Multi-sensor gradient estimation method and device for intelligent electric vehicle

Abstract

The invention discloses an intelligent electric vehicle multi-sensor gradient estimation method and device, and belongs to the technical field of intelligent electric commercial vehicle gradient estimation. Data are synchronously acquired through a laser radar and an inertial measurement unit, and data alignment is realized by using a synchronizing signal; performing motion distortion correction on the point cloud data based on the pose information of the inertial measurement unit to obtain accurate point cloud information; voxel filtering preprocessing is carried out on the corrected point cloud, features are extracted, and data dimension reduction and association construction are realized; point cloud and inertial measurement data are optimized through a multi-sensor fusion algorithm, a slope angle is calculated according to a rotation matrix in an optimal pose, and complete state estimation including a carrier pose, a speed, sensor deviation and an accurate slope angle is output after filtering. According to the method, high-precision and robust estimation of the gradient of the vehicle in a complex driving environment is realized, and the accuracy of vehicle state sensing and the system stability are improved.

Description

Technical Field

[0001] This invention belongs to the field of slope estimation technology for intelligent electric commercial vehicles, and in particular relates to a multi-sensor slope estimation method and device for intelligent electric vehicles. Background Technology

[0002] Road gradient is a key parameter characterizing the longitudinal inclination of a road and is one of the important road geometric features affecting vehicle driving behavior. In the integrated control system of intelligent electric vehicles, accurate road gradient information plays a crucial role in improving vehicle driving stability, driving safety, and energy management economy. For example, in the vehicle controller, gradient is one of the core inputs for calculating longitudinal load transfer, optimizing drive / brake torque distribution, and predicting driving range.

[0003] Currently, the mainstream method for real-time road slope estimation in the industry relies on mathematical processing of data collected by onboard inertial measurement units (IMUs). These units provide information on the vehicle's three-axis acceleration and angular velocity. Based on this, algorithms such as integration or filtering are used to extract attitude angle changes from the measurement data, thereby calculating the road slope. However, this type of estimation scheme based on IMUs suffers from several inherent technical bottlenecks: First, the output of the inertial measurement unit (IMU) is specific force information, which is the composite of a specific acceleration and gravitational acceleration in its carrier coordinate system. When the vehicle is accelerating or braking, it generates significant longitudinal dynamic acceleration. This dynamic component is coupled with the component of gravitational acceleration along the vehicle's longitudinal axis, making it difficult to separate effectively and in real-time in the time or frequency domain. This results in significant deviations in the gradient estimate during vehicle acceleration and deceleration.

[0004] Secondly, inertial measurement unit (IMU) data introduces accumulated errors during integration. These errors diverge over time, causing the slope estimation output to drift and compromising long-term accuracy. To suppress such errors, existing technologies typically require the introduction of external reference signals, such as the Global Positioning System (GPS), for periodic correction. While this sensor fusion strategy is feasible on open roads, its correction function fails in environments with weak or no GPS signals, such as tunnels, underground parking garages, and urban canyons. Once external correction is lost, the accumulated errors of the IMU cannot be effectively constrained, causing the road slope estimation accuracy to deteriorate rapidly in a short period, failing to meet the stringent requirements of intelligent vehicle control systems for continuous and reliable parameter input.

[0005] Therefore, there is an urgent need in this field for a new technology that can overcome the dependence on GPS signals and provide stable and accurate road slope information under complex dynamic conditions of vehicles. Summary of the Invention

[0006] This invention discloses a multi-sensor slope estimation method and device for intelligent electric vehicles. Its purpose is to address the problem of insufficient estimation accuracy of a single sensor in complex environments by fusing data from lidar and inertial measurement units. Through synchronous acquisition and data alignment, point cloud motion distortion correction, voxel filtering and feature extraction, and multi-sensor fusion optimization, the final output is a complete state estimate including vehicle pose, velocity, sensor bias, and precise slope angle. This achieves high-precision and robust perception of vehicle slope, improving the environmental adaptability and system stability of intelligent vehicles.

[0007] To achieve the above objectives, this invention provides a multi-sensor slope estimation method for intelligent electric vehicles. The method includes: acquiring point cloud data and inertial measurement data, and aligning the point cloud data and inertial measurement data by sending a PPS signal to a lidar sensor for synchronization locking; using pose data from the inertial measurement data as a reference to correct distortion in the point cloud data to obtain corrected point cloud data; preprocessing the corrected point cloud data using voxel filtering, and performing feature extraction and dimensionality reduction on the point cloud data in the preprocessed voxels to obtain dimensionality-reduced point cloud data; processing the dimensionality-reduced point cloud data and inertial measurement data using a multi-sensor fusion optimization algorithm, and using the rotation matrix in the estimated optimal pose for slope angle conversion and filtering to obtain a complete state estimate of the vehicle's optimal pose, speed, sensor bias, and optimized slope angle.

[0008] Furthermore, the steps of acquiring point cloud data and inertial measurement data, and aligning the point cloud data and inertial measurement data by sending a PPS signal to the lidar for synchronization locking include: acquiring point cloud data through an onboard lidar; acquiring the three-axis acceleration and angular velocity around the three axes in a Cartesian coordinate system through an inertial measurement unit as inertial measurement data; and aligning the point cloud data and inertial measurement data by sending a PPS signal to the lidar for synchronization locking.

[0009] Furthermore, the step of using pose data from inertial measurement data as a reference to correct the distortion of point cloud data to obtain corrected point cloud data includes: calculating the trajectory of motion along the laser point using high-frequency inertial measurement data; interpolating the corresponding motion amount of each laser point relative to the frame start time at the acquisition instant on the trajectory of motion along the laser point to obtain reference pose data; and based on the reference pose data, uniformly correcting all points to the same coordinate system at the same time through coordinate transformation to eliminate the distortion of point cloud data and obtain corrected point cloud data.

[0010] Furthermore, the steps of preprocessing the corrected point cloud data using voxel filtering and extracting features and reducing the dimensionality of the point cloud in the preprocessed voxels to obtain dimensionality-reduced point cloud data include: reducing the number of points in the point cloud data by voxel filtering to obtain preprocessed point cloud data; extracting the most critical feature points by calculating the curvature of the preprocessed point cloud data; and then converting the preprocessed point cloud data into edge points and planar points containing geometric information.

[0011] Furthermore, the steps of processing the dimensionality-reduced point cloud data and inertial measurement data through a multi-sensor fusion optimization algorithm, and using the rotation matrix in the estimated optimal pose to perform slope angle conversion and filtering to obtain a complete state estimate of the carrier's optimal pose, velocity, sensor deviation, and optimized slope angle include: constructing an optimization problem by fusing constraints of inertial measurement data and feature matching constraints of lidar, solving for the optimal pose; parsing the pitch angle from the optimal pose and filtering it to obtain the optimized slope angle.

[0012] In another aspect, the present invention provides a multi-sensor slope estimation device for intelligent electric vehicles. The device includes: a data acquisition module for acquiring point cloud data and inertial measurement data, and aligning the point cloud data and inertial measurement data by sending a PPS signal to a lidar for synchronization locking; a distortion correction module for using pose data from the inertial measurement data as a reference to correct the distortion of the point cloud data, thereby obtaining corrected point cloud data; an extraction and dimensionality reduction module for preprocessing the corrected point cloud data using voxel filtering, and performing feature extraction and dimensionality reduction on the point cloud data in the preprocessed voxels, thereby obtaining dimensionality-reduced point cloud data; and an optimization output module for processing the dimensionality-reduced point cloud data and inertial measurement data using a multi-sensor fusion optimization algorithm, and using the rotation matrix in the estimated optimal pose for slope angle conversion and filtering, thereby obtaining a complete state estimate of the vehicle's optimal pose, speed, sensor deviation, and optimized slope angle.

[0013] Furthermore, the data acquisition module includes: acquiring point cloud data via an onboard lidar; acquiring triaxial acceleration and angular velocity around the three axes in a Cartesian coordinate system via an inertial measurement unit as inertial measurement data; and aligning the point cloud data with the inertial measurement data by sending a PPS signal to the lidar for synchronization locking.

[0014] Furthermore, the distortion correction module includes: calculating the trajectory of motion along the laser point using high-frequency inertial measurement data; interpolating the corresponding motion amount of each laser point relative to the frame start time at the acquisition instant on the trajectory of motion along the laser point to obtain reference pose data; and based on the reference pose data, uniformly correcting all points to the same coordinate system at the same time through coordinate transformation to eliminate the distortion of the point cloud data and obtain the corrected point cloud data.

[0015] Furthermore, the extraction and dimensionality reduction module includes: reducing the number of points in the point cloud data through voxel filtering to obtain preprocessed point cloud data; extracting the most critical feature points by calculating the curvature of the preprocessed point cloud data; and then converting the preprocessed point cloud data into edge points and planar points containing geometric information.

[0016] Furthermore, the optimization output module includes: constructing an optimization problem by fusing constraints from inertial measurement data and feature matching constraints from lidar, solving for the optimal pose; and extracting the pitch angle from the optimal pose and filtering it to obtain the optimized slope angle.

[0017] Compared with the prior art, the technical solution provided by the present invention has at least the following beneficial effects: 1. Multi-sensor data fusion and collaboration: High-precision alignment of lidar point cloud data and inertial measurement unit data is achieved through a synchronization locking mechanism, making full use of the characteristics of different sensors to build a unified state estimation framework, effectively overcoming the perception limitations of a single sensor in complex scenarios.

[0018] 2. Point cloud data quality optimization: Motion distortion correction of point cloud data is performed using pose information provided by the inertial measurement unit, which significantly improves the accuracy and purity of point cloud data, laying a reliable data foundation for subsequent feature extraction and state estimation.

[0019] 3. Improved data processing efficiency and correlation: By performing voxel filtering preprocessing and feature extraction dimensionality reduction on point cloud data, the amount of data computation is greatly reduced while retaining key environmental structural information, and the spatiotemporal correlation between point cloud data is enhanced.

[0020] 4. High-precision estimation of complete state: Through multi-sensor fusion optimization algorithm, the optimal pose, velocity, sensor deviation and accurate slope angle of the vehicle are solved simultaneously, realizing high-precision and robust estimation of the complete state of the vehicle in complex driving environment, which significantly improves the environmental perception capability and overall control stability of intelligent vehicle system.

[0021] This invention discloses a multi-sensor slope estimation method and device for intelligent electric vehicles. Through deep fusion and collaborative processing of multi-source sensor data, it effectively addresses the inherent limitations of insufficient estimation accuracy and poor reliability of single sensors in complex driving environments. It achieves end-to-end optimization from synchronous data acquisition, point cloud distortion correction, feature extraction and dimensionality reduction to fusion state estimation, significantly improving the accuracy of slope angle measurement and the robustness of the system. While enhancing the vehicle's environmental perception capabilities, it provides reliable state support for the intelligent vehicle's attitude control and safe driving, improving the system's adaptability and stability in complex scenarios. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of a multi-sensor slope estimation method for intelligent electric vehicles provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the specific steps of a multi-sensor slope estimation method for intelligent electric vehicles provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the structure of a multi-sensor slope estimation device for intelligent electric vehicles provided in an embodiment of the present invention. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] To address the shortcomings of the existing technologies, this invention provides a multi-sensor slope estimation method and apparatus for intelligent electric vehicles. The method involves acquiring data from LiDAR and inertial measurement units (IMUs), performing motion distortion correction on the LiDAR point cloud using IMU data, performing dimensionality reduction and feature extraction on the point cloud data, conducting joint optimization estimation using IMU and LiDAR, performing slope angle conversion and filtering, and effectively improving the stability and accuracy of slope estimation for intelligent electric vehicles by fusing IMU and LiDAR data for multi-sensor joint optimization estimation.

[0026] The core idea of ​​this invention is to construct a collaborative perception framework that deeply integrates LiDAR and inertial measurement units, mapping point cloud spatial features and inertial motion data to a unified state estimation system to achieve accurate calculation of vehicle attitude and terrain slope. Motion distortion correction of the point cloud based on high-frequency inertial data effectively overcomes the inherent limitation of insufficient perception accuracy of a single sensor in dynamic and complex environments. By performing voxel filtering and geometric feature extraction on the corrected point cloud, data processing efficiency is significantly improved while preserving key environmental structural information. Finally, through a multi-sensor fusion optimization algorithm, a complete state estimate including vehicle pose, velocity, sensor bias, and precise slope angle is simultaneously output, forming a full-process optimization from data preprocessing and feature association to fusion calculation. This significantly improves the slope perception accuracy and system robustness of the vehicle in complex driving environments, providing reliable state support for the chassis control and autonomous navigation of intelligent vehicles.

[0027] The following description, with reference to the accompanying drawings, illustrates a multi-sensor slope estimation method and apparatus for intelligent electric vehicles according to an embodiment of the present invention.

[0028] Example 1 This embodiment provides a multi-sensor slope estimation method for intelligent electric vehicles. For example... Figure 1 As shown, the method includes the following steps: S1 collects point cloud data and inertial measurement data, and synchronizes and locks the point cloud data with the inertial measurement data by sending a PPS signal to the lidar.

[0029] Specifically, point cloud data is collected by vehicle-mounted lidar; the three-axis acceleration and angular velocity around the three axes in the Cartesian coordinate system are collected by inertial measurement unit (IMU), i.e., IMU data; and the point cloud data and IMU data are aligned by sending PPS signals to the lidar for synchronization locking.

[0030] Specifically, its core objective is to resolve the spatiotemporal inconsistency between lidar and inertial measurement units caused by their different sampling mechanisms. Through hardware synchronization and data alignment, it provides precisely matched raw data for subsequent point cloud distortion correction and fusion estimation.

[0031] Specifically, a multi-sensor collaborative acquisition mechanism is first established. The vehicle-mounted lidar continuously scans the surrounding environment to acquire point cloud data containing spatial three-dimensional coordinates and reflection intensity. Simultaneously, an inertial measurement unit acquires data at a higher frequency of the vehicle's three-axis linear acceleration and angular velocity around the three axes in a Cartesian coordinate system. These data collectively constitute the raw observation information for subsequent state estimation.

[0032] Crucially, to address the potential time discrepancies introduced by independent sampling from the two types of sensors, a hardware synchronization mechanism is employed to achieve precise timescale alignment. Specifically, a PPS-level signal is sent to the lidar as a synchronization trigger reference, ensuring that the acquisition time of each point cloud data frame is synchronized with the inertial measurement unit's time reference at the microsecond level. This hardware-level synchronization locking mechanism fundamentally guarantees the consistency of data from different modal sensors across the time dimension.

[0033] Furthermore, based on time synchronization, spatial correspondence is established through coordinate transformation. Point cloud data in the lidar coordinate system is uniformly transformed to the carrier coordinate system where the inertial measurement unit resides, based on the pose reference at the synchronization moment. This data alignment process effectively eliminates spatial measurement deviations caused by differences in sensor installation positions, forming a spatiotemporally unified observation data sequence, laying a reliable data foundation for constructing a high-precision state estimation model.

[0034] Specifically, the multi-source sensor data, after being synchronized and aligned, retains the high-precision spatial information of the lidar and integrates the high-frequency motion characteristics of the inertial measurement unit, providing the necessary data support for subsequent accurate point cloud distortion correction and multi-sensor fusion optimization.

[0035] S2 uses the pose data in the inertial measurement data as a reference to correct the distortion of the point cloud data, so as to obtain the corrected point cloud data.

[0036] Specifically, the motion trajectory is obtained by integrating high-frequency inertial measurement unit data; the motion of each laser point relative to the start of the frame at the moment of acquisition is accurately interpolated, i.e., the pose data is used as a reference; all points are uniformly corrected to the same coordinate system at the same moment through coordinate transformation, thereby eliminating the distortion of point cloud data and obtaining more accurate distortion-free point cloud data.

[0037] Specifically, by fusing high-frequency motion observation data from the inertial measurement unit, motion compensation is performed on the point cloud data to obtain a point cloud that accurately reflects the instantaneous spatial geometric relationship.

[0038] Specifically, the motion trajectory is first reconstructed based on data from the inertial measurement unit (IMU). Using angular velocity and linear acceleration data acquired by the IMU at frequencies of several hundred hertz, the six-degree-of-freedom pose changes of the carrier at each moment are integrated in real time using inertial navigation calculation methods, forming a continuous and smooth motion trajectory. This trajectory accurately describes the continuous motion state of the lidar during the scanning of a frame of point cloud.

[0039] Crucially, given the time required for a single frame scan by a LiDAR system, precise timestamp matching and motion interpolation are implemented. Since each LiDAR point within a single point cloud frame is acquired at different times, it is necessary to accurately calculate the relative motion of each LiDAR point at its acquisition time relative to the start time of the frame. Using the continuous motion trajectory provided by the inertial measurement unit, algorithms such as quaternion spherical linear interpolation are employed to interpolate the precise pose transformation matrix of each LiDAR point at the moment of acquisition.

[0040] Furthermore, based on the pose transformation relationship obtained through interpolation, the original point cloud undergoes coordinate normalization. Laser points distributed in coordinate systems at different times are uniformly projected onto the same reference time (usually the start or end time of the frame) through rigid body transformation. This coordinate normalization process effectively eliminates geometric distortions such as stretching and twisting of the point cloud caused by carrier motion, enabling the corrected point cloud to accurately reflect the real environmental structure at that moment.

[0041] Specifically, point cloud data corrected for motion distortion not only maintains the original spatial resolution but, more importantly, eliminates motion artifacts, significantly improving the geometric accuracy and spatial consistency of the point cloud data. This high-quality point cloud provides a reliable data foundation for subsequent processing steps such as feature extraction and point cloud matching, and is an important guarantee for achieving high-precision state estimation.

[0042] S3. The corrected point cloud data is preprocessed using voxel filtering, and the point cloud in the preprocessed voxels is subjected to feature extraction and dimensionality reduction to obtain the dimensionality-reduced point cloud data.

[0043] Specifically, the point cloud data is first reduced in number by voxel filtering, i.e., voxel filtering is used to preprocess the point cloud data; the most critical feature points are extracted by calculating curvature; and then the massive amount of original point cloud is transformed into a small number of edge points and planar points rich in geometric information, i.e., feature extraction and dimensionality reduction are performed on the point cloud in the preprocessed voxels to obtain the correlation between point cloud data.

[0044] Specifically, the point cloud data after motion distortion correction is spatially meshed. The three-dimensional space is divided into a uniform voxel grid, with each voxel representing a cubic spatial unit of a defined size. In one specific implementation, the voxel size can be adjusted according to the point cloud density and application scenario, typically set to a cubic grid between 0.1 meters and 0.3 meters.

[0045] Crucially, point clouds within each voxel undergo aggregation sampling. For all spatial points falling within the same voxel grid, their geometric centers are calculated or a centroid-weighted algorithm is applied to aggregate multiple point clouds within that voxel into a representative feature point. This voxel filtering preprocessing method reduces the amount of original point cloud data by 70%-90% while preserving the main structural features of the environment, significantly reducing subsequent computational complexity.

[0046] Furthermore, geometric feature analysis is performed on the downsampled point cloud. Candidate points with significant geometric features are identified by calculating the curvature distribution characteristics of each point within its local neighborhood. Specifically, a curvature estimation algorithm based on eigenvalue decomposition is used to analyze the covariance matrix of the neighboring point cloud for each point, quantifying the curvature degree of that point through the relative magnitudes of the matrix eigenvalues.

[0047] Specifically, point clouds are classified into different geometric feature types based on curvature analysis results. Points with curvature values ​​greater than a set threshold are identified as edge feature points, typically located in areas with drastic curvature changes, such as object contour boundaries. Points with curvature values ​​less than the set threshold are identified as planar feature points, mainly distributed in areas with gentle curvature changes, such as flat surfaces. This classification method transforms massive amounts of raw point clouds into a small set of feature points rich in geometric structural information.

[0048] Specifically, the spatial relationships between point cloud data are established through the above processing. Edge feature points and planar feature points not only retain key geometric structural information of the environment, but more importantly, establish intra-frame and inter-frame data relationships through their feature descriptors. This provides reliable matching primitives for subsequent multi-sensor fusion optimization, thereby significantly improving the processing efficiency of the entire system while maintaining accuracy.

[0049] S4 processes the dimensionality-reduced point cloud data and inertial measurement data through a multi-sensor fusion optimization algorithm, and uses the rotation matrix in the estimated optimal pose to perform slope angle conversion and filtering, thereby obtaining a complete state estimate of the carrier's optimal pose, velocity, sensor deviation, and optimized slope angle.

[0050] Specifically, by fusing the pre-integration constraints of the inertial measurement unit data and the feature matching constraints of the lidar, an optimization problem is constructed to solve for the optimal pose; the pitch angle is then extracted from this pose and filtered to obtain accurate slope information.

[0051] Specifically, a multi-sensor fusion optimization framework is first established, deeply fusing lidar feature point clouds with inertial measurement unit data. In one specific implementation, a graph optimization-based state estimation model is constructed, where nodes represent the carrier's state variables and edges represent observation constraints from various sensors.

[0052] Crucially, the inertial measurement unit (IMU) data undergoes pre-integration to form relative motion constraints. By pre-integrating the IMU observations between adjacent time points, the relative pose change of the carrier during this period is obtained, serving as the IMU constraint edges in the optimization problem. This pre-integration method avoids repeated integration during the optimization process, significantly improving computational efficiency.

[0053] Furthermore, spatial geometric constraints for lidar feature matching are established. Edge and planar feature points extracted in the current frame are matched with corresponding features in the local map. The observation constraint edges for the lidar are constructed by calculating the distances from points to edge lines and points to planes. These spatial geometric constraints provide accurate position and attitude observation information for the optimization problem.

[0054] Specifically, an overall optimization problem is constructed by integrating the aforementioned multi-source constraints. The relative motion constraints provided by the inertial measurement unit's pre-integration and the absolute pose constraints provided by the lidar feature matching are jointly incorporated into the optimization framework. Considering the sensor's time synchronization deviation and measurement noise characteristics, a complete nonlinear least squares problem is constructed. Iterative solutions using optimization algorithms such as the Gauss-Newton method or the Levenberg-Marquardt method are employed to obtain the optimal pose estimate of the carrier.

[0055] Specifically, the slope information is calculated based on the optimal pose obtained through optimization. The pitch angle of the vehicle coordinate system relative to the horizontal plane is extracted from the estimated optimal rotation matrix. Combined with the vehicle's kinematic model, the estimated slope angle is smoothed using Kalman filtering or complementary filtering to eliminate high-frequency noise interference, resulting in a stable and reliable accurate slope angle output. Simultaneously, this optimization process also estimates the vehicle's three-dimensional velocity, accelerometer and gyroscope zero-bias errors, forming a complete state estimation output including pose, velocity, sensor bias, and slope angle, providing comprehensive state information support for the intelligent vehicle's slope perception and control decisions.

[0056] This invention discloses a multi-sensor slope estimation method for intelligent electric vehicles. Through deep fusion and collaborative processing of multiple sensor sources, it effectively addresses the inherent limitations of single sensors in terms of insufficient estimation accuracy and poor reliability under complex driving environments. This method achieves end-to-end optimization from synchronous data acquisition, point cloud distortion correction, feature extraction and dimensionality reduction to fusion state estimation, significantly improving the accuracy of slope angle measurement and the robustness of the system. While enhancing the vehicle's environmental perception capabilities, it provides reliable state support for the intelligent vehicle's attitude control and safe driving, improving the system's adaptability and stability in complex scenarios.

[0057] Example 2 This invention also provides specific steps for a multi-sensor slope estimation method for intelligent electric vehicles, such as... Figure 2As shown, it includes: S101 performs data acquisition for lidar and inertial measurement unit.

[0058] Specifically, point cloud data is collected via an onboard LiDAR with a horizontal viewing angle of 360 degrees and a vertical viewing angle of 26 degrees, at a frequency of 10Hz. The LiDAR is positioned at the center of the vehicle roof without any tilt. The inertial measurement unit (IMU) collects the three-axis acceleration and angular velocity around the three axes in a Cartesian coordinate system at a frequency of 100Hz. The IMU synchronizes and locks with the LiDAR by sending a PPS signal, aligning the point cloud data with the IMU data. The collected point cloud data is transmitted to the domain controller via a network port, and the inertial navigation system transmits the data to the domain controller via a serial port for further processing.

[0059] S102 uses inertial measurement unit data to perform motion distortion correction processing on lidar point clouds.

[0060] Specifically, the acquisition frequency of lidar is generally low, meaning there are significant time differences in the data acquisition process. The data acquired by lidar will undergo varying degrees of motion distortion as the vehicle speed increases. Therefore, high-frequency pose data from the inertial measurement unit is used to eliminate point cloud motion distortion.

[0061] Furthermore, the inertial measurement unit data is triaxial acceleration. With angular velocity Its state is defined as: (1) in, , , , , They represent t The position, velocity, angle, acceleration offset, and angular velocity offset at any given moment.

[0062] Specifically, no. j The state of the frame inertial measurement unit can be determined by the first... i The frame inertial measurement unit (IMU) state and IMU data are discretely estimated, and the expression is: (2) (3) (4) in, The current gravity vector, This is the rotation matrix for transforming from the inertial measurement unit coordinate system to the world coordinate system. The corresponding time difference, symbol This is the symbol for quaternion multiplication.

[0063] Furthermore, suppose the timestamp of the first laser point acquired by a lidar in a certain frame is... The timestamp of the last laser dot is any one of the laser points p The time is , obtain Two adjacent frames and The data from the inertial measurement unit can be used to obtain the corresponding states through formulas (2), (3), and (4), and then obtain the results respectively. Compared to and Compared to pose changes and .

[0064] Furthermore, based on lidar points p time Pose interpolation is performed using the ratio of the data at times k and k+1 adjacent to the inertial measurement unit, i.e. (5) (6) Since the actual lidar point is located in the lidar coordinate system of the current frame, coordinate system transformation is required using calibrated extrinsic parameter data. Therefore, any lidar point... p Unified to The distortion-corrected pose transformation at time t is: (7) in, This is the extrinsic parameter matrix from the inertial measurement unit to the lidar. This is the extrinsic parameter matrix for laser radar to the inertial measurement unit.

[0065] S103, Dimensionality Reduction and Feature Extraction of Point Cloud Data.

[0066] Specifically, since the raw point cloud data is large, direct processing would waste computing resources, so voxel filtering is used to preprocess the raw point cloud.

[0067] Furthermore, the maximum and minimum values ​​of the point cloud dataset are taken along the three coordinate axes of the Cartesian coordinate system. , , , , , Set the side length of the voxel. The three-dimensional space occupied by the point cloud is uniformly divided into A voxel of equal size, i.e. (8) in, This indicates a round-down operation.

[0068] The point cloud is numbered according to its voxel, and the point cloud in each voxel is divided: (9) Dimensionality reduction is achieved by center sampling of the point cloud in each voxel.

[0069] Furthermore, after preprocessing, the number of point clouds is significantly reduced, but it is still insufficient to meet the requirements for real-time slope estimation. Therefore, feature extraction of point cloud data is used to transform the point cloud into features with obvious physical meaning, further reducing the data dimensionality.

[0070] Specifically, let Indicates the first in lidar k Line 1 i One point, Indicates the first in lidar k Line 1 j One point, Indicates the point in the lidar i The set of adjacent points in the same row, with a size of 10, and ensuring that the points are adjacent. The number of curves on both sides is the same, and the local curvature is defined as: (10) The 360-degree horizontal field of view is divided into 6 equal parts. The point cloud of each region is sorted according to its local curvature. The top 5 points with the highest local smoothness values ​​in descending order are selected as edge feature points and represented by a set. The top 20 points with the highest local curvature values ​​in ascending order are selected as planar feature points, and a set is used. express.

[0071] S104, Joint optimization estimation of inertial measurement unit and lidar multi-sensor.

[0072] Specifically, the optimization objective is defined as follows: k Vehicle six degrees of freedom pose at all times : (11) in, Represents the rotation matrix. This represents the translation matrix.

[0073] Furthermore, a joint optimization equation is constructed: (12) in, and These represent the observation error of the inertial measurement unit and the observation error of the lidar, respectively.

[0074] (13) in, , , , , These represent the current position, velocity, angle, acceleration offset deviation, and angular velocity offset deviation, respectively. , , They represent in k -1 to k Displacement, velocity, and angular velocity at any given moment.

[0075] (14) in, and These are the edge feature residuals and the planar feature residuals, respectively. and These are the weights of the residuals for edge features and planar features, respectively.

[0076] Specifically, the current feature point exist k -1 is the time step to find the nearest neighbor with the same feature.

[0077] Furthermore, if the point i If it belongs to the edge feature point, then the point is calculated using equation (15). i arrive k The distance between the 5 nearest neighbors at time -1: (15) Furthermore, if the point i If it belongs to a planar feature point, then the point is calculated using equation (16). i arrive k The distance between the 5 nearest neighbors formed by the plane at time -1: (16) The Levenberg-Marquardt method was used to iteratively calculate the defined joint optimization equation (12), and the results were obtained. .

[0078] S105, slope angle conversion and filtering.

[0079] Specifically, for the calculated rotation matrix After angle transformation, we get: (17) in, for The element in the third row and first column.

[0080] Furthermore, the slope angle is separated using a second-order Butterworth low-pass filter. slope The design cutoff frequency is 1Hz. , , , , The filtering formula is: (18) This invention relates to specific steps of a multi-sensor slope estimation method for intelligent electric vehicles. The specific steps include: acquiring data from LiDAR and an inertial measurement unit (IMU); performing motion distortion correction processing on the LiDAR point cloud using IMU data; dimensionality reduction and feature extraction of the point cloud data; joint optimization estimation using the IMU and LiDAR sensors; and slope angle conversion and filtering. This method, by fusing IMU and LiDAR data for multi-sensor joint optimization estimation, effectively improves the stability and accuracy of slope estimation for intelligent electric vehicles.

[0081] Example 3 This invention also provides a multi-sensor slope estimation device 10 for intelligent electric vehicles, such as... Figure 3 As shown, the device includes: The data acquisition module 100 is used to acquire point cloud data and inertial measurement data, and to synchronize and lock the point cloud data with the inertial measurement data by sending a PPS signal to the lidar.

[0082] Specifically, point cloud data is collected by vehicle-mounted lidar; the three-axis acceleration and angular velocity around the three axes in the Cartesian coordinate system are collected by inertial measurement unit as inertial measurement data; and the point cloud data and inertial measurement data are aligned by sending PPS signals to lidar for synchronization locking.

[0083] The distortion correction module 200 is used to correct the distortion of point cloud data by using the pose data in the inertial measurement data as a reference, so as to obtain the corrected point cloud data.

[0084] Specifically, the trajectory of the motion along the laser point is calculated using high-frequency inertial measurement data; the motion of each laser point relative to the start of the frame at the moment of acquisition is interpolated on the trajectory of the motion along the laser point to obtain reference pose data; based on the reference pose data, all points are uniformly corrected to the same coordinate system at the same moment through coordinate transformation to eliminate the distortion of the point cloud data and obtain the corrected point cloud data.

[0085] The extraction and dimensionality reduction module 300 is used to preprocess the corrected point cloud data using voxel filtering, and to extract and reduce the dimensionality of the point cloud in the preprocessed voxels to obtain the dimensionality-reduced point cloud data.

[0086] Specifically, the number of points in the point cloud data is reduced by voxel filtering to obtain preprocessed point cloud data; the most critical feature points are extracted by calculating the curvature of the preprocessed point cloud data; and then the preprocessed point cloud data is transformed into edge points and planar points containing geometric information.

[0087] The optimized output module 400 is used to process the dimensionality-reduced point cloud data and inertial measurement data through a multi-sensor fusion optimization algorithm, and to perform slope angle conversion and filtering using the rotation matrix in the estimated optimal pose, so as to obtain a complete state estimate of the carrier's optimal pose, velocity, sensor deviation and optimized slope angle.

[0088] Specifically, by fusing constraints from inertial measurement data and feature matching constraints from lidar, an optimization problem is constructed to solve for the optimal pose; the pitch angle is then extracted from the optimal pose and filtered to obtain the optimized slope angle.

[0089] This invention discloses a multi-sensor slope estimation device for intelligent electric vehicles. Through a modular architecture, it achieves collaborative processing and deep fusion of data from multiple sensors, effectively overcoming the systemic deficiency of insufficient perception accuracy of a single sensor in complex driving environments. The device achieves automated conversion from raw data to complete state estimation through end-to-end processing including data acquisition, distortion correction, feature extraction, and fusion optimization, significantly improving the accuracy of slope angle measurement and the robustness of the system. While enhancing the vehicle's environmental perception capabilities, it provides reliable hardware support for the attitude control and safe driving of intelligent vehicles, effectively improving the adaptability and operational stability of autonomous driving systems in complex road conditions.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0091] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. An intelligent electric vehicle multi-sensor slope estimation method, characterized in that, The method comprises the following steps: S1, collecting point cloud data and inertial measurement data, and synchronously locking by sending a PPS signal to the laser radar to align the point cloud data and the inertial measurement data; S2, correcting the distortion of the point cloud data by taking the pose data in the inertial measurement data as a reference to obtain corrected point cloud data; S3, pre-processing the corrected point cloud data by voxel filtering, and extracting features and reducing dimensions of the point cloud in the pre-processed voxel to obtain reduced dimension point cloud data; S4, processing the reduced dimension point cloud data and the inertial measurement data by a multi-sensor fusion optimization algorithm, and performing slope angle conversion and filtering by using the rotation matrix in the estimated optimal pose to obtain the complete state estimation of the optimal pose, speed, sensor deviation and optimized slope angle of the carrier.

2. The method of claim 1, wherein, The S1 comprises: collecting point cloud data by a vehicle-mounted laser radar; collecting three-axis acceleration and angular velocity around the three axes in the Cartesian coordinate system as inertial measurement data by an inertial measurement unit; synchronously locking by sending a PPS signal to the laser radar to align the point cloud data and the inertial measurement data.

3. The method of claim 2, wherein, The S2 comprises: calculating the trajectory along which the laser point moves by using high-frequency inertial measurement data; interpolating the motion amount of each laser point at the corresponding collection moment relative to the frame start moment on the trajectory along which the laser point moves to obtain reference pose data; correcting all points to the same time coordinate system based on the reference pose data to eliminate the distortion of the point cloud data and obtain corrected point cloud data.

4. The method of claim 3, wherein, The S3 comprises: reducing the number of points by voxel filtering the point cloud data to obtain pre-processed point cloud data; extracting the most critical feature points by calculating the curvature of the pre-processed point cloud data, and then converting the pre-processed point cloud data into edge points and plane points containing geometric information.

5. The method of claim 4, wherein, The S4 comprises: constructing an optimization problem by fusing the constraints of the inertial measurement data and the feature matching constraints of the laser radar to solve the optimal pose; analyzing the pitch angle from the optimal pose and filtering to obtain the optimized slope angle.

6. An intelligent electric vehicle multi-sensor slope estimation device, characterized by, The method comprises the following steps: a data collection module for collecting point cloud data and inertial measurement data, and synchronously locking by sending a PPS signal to the laser radar to align the point cloud data and the inertial measurement data; a correction distortion module for taking the pose data in the inertial measurement data as a reference to correct the distortion of the point cloud data to obtain corrected point cloud data; an extraction and dimension reduction module for pre-processing the corrected point cloud data by voxel filtering, and extracting features and reducing dimensions of the point cloud in the pre-processed voxel to obtain reduced dimension point cloud data; an optimization output module for processing the reduced dimension point cloud data and the inertial measurement data by a multi-sensor fusion optimization algorithm, and performing slope angle conversion and filtering by using the rotation matrix in the estimated optimal pose to obtain the complete state estimation of the optimal pose, speed, sensor deviation and optimized slope angle of the carrier.

7. The apparatus of claim 6, wherein, The data collection module is further configured to: collect point cloud data by a vehicle-mounted laser radar; The three-axis acceleration and angular velocity around the three axes in the Cartesian coordinate system are collected by the inertial measurement unit as inertial measurement data; The PPS signal is sent to the laser radar for synchronization locking, and the point cloud data and the inertial measurement data are aligned.

8. The apparatus of claim 6, wherein, The correction distortion module is further configured to: Calculate the trajectory of the laser point by using the high-frequency inertial measurement data; Interpolate the corresponding motion amount of each laser point relative to the frame start time at the collection moment on the trajectory of the laser point to obtain the reference pose data; Based on the reference pose data, all points are uniformly corrected to the coordinate system at the same moment through coordinate transformation, the distortion of the point cloud data is eliminated, and the corrected point cloud data is obtained.

9. The apparatus of claim 6, wherein, The extraction dimension reduction module is further configured to: Reduce the number of points by voxel filtering the point cloud data to obtain the pretreated point cloud data; Extract the most critical feature points by calculating the curvature of the pretreated point cloud data, and then convert the pretreated point cloud data into edge points and plane points containing geometric information.

10. The apparatus of claim 6, wherein, The optimization output module is further configured to: Construct an optimization problem by fusing the constraints of the inertial measurement data and the feature matching constraints of the laser radar, and solve the optimal pose; Parse the pitch angle from the optimal pose and filter it to obtain the optimized slope angle.

Images