Lightweight map processing method and device for vehicle positioning

Abstract

The invention discloses a lightweight map processing method and device for vehicle positioning, and the method comprises the steps: obtaining sensor data of a vehicle, and building a lightweight frame of the vehicle according to the sensor data and road boundary information obtained by a preset map; converting a laser radar point corresponding to the lightweight frame into a UTM coordinate according to the current position and direction of the vehicle of the GNSS data, calculating the conversion from the WGS coordinate to the UTM coordinate, obtaining a laser radar frame from the conversion from the UTM coordinate to the laser radar coordinate, obtaining a boundary point in the laser radar frame by adopting a spatial index according to the laser radar point, and obtaining the boundary point according to the boundary point. According to the method and the system, the boundary points are obtained, point cloud data of 3D laser radar data are converted into navigation coordinates according to the boundary points to generate a lightweight map, the point cloud map density is reduced, the calculation amount of a map matching algorithm is also reduced, dead reckoning is combined with NDT map matching under a Pareto optimization algorithm, and the vehicle positioning precision is improved.

Description

Technical Field

[0001] This invention belongs to the field of map matching technology, and particularly relates to a lightweight map processing method and apparatus for vehicle positioning. Background Technology

[0002] During vehicle operation, a range of sensor modes are used to support positioning, including Global Navigation Satellite System (GNSS), Inertial Measurement Unit (IMU), cameras, lidar, and radar. Each sensor contributes to a rich stream of data, but their performance is often affected by weather conditions such as heavy rain, fog, or snow, which can degrade the quality and reliability of sensor data. For example, GNSS operating in challenging environments such as urban canyons or forests is susceptible to signal attenuation and multipath errors; lidar systems, on the other hand, are affected by beam scattering under rainy conditions, further leading to lower target detection accuracy. These environmental parameters introduce variations in sensor data, increasing the computational burden of real-time processing and reducing overall positioning accuracy.

[0003] To overcome these limitations of single-sensor systems, multi-sensor fusion has recently gained significant attention. In fact, it can address the specific errors of each sensor, such as the NLOS problem in GNSS, biases in IMU measurements, and accumulated errors in visual or LiDAR odometry systems. These are typically integrated into an INS (Inertial Navigation System), fused with GNSS data via Kalman filtering, to estimate the vehicle's position, velocity, and heading. Integration works well under normal GNSS conditions; however, when GNSS is rejected or the signal is weak, positioning drift occurs due to the accumulation of errors in the inertial navigation system. These errors can be mitigated using other sensors, such as the odometer, or auxiliary GNSS signals (e.g., Doppler frequencies). In situations where GNSS signals are unavailable or unreliable, high-definition (HD) maps have become a crucial asset for enhanced positioning. These maps, generated from dense point cloud data maps (DPCD-maps) produced by LiDAR, provide highly accurate global location information. On the other hand, real-time map matching is computationally expensive, posing a significant challenge. Traditional positioning methods based on extended Kalman filters often require complex linearization and simplification, reducing flexibility and robustness in dynamic environments. Therefore, there is an urgent need to provide a lightweight map processing method for vehicle positioning to solve the aforementioned technical problems. Summary of the Invention

[0004] In view of this, the present invention provides a lightweight map processing method and apparatus for vehicle positioning, and proposes a new lightweight map generation method, which greatly reduces the point cloud map density and the computational load of map matching algorithms. Specifically, the following technical solutions are adopted to achieve this.

[0005] In a first aspect, the present invention provides a lightweight map processing method for vehicle positioning, comprising the following steps: The system acquires vehicle sensor data and creates a lightweight frame for the vehicle based on the sensor data and road boundary information obtained from a preset map. The sensor data includes 3D LiDAR data and GNSS data. The road boundary information is represented by a list of four discrete points for each boundary. Two sets of discrete points are imported, one for vegetation and / or the other for the road boundary, with the leftmost lane boundary being the first set. , and the rightmost lane boundary , ; Based on the vehicle's current position and orientation in the GNSS data, the lidar points corresponding to the lightweight frame are converted into UTM coordinates. Specifically, the nearest lateral position point is found from the four boundary lists and marked as a lidar point, and the corresponding lidar point is deleted from the four discrete point lists. The WGS coordinates are converted to UTM coordinates, and the coordinates of the lidar points are transformed according to the UTM coordinates to obtain lidar frames. The WGS coordinates are World Geodetic System coordinates, and the UTM coordinates are Universal Mercator coordinates. Based on the LiDAR points, spatial indexing is used to obtain boundary points in the LiDAR frame, and the 3D LiDAR data is converted into point cloud data into navigation coordinates based on the boundary points to generate a lightweight map.

[0006] Secondly, the present invention also provides a lightweight map processing device for vehicle positioning, comprising: This invention provides a lightweight map processing method and apparatus for vehicle positioning. It acquires vehicle sensor data and creates a lightweight frame for the vehicle based on road boundary information obtained from the sensor data and a preset map. The method converts the corresponding LiDAR points in the lightweight frame to UTM coordinates based on the vehicle's current position and direction from GNSS data, converts WGS coordinates to UTM coordinates, and performs coordinate transformation on the LiDAR points based on the UTM coordinates to obtain a LiDAR frame. Boundary points are obtained in the LiDAR frame using spatial indexing based on the LiDAR points, and the 3D LiDAR data is converted from point cloud data to navigation coordinates based on the boundary points to generate a lightweight map. This reduces the point cloud map density and the computational load of the map matching algorithm. By combining dead reckoning with NDT map matching under Pareto optimization, it achieves the fusion of distance, speed, and direction without any linearization processing, thus improving vehicle positioning accuracy. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A flowchart of the lightweight map processing method for vehicle positioning provided by the present invention; Figure 2 This is a structural block diagram of the lightweight map processing device for vehicle positioning provided by the present invention. Detailed Implementation

[0009] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0010] See Figure 1 A lightweight map processing method for vehicle positioning, characterized by comprising the following steps: S1: Acquire vehicle sensor data and create a lightweight frame for the vehicle based on the sensor data and road boundary information obtained from a preset map. The sensor data includes 3D LiDAR data and GNSS data. The road boundary information is represented by a list of four discrete points for each boundary. Two sets of discrete points are imported, one for vegetation and / or the other for the road boundary. The leftmost lane boundary is also included. , and the rightmost lane boundary , ; S2: Based on the vehicle's current position and orientation in the GNSS data, the lidar points corresponding to the lightweight frame are determined, wherein the nearest lateral position point is found from the four boundary lists and marked as a lidar point, and the corresponding lidar point is deleted from the four discrete point lists. S3: Convert WGS coordinates to UTM coordinates, and perform coordinate transformation on the lidar points according to the UTM coordinates to obtain lidar frames, where WGS coordinates are World Geodetic System coordinates and UTM coordinates are Universal Mercator coordinates. S4: Based on the LiDAR points, spatial indexing is used to obtain boundary points in the LiDAR frame, and the 3D LiDAR data is converted into point cloud data into navigation coordinates based on the boundary points to generate a lightweight map.

[0011] In this embodiment, a point cloud map generation technique using GNSS, INS, and 3D LiDAR data is employed, retaining only basic features such as buildings and roads while excluding seasonal variables. This lightweight map is then combined with a dead reckoning model and a map matching algorithm based on Normal Distribution Transform (NDT). The Pareto optimization method is used to minimize the mean and variance of positioning errors, replacing the traditional Kalman filtering method and providing a robust solution to the multi-objective challenges of positioning accuracy and reliability. Based on the coarse data provided by the lightweight map, vegetation and road boundary information on the left and right sides of the vehicle can be segmented. Each segment is constrained by boundaries considered to be close to the road, depicting the edges of vegetation or other unwanted features. The leftmost and rightmost boundaries of the boundary regions contain these features. Boundary information is given as a list of four discrete points defining their respective boundaries. Then, based on the vehicle's current position and orientation from the GNSS data, the LiDAR points are converted to UTM coordinates (East and North). For each translated LiDAR point, the nearest lateral location point is found from the four boundary lists; this area is the region enclosed by the two points on the left and right boundary lists. These LiDAR points are then marked as removed from the original list, resulting in “brighter” but relevant LiDAR point frames.

[0012] It is noteworthy that prior information from a predefined map is used to segment non-road areas and vegetation, while low-cost GNSS is used to reduce the point cloud data of the LiDAR frame. The lightweight frame concept proposed in this invention is highly effective in reducing computational costs while maintaining the accuracy of NDT-based map matching algorithms. It also implements automatic road and vegetation identification algorithms to segment the region of interest for the NDT map matching algorithm. The transformation from World Geodetic System (WGS) coordinates to Universal Lateral Mercator (UTM) coordinates, and then from UTM to LiDAR coordinates, is pre-calculated, which prevents repeated transformation calculations in each iteration and greatly reduces computational overhead. The algorithm pre-stores these transformations, avoiding recalculation for each LiDAR frame. In the context of assisted driving vehicle navigation, accurate and reliable positioning is required. Most traditional positioning methods combine various sensor input data to continuously estimate the vehicle's position and orientation. The most widely used is the dead reckoning model, which utilizes airborne sensors to measure wheel speed and yaw rate. This model calculates the position and heading in navigation coordinates and provides a basic reference for data from other sensors, such as LiDAR.

[0013] It should be understood that by acquiring vehicle sensor data and creating a lightweight frame for the vehicle based on the sensor data and road boundary information obtained from a preset map, the LiDAR points corresponding to the lightweight frame are converted to UTM coordinates based on the vehicle's current position and direction from the GNSS data. WGS coordinates are then converted to UTM coordinates, and the LiDAR points are further transformed using the UTM coordinates to obtain a LiDAR frame. Boundary points are obtained from the LiDAR frame using spatial indexing based on the LiDAR points, and the 3D LiDAR data is converted from point cloud data to navigation coordinates based on these boundary points to generate a lightweight map. This reduces the density of the point cloud map and the computational load of the map matching algorithm. Under the Pareto optimization algorithm, dead reckoning is combined with NDT map matching, achieving the fusion of distance, speed, and direction without any linearization processing, thus improving vehicle positioning accuracy.

[0014] Optionally, converting the 3D LiDAR data into point cloud data and then into navigation coordinates based on the boundary points includes: The point cloud data of the lightweight frame is converted into navigation coordinates, and the converted point clouds are then superimposed. The corresponding expression is: (1) in, These are the coordinates of each point in the lightweight frame. It is the translation vector from the lidar frame to the navigation frame. It is the rotation matrix between the lidar coordinates and the navigation coordinates. The definition expression is: (2) in, , and It is heading, pitch, and roll information. and After point clouds exist in the navigation frames, lightweight frames are accumulated together to form a lightweight map.

[0015] In this embodiment, based on each translated LiDAR point, the nearest lateral position point is found from the four boundary lists. This region is the area enclosed by two points on the left and right boundary lists. These LiDAR points are then marked as removed from the original lists, thus generating a LiDAR point frame. Notably, prior information from predefined maps (such as map navigation software) is used to segment non-road areas and vegetation, while low-cost GNSS is used to reduce the point cloud data of the LiDAR frame. The lightweight frame concept proposed in this invention is highly effective in reducing computational costs while maintaining the accuracy of NDT-based map matching algorithms. It also implements automatic road and vegetation identification algorithms, which can be integrated into the lightweight frame to segment the region of interest for the NDT map matching algorithm. Using the Applanix-LVX system with RTK correction installed on high-end vehicles, the 3D LiDAR coordinates are converted into navigation coordinates. This precise conversion is used to convert the point cloud data of the lightweight frame into navigation coordinates. These converted point clouds are stacked to create a lightweight map.

[0016] Optionally, determining the lidar point corresponding to the lightweight frame based on the vehicle's current position and orientation in the GNSS data includes: The vehicle's position is estimated based on its current location, speed, and direction to establish a dead reckoning model. Wheel speed and yaw rate from onboard sensors are used to calculate the position in navigation coordinates, as well as the vehicle's heading between LiDAR coordinates and navigation coordinates. hour along The expression for dead reckoning of axis position is: (3) in, Indicates time k along The dead coordinates of the axis position. This represents the heading angle of the vehicle at time k-1. Indicates time difference, for Time position estimate, The time interval between two measurements. for Observe the speed at all times. yes Observe the azimuth angle at all times. The true location at any moment Represented as: (4) in, for The true speed of a moment for The true azimuth angle at any given moment, measured from the wheel speed sensor. and direction , It is the vehicle's heading angle. It is the heading angle. It is the vehicle's sideslip angle. The expression is: (5) in, It is the yaw rate, the measured value is... and Given the corresponding noise term and The damage is such that the measured values ​​follow a variance of . and The zero-mean Gaussian distribution; Based on the given dead reckoning data and the observed measured velocity and the direction being measured The expressions for calculating the eastward and northward velocities are as follows: (6) (7) in, It is the speed being measured. Indicates the direction being measured.

[0017] In this embodiment, the dead reckoning model estimates the vehicle's position based on its previous position, speed, and direction to establish the dead reckoning model. It utilizes wheel speed and yaw rate provided by onboard sensors similar to an inertial navigation system to calculate the position in navigation coordinates and the vehicle's heading between LiDAR and navigation coordinates. The yaw rate is negligible due to the vehicle's small lateral distance, meaning a small sideslip angle, and can be approximated by the heading angle. The measurements are given by and , corrupted by corresponding noise terms and , which follow zero-mean Gaussian distributions with variances and , respectively. To calculate the eastward and northward speeds, these components are first derived from the given dead reckoning data, using the observed speed and observed vehicle direction. The tire slip ratio is not considered; without it, speed errors occur because the tire rotation speed is not exactly the same as the wheel speed, leading to accumulated position errors. Simultaneously, the yaw and sideslip angles of the yaw rate gyroscope are not considered when solving for the heading angle, resulting in heading errors. Therefore, the increase in position error is purely a matter of time. This further justifies deploying the map matching algorithm, as it reduces accumulated error and establishes a Pareto-based approach to estimate the inherent errors of the dead reckoning model itself. Furthermore, the dead reckoning model operates at a frequency dependent only on two sensors, wheel speed and yaw rate sensors. It is set to 50 Hz, higher than the frequency of the map matching algorithm.

[0018] Optionally, the deviation in dead reckoning is quantified by the difference between the estimated position and the actual position, and the deviation is calculated by assessing the expected difference between the dead reckoning estimate and the actual position. The corresponding mathematical expression is: (8) The expression to be replaced is and for: (9) To further simplify: (10) Among them, preset It is zero, in time The estimate is unbiased, with a preset direction angle. Additive white Gaussian noise If destruction occurs, the observed direction is represented as: (11) in, Indicates variance For zero-mean Gaussian noise, the expected value of the noise angle cosine can be approximated by utilizing the properties of the Gaussian distribution, thus approximating the expected value of the low-noise variance as: (12) Substituting formula (12) into formula (10) to calculate the deviation, the result is: (13) By introducing correction factors based on system characteristics or noise factors, the deviation is rewritten as: (14) in, Considered as adjustments related to system characteristics or empirical calibration, the deviation in the dead reckoning model is based on the expected value of the deviation between the estimated position and the actual position.

[0019] Optionally, the variance of the estimation error in the dead reckoning model can be determined and evaluated by analyzing the propagation of velocity and orientation measurement noise within the system. The change in position estimation is influenced by the measured velocity. and direction The variance of dead reckoning position estimates can be expressed as: (15) Utilizing the characteristics of Gaussian noise, the variance of the product of velocity and the cosine of the direction angle is: (16) Among them, Indicates the expected value. Approximate to The subsequent approximations are applicable and should be used: (17) In this process, variance is incorporated into the fusion process, and dead reckoning estimates are weighted relative to distance measurements.

[0020] In this embodiment, the deviation in dead reckoning estimation is quantified by the difference between the estimated position and the true position. To calculate the deviation, the expected difference between the dead reckoning estimate and the true position is evaluated. The deviation in the dead reckoning model is based on the expected value of the difference between the estimated and true positions, which is a function of the nonlinear effect of noise on the cosine function in the heading angle. This deviation term depends on the vehicle speed, time interval, the heading angle itself, and variations in noise in the heading measurement. The error state-constrained Pareto method aims to determine ideal fusion parameters and strike a balance between minimizing position and heading errors in the estimation and managing noise variance. The goal of this method is to reduce the mean square error of the estimation while keeping the estimation error within acceptable limits and considering the uncertainties caused by sensor noise.

[0021] It should be noted that the goal of the error state-constrained Pareto method is to estimate the position and heading errors in the dead reckoning model and obtain error measurements of the northeast position, velocity, and heading. This method optimizes the fusion parameters to minimize the estimation error and noise variance while ensuring that the error state constraints are met, maintaining a balance between two competing objectives: the northeast position. Using this optimization framework, the fusion parameters are selected to reduce positioning and heading errors while controlling the overall noise variance. The error constraints ensure that the estimation error does not exceed a certain maximum value and that the sum of the sensor fusion weights does not exceed 1. This achieves a balance between reducing the mean square error (MSE) of the estimation and minimizing the uncertainty caused by sensor noise, thus ensuring robust positioning of the assisted driving vehicle.

[0022] Optionally, the WGS coordinates are transformed to UTM coordinates, and the lidar points are then transformed according to the UTM coordinates to obtain lidar frames, including: The Normal Distribution Transform (NDT) algorithm is used to convert the lidar point cloud into navigation coordinates to calculate the reference map. Based on the reference map, the lidar frame is aligned with the navigation frame to achieve map matching. Formula (18) represents the lidar frame. To navigation frame Transformation: (18) in, Indicates intra-frame points coordinates This represents the corresponding point in the lidar coordinate system. This represents the rotation matrix of the lidar relative to the navigation frame direction. The rotation matrix represents the vector of translation between the lidar and navigation coordinate systems. Defined as: (19) in, , and Indicates roll angle, pitch angle, and yaw angle. , , , and , ; In this embodiment, an Applanix-LVX system with RTK correction, installed on the vehicle, is used to convert 3D LiDAR coordinates to navigation coordinates. This precise transformation converts point cloud data from lightweight frames into navigation coordinates. These converted point clouds are then superimposed to form a lightweight map. Since seasonally changing vegetation has been removed, map density is reduced. Furthermore, because vegetation is constantly changing, the map may not necessarily be updated. For the former, the advantage is that the computational load on the map matching algorithm is significantly reduced based on the lower-density light frames; for the latter, since neither the map nor the frame has changing vegetation, positioning accuracy can be improved.

[0023] Optionally, each lidar (scanning) point Converted into navigation frames and aggregated to form a point cloud map : (20) in, It is an aggregation to generate a point cloud map. Total number of points, for point cloud map Optimization includes: Vegetation removal: Filter out points corresponding to vegetation; Map downsampling: Downsamples the map and retains the points most important for accurate positioning.

[0024] In this embodiment, the Normal Distribution Transform (NDT) algorithm utilizes a reference map to calculate the reference map by converting the LiDAR point cloud into navigation coordinates. This transformation aligns the LiDAR with the navigation frame, thereby providing appropriate map matching for positioning.

[0025] Optionally, the normal distribution transformation (NDT) algorithm is used to convert the lidar point cloud into navigation coordinates to calculate the reference map, including: The vehicle's position is calculated using a given dead reckoning model. The estimated position at that time is: (twenty one) in, express The estimated eastward and northward positions at any given time. express The estimated speed at any time express The estimated course at any given time express Time and The time interval is determined by finding the optimal alignment transformation between the current lidar scan and the reference image. The transformation expression is: (twenty two) in, For rotation angle, , These are the translation components in the east and north directions, respectively. The NDT map matching process determines the optimal transformation. This will reduce the number of current scan points. and reference image The problem of minimizing the difference between them is described as follows: (twenty three) in, Let i represent the i-th point on the lidar frame, where Should be ; Indicates the weights of the NDT estimate. This indicates the number of points in the current scan, determining the optimal transformation. Used to correct dead reckoning position estimates, after implementing NDT-based map matching, the corrected position estimate can be expressed as: (twenty four) A weighted fusion algorithm is used to fuse dead reckoning and NDT estimates. Timing position estimation Represented as: (25) in, and Indicates the fusion weights, satisfying the conditions. The difference in the merged location estimates is expressed as: (26) in, Indicates that the vehicle is in The actual location at any given moment.

[0026] In this embodiment, the following operations were performed to optimize the map and make the positioning usable: (1) Vegetation removal: points corresponding to vegetation were filtered out. This is an important stage to reduce map density and maintain uniformity across different seasons. Eliminating seasonal vegetation enhances the reliability of the map positioning under a wide range of environmental conditions; (2) Map downsampling: by downsampling the map, only the points most important for accurate positioning are retained, which is another way to further reduce computational complexity. Downsampling technology ensures that the map retains key features, provides effective navigation, and still maintains computational efficiency, generating a lightweight point cloud map. By minimizing the existence of variable features such as seasonal vegetation, the stability of the reference map over time can be maintained.

[0027] Optionally, the uncertainty estimates in the NDT correction estimate are combined with the uncertainty estimates in the dead reckoning estimate, given the estimation error. exist The covariance at time k+1 can be propagated using formula (27): (27) in, This represents the covariance at time k+1. The covariance estimated for dead reckoning. The covariance is based on NDT map matching estimation. The covariance estimated for dead reckoning. The cross-covariance between the NDT-based estimate and the dead reckoning estimate is calculated using the following formula: (28) in, This represents the estimated position after dead reckoning. The expected value estimated by dead reckoning. This represents the expected value estimated after NDT correction. This is the expected value estimated after NDT correction. This represents the expectation operator for calculating the mean of a random variable. It is the cross covariance of the two estimation errors to reflect the linear relationship between them.

[0028] In this embodiment, the transformed point cloud is accumulated to generate a lightweight map, which serves as a reference for the NDT algorithm. This map is optimized by removing vegetation to reduce density, thereby improving computational efficiency. Since these steps are for map creation, where the LiDAR scans are matched, high-precision location correction and localization are allowed through an NDT-based map matching process. The Normal Distribution Transform (NDT) algorithm aligns the current LiDAR scan with the reference map to find the optimal transformation, minimizing the difference between the transformed scan points and the reference map points.

[0029] Optionally, the error of the fused location estimation is obtained. Time error The norm is less than or equal to Error at time, system state transition matrix The spectral radius is less than 1: and (29) in, Let be the spectral radius of the system state transition matrix. and This represents the error between the fused estimate and the true location, ensuring... This means that the estimation error will not diverge over time.

[0030] In this embodiment, how uncertainty in one estimate interacts with error in another, and the cross-covariance, are crucial for understanding and mitigating these compound uncertainties that combine multiple estimation techniques. Stability is critical for the sustained performance of assisted driving vehicles, especially in dynamic and uncertain environments. Notably, combining the map generation process with NDT-based map matching improves positioning accuracy by preserving the reference map in an accurate and computationally efficient manner. By reducing map density and minimizing unnecessary elements, the computational load is reduced, allowing for real-time processing. This fusion of NDT-corrected estimation and dead reckoning yields a more robust and stable position estimate—essential for autonomous navigation and vehicle localization. In fact, the extended accuracy of this approach not only directly contributes to improving the accuracy of position estimates but also provides additional benefits to the reliability of the entire autonomous setup by making it more robust to environmental changes or sensor noise. Therefore, this joint approach offers a broad solution to the positioning challenges in assisted driving vehicle applications, significantly improving upon traditional single-sensor or single-method approaches.

[0031] Specifically, multiple data acquisition systems were employed to capture point cloud data for high-precision mapping and positioning. The HDL-32E 3D LiDAR sensor mounted on the vehicle provided the required high-frequency point cloud data at 20 Hz, while algorithms developed by ROS processed this data in real time on an industrial-grade PC. Vehicle attitude updates were performed in response to positioning at 5 Hz, while ground truth was determined by an integrated GNSS and IMU system, and a planar LVX. For large-scale data collection, the RIEGL VMX-2HA mobile laser scanning system was used, providing highly accurate 3D data with an accuracy of 8 mm and 5 mm, acquiring up to 1.1 million measurement data points per second due to its dual-scan mechanism. Fixed GPS stations were also used to establish control points to enhance georeferencing, ensuring precise alignment of the data with real-world coordinates. The MLS system operated at speeds up to 60 km / h, achieving a point density of approximately 7500-8500 points per square meter. This extremely high point density allowed even the finest details to be represented in the data, providing the granularity required for mapping and subsequent analysis. Four different independent MLS datasets were used and provided in LAS file format. Dataset 1 is a 30-meter-long urban road corridor segment. The study area contains 1,827,963 points, with a density of 8,500 points / m. 2 Dataset 2 consists of other road segments with a length of 176.5 m, containing 9,850,578 points. These datasets span different segments of the highway, each with different localization and mapping challenges. The high level of point density in these datasets allows for thorough detail scanning of road markings, curbs, and even building facades, which can easily capture details and provide robust analytics for applications in autonomous navigation. Specifically, real-time performance is the most critical issue in assisted driving vehicle localization. The rotation rate of the Velodyne LiDAR and the update frequency of the algorithm ensure that the vehicle's attitude is generated at a frequency of 5Hz, thereby achieving continuous and responsive localization. The computational burden of processing large datasets in real time using an industrial PC is mitigated, ensuring both the accuracy and speed of the algorithm. A powerful data acquisition and localization system is proposed, integrating a high-performance LiDAR, a precision GNSS / IMU, and a robust data processing platform. These datasets are invaluable for further validation and continuous improvement of the localization algorithm in challenging real-world scenarios.

[0032] Testing was conducted on a loop of approximately 2.7 km, with a 0.5 km section specifically selected for analysis due to its complex environment, including highly complex architecture and diverse vegetation. The performance of the brightness map was assessed by identifying five invariant features from 1D, 2D, and 3D structures. Regions 1 and 2, containing buildings and vegetation, were considered, and the impact of the environment on algorithm accuracy was analyzed in detail. These tests were performed under conditions common in the region, including heavy rain and fog. These adverse conditions introduced significant variability into the LiDAR point cloud, primarily due to the scattering effect caused by raindrops and reduced visibility under fog conditions. The density of raindrops and the diffusion effect of light in fog necessitate frequent map updates and adaptations for accurate real-time localization. Indeed, the ability to consider and adapt to such environmental changes is a crucial factor in evaluating the algorithm.

[0033] To simulate real-world driving conditions, a shuttle bus traveled along a circular road under normal acceleration, deceleration, and steering patterns. During testing, the longitudinal speed was maintained between 0 and 25 km / h by adhering to the speed limits of the circular road. Therefore, the test focused on the robustness of the map matching algorithm under adverse weather conditions, particularly rain, which affects the data acquisition from the LiDAR sensors and the processing required for accurate vehicle localization. The algorithm was initially executed in clear weather to provide a baseline for its practical localization performance, and this result has been used as a reference for further evaluation. Subsequent experiments were conducted to test the combination of rainfall with LiDAR scanning and feature extraction. The test also explored how weather affects the ability to distinguish between static features (such as buildings) and dynamic environmental features (such as vegetation). Considering their appearance properties in rain, this part of the evaluation is crucial for determining whether cost-effective navigation can be relied upon in variable weather conditions, especially in areas where heavy rainfall is expected.

[0034] The proposed brightness map generation was tested under adverse weather conditions, specifically during a period of clear skies followed by heavy rain. Two types of point cloud maps were created: a dense map incorporating data from all available LiDAR frames, and a lightweight map that retained only invariant features such as buildings and roads, while filtering out transient objects like vegetation. The dense map contained all the information from the LiDAR points and provided a complete representation of the environment. Notably, the lightweight map retained less point density because weather-sensitive points were removed. Compared to the dense map, vegetation, characterized by areas 1 and 2, showed a greater reduction in point density in the brightness map. This was particularly beneficial during heavy rain, as this often worsens LiDAR data due to scattering and reflection effects. The brightness map, by removing these weather-sensitive points, presents fewer artifacts and provides a clearer view of the campus infrastructure.

[0035] Therefore, the size of the point cloud map was reduced from 118 MB in the dense map to 48.3 MB in the lightweight map, representing a 59% reduction in density. Furthermore, although vegetation changes necessitate seasonal updates to the dense map, the coarse localization map generated based on a low-cost GNSS receiver exhibits good robustness to seasonal and weather changes. This allows for the filtering out of different vegetation points, resulting in a denser map and reduced update frequency, making the map matching algorithm more efficient, reliable, and robust even under very severe weather conditions such as heavy rain.

[0036] Specifically, processing dense LiDAR frames effectively filters out vegetation points, thereby reducing point cloud density. This filtering step is crucial because the ability of LiDAR to capture accurate data is severely impacted during heavy rain. LiDAR scans obtained under heavy rain conditions appear relatively sparse and darker, and the presence of noise points increases compared to LiDAR scans collected in clear weather. Notably, these noise points are primarily generated by raindrops reflecting off the laser beam, accounting for only 1.22% of the total points received per scan—a negligible number. Under heavy rain conditions, the number of measurement points decreases, intensity values ​​drop, and some noise points randomly exist in the scan due to laser reflection from raindrops. Table 1 summarizes the statistics for three other rainfall scenarios in the environment, showing that, on average, 23% of points are lost per scan cycle. However, even after losing these points, the impact of noise points generated by reflected raindrops is very small, further reinforcing the strength of the filtering method under weather conditions.

[0037] Table 1 Rainfall Impact Assessment The reduction in the number of receiver points not only affects the quantity and quality of features relevant to good localization that can be extracted, but also the cumulative ground reflectance features from approximately 200 scans during the localization test. In contrast, some areas showed no ground reflectance features during rainfall, while others remained unchanged. This is because of the varying permeability of surface materials; this is why some areas, such as roadside grass, easily allow groundwater to seep into the soil and show less change in reflectance during rainfall. Table 2 shows the corresponding number of features for each scenario, with the number of observable features decreasing significantly as rainfall intensity increases. However, features less affected by rainfall show minimal change; while falling raindrops may absorb some lidar points, they are unlikely to absorb all points impacting the surface.

[0038] When responding to current fair weather conditions, the LL method has the shortest processing time, averaging 107.5 ms, while the LD and DD methods have higher processing times, at 136.1 ms and 243.8 ms, respectively. The processing time increases for all methods as rain increases from light to heavy, with the DD method reaching a critical peak of 284.7 ms during heavy rain. This behavior changes beneficially when using the ensemble Pareto method: efficiency is significantly improved across various methods and conditions. This is particularly important when considering the DD method for heavy rain, where the processing time decreases from 284.7 ms to 104.4 ms. The optimization highlights the efficiency of the Pareto method in finding a suitable balance between computational efficiency and accuracy; therefore, it may be suitable for real-time map matching under adverse weather conditions.

[0039] Table 2. Characteristic observations under different weather conditions After preprocessing, an ensemble localization algorithm was implemented using lightweight and dense maps generated under heavy rain conditions. Lightweight frames and lightweight maps are denoted as LL, lightweight frames and dense point cloud maps as LD, and dense frames and dense point cloud maps as DD. Lightweight frame images can be accurately registered onto lightweight frame images or dense maps, similar to dense frame image registration on dense maps. When responding to the current fair weather conditions, the LL method has the shortest processing time, averaging 107.5 ms, while the LD and DD methods have higher processing times, at 136.1 ms and 243.8 ms, respectively. The processing time of all methods increases from light rain to heavy rain, with the DD method reaching a critical peak of 284.7 ms during heavy rain. This behavior changes beneficially when using the ensemble Pareto method: efficiency is significantly improved across various methods and conditions. When the DD method is considered for heavy rain, the processing time for heavy rain decreases from 284.7 ms to 104.4 ms. The optimization emphasizes the efficiency of the Pareto method in finding a proper balance between computational efficiency and accuracy; therefore, it may be suitable for real-time map matching under adverse weather conditions.

[0040] Table 3 summarizes the comparison of processing times under different environmental conditions and map processing methods, while emphasizing the efficiency of individual and combined Pareto methods. The data shows that the Lightweight Frame and Lightweight Map (LL) method has the shortest processing time under good weather conditions, averaging 107.5 ms with a root mean square (RMS) of 32.1 ms. In contrast, the Lightweight Frame and Dense Map (LD) method has an average processing time of 136.1 ms and an RMS of 47.4 ms, while the Dense Frame and Dense Map (DD) method has the longest processing time, averaging 243.8 ms with an RMS of 65.3 ms. The processing time increases for each method as the weather worsens. During heavy rain, the LL method averages 125.2 ms with a standard deviation of 37.39 ms. On the other hand, the average times for the LD and DD methods under the same weather conditions are 155.5 ms and 284.7 ms, respectively, with root mean square values ​​of 54.3 ms and 76.2 ms, respectively. This increase is due to the increased complexity and noise caused by severe weather, which will ultimately affect frame quality and map matching accuracy.

[0041] The integrated Pareto method reduced processing time under all conditions and methods. Specifically, under heavy rain conditions, the average time for the LL method decreased from 125.2 ms to 70.8 ms, the LD method from 155.5 ms to 81.5 ms, and the DD method from 284.7 ms to 104.4 ms. This pattern remained consistent regardless of weather changes. For example, under favorable weather conditions, the LL method reduced time to 60.8 ms, the LD method to 71.5 ms, and the DD method to 89.3 ms. This not only demonstrates accuracy but also proves the high processing efficiency of the method employed in this paper. In summary, the data above shows that lightweight frames reduce computation time by more than 43.4%, and it also indicates a slight reduction in the root mean square value for both the LL and LD methods. While the use of lightweight maps did not significantly reduce processing time, it did effectively reduce map size. Currently, the integrated Pareto method effectively optimizes real-time map matching systems, especially under challenging conditions, because it balances computational performance and accuracy.

[0042] Table 3 Comparison of Processing Times Table 4 shows a comparison of the positioning estimation errors of different settings in the map matching algorithm under different weather conditions, using three different methods: (1) Lightweight Frame and Lightweight Map (LL), (2) Lightweight Frame and Dense Map (LD), and (3) Dense Frame and Dense Map (DD). For this comparison, error metrics will include AME and RMSE. Since positioning performance is evaluated as balanced under different environmental conditions, an integrated Pareto method is applied to optimize the trade-off between these error metrics.

[0043] Table 4 Comparison of Positioning Estimation Errors By fusing the dead reckoning model with an NDT-based map matching algorithm for position estimation, thus, for a given time step... The error can be Defined as the estimated position used in dead reckoning and NDT The estimated location used by NDT With respect to the actual location of the vehicle The difference between them: (30) in, For time step The estimated position of the vehicle in the east and north directions at that time. To estimate the vehicle's true location in the east and north directions, this method uses a fusion of dead reckoning estimation and NDT-based map matching to determine the location. and NDT-based map locations and fusion weight and , so that: (31) in, The dead reckoning position estimate is derived from... Calculated; yes The speed of time, It is the time interval between time steps. yes Estimated heading angle at that time It is an NDT-based location estimate obtained by matching LiDAR scans with a reference (land) map, utilizing the ground truth east and north positions of the Applanix LVX system. and The results show that the lidar frame achieved a positioning accuracy comparable to the original dense point cloud.

[0044] The above comprehensive analysis of the error performance of various sensor fusion methods reveals clear patterns in error distribution across ranging, dead reckoning, and the proposed Pareto sensor fusion method. In the LL scenario, the proposed Pareto sensor fusion exhibits a lower cumulative error distribution compared to ranging and dead reckoning, indicating its superior performance in minimizing absolute position errors. For the LD scenario, although the error level is higher than in LL, the error level in the empirical cumulative distribution function (CDF) of the simulated error data is also more favorable for the proposed Pareto sensor fusion. With further reductions in the error rate, the DD method extends the advantages of the proposed Pareto sensor fusion, achieving the lowest cumulative error among all methods. These observations are further confirmed by the root mean square error (RMSE) values, as the proposed Pareto sensor fusion produces the minimum RMSE value in each case, confirming its ability to generate accurate and reliable position estimates. These results further enhance the strength of the proposed Pareto sensor fusion, effectively improving error performance across different orders of magnitude and complexities of error in sensor data.

[0045] Under favorable weather conditions, the lightweight frame versus brightness map setup exhibits the lowest positioning error, thus providing a fair baseline for positional accuracy: it balances computational efficiency with reduced frame and map density to avoid overfitting. However, under favorable conditions, denser setups that provide detailed environmental features do not offer significant accuracy improvements. Positioning performance deteriorates across all configurations as weather conditions worsen. Moderate and heavy rain cause substantial increases in positioning error, with more pronounced increases observed in denser configurations (lightweight frame versus dense map and dense frame versus dense map). This is likely because the LiDAR sensor interacts more with the rain, meaning transient environmental features such as raindrops and wet surfaces introduce noise.

[0046] In light rain, both AME and RMSE increased by approximately 50% due to the adverse effects of severe weather on positioning accuracy. Despite these increases, the Pareto method remained resilient, limiting the errors of LL and LD to approximately 0.0161 m and the error of DD to approximately 0.0168 m. This validated the reliability of the method in dynamic environments. Under moderate rain conditions, the performance gap widened, with AME values ​​ranging from 0.0498 to 0.0515 m and RMSE values ​​ranging from 0.0474 to 0.0485 m. The Pareto method continued to demonstrate superior performance, particularly the DD method, which benefited the most from optimization and exhibited the lowest error of 0.0278 m.

[0047] It can be seen that the most challenging condition for heavy rain is a sharp increase in AME and RMSE. AME varies between 0.1002 and 0.1022 m, while RMSE is slightly lower, between 0.0934 and 0.0965 m. Under these adverse conditions, the integrated Pareto method shows a significant reduction in the range of error increase within a relatively low range (between 0.0564 and 0.0566 m). This indicates that, compared to other methods, the Pareto method provides stable and more accurate positioning under harsh environmental conditions. Therefore, under severe weather conditions, frame and map density do not affect positioning accuracy as much as the optimization methods used, making robust optimization techniques crucial in harsh environments.

[0048] As can be expected from the relationship between the two metrics, RMSE is slightly smaller under all conditions. Due to the larger error, RMSE describes the consistency of error values, while AME provides a broader view of the error magnitude over time. However, the most important conclusion is that the ensemble Pareto method consistently outperforms both AME and RMSE, thus highlighting the efficiency of this multi-objective optimization strategy in minimizing localization errors.

[0049] Severe and adverse weather conditions, especially rain, have a more profound impact on both methods and metrics, leading to higher localization errors. However, the integrated Pareto method results in a slightly slower increase in the error rate, which may indicate its robustness to changes in environmental conditions. This could be explained by its ability to simultaneously optimize multiple objectives, such as accuracy and variance. Furthermore, different mapping methods, including lightweight frames and luminance maps (LL), lightweight frames and dense maps (LD), and dense frames and dense maps (DD), show little difference in clear weather conditions but exhibit higher degradation under rainy conditions. Compared to LL and DD, the LD method has slightly lower AME and RMSE values, indicating that the combination of lightweight frames and dense maps achieves an effective balance between computational efficiency and information richness, especially in light to moderate rain scenarios.

[0050] Therefore, comparative analysis confirms the effectiveness of the integrated Pareto method in reducing positioning errors under different methods and environmental conditions. While AME and RMSE remain useful metrics, the Pareto optimization method performs better across a wide range of conditions, exhibiting lower errors and significantly better robustness to adverse weather conditions. Considering that reliability and adaptability are the two most important issues, this highlights the potential of the integrated Pareto method as a real-world positioning solution for assisted driving vehicles. Of course, in more complex environments (such as canyons or dynamic obstacles), further refinement can be achieved through more focused methods and applications, thus partially extending its practical applicability.

[0051] Lightweight frames with lightweight map configurations remain fairly stable with low positioning errors even in light rain; however, they suffer some accuracy loss even in moderate and heavy rain. Nevertheless, they still outperform denser, more noise-resistant positioning methods. While dense maps may offer benefits under ideal conditions, lighter approaches (especially lightweight frames with lightweight maps) may provide stronger positioning performance in adverse weather conditions such as heavy rain, as noise can compromise the integrity of dense point clouds.

[0052] The east and north position errors of the three selected frame and map density configurations: Light frame with Light map (LL), Light frame with Dense map (LD), and Dense frame with Dense map (DD). The results show all four different weather conditions: sunny, light rain, moderate rain, and heavy rain. The variation of the east position error of the configuration over time reflects the fluctuation of positioning accuracy under environmental changes. (1) LL (Light Frame with Light Map) configuration, in which the LL (Light Frame with Light Map) configuration performs stably under good weather conditions, with the east position error generally less than 0.5 meters. Its performance is roughly the same under light rain conditions. However, under light rain and heavy rain conditions, its east position error begins to deteriorate, sometimes exceeding 0.75 meters. Although the LL configuration is robust under good conditions, its sparse mapping cannot capture and compensate for dynamic environmental changes (such as raindrops and wet surfaces), resulting in a large position accuracy deviation. (2) LD (Light Frame with Dense Map) configuration, in which the LD configuration has a slightly higher error under severe weather conditions. The performance is relatively stable in sunny weather, with a position error of around -0.25 to 0.25 m. Light rain and heavy rain cause large fluctuations in power and position error. In most heavy rain conditions, the positioning error exceeds 0.5 m. The increased map accuracy captures transient features such as raindrops and reflections that the positioning algorithm cannot filter out uniformly, thus reducing accuracy. (3) DD (Dense Frame with Dense Map) configuration, where the maximum position error value under different weather conditions is the DD configuration. In good weather, the error is very close to the values ​​of the other two configurations, mostly less than 0.5 m. However, as the rain begins to get heavier, these position errors become unstable, mostly exceeding 0.75 m, and approaching 1 m in heavy rain. The combination of dense frames and dense maps greatly amplifies environmental noise and captures too many transient or irrelevant features, such as artifacts caused by rain. This will lead to overfitting and a serious loss of positioning accuracy.

[0053] Comparative analysis shows that the LL configuration exhibits the most stable performance, as the location error remains constant and lower compared to other configurations, regardless of whether it's sunny or in light rain. This means that the frame and map density in LL optimally balance environmental performance, thus resisting noise and making it the most reliable for general use, especially under adverse conditions. Conversely, LD and DD configurations can provide finer environmental identification details but are more susceptible to environmental noise, particularly during heavy rain. Dense configurations introduce redundant data points, significantly increasing noise and reducing positioning accuracy. This means that excessive detail in the environmental model is detrimental under dynamic and noisy conditions, as the positioning system cannot filter out useful information from the noise. Therefore, LL performs best under all weather conditions, especially under highly disturbed conditions such as heavy rain. While higher-density LD and DD can capture more detail, they are also more susceptible to noise, leading to significant errors during weather degradation. Appropriate selection of frame and map accuracy is crucial for achieving optimal positioning accuracy, at least in noisy environments, especially in highly dynamic and challenging outdoor conditions.

[0054] By comparing different analyses, it can be observed that rainfall significantly increases positioning errors across all configurations, with the severity of these errors increasing with rainfall intensity. However, the most consistent performance is provided by the LL configuration, as the error deviation is smaller in both sunny and rainy conditions. The lightweight frame and map settings are more effective in handling noisy and dynamic environments without capturing irrelevant details that interfere with positioning accuracy. However, in heavy rain, the LD setting is more sensitive to rain, while the dense map provides more environmental detail. It may be useful under ideal conditions, but in severe weather, it introduces more noise, leading to greater positioning errors. This results in the highest positional error across all weather types in the DD configuration, which uses both dense frames and a dense map. Such a configuration, especially in rainy conditions, captures too much noise, making it unable to maintain good positioning and thus the least suitable configuration for use in challenging environments. Therefore, the LL configuration is the most robust and reliable in positioning performance under various weather conditions, especially in rainy conditions. Denseer configurations, whether involving dense maps or frames, or both, are more prone to noise and positional errors, especially in rainy conditions. These results bring us back to the question of optimizing positioning systems by balancing environmental detail and noise immunity. Therefore, lightweight frames and map settings may perform better when dealing with dynamic and challenging outdoor conditions, which are often accompanied by rain or other types of environmental disturbances.

[0055] It is worth noting that these results provide important insights into the performance and optimization possibilities of sensor fusion methods in handling time and location estimation errors. Removing vegetation from LiDAR point clouds to reduce point cloud density and improve computational efficiency is a highly effective strategy. This reduction, especially for lightweight frame configurations, significantly reduces the processing load required by the mapping matching algorithm. Although vegetation removes some low-quality features, it does not significantly affect localization accuracy because structural features (such as buildings) dominate the map matching process. These features are more stable and invariant, thus providing a more reliable reference for localization.

[0056] Calculations show that using lightweight maps alone offers limited benefits in terms of increased processing time, while combining them with lightweight frame generation reduces computational requirements and further mitigates challenges under adverse weather conditions. This is evident in the demonstrated examples using moderate and heavy rain, where processing times are significantly lower than using dense frames and maps. As vehicle-driving systems evolve, computational complexity increases. Optimizing each subsystem becomes even more critical as sensor fusion systems begin to integrate numerous features and functions. By simplifying maps to reduce computational load, this invention, in addition to enhancing real-time processing capabilities, opens new avenues for deploying complex localization and map-matching algorithms in challenging dynamic environments.

[0057] Considering the variations in sensor data caused by rain, localization in rain-affected environments is one of the most challenging tasks for any assisted driving positioning system. LiDAR point cloud data exhibits high variability under rainfall conditions because each beam scatters and attenuates the signal. These effects reduce the accuracy of map matching algorithms, especially those relying on dense point clouds representing transient environmental features such as vegetation. These features, like vegetation, are highly sensitive to rain, thus increasing the variability of point-to-point localization performance during moderate to heavy rain. The proposed Pareto sensor fusion method achieves optimal map generation by intelligently excluding these rain-sensitive features. Therefore, the reference map will be highly stable and reliable because it filters out vegetation-related points, which tend to fluctuate with weather conditions and seasons. This reduction in point cloud density improves the efficiency of the process and reduces the adverse effects of rainfall on LiDAR-based map matching.

[0058] Integrating lightweight frames and lightweight maps enhances the system's robustness in rain-affected environments. This setup strikes a good balance during light to moderate rainfall, minimizing computational load without compromising positioning accuracy. When rain is heavy, the additional contribution of instantaneous point cloud removal becomes more critical. While dense point clouds can offer higher spatial resolution, increased LiDAR signal distortion makes them more susceptible to noise and errors under heavy rainfall conditions. The resulting lightweight map is better suited to these conditions, focusing on preserving more stable and rain-resistant features such as building and road boundaries, which serve as reliable markers for positioning.

[0059] Therefore, rain-affected environments need to be further processed in a more automated and adaptive manner. In practice, this would involve manually removing rain-sensitive features, such as vegetation, using tools combined with coarse GNSS data. Automating this process using machine learning and computer vision would significantly improve accuracy and efficiency. For example, deep learning models can be used to identify and exclude rain-affected elements in LiDAR point clouds in real time. This would provide more consistent LiDAR performance under different weather conditions without human intervention. Such adaptive filtering can also further enhance the system's ability to maintain high positioning accuracy under heavy rainfall conditions, thereby improving the reliability of vehicle navigation in rain-affected environments. Furthermore, integrating data from weather forecasts or rainfall sensors into the positioning framework allows for dynamic adjustment of the point cloud filtering process based on real-time weather conditions. While proposed Pareto sensor fusion methods have significantly improved positioning performance under rain-affected conditions, considerable room for optimization remains. Automating the removal of rain-sensitive features and using adaptive filtering techniques with real-time environmental data hold great promise for further enhancing the robustness and accuracy of future autonomous positioning systems.

[0060] The system would be even stronger if GNSS data were also considered in map matching under good signal conditions to provide large-scale resilience to cope with increasingly larger dead reckoning errors over time. This hybrid approach, fusing GNSS with lidar and inertial sensors within a consensus framework, has the potential to further expand the system's operational range by improving its adaptability to real-world conditions. The Pareto-optimized positioning framework optimally balances accuracy and computational efficiency under different conditions. This multi-objective approach excels at providing high adaptability and accuracy, especially in complex urban environments. Combining Pareto optimization with lightweight map and sensor fusion enhances the system's resilience and accuracy, making it suitable for real-time, weather-variable positioning in assisted driving applications. The Pareto-optimized positioning method shows significant improvements across all metrics, with a 40-45% reduction in RMSE (from 0.0965 m to 0.0566 m in heavy rain), significantly improving the positioning flexibility and accuracy of traditional methods. By using a lightweight map that retains only essential invariant features, the method reduces point cloud density by 59%, thereby reducing computational load and improving performance under adverse weather conditions. Meanwhile, the Pareto optimized lightweight frame and lightweight map (LL) method reduces the processing time from 107.5 ms to 60.8 ms under clear weather conditions and from 284.7 ms to 104.4 ms under heavy rain conditions, demonstrating the efficiency and accuracy of the method. The Pareto method is effective for real-time applications in dynamic and climate change environments, while pursuing higher positioning accuracy and higher processing efficiency.

[0061] See Figure 2 The present invention also provides a lightweight map processing device for vehicle positioning, comprising: The data acquisition module is used to acquire vehicle sensor data and create a lightweight frame of the vehicle based on the sensor data and road boundary information obtained from a preset map. The sensor data includes 3D LiDAR data and GNSS data. The road boundary information is represented by a list of four discrete points for each boundary. Two sets of discrete points are imported, one for vegetation and / or the other for the road boundary, with the leftmost lane boundary being the first set. , and the rightmost lane boundary , ; The location marking module is used to determine the lidar point corresponding to the lightweight frame based on the vehicle's current position and orientation in the GNSS data. Specifically, it finds the nearest lateral position point from the four boundary lists and marks it as a lidar point, and deletes the corresponding lidar point from the four discrete point lists. The coordinate transformation module is used to transform WGS coordinates to UTM coordinates, and to perform coordinate transformation on the lidar points according to the UTM coordinates to obtain lidar frames. Here, WGS coordinates are World Geodetic System coordinates, and UTM coordinates are Universal Mercator coordinates. The map generation module uses spatial indexing to obtain boundary points in the LiDAR frame based on the LiDAR points, and converts the 3D LiDAR data into point cloud data and navigation coordinates based on the boundary points to generate a lightweight map.

[0062] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0063] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0064] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A lightweight map processing method for vehicle positioning, characterized in that, Includes the following steps: The system acquires vehicle sensor data and creates a lightweight frame for the vehicle based on the sensor data and road boundary information obtained from a preset map. The sensor data includes 3D LiDAR data and GNSS data. The road boundary information is represented by a list of four discrete points for each boundary. Two sets of discrete points are imported, one for vegetation and / or the other for the road boundary, with the leftmost lane boundary being the first set. , and the rightmost lane boundary , ; The lidar point corresponding to the lightweight frame is determined based on the vehicle's current position and orientation in the GNSS data. Specifically, the nearest lateral position point is found from the four boundary lists and marked as a lidar point, and the corresponding lidar point is deleted from the four discrete point lists. The WGS coordinates are converted to UTM coordinates, and the coordinates of the lidar points are transformed according to the UTM coordinates to obtain lidar frames. The WGS coordinates are World Geodetic System coordinates, and the UTM coordinates are Universal Mercator coordinates. Based on the LiDAR points, spatial indexing is used to obtain boundary points in the LiDAR frame, and the 3D LiDAR data is converted into point cloud data into navigation coordinates based on the boundary points to generate a lightweight map.

2. The lightweight map processing method for vehicle positioning according to claim 1, characterized in that, Based on the boundary points, the 3D LiDAR data is converted into point cloud data and then into navigation coordinates, including: The point cloud data of the lightweight frame is converted into navigation coordinates, and the converted point clouds are then superimposed. The corresponding expression is: （1） in, These are the coordinates of each point in the lightweight frame. It is the translation vector from the lidar frame to the navigation frame. It is the rotation matrix between the lidar coordinates and the navigation coordinates. The definition expression is: （2） in, , and It is heading, pitch, and roll information. and After point clouds exist in the navigation frames, lightweight frames are accumulated together to form a lightweight map.

3. The lightweight map processing method for vehicle positioning according to claim 1, characterized in that, Based on the vehicle's current position and orientation from the GNSS data, the lidar points corresponding to the lightweight frame are converted into UTM coordinates, including: The vehicle's position is estimated based on its current location, speed, and direction to establish a dead reckoning model. Wheel speed and yaw rate from onboard sensors are used to calculate the position in navigation coordinates, as well as the vehicle's heading between LiDAR coordinates and navigation coordinates. hour along The expression for dead reckoning of axis position is: （3） in, Indicates time k along The dead coordinates of the axis position. This represents the heading angle of the vehicle at time k-1. Indicates time difference, for Time position estimate, The time interval between two measurements. for Observe the speed at all times. yes Observe the azimuth angle at all times. The true location at any moment Represented as: （4） in, for The true speed of a moment for The true azimuth angle at any given moment, measured from the wheel speed sensor. and direction , , It is the heading angle. It is the vehicle's sideslip angle. The expression is: （5） in, It is the yaw rate, the measured value is... and Given the corresponding noise term and The damage is such that the measured values ​​follow a variance of . and The zero-mean Gaussian distribution; Based on the given dead reckoning data and the observed measured velocity and the direction being measured The expressions for calculating the eastward and northward velocities are as follows: （6） （7） in, It is the speed being measured. Indicates the direction being measured.

4. The lightweight map processing method for vehicle positioning according to claim 3, characterized in that, Also includes: The deviation in dead reckoning is quantified by the difference between the estimated position and the actual position. The deviation is calculated by assessing the expected difference between the dead reckoning estimate and the actual position. The corresponding mathematical expression is: （8） The expression to be replaced is and for: （9） To further simplify: （10） Among them, preset It is zero, in time The estimate is unbiased, with a preset direction angle. Additive white Gaussian noise If destruction occurs, the observed direction is represented as: （11） in, Indicates variance For zero-mean Gaussian noise, the expected value of the noise angle cosine can be approximated by utilizing the properties of the Gaussian distribution, thus approximating the expected value of the low-noise variance as: （12） Substituting formula (12) into formula (10) to calculate the deviation, the result is: （13） By introducing correction factors based on system characteristics or noise factors, the deviation is rewritten as: （14） in, Considered as adjustments related to system characteristics or empirical calibration, the deviation in the dead reckoning model is based on the expected value of the deviation between the estimated position and the actual position.

5. The lightweight map processing method for vehicle positioning according to claim 3, characterized in that, The variance of the estimation error in dead reckoning models can be determined and evaluated by analyzing the propagation of velocity and orientation measurement noise within the system. Changes in position estimation are influenced by the measured velocity. and direction The variance of dead reckoning position estimates can be expressed as: （15） Utilizing the characteristics of Gaussian noise, the variance of the product of velocity and the cosine of the direction angle is: (16) Among them, Indicates the expected value. Approximate to The subsequent approximations are applicable and should be used: （17） In this process, variance is incorporated into the fusion process, and dead reckoning estimates are weighted relative to distance measurements.

6. The lightweight map processing method for vehicle positioning according to claim 1, characterized in that, Calculate the transformation from WGS coordinates to UTM coordinates, and then transform the UTM coordinates to LiDAR coordinates to obtain the LiDAR frame, including: The Normal Distribution Transform (NDT) algorithm is used to convert the lidar point cloud into navigation coordinates to calculate the reference map. Based on the reference map, the lidar frame is aligned with the navigation frame to achieve map matching. Formula (18) represents the lidar frame. To navigation frame Transformation: （18） in, Indicates intra-frame points coordinates This represents the corresponding point in the lidar coordinate system. This represents the rotation matrix of the lidar relative to the navigation frame direction. The rotation matrix represents the vector of translation between the lidar and navigation coordinate systems. Defined as: （19） in, , and Indicates roll angle, pitch angle, and yaw angle. , , , and , ; Each lidar point Converted into navigation frames and aggregated to form a point cloud map : （20） in, It is an aggregation to generate a point cloud map. Total number of points, for point cloud map Optimization includes: Vegetation removal: Filter out points corresponding to vegetation; Map downsampling: Downsamples the map and retains the points most important for accurate positioning.

7. The lightweight map processing method for vehicle positioning according to claim 6, characterized in that, The Normal Distribution Transform (NDT) algorithm is used to convert LiDAR point clouds into navigation coordinates to calculate a reference map, including: The vehicle's position is calculated using a given dead reckoning model. The estimated position at that time is: （21） in, express The estimated eastward and northward positions at any given time. express The estimated speed at any time express The estimated course at any given time express Time and The time interval is determined by finding the optimal alignment transformation between the current lidar scan and the reference image. The transformation expression is: （22） in, For rotation angle, , These are the translation components in the east and north directions, respectively. The NDT map matching process determines the optimal transformation. This will reduce the number of current scan points. and reference image The problem of minimizing the difference between them is described as follows: （23） in, Let i represent the i-th point on the lidar frame, where Should be ; Indicates the weights of the NDT estimate. This indicates the number of points in the current scan, determining the optimal transformation. Used to correct dead reckoning position estimates, after implementing NDT-based map matching, the corrected position estimate can be expressed as: （24） A weighted fusion algorithm is used to fuse dead reckoning and NDT estimates. Timing position estimation Represented as: （25） in, and Indicates the fusion weight, satisfying the condition. The difference in the merged location estimates is expressed as: （26） in, Indicates that the vehicle is in The actual location at any given moment.

8. The lightweight map processing method for vehicle positioning according to claim 7, characterized in that, The uncertainty estimates in NDT correction estimation are combined with the uncertainty estimates in dead reckoning estimation, given the estimation error. exist The covariance at time k+1 can be propagated using formula (27): （27） in, This represents the covariance at time k+1. The covariance estimated for dead reckoning. The covariance is based on NDT map matching estimation. The covariance estimated for dead reckoning. The cross-covariance between the NDT-based estimate and the dead reckoning estimate is calculated using the following formula: （28） in, This represents the estimated position after dead reckoning. The expected value estimated for dead reckoning. This represents the expected value estimated after NDT correction. This is the expected value estimated after NDT correction. This represents the expectation operator for calculating the mean of a random variable. It is the cross covariance of the two estimation errors to reflect the linear relationship between them.

9. The lightweight map processing method for vehicle positioning according to claim 8, characterized in that, Obtain the error of the fused location estimation. Time error The norm is less than or equal to Error at time, system state transition matrix The spectral radius is less than 1: and (29) in, Let be the spectral radius of the system state transition matrix. and This represents the error between the fused estimate and the true location, ensuring... This means that the estimation error will not diverge over time.

10. A lightweight map processing device for vehicle positioning, characterized in that, The lightweight map processing method for vehicle positioning as described in any one of claims 1-9 includes: The data acquisition module is used to acquire vehicle sensor data and create a lightweight frame of the vehicle based on the sensor data and road boundary information obtained from a preset map. The sensor data includes 3D LiDAR data and GNSS data. The road boundary information is represented by a list of four discrete points for each boundary. Two sets of discrete points are imported, one for vegetation and / or the other for the road boundary, with the leftmost lane boundary being the first set. , and the rightmost lane boundary , ; The location marking module is used to determine the lidar point corresponding to the lightweight frame based on the vehicle's current position and orientation in the GNSS data. Specifically, it finds the nearest lateral position point from the four boundary lists and marks it as a lidar point, and deletes the corresponding lidar point from the four discrete point lists. The coordinate transformation module is used to transform WGS coordinates to UTM coordinates, and to perform coordinate transformation on the lidar points according to the UTM coordinates to obtain lidar frames. Here, WGS coordinates are World Geodetic System coordinates, and UTM coordinates are Universal Mercator coordinates. The map generation module uses spatial indexing to obtain boundary points in the LiDAR frame based on the LiDAR points, and converts the 3D LiDAR data into point cloud data and navigation coordinates based on the boundary points to generate a lightweight map.

Images