]

[0106] in, This represents the unique global candidate position, selected from the sub-trajectory. i u This represents the index of a unique global position sequence whose length is related to the length of the input frame.

[0107] Furthermore, the first candidate transformation matrix The calculation is as follows: (21)

[0108]

[0109] in, Let SE(2) represent the i-th transformation matrix estimated by the i-th global candidate sub-trajectory in the calculation, and let i(i) represent the horizontal axis, vertical axis, and yaw. u) represents the i-th candidate sub-trajectory in the global candidate trajectory u The key non-static frame sequence P of the odometer corresponding to each point O index, l u is the length of the global candidate sub-trajectory set and its corresponding keyframe odometry subsequence.

[0110] Optionally, the key non-static frame sequence of the odometer can be understood as the vehicle's motion pattern. The key non-static frame sequence of the odometer can be obtained by sampling from the original odometer sequence or by obtaining it directly from the odometer.

[0111] For example, determining the key non-static frame sequence of the odometer includes: acquiring the pose of each location point through a lidar odometer to obtain the original odometer sequence; and sampling from the original odometer sequence to obtain the key non-static frame sequence of the odometer.

[0112] Here, pose can include the horizontal axis, the vertical axis, and the yaw angle.

[0113] For example, the original odometer sequence P o for

[0114] Among them, l o Represents the original odometer sequence P o The length of the input lidar frames, i.e., the total number of input lidar frames. The original odometry sequence P represents the estimated pose of the LiDAR in the i-th frame. o It is represented by a local coordinate system, the origin of which is fixed at the initial frame.

[0115] Here, the sampling step size can be any suitable value.

[0116] It should be noted that, although each local odometry pose Each corresponds to a global position estimate P. i However, the LiDAR-standard definition electronic map descriptor matching results are limited by the discrete resolution (Pres) in the standard definition electronic map road node interpolation process, making each estimated global location P... i All fall on fixed discrete locations. When the CAV is in a waiting state, moving slowly, or has low descriptor matching accuracy, the global position estimate may still fluctuate significantly even if the local odometry remains unchanged.

[0117] Since the prior descriptor database Dprior is generated by interpolating the road network node Pprior with the minimum interpolation resolution Pres, it is difficult for the global position estimation sequence to match the frame-by-frame odometry pose sequence which is smaller than the interpolation resolution. To alleviate the mismatch between global position estimation and local odometry, the minimum effective driving distance equal to Pres is defined as the segmentation threshold. The local odometry sequence is then segmented according to this threshold, thereby removing stationary or low-speed frames. In other words, the minimum difference resolution is used as the sampling step size to sample from the original odometry sequence to obtain the keyframe odometry sequence.

[0118] For example, a minimum driving distance threshold equal to Pres is applied to the local odometry trajectory, and a reference motion mode sequence (also known as a keyframe odometry sequence) is constructed using the Adaptive Key Frame Sampling (AKFS) method.

[0119] Furthermore, key non-static frame sequences of the odometer are selected from the original odometer sequence, i M The following formula (22) must be satisfied:

[0120]

[0121] Among them, i k Indicates the index of the selected keyframe, i k Greater than or equal to 0 Less than or equal to 1 o And the index sequence i k Strictly monotonically increasing, This represents the previous pose in the key non-static frame sequence of the odometer, where j is the original odometer index greater than i. k Frame index, This indicates the pose of frame j and frame i. k The distance between the poses of the frames will be greater than P. res The pose with the smallest index among all frames is used as the pose of the current keyframe. M The maximum index value that strictly corresponds to the original frame index.

[0122] After performing AKFS processing on the raw CAV odometer data, the key non-static frame sequence of the odometer data was extracted. The spacing conditions between adjacent elements are basically consistent with the node spacing of the road network in standard definition electronic maps. This keyframe sequence can be used as a benchmark for establishing cross-modal spatiotemporal associations.

[0123] In this embodiment of the invention, by sampling from the original odometry sequence, a key non-static odometry frame sequence is obtained, static or low-speed frames are removed, the difference between the local odometry and the global pose sequence is reduced, the mismatch between the global position estimation and the local odometry is alleviated, and a consistent association between the odometry sequence and the global position estimation sequence is established.

[0124] Optionally, the global candidate sub-trajectory may or may not contain overlapping positions. The global candidate sub-trajectory is a part of the global trajectory.

[0125] Optionally, global candidate sub-trajectories can be obtained by segmenting the global trajectory or by the continuity between multiple location points.

[0126] For example, determining the global candidate sub-trajectory includes: determining a plurality of global candidate locations in the online open map; and determining at least one of the global candidate sub-trajectory from the plurality of global candidate locations based on the continuity of each of the global candidate locations.

[0127] Here, the continuity of global candidate positions can be determined based on the distance between two adjacent global candidate positions, or it can be determined based on temporal continuity, directional continuity, etc.

[0128] Figure 3 This is a schematic diagram illustrating the inconsistency between the odometry pose sequence and the global position estimation sequence provided by this invention, as shown below. Figure 3 As shown, black dots represent the road network of the standard definition electronic map, and red lines represent the global initial position estimation sequence corresponding to the first 94 LiDAR frames, arranged in chronological order. The blue line represents the advanced Lidar Odometry and Mapping (A-LOAM) odometry of the first 94 frames. For ease of visualization, the actual pose is projected onto the UTM coordinate system. It can be seen that due to errors in the descriptor matching process, the original odometry P... o The corresponding global position estimation sequence P * There is a discontinuity.

[0129] It should be noted that, under ideal conditions, the global position estimate obtained through descriptor matching should remain within the same road segment across multiple consecutive frames. Position estimation deviations caused by descriptor matching errors do not alter the overall motion trend of the CAV; such deviations are considered outliers in the position estimation sequence.

[0130] The fundamental assumption of global inter-frame motion consistency requires that the global position of the CAV should not exhibit abrupt transitions to different road segments. To ensure that the extracted key candidate sequences accurately reflect the vehicle motion modes, outliers caused by descriptor mismatches must be eliminated.

[0131] To eliminate inconsistent candidate points caused by descriptor matching errors, a Spatiotemporal Sliding Window Detection (STSWD) mechanism is introduced to evaluate the continuity of the Top-1 global position estimation. The STSWD mechanism only retains candidate positions that present a stable spatial pattern in consecutive frames and conform to the expected motion mode of CAV.

[0132] The candidate position set for continuous verification of the Top-1 matching results can be defined by the following formula (23):

[0133]

[0134] Where, j m Indicates the index of the global location estimation sequence. Let i represent the m-th key candidate location in the global location estimation sequence, M represent the maximum length of the candidate location sequence under STSWD constraints, and i s Indicates the index of the candidate position set.

[0135] For a minimum usable length of l m For each sliding window, spatial continuity can be verified by evaluating the Euclidean distance between adjacent positions, corresponding to index j. m The following formula (24):

[0136]

[0137] Where, d min This represents the minimum inter-frame displacement threshold of the keyframe sequence. The window is retained as a global candidate sub-trajectory only when all paired distances satisfy the constraints.

[0138] As the length of the input LiDAR frame increases, multiple candidate sequences that satisfy the STSWD mechanism may appear in the LiDAR-standard definition electronic map matching results. The number of candidate location sets that are located on the same road segment and pass the spatiotemporal continuity verification will increase with the increase of matching candidate points.

[0139] Figure 4 This is a schematic diagram of a global position estimation sequence obtained by matching LiDAR-standard definition electronic map descriptors and performing motion continuity checks, as provided by the present invention. Figure 4As shown, by performing context matching within the same road network topology, candidate subsequences can be obtained from the global location estimation results. These subsequences exhibit motion patterns similar to the keyframe odometry trajectories, where... Figure 4 The colored lines in the diagram represent candidate subsequences.

[0140] The set of all global candidate trajectories extracted based on STSWD constraints The following formula (25):

[0141]

[0142] in, This represents the temporal global candidate sub-trajectory that satisfies the spatial continuity condition. Note that... and its sub-trajectories All are arranged strictly in chronological order. Here, The global candidate sub-trajectories may contain overlapping positions.

[0143] Furthermore, to ensure the accuracy of the correlation between the key non-static frame sequence of odometry and the global candidate sub-trajectories, it is possible to eliminate... Redundant candidate information among global candidate sub-trajectories.

[0144] For example, after eliminating overlapping coordinates, i.e. redundant candidate information, between global candidate sub-trajectories, a global position sequence that is temporally arranged and spatially non-redundant is obtained. The following formula (20):

[0145] It should be noted that after performing the uniqueness operation to eliminate redundant and overlapping information, Candidate position nodes and their original sub-trajectory relationships will be cleared. Therefore, the unified sequence needs to be re-segmented into a unique set of non-overlapping sub-trajectories that satisfy the constraints of temporal order and spatial independence. The following formula (26):

[0146]

[0147] in, Let i represent the i-th non-overlapping sub-trajectory.

[0148] further, The location point in the image is obtained using the following formula (27):

[0149]

[0150] Among them, [s i ,e i ] represents its index range in the unified sequence, and τ is the inter-segment spatial threshold for determining the independence of sub-trajectories. This represents the i-th unique sub-track sequence.

[0151] Figure 5 This invention provides a global candidate trajectory available as a sub-trajectory. One of the schematic diagrams, such as Figure 5 As shown, the colored lines represent the original output of STSWD, and there is spatial overlap.

[0152] Figure 6 This invention provides a global candidate trajectory available as a sub-trajectory. The second diagram, as shown Figure 6 As shown, the red line represents the global candidate sub-trajectory set segmented by time sequence, and the purple line represents the ground truth. The visualization results are limited to the coordinate range [455500, 455600] × [5425800, 5425940].

[0153] In this embodiment of the invention, the global inter-frame motion consistency is maintained based on the continuity of global candidate positions, ensuring that the extracted key candidate sequences accurately reflect the vehicle motion modes.

[0154] Step 202: Based on the KD tree and the road network in the standard definition electronic map, perform a consistency check on each of the first candidate transformation matrices, and select at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices.

[0155] Here, the KD tree can be constructed using the full set of road nodes Pprior from a standard definition electronic map, which is used for efficient searching of multidimensional data.

[0156] Optionally, a better feasible transformation matrix can be searched from the first candidate transformation matrix based on the KD tree. This can be done by searching the standard definition electronic map road network for the point closest to the global position corresponding to the odometer pose, i.e., the nearest neighbor of the global position. Then, based on the nearest neighbor, the first candidate transformation matrix that meets the conditions can be determined from each first candidate transformation matrix to obtain the feasible transformation matrix.

[0157] For example, the step of performing a consistency check on each of the first candidate transformation matrices based on the KD tree and the road network in the standard definition electronic map, and selecting at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices, includes: mapping each odometer point in the original odometer sequence to a global coordinate system based on the first candidate transformation matrix to obtain the global location point corresponding to each odometer point; finding the first nearest neighbor point corresponding to each global location point in the road network in the standard definition electronic map based on the KD tree; determining the first average distance between all global location points and the first nearest neighbor point corresponding to the global location point; and performing a consistency check on each of the first candidate transformation matrices based on the first average distance, and selecting at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices.

[0158] It should be noted that the consistency check can be performed by comparing the first average distance with a preset threshold. If the distance is less than the preset threshold, the consistency check passes; otherwise, it fails.

[0159] For example, mapping the original odometry sequence to the global coordinate system using different first candidate transformation matrices will result in different global trajectories. Each location point in each global trajectory will have a corresponding nearest neighbor in the road network. The first average distance between all location points in each global trajectory and their corresponding nearest neighbors will be calculated. The first transformation corresponding to the first average distance being less than the distance threshold will be determined as a feasible transformation matrix.

[0160] Optionally, there can be one or more feasible transformation matrices. The feasible transformation matrix is ​​the first candidate transformation matrix that passes the motion consistency check.

[0161] For example, the first average distance The calculation is as follows (28):

[0162]

[0163] Where, N i Indicates the length of the original odometer sequence. This represents the nearest neighbor of the j-th transformed odometer point to the global location point in the global coordinate system in the standard definition electronic map road network, obtained through a KD-tree search constructed by Pprior.

[0164] In another example, if the first average error exceeds the distance threshold, the global candidate sub-trajectory and its corresponding transformation matrix are rejected, and the next candidate sub-trajectory is evaluated instead.

[0165] For example, the set of all transformation matrices passed by the ETMVM (Electronic Motion Consistency Verification). The feasible transformation matrix is ​​as follows: (29)

[0166]

[0167] Here, the set of transformation matrices It includes multiple feasible transformation matrices.

[0168] It should be noted that finding a feasible transformation matrix is ​​an iterative process until a sub-trajectory that can generate global consistency and reasonable spatial alignment is identified. In extreme cases, if none of the global candidate sub-trajectories meet the acceptance criteria, it is determined that reliable global localization cannot be achieved.

[0169] If set If the value is empty, it indicates that there are no valid transformation candidates. In this case, global positioning is determined to have failed, and the current local keyframe odometry and its corresponding LiDAR-standard definition electronic map position candidates cannot generate a globally consistent positioning result that satisfies the continuous motion constraint.

[0170] In this embodiment of the invention, a KD tree is used to quickly evaluate the Euclidean distance between the input candidate set and the road node set. An estimated transformation matrix verification mechanism is used to determine the feasible transformation solution, so as to avoid poor alignment between the transformed local trajectory and the road network, which would ultimately lead to global continuous positioning failure.

[0171] Step 203: Optimize the global position of the vehicle for autonomous driving based on the feasible transformation matrices.

[0172] Optionally, the original odometer sequence can be directly mapped to the global coordinate system based on the feasible transformation matrix to obtain the global trajectory, which is used for global vehicle positioning. Alternatively, the second transformation matrix can be further optimized to obtain the target transformation matrix, which is then used for global vehicle positioning.

[0173] like If the set is not empty, each feasible transformation matrix in the set can be used to project the local keyframe odometry sequence onto the global coordinate system.

[0174] Figure 7 This is one of the projection results of the transformation matrix for estimating the correspondence between global candidate sub-trajectories and local keyframe odometry provided by this invention, such as... Figure 7 As shown, green represents the ground truth and red represents the rejected trajectory. The erroneous transformation caused the projected odometer of the first 1296 frames to deviate significantly from the road network, which is a failure case.

[0175] Figure 8This is the second projection result of the transformation matrix for estimating the correspondence between global candidate sub-trajectories and local keyframe odometry provided by this invention, such as... Figure 8 As shown, blue indicates an acceptable trajectory. The feasible transformation aligns the projected odometers of the first 1337 frames with the road network, which is a successful case.

[0176] In another embodiment, if If the value is not empty, the transformation matrix can be re-evaluated based on KD nearest neighbor search and least squares error calculation to obtain the target transformation matrix, and further optimize the consistency between the transformed global trajectory and the standard definition electronic map road network.

[0177] For example, optimizing the global position of the vehicle's autonomous driving based on each of the feasible transformation matrices includes: mapping the original odometer sequence to a global coordinate system based on each of the feasible transformation matrices to obtain a first global trajectory corresponding to each of the feasible transformation matrices; finding a second nearest neighbor point corresponding to each location point in the first global trajectory in the road network of the standard definition electronic map based on the KD tree; determining a target transformation matrix based on the second nearest neighbor point; and optimizing the global position of the vehicle's autonomous driving based on the target transformation matrix.

[0178] It should be noted that using the feasible transformation matrix to map the original odometer sequence to the global coordinate system can be understood as mapping each position in the original odometer sequence to the global coordinate system one by one to obtain each global position in the first global trajectory. Then, based on the KD tree, the second nearest point corresponding to each position point in the first global trajectory is found in the road network. Based on the second nearest point, the target transformation matrix is ​​determined, and based on the target transformation matrix, the vehicle is globally located.

[0179] Optionally, the average error between each first global trajectory and the corresponding second neighbor trajectory can be calculated based on the second nearest point, and then the target transformation matrix can be determined based on the average error. Alternatively, the second candidate transformation matrix between the current odometry key non-static frame sequence and at least one global candidate sub-trajectory can be re-determined using the least squares error, and the target transformation matrix can be determined based on the second candidate transformation matrix and the second nearest neighbor.

[0180] It should be noted that if T^ is not empty, each feasible transformation matrix in the set can be used to project the local keyframe odometry sequence to the global coordinate system. Subsequently, based on PKDT nearest neighbor search and least squares error calculation, the consistency between the transformed trajectory and the standard definition electronic map road network is re-evaluated.

[0181] For example, determining the target transformation matrix based on the second nearest neighbor includes: determining a second candidate transformation matrix between the odometry key non-static frame sequence and at least one of the global candidate sub-trajectories based on minimizing the least squares estimation error; mapping the original odometry sequence to a global coordinate system based on each of the second candidate transformation matrices to obtain at least one second global trajectory; determining a second average distance between each position point in the second global trajectory and the second nearest neighbor corresponding to each position point in each of the first global trajectories; and determining the second candidate transformation matrix corresponding to the minimum second average distance as the target transformation matrix.

[0182] It should be noted that the key non-static frame sequence of the odometer and each global candidate sub-trajectory have a corresponding second candidate transformation matrix. Using each second candidate transformation matrix, the original odometer charge is mapped to the global coordinate system to obtain the corresponding second global trajectory point. The average error between all position points in the second global trajectory point and the second nearest neighbor point corresponding to each position point in the first global trajectory is used to determine the target transformation matrix.

[0183] For example, the target transformation matrix T * The calculation is as follows: (30)

[0184]

[0185] Where T represents the second candidate transformation matrix, and arg represents the second candidate transformation matrix corresponding to the minimum second average distance. This represents the trajectory of the second nearest neighbor point corresponding to the first global trajectory.

[0186] It should be noted that the target transformation matrix T * It may change as the number of input frames accumulates. An online operation mode can be adopted, and the global localization result gradually converges to the true value as the number of sensing frames increases.

[0187] In this embodiment of the invention, by verifying the average distance between the local laser odometry and the road network nodes accessible to the vehicle under the action of the feasible transformation matrix online, the solution with the lowest distance is selected as the real-time optimal transformation matrix, thereby eliminating the difference between the actual observation distance of the autonomous vehicle and the prior position of the road node on the standard definition electronic map and improving the positioning accuracy.

[0188] In this embodiment of the invention, a first candidate transformation matrix is ​​determined between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in the standard definition electronic map. Based on the KD tree and the road network in the standard definition electronic map, a consistency check is performed on each of the first candidate transformation matrices, and at least one feasible transformation matrix that satisfies the road network shape consistency check is selected from each of the first candidate transformation matrices. Based on each feasible transformation matrix, the global position of the vehicle's autonomous driving is optimized. Thus, a KD tree is constructed based on the vehicle-accessible road network nodes in the static standard definition electronic map, and a motion consistency check is performed on the local laser odometer sequence of the first candidate transformation matrix based on the KD tree. The first candidate transformation matrix that passes the consistency check is used as a feasible transformation matrix. The global position of the vehicle's autonomous driving is optimized based on the optimized transformation matrix, eliminating the difference between the actual observed distance of the autonomous vehicle and the prior position of the road node in the standard definition electronic map, and improving the positioning accuracy.

[0189] In this embodiment of the invention, a KD tree is constructed based on the vehicle-accessible road network nodes in a static standard definition electronic map, and motion consistency verification is performed on the local laser odometry sequence under the action of the candidate solution, i.e., the first candidate transformation matrix, based on the KD tree. For feasible transformation matrices that pass the consistency verification, the feasible transformation matrix is ​​used as a feasible solution of the global transformation matrix and constitutes a feasible solution set. By verifying the average distance between the local laser odometry and the vehicle-accessible road network nodes under the action of the feasible transformation matrix online, the solution with the lowest distance is selected as the real-time optimal transformation matrix, i.e., the target transformation matrix, to eliminate the difference between the actual observation distance of the autonomous vehicle and the prior position of the road nodes in the standard definition electronic map, thereby improving the positioning accuracy.

[0190] The following describes the global location optimization system for autonomous driving based on road network consistency verification provided by the present invention. The global location optimization system for autonomous driving based on road network consistency verification described below can be referred to in correspondence with the global location optimization method for autonomous driving based on road network consistency verification described above.

[0191] Figure 9 This is a schematic diagram of the structure of the autonomous driving global location optimization system based on road network consistency verification provided by the present invention, as shown below. Figure 9 As shown, the autonomous driving global location optimization system 900 based on road network consistency verification includes the following modules:

[0192] The first determining module 910 is used to determine the first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in the standard definition electronic map.

[0193] The second determining module 920 is used to perform consistency checks on each of the first candidate transformation matrices based on the KD tree and the road network in the standard definition electronic map, and select at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices.

[0194] The optimization module 930 is used to optimize the global position of the vehicle for autonomous driving based on the feasible transformation matrices.

[0195] In another embodiment, the second determining module is specifically configured to: map each odometer point in the key non-static frame sequence of the odometer to a global coordinate system based on the first candidate transformation matrix, to obtain a global location point corresponding to each odometer point; search for the first nearest neighbor point corresponding to each global location point in the road network of the standard definition electronic map based on the KD tree; determine the first average distance between all global location points and the first nearest neighbor point corresponding to each global location point; and perform a consistency check on each of the first candidate transformation matrices based on the first average distance, and select at least one feasible transformation matrix that satisfies the road network morphology consistency check from each of the first candidate transformation matrices.

[0196] In another embodiment, the optimization module 930 is specifically used for: mapping the original odometer sequence to a global coordinate system based on each of the feasible transformation matrices to obtain a first global trajectory corresponding to each of the feasible transformation matrices; searching for a second nearest neighbor point corresponding to each location point in the first global trajectory in the road network of the standard definition electronic map based on the KD tree; determining a target transformation matrix based on the second nearest neighbor point; and optimizing the global position of the vehicle's autonomous driving based on the target transformation matrix.

[0197] In another embodiment, the optimization module 930 is further configured to: determine a second candidate transformation matrix between the odometry key non-static frame sequence and at least one of the global candidate sub-trajectories based on minimizing the least squares estimation error; map the original odometry sequence to a global coordinate system based on each of the second candidate transformation matrices to obtain at least one second global trajectory; determine a second average distance between each position point in the second global trajectory and the second nearest neighbor point corresponding to each position point in each of the first global trajectories; and determine the second candidate transformation matrix corresponding to the minimum second average distance as the target transformation matrix.

[0198] In another embodiment, the autonomous driving global position optimization system 900 based on road network consistency verification further includes an odometer key non-static frame sequence determination module, specifically used for: acquiring the pose of each location point through a lidar odometer to obtain the original odometer sequence; and sampling from the original odometer sequence to obtain the odometer key non-static frame sequence.

[0199] In another embodiment, the autonomous driving global location optimization system 900 based on road network consistency verification further includes a global candidate sub-trajectory determination module, specifically used for: determining multiple global candidate locations in the standard definition electronic map; and determining at least one global candidate sub-trajectory from the multiple global candidate locations based on the continuity of each of the global candidate locations.

[0200] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 10 As shown, the electronic device may include a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communications interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 can call logical instructions in the memory 1030 to execute an autonomous driving global position optimization method based on road network consistency verification. The method includes: determining a first candidate transformation matrix between a key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in a standard definition electronic map; performing a consistency verification on each of the first candidate transformation matrices based on a KD tree and the road network in the standard definition electronic map, and selecting at least one feasible transformation matrix that satisfies the road network shape consistency verification from each of the first candidate transformation matrices; and optimizing the global position of the vehicle's autonomous driving based on each feasible transformation matrix.

[0201] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0202] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the autonomous driving global position optimization method based on road network consistency verification provided by the above methods. The method includes: determining a first candidate transformation matrix between a key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in a standard definition electronic map; performing a consistency verification on each of the first candidate transformation matrices based on a KD tree and the road network in the standard definition electronic map; selecting at least one feasible transformation matrix from each of the first candidate transformation matrices that satisfies the road network shape consistency verification; and optimizing the global position of the vehicle for autonomous driving based on each of the feasible transformation matrices.

[0203] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the autonomous driving global position optimization method based on road network consistency verification provided by the above methods. The method includes: determining a first candidate transformation matrix between a key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with the same shape in a standard definition electronic map; performing a consistency verification on each of the first candidate transformation matrices based on a KD tree and the road network in the standard definition electronic map, and selecting at least one feasible transformation matrix from each of the first candidate transformation matrices that satisfies the road network shape consistency verification; and optimizing the global position of the vehicle for autonomous driving based on each of the feasible transformation matrices.

[0204] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0205] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0206] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A global location optimization method for autonomous driving based on road network consistency verification, characterized in that, include: Determine the first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with a matching spatiotemporal motion pattern in the standard definition electronic map; Based on the first candidate transformation matrix, each odometer point in the original odometer sequence is mapped to the global coordinate system to obtain the global position point corresponding to each odometer point; Based on the KD tree, the first nearest neighbor point corresponding to each global location point is found in the road network of the standard definition electronic map; the KD tree is constructed using the full set of road nodes of the standard definition electronic map; Determine the first average distance between all global location points and their corresponding first nearest neighbors; Based on the first average distance, the first candidate transformation matrix is ​​subjected to a road network morphology consistency check to obtain at least one feasible transformation matrix that satisfies the road network morphology consistency check among the first candidate transformation matrices. Based on each of the feasible transformation matrices, the original odometer sequence is mapped to the global coordinate system to obtain the first global trajectory corresponding to each of the feasible transformation matrices; Based on the KD tree, find the second nearest neighbor point corresponding to each location point in the first global trajectory in the road network of the standard definition electronic map; Based on the second nearest neighbor, determine the target transformation matrix; Based on the target transformation matrix, the global position of the vehicle for autonomous driving is optimized.

2. The autonomous driving global location optimization method based on road network consistency verification according to claim 1, characterized in that, Determining the target transformation matrix based on the second nearest neighbor includes: Based on minimizing the least squares estimation error, a second candidate transformation matrix is ​​determined between the odometry key non-static frame sequence and at least one of the global candidate sub-trajectories; Based on each of the second candidate transformation matrices, the original odometry sequence is mapped to the global coordinate system to obtain at least one second global trajectory; Determine the second average distance between each location point in the second global trajectory and the second nearest neighbor point corresponding to each location point in the first global trajectory; The second candidate transformation matrix corresponding to the minimum second average distance is determined as the target transformation matrix.

3. The autonomous driving global location optimization method based on road network consistency verification according to claim 1, characterized in that, Determining the key non-static frame sequence of the odometer includes: The pose of each location point is acquired by a lidar odometer to obtain the original odometer sequence; The key non-static frame sequence of the odometer is obtained by sampling from the original odometer sequence.

4. The autonomous driving global location optimization method based on road network consistency verification according to claim 1, characterized in that, Determining the global candidate sub-trajectory includes: Identify multiple global candidate locations in the standard definition electronic map; Based on the continuity of each of the global candidate positions, at least one of the global candidate sub-trajectories is determined from the plurality of global candidate positions.

5. An autonomous driving global location optimization system based on road network consistency verification, characterized in that, include: The first determining module is used to determine a first candidate transformation matrix between the key non-static frame sequence of the odometer and at least one global candidate sub-trajectory with a matching spatiotemporal motion pattern in the standard definition electronic map. The second determining module is used to map each odometer point in the original odometer sequence to a global coordinate system based on the first candidate transformation matrix, to obtain the global location point corresponding to each odometer point; to find the first nearest neighbor point corresponding to each global location point in the road network of the standard definition electronic map based on a KD tree; the KD tree is constructed using the full set of road nodes of the standard definition electronic map; to determine the first average distance between all global location points and the first nearest neighbor point corresponding to each global location point; and to perform a road network morphology consistency check on the first candidate transformation matrix based on the first average distance, to obtain at least one feasible transformation matrix among the first candidate transformation matrices that satisfies the road network morphology consistency check. The optimization module is used to map the original odometer sequence to a global coordinate system based on each feasible transformation matrix to obtain a first global trajectory corresponding to each feasible transformation matrix; based on the KD tree, to find the second nearest neighbor point corresponding to each position point in the first global trajectory in the road network of the standard definition electronic map; based on the second nearest neighbor point, to determine the target transformation matrix; and based on the target transformation matrix, to optimize the global position of the vehicle's autonomous driving.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the global location optimization method for autonomous driving based on road network consistency verification as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the global location optimization method for autonomous driving based on road network consistency verification as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the global location optimization method for autonomous driving based on road network consistency verification as described in any one of claims 1 to 4.

Images