Road surface unevenness extraction method and electronic equipment thereof

Abstract

The embodiment of the invention provides a road surface unevenness extraction method, device and system. The method comprises the following steps: acquiring original point cloud data of a current vehicle; performing anti-shake compensation processing on the original point cloud data to obtain target point cloud data; preprocessing the target point cloud data to obtain feature point cloud data; and performing point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain target unevenness data of the road surface where the current vehicle is located. The method comprises the following steps: obtaining point cloud data, carrying out anti-shake compensation to obtain target point cloud data, carrying out classified calculation on point cloud through a preset unevenness model, accurately extracting current road surface unevenness information, reducing the influence of vehicle vibration on the data, improving the stability of the point cloud, and realizing high-precision, low-noise and anti-shake road unevenness detection. Stable and accurate point cloud data are obtained, so that the vehicle can adapt to complex terrains, and the performance is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and particularly to a method, device, system, electronic device and storage medium for extracting road surface unevenness. Background Art

[0002] In application scenarios such as intelligent transportation, autonomous driving, and road detection, obtaining accurate road surface unevenness information is of great significance for improving driving safety, optimizing suspension system adjustment, and reducing vehicle wear.

[0003] Existing road surface unevenness detection methods have problems such as low efficiency of manual inspection, jitter of lidar data, serious interference of point cloud noise, and insufficient accuracy of analysis models. Summary of the Invention

[0004] Embodiments of the present invention provide a method for extracting road surface unevenness to solve the problems of low efficiency of manual inspection, jitter of lidar data, serious interference of point cloud noise, and insufficient accuracy of analysis models existing in existing road surface unevenness detection methods.

[0005] In a first aspect, embodiments of the present invention provide a method for extracting road surface unevenness, and the method includes the following steps: Obtain the original point cloud data of the current vehicle; Perform anti-jitter compensation processing on the original point cloud data to obtain target point cloud data; Perform preprocessing on the target point cloud data to obtain feature point cloud data; Perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain target unevenness data of the road surface where the current vehicle is located.

[0006] Optionally, the original point cloud data includes the attitude data and point cloud data of the vehicle, and obtaining the original point cloud data of the current vehicle includes: Scan the current vehicle in the horizontal and vertical angle directions through a lidar to determine the three-dimensional coordinates in the corresponding directions; Generate sparse point cloud data based on the three-dimensional coordinates; Realtime detect the acceleration and angular velocity of the current vehicle through an IMU sensor to determine the attitude data of the current vehicle, and the attitude data includes the degree of bumpiness, steering data, and acceleration condition.

[0007] Optionally, performing anti-jitter compensation processing on the original point cloud data to obtain target point cloud data includes: Perform anti-jitter compensation processing on the original point cloud data through a preset anti-jitter compensation algorithm to obtain a first compensation parameter; Obtain the vibration parameters of the current vehicle through an IMU sensor; Based on the ARIMA model, establish a motion state equation and the vibration parameters, and perform prediction processing on the original point cloud data to determine the second compensation parameter; Based on the first compensation parameter and the second compensation parameter, process the original point cloud data to obtain the target point cloud data.

[0008] Optionally, the preprocessing of the target attitude data and the target point cloud data to obtain the feature point cloud data includes: Denoise the target point cloud data through a median filtering algorithm to obtain the first feature point cloud data; Calculate the point cloud rotation matrix R and the translation amount t by using the ICP algorithm and the bidirectional KD algorithm on the first feature point cloud data; Register the first feature point cloud data according to the point cloud rotation matrix R and the translation amount t to generate the second feature point cloud data; Perform point cloud segmentation processing on the second feature point cloud data through a point cloud segmentation algorithm to obtain the third feature point cloud data.

[0009] Optionally, the method of performing point cloud segmentation processing on the second feature point cloud data through a point cloud segmentation algorithm to obtain the third feature point cloud data further includes: Extract the domain features of the second feature point cloud data to determine each center point and its adjacent point cloud group in the second feature point cloud data; Generate corresponding feature coordinate groups based on each center point and its adjacent point cloud group, and the feature coordinate groups include local polar coordinates, Cartesian coordinates, and position feature coordinates; Perform point feature fusion processing on each feature coordinate group based on a point cloud semantic segmentation algorithm to obtain the eigenvalue vector corresponding to each feature coordinate group; Use the eigenvalue vector as the third feature point cloud data.

[0010] Optionally, the method of performing point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located includes: Perform feature enhancement processing on the feature point cloud data through a preset hybrid pooling layer to obtain the first point cloud classification data; Perform local feature expression enhancement processing on the first point cloud classification data through a preset spatial layer to obtain the second point cloud classification data; Process the second point cloud classification data through a preset hierarchical weight to output the target unevenness data of the road surface where the current vehicle is located.

[0011] In a second aspect, an embodiment of the present invention further provides a road surface unevenness extraction device, and the road surface unevenness extraction device includes: A first acquisition module, configured to acquire the original point cloud data of the current vehicle; A second acquisition module, configured to perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data; A third acquisition module, configured to perform preprocessing on the target point cloud data to obtain feature point cloud data; A fourth acquisition module, configured to perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain target unevenness data of the road surface where the current vehicle is located.

[0012] In a third aspect, an embodiment of the present invention provides a road surface unevenness extraction system, where the road surface unevenness extraction system includes: a road surface unevenness extraction device, a server, and a vehicle device.

[0013] In a fourth aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the steps in the road surface unevenness extraction method provided by the embodiment of the present invention are implemented.

[0014] In a fifth aspect, an embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps in the road surface unevenness extraction method provided by the embodiment of the invention are implemented.

[0015] In the embodiment of the present invention, the original point cloud data of the current vehicle is acquired; anti-shake compensation processing is performed on the original point cloud data to obtain target point cloud data; preprocessing is performed on the target point cloud data to obtain feature point cloud data; point cloud classification processing is performed on the feature point cloud data through a preset unevenness model to obtain target unevenness data of the road surface where the current vehicle is located. By acquiring point cloud data and performing anti-shake compensation, target point cloud data is obtained, and then point cloud classification calculation is performed through a preset unevenness model to accurately extract the current road surface unevenness information, reduce the influence of vehicle vibration on the data, improve the stability of the point cloud, achieve high-precision, low-noise, and anti-shake road unevenness detection, obtain stable and accurate point cloud data, enable the vehicle to adapt to complex terrains, and improve performance. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is the architecture diagram of a road unevenness extraction system adopted in an embodiment of the present invention; Figure 2 is the flowchart of a road unevenness extraction method provided by an embodiment of the present invention; Figure 3 is the structural schematic diagram of a small three-degree-of-freedom motion platform provided by an embodiment of the present invention; Figure 4 is the structural schematic diagram of another road unevenness extraction device provided in an embodiment of the present invention; Figure 5 is the structural schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] As Figure 1 shown, Figure 1 is the architecture diagram of a road unevenness extraction system 100 provided by an embodiment of the present invention. The road unevenness extraction system includes: a road unevenness extraction device 400, a server 101, and a vehicle device 102. Among them, the road unevenness extraction device 400 further includes a first acquisition module, which can be used to acquire the original point cloud data of the current vehicle; a second acquisition module, which can be used to perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data; a third acquisition module, which can be used to perform preprocessing on the target point cloud data to obtain feature point cloud data; a fourth acquisition module, which can be used to perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located.

[0020] Specifically, the above original point cloud data can be a set of three-dimensional space coordinate points obtained by a lidar, combined with the attitude data of the vehicle provided by an IMU (inertial measurement unit) sensor. It can be used as the basic input for subsequent anti-shake compensation, preprocessing, and unevenness data analysis.

[0021] It can be understood that the lidar installed on the small three-degree-of-freedom motion platform can perform horizontal and vertical scans on the road surface where the vehicle is located to obtain the three-dimensional coordinates (X, Y, Z) reflected by each laser beam. According to the initially obtained reflected three-dimensional coordinates, sparse point cloud data is formed to identify the initial state of the road surface.

[0022] The above IMU (Inertial Measurement Unit) sensor can be used to detect the acceleration and angular velocity data of a vehicle. In this embodiment, the motion state of the vehicle, such as the vehicle vibration intensity, roll angle, acceleration and deceleration conditions, etc., can be characterized by the acceleration and angular velocity of the vehicle, so as to verify the attitude data of the vehicle. The scanning error of the lidar can also be corrected through the attitude data of the vehicle to improve the stability of the point cloud data.

[0023] More specifically, the above-mentioned small three-degree-of-freedom motion platform can be illustrated as follows Figure 3 The above-mentioned small three-degree-of-freedom motion platform includes: It is composed of a triangular fixed base and a circular movable platform. The fixed base and the movable platform are connected by 3 motion chains and 1 fixed chain. Each motion chain is equipped with an electric cylinder, and each actuator is composed of a servo motor and a ball hinge mechanism. Driven by the three motion chains, the movable platform can achieve rotation along the x, y, and z axes.

[0024] Among them, the outer contour of the fixed base is an equilateral triangle, and inside is the connection between the center point of the equilateral triangle and its endpoints respectively; each endpoint is connected to the lower Hooke joint of the motion chain, and the other end is connected to the upper Hooke joint on the movable platform; make a concentric circle on the circular movable platform, the radius of the concentric circle is 0.8 times the radius of the outer circle, and make an equilateral triangle on this concentric circle, the center point of which is the same as the center of the circle, and the endpoints of the equilateral triangle are the upper Hooke joint points; the center point of the fixed base and the center point of the movable platform are connected by a fixed chain.

[0025] The center of gravity of the above-mentioned lidar coincides with the center of the above-mentioned small three-degree-of-freedom motion platform. Its motion chains are symmetrically distributed. The three motion chains share parameters and weights and use the same algorithm, so that the efficiency is improved. In this embodiment, the above-mentioned lidar and the above-mentioned small three-degree-of-freedom motion platform can be built in the middle position of the rear crossbeam of the vehicle. Set the plane initial angle of the above-mentioned small three-degree-of-freedom motion platform to 45°, and this angle is the included angle between the movable platform and the horizontal plane; the above-mentioned lidar can be used to obtain the original point cloud data, and the horizontal field of view range of the above-mentioned lidar is close to 180°, and the vertical line of sight range is close to 40°, including a 15° upward scanning angle and a 25° downward scanning angle to ensure better scanning of the road surface information.

[0026] The above anti-shake compensation process can be a process of eliminating the errors generated in the point cloud data caused by factors such as the vehicle's own vibration and movement. Specifically, it can be the dynamic correction of the original point cloud data to eliminate the point cloud jitter and errors caused by factors such as vehicle movement, bumping, vibration, acceleration or steering, so as to obtain more stable and accurate point cloud data, reduce motion artifacts, improve the accuracy of the point cloud data, and ensure the reliability of road unevenness extraction.

[0027] Specifically, after the above-mentioned anti-shake compensation determines the acceleration, angular velocity, and attitude data of the vehicle through the above-mentioned IMU sensor, the Kalman filter (UKF) and Sage-Husa adaptive filtering algorithm can be used to reduce sensor noise and improve the accuracy of attitude data. Additionally, the measurement error caused by the jitter of the lidar itself can be eliminated through the hinge motion compensation of the above-mentioned small three-degree-of-freedom motion platform.

[0028] More specifically, the hinge motion compensation of the above-mentioned small three-degree-of-freedom motion platform can be achieved by setting the initial zero pose of the moving platform , converting it into a dual quaternion, and using it as the initial value for Newton iteration calculation; based on the forward kinematic model, the attitude information of the current moving platform is obtained according to the length changes of the motion branches and the geometric center parameters of the upper and lower Hooke joints; the Jacobian matrix and the objective function are calculated based on the length changes of the motion branches and the geometric center parameters of the upper and lower Hooke joints; the pose error of the iteration between the current pose and the ideal pose is measured based on the objective function , constructing an iteration sequence and calculating the iterative pose; The pose error of the iteration should satisfy the following formula: ; where is a set threshold, and the output pose is converted into Euler angles. By constructing a Newton iteration sequence, the position and attitude of the moving platform are efficiently solved. The motor driver receives the control signal, drives the servo motor, and the servo motor drives the motion branches to work. The motion branches then achieve their predetermined motion trajectories through the extension or shortening of the electric cylinders. Repeating the above steps can correct the data collected by the lidar in real time during its operation.

[0029] The above-mentioned target point cloud data can refer to the point cloud data after the above-mentioned anti-shake compensation processing, that is, the jitter and error caused by factors such as vehicle movement, vibration, and tilt are eliminated, making it more stable and accurate, and capable of truly reflecting the road surface characteristics. It can be understood that the above-mentioned target point cloud data has improved in terms of accuracy, continuity, and stability compared to the original point cloud data, providing reliable input for subsequent point cloud preprocessing, roughness analysis, and classification.

[0030] The above-mentioned preprocessing process can refer to a series of optimization processes performed on the target point cloud data after it is obtained to remove redundant information, reduce noise, and improve data quality, making it more suitable for road surface roughness calculation and classification. Specifically, methods such as median filtering and voxel grid downsampling can be used to denoise the above-mentioned target point cloud data, and the corresponding point cloud rotation matrix R and translation vector t are calculated through the ICP algorithm. The preprocessed point cloud data is aligned through the point cloud rotation matrix R and translation vector t to ensure that the data matches the real road.

[0031] More specifically, the above preprocessing process can be carried out through the following steps: Downsample through a voxel grid to reduce redundant information in the point cloud data, remove non-ground points and outliers in the point cloud data based on the median filtering algorithm, and define the point set X. , calculate the centroid of the point cloud based on the following formula;

[0032] Calculate the eigenvalues by constructing the covariance matrix of the point set:

[0033] Use the eigenvalues and eigenvectors to calculate the information entropy and select clusters, extract the central key points of each category, and for each point in the source point cloud based on the KD-tree algorithm , find the nearest neighbor point in the target point cloud , remove point pairs with a distance exceeding the threshold or a large difference in the normal direction.

[0034] Calculate the centers of the source point cloud and the target point cloud: ;

[0035] Calculate the decentralized coordinates: ;

[0036] Construct the covariance matrix:

[0037] Perform SVD (Singular Value Decomposition) decomposition on H:

[0038] Where U is the left singular vector matrix, V is the left singular vector matrix, and A is the diagonal matrix with elements being the singular values; Calculate the rotation matrix:

[0039] Calculate the translation vector: -R

[0040] Update the source point cloud: +t Solve for the optimal R (rotation matrix) and t (translation vector) through the following steps, and apply the estimated transformation matrix T = {R, t} to the current frame point cloud to eliminate the relative motion with the previous frame. The point-to-plane ICP algorithm improved based on the bidirectional KD-tree fuses point cloud data with different coordinate systems into a complete scene to achieve precise point cloud registration.

[0041] The above preset roughness model can refer to a mathematical or deep learning model used to calculate and evaluate road roughness. This model is based on the preprocessed feature point cloud data, extracts road surface features and quantifies the roughness level, providing an accurate road condition assessment for vehicle driving. It can be understood that the above preset roughness model can extract the characteristic information of road conditions by calculating road surface height changes, slopes, curvatures, etc., and identify and extract the roughness data of road conditions through methods such as feature enhancement of the mixing pool layer, growth of spatial feature expression, and calculation of hierarchical weights.

[0042] The above point cloud classification process can refer to classifying the feature point cloud data according to the preset roughness model to obtain the classification result of road roughness. This process maps the road surface features to different roughness categories to help analyze the road surface condition.

[0043] Specifically, the above classification process can be implemented through the following steps: Based on the feature point cloud data, generate the local polar coordinates, local Cartesian product coordinates, and point features of the domain points to better associate the position and orientation information of the neighborhood points within the group; Based on the domain feature enhancement point cloud semantic segmentation method of PointNet++, enhance the local position information, and use a hybrid pooling method that combines max pooling and self-attention pooling to fuse the point features of each group into a feature value vector. The calculation formulas of max pooling and self-attention pooling are as follows:

[0044]

[0045] Among them, and respectively represent the results of max pooling and self-attention pooling for the i-th group, represents the feature vector of all points within the group, is the learned attention score.

[0046] Then, fuse the two pooling methods of max pooling and self-attention pooling, and use the MLP (multi-layer perceptron) operation to align the point feature lengths. The formula is:

[0047] It represents the final point feature grouped with the i-th point as the center point.

[0048] The process of classification can be specifically achieved through the following steps: Based on the local features extracted by the improved Pointnet++ model, the inverse density features are extracted from the point cloud using a three-layer MLP transformation. The inverse density factors are obtained by grouping the local regions, and they are weighted with the original point cloud features to obtain the point cloud features with density information; Based on the kernel density estimation method, the density features of each point are calculated to better estimate the point cloud distribution in the local region. At the point ( , ) the kernel density calculation formula is:

[0049] where n is the number of points in the local region, is the current point and the square of the distance from other points in the local region; bandwidth is the bandwidth, which determines the scope of the kernel function; The density distribution is obtained through kernel density estimation, and the non-linear transformation is realized using a multi-layer perceptron to obtain the inverse density factor. Based on the neural network learning, the inverse density factor increases the feature weight of the edge points, and its inverse density factor is expressed by the formula:

[0050] The smaller the point density, the larger the inverse density factor, indicating that the point is more at the edge and its contribution to the features of the local region is greater; Subsequently, the attention mechanism module and the spatial attention mechanism module in the preset unevenness model are used to compress the feature map in two different dimensions, and the features are sent to a shared network composed of a multi-layer perceptron and a hidden layer to generate channel attention (C is the number of channels), and the calculation formula is:

[0051] In the formula: is the output weight of the feature F after spatial attention, is the sigmoid activation function; MLP is the multi-layer perceptron; , is the hidden layer weight of the MLP, r is the dimensionality reduction coefficient; , is the output layer weight of the MLP, AvgPool is the average pooling, and MaxPool is the maximum pooling.

[0052] Based on spatial attention to enhance the network's focus on the target point cloud and reduce the interference of noise. Use the max pooling layer to compress the point cloud features and output them to the next set sampling layer, and then output the classification result through the fully connected layer to obtain the target unevenness data of the road surface where the current vehicle is located.

[0053] The above road unevenness extraction system uses lidar and IMU sensors to collect point cloud data during vehicle driving, eliminates errors caused by vehicle vibration through anti-shake compensation, generates more stable target point cloud data, performs denoising, ICP point cloud registration, and deep learning segmentation on the target point cloud data, extracts road feature points, and then calculates parameters such as road height change, slope, and curvature based on a preset unevenness model, and classifies the road unevenness through the CBAM attention mechanism + MLP classification, outputting the target unevenness data of the road surface where the vehicle is located, improving the stability and classification accuracy of the point cloud data, and providing efficient and accurate road unevenness detection for autonomous driving and intelligent transportation.

[0054] As Figure 2 shown, Figure 2 is a flowchart of a road unevenness extraction method provided by an embodiment of the present invention. The road unevenness extraction method includes the steps: 201. Obtain the original point cloud data of the current vehicle.

[0055] In an embodiment of the present invention, the above road unevenness extraction method can be applied to a road unevenness extraction platform. The above road unevenness extraction system has functions such as road unevenness data processing, road unevenness data transceiver, and road unevenness data memory storage, and can be constructed based on a server or a server cluster. The above server or server cluster can be an electronic device with road unevenness data processing capabilities.

[0056] The above original point cloud data can be a set of three-dimensional space coordinate points obtained by lidar and combined with the attitude data of the vehicle provided by the IMU (inertial measurement unit) sensor. It can be used as the basic input for subsequent anti-shake compensation, preprocessing, and unevenness data analysis.

[0057] It can be understood that the lidar set on a small three-degree-of-freedom motion platform can be used to horizontally and vertically scan the road surface where the vehicle is located to obtain the three-dimensional coordinates (X, Y, Z) reflected by each laser beam. Based on the initially obtained reflected three-dimensional coordinates, sparse point cloud data is formed to identify the initial state of the road surface.

[0058] 202. Perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data.

[0059] In an embodiment of the present invention, the above anti-shake compensation process may be a process of eliminating errors in the point cloud data caused by factors such as the vibration and movement of the vehicle itself. Specifically, it may be a dynamic correction of the original point cloud data to eliminate the jitter and errors of the point cloud caused by factors such as vehicle movement, bumpiness, vibration, acceleration, or steering, so as to obtain more stable and accurate point cloud data, reduce motion artifacts, improve the accuracy of the point cloud data, and ensure the reliability of road surface unevenness extraction.

[0060] Specifically, after the above IMU sensor determines the acceleration, angular velocity, and attitude data of the vehicle, the above anti-shake compensation can reduce sensor noise and improve the accuracy of the attitude data through the Kalman filter (UKF) and Sage-Husa adaptive filtering algorithms. It can also eliminate the measurement error caused by the jitter of the lidar through the hinge motion compensation of the above small three-degree-of-freedom motion platform.

[0061] The above target point cloud data may refer to the point cloud data after the above anti-shake compensation process, that is, the jitter and errors caused by factors such as vehicle movement, vibration, and tilt are eliminated, making it more stable and accurate, and capable of truly reflecting the road surface characteristics. It can be understood that the above target point cloud data has improved in terms of accuracy, continuity, and stability compared to the original point cloud data, providing reliable input for subsequent point cloud preprocessing, unevenness analysis, and classification.

[0062] 203. Perform preprocessing on the target point cloud data to obtain feature point cloud data.

[0063] In an embodiment of the present invention, the above preprocessing process may refer to a series of optimization processes performed on the target point cloud data after it is obtained to remove redundant information, reduce noise, improve data quality, and make it more suitable for road surface unevenness calculation and classification. Specifically, methods such as median filtering and voxel grid downsampling can be used to denoise the above target point cloud data, and the corresponding point cloud rotation matrix R and translation vector t can be calculated through the ICP algorithm. The preprocessed point cloud data is aligned through the point cloud rotation matrix R and translation vector t to ensure that the data matches the real road.

[0064] 204. Perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located.

[0065] In an embodiment of the present invention, the above-mentioned preset unevenness model may refer to a mathematical or deep learning model used to calculate and evaluate road unevenness. Based on the preprocessed feature point cloud data, this model extracts road features and quantifies the unevenness level, providing an accurate road condition assessment for vehicle driving. It can be understood that the above-mentioned preset unevenness model can extract the characteristic information of road conditions by calculating road height changes, slopes, curvatures, etc., and identify and extract the unevenness data of road conditions through methods such as feature enhancement in the hybrid pooling layer, spatial feature expression growth, and hierarchical weight calculation.

[0066] The above-mentioned point cloud classification process may refer to classifying the feature point cloud data according to the preset unevenness model to obtain the classification result of road unevenness. This process maps the road surface features to different unevenness categories to help analyze the road surface condition.

[0067] In an embodiment of the present invention, the original point cloud data of the current vehicle is obtained; anti-shake compensation processing is performed on the original point cloud data to obtain target point cloud data; preprocessing is performed on the target point cloud data to obtain feature point cloud data; point cloud classification processing is performed on the feature point cloud data through the preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located. By obtaining point cloud data and performing anti-shake compensation to obtain target point cloud data, and then classifying and calculating the point cloud through the preset unevenness model, the unevenness information of the current road surface is accurately extracted, reducing the impact of vehicle vibration on the data, improving the stability of the point cloud, realizing high-precision, low-noise, and anti-shake road unevenness detection, obtaining stable and accurate point cloud data, enabling the vehicle to adapt to complex terrains, and improving performance.

[0068] Optionally, in the step of obtaining the original point cloud data of the current vehicle, the current vehicle can also be scanned in the horizontal and vertical angle directions by a lidar to determine the three-dimensional coordinates in the corresponding directions; based on the three-dimensional coordinates, sparse point cloud data is generated; the acceleration and angular velocity of the current vehicle are detected in real time through an IMU sensor to determine the attitude data of the current vehicle.

[0069] In an embodiment of the present invention, the above-mentioned original point cloud data may include but is not limited to the attitude data and point cloud data of the vehicle. The above-mentioned attitude data may include but is not limited to the degree of bumpiness, steering data, and acceleration conditions.

[0070] The above-mentioned sparse point cloud data may refer to point cloud data with a relatively low distribution density of points, a relatively large distance between points, and that does not fully cover the target surface or area. It is usually caused by factors such as sensor resolution, scanning distance, sampling method, or data dimensionality reduction. Therefore, it is necessary to preprocess the sparse point cloud data to enhance its stability and reliability.

[0071] In a possible embodiment, the above road surface unevenness extraction system acquires the vehicle attitude and the original point cloud data through lidar scanning and IMU sensor measurement. Specifically, the above lidar scans in the horizontal and vertical directions of the vehicle, determines the three-dimensional coordinates based on the time-of-flight (ToF) method, and generates sparse point cloud data; the above IMU sensor records the acceleration and angular velocity in real time, and uses the Kalman filter + quaternion fusion algorithm to calculate the pitch angle, roll angle, and yaw angle of the vehicle, obtains accurate vehicle attitude data, and can also compensate the point cloud coordinates in combination with the rotation matrix R and the translation vector t to eliminate the drift caused by vehicle movement, and finally generates stable and high-precision target point cloud data, providing accurate data support for road unevenness detection, autonomous driving perception, and intelligent transportation systems.

[0072] Optionally, in the step of performing anti-shake compensation processing on the original point cloud data to obtain the target point cloud data, it further includes performing anti-shake compensation processing on the original point cloud data through a preset anti-shake compensation algorithm to obtain the first compensation parameter; obtaining the vibration parameters of the current vehicle through the IMU sensor; establishing a motion state equation and vibration parameters based on the ARIMA model, performing prediction processing on the original point cloud data to determine the second compensation parameter; and processing the original point cloud data based on the first compensation parameter and the second compensation parameter to obtain the target point cloud data.

[0073] In the embodiment of the present invention, the above preset anti-shake compensation algorithm may include, but is not limited to, algorithms such as Kalman filter and quaternion fusion algorithm that use the vehicle's own vibration parameters for anti-shake.

[0074] The above vibration parameters may include, but are not limited to, vibration characteristics used to quantify the influence of the road surface on the vehicle during driving, such as acceleration, angular velocity, frequency analysis, etc.

[0075] The above first compensation parameter may be a point cloud error correction amount calculated based on a preset anti-shake compensation algorithm, used to eliminate the point cloud jitter caused by vehicle attitude changes (such as bumps, steering, acceleration, etc.), ensure that the point cloud data is as aligned as possible with the real road surface. Specifically, the acceleration, angular velocity, and attitude information (pitch angle, roll angle, yaw angle) of the vehicle can be obtained through the IMU sensor, and the unscented Kalman filter (UKF) + Sage-Husa adaptive filter can also be used to optimize the IMU data and reduce measurement noise. The attitude error compensation matrix can also be calculated:

[0076] Among them, is the rotation matrix provided by the IMU sensor (representing the tilt state of the vehicle), is the translation offset calculated by the IMU.

[0077] Through inverse kinematics calculation, the point cloud data scanned by the lidar is corrected for rotation and translation:

[0078] Among them, is the original point cloud data, is the corrected point cloud data.

[0079] The above second compensation parameter can be the compensation amount of the motion trend of the point cloud data calculated by the ARIMA prediction model based on the vibration parameters measured by the IMU sensor, which is used to predict the motion trend of the vehicle in the short term in the future and correct the potential errors of the point cloud data.

[0080] In a possible embodiment, the above road surface unevenness extraction system obtains the original point cloud data through lidar scanning, and the IMU sensor synchronously records the acceleration, angular velocity and vehicle attitude information to generate vibration parameters. According to the above preset anti-shake compensation algorithm, the first compensation parameter is calculated, and the attitude error is corrected by using the IMU data. Combining with the ARIMA prediction model, the second compensation parameter is calculated, the point cloud drift is predicted based on the vehicle motion trend, and the original point cloud data is optimized by fusing the first compensation parameter and the second compensation parameter to generate stable and high-precision target point cloud data, reduce the point cloud jitter error, improve the data continuity, and provide accurate basic input data for road unevenness detection, autonomous driving environment perception and intelligent transportation applications.

[0081] Optionally, in the step of preprocessing the target attitude data and the target point cloud data to obtain the feature point cloud data, it further includes denoising the target point cloud data through the median filtering algorithm to obtain the first feature point cloud data; calculating the point cloud rotation matrix R and the translation amount t for the first feature point cloud data through the ICP algorithm and the bidirectional KD algorithm; registering the first feature point cloud data according to the point cloud rotation matrix R and the translation amount t to generate the second feature point cloud data; performing point cloud segmentation processing on the second feature point cloud data through the point cloud segmentation algorithm to obtain the third feature point cloud data.

[0082] In the embodiment of the present invention, the above first feature point cloud data may refer to the point cloud data obtained after denoising the target point cloud data, and its purpose is to eliminate noise points, abnormal points and outliers, improve the quality of the point cloud data, and provide high-precision input for subsequent point cloud registration and segmentation.

[0083] The above second feature point cloud data may refer to the data after aligning and completing point cloud registration by calculating the ICP (Iterative Closest Point) and bidirectional KD-tree (KD-Tree) algorithms for the first feature point cloud data. Its purpose is to keep the point cloud data spatially consistent in different time frames or different coordinate systems, and ensure that the point cloud data can be accurately aligned with the road surface.

[0084] The above-mentioned third feature point cloud data may refer to the point cloud data with specific categories and spatial features after point cloud segmentation processing, which can be used for further unevenness calculation and road analysis.

[0085] The above-mentioned point cloud segmentation processing may refer to dividing the point cloud data into different categories or regions according to the spatial features, geometric information or deep learning methods of the point cloud, so as to facilitate subsequent unevenness detection and classification. Specifically, based on the deep learning segmentation of PointNet++, local polar coordinates + Cartesian coordinate fusion can be used to extract point cloud features, and combined with max pooling + self-attention pooling (Hybrid Pooling) to improve the segmentation accuracy.

[0086] In a possible embodiment, the above-mentioned road surface unevenness extraction system gradually optimizes the target point cloud data through median filtering denoising, ICP registration and point cloud segmentation, and improves the accuracy of road unevenness detection. Specifically, the median filtering algorithm is used to remove noise points to obtain the first feature point cloud data, the ICP algorithm is combined with a bidirectional KD tree to calculate the rotation matrix RRR and translation vector ttt of the point cloud, and the point cloud is aligned and registered to generate the second feature point cloud data. Based on the point cloud segmentation algorithm, key points on the road surface are extracted to obtain the third feature point cloud data, effectively identifying the road area and improving the stability, alignment accuracy and classification ability of the point cloud data.

[0087] Optionally, in the step of performing point cloud segmentation processing on the second feature point cloud data through a point cloud segmentation algorithm to obtain the third feature point cloud data, it further includes extracting neighborhood features of the second feature point cloud data to determine each center point and its neighboring point cloud group in the second feature point cloud data; generating corresponding feature coordinate groups based on each center point and its neighboring point cloud group; performing point feature fusion processing on each feature coordinate group according to the point cloud semantic segmentation algorithm to obtain a feature value vector corresponding to each feature coordinate group; and using the feature value vector as the third feature point cloud data.

[0088] In the embodiment of the present invention, the above-mentioned neighborhood feature extraction may refer to determining the neighborhood features of each point based on the spatial relationship of the point cloud, which is used to capture local geometric information, thereby enhancing the expression ability of the point cloud data.

[0089] The above-mentioned point cloud group may refer to a set of neighboring points obtained based on neighborhood feature extraction, which is used to represent the local spatial structure of the point cloud.

[0090] The above-mentioned feature coordinate group may refer to performing different coordinate representations on the points in the point cloud group to enhance the spatial distinguishability of the point cloud features, and specifically may include but is not limited to local polar coordinates, Cartesian coordinates, and position feature coordinates.

[0091] The above point cloud semantic segmentation algorithm can refer to classifying point cloud data and assigning semantic labels based on deep learning or traditional feature analysis methods. Generally speaking, it can be calculated based on deep learning (models such as PointNet++ and RandLA-Net), using MLP (Multi-Layer Perceptron) + attention mechanism to extract point cloud features and perform classification. It can also be based on geometric features, using features such as point cloud curvature and normal vector for clustering segmentation.

[0092] The above feature fusion processing can refer to fusing multiple feature coordinate groups to generate a point cloud feature vector with more global and local information. Specifically, the feature fusion processing through feature coordinate groups can be obtained through the following steps: Based on the feature point cloud data, generate the local polar coordinates, local Cartesian product coordinates, and point features of the neighborhood points to better associate the position and orientation information of the neighborhood points within the group; Based on the neighborhood feature enhanced point cloud semantic segmentation method of PointNet++, enhance the local position information, and use a hybrid pooling method that combines max pooling and self-attention pooling to fuse the point features of each group into a feature value vector. The calculation formulas of max pooling and self-attention pooling are as follows:

[0093]

[0094] Among them, and respectively represent the results of the i-th group's max pooling and self-attention pooling, represents the feature vector of all points within the group, is the learned attention score.

[0095] Then, fuse the two pooling methods of max pooling and self-attention pooling, and use MLP (Multi-Layer Perceptron) operation to align the point feature lengths. The formula is:

[0096] represents the final point feature grouped with the i-th point as the center point.

[0097] The above feature value vector can refer to the fused point cloud feature representation for semantic segmentation or classification tasks.

[0098] Optionally, in the step of performing point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located, it further includes performing feature enhancement processing on the feature point cloud data through a preset mixing pool layer to obtain the first point cloud classification data; performing local feature expression enhancement processing on the first point cloud classification data through a preset spatial layer to obtain the second point cloud classification data; and processing the second point cloud classification data through a preset hierarchical weight to output the target unevenness data of the road surface where the current vehicle is located.

[0099] In the embodiment of the present invention, the above-mentioned feature enhancement processing refers to optimizing the feature point cloud data through a specific algorithm so that it can more comprehensively express the unevenness information of the road surface.

[0100] The above-mentioned first point cloud classification data refers to the preliminary classification data obtained after feature enhancement processing and is used for subsequent local feature optimization.

[0101] The above-mentioned preset spatial layer refers to a computational layer used to enhance the local feature expression of point cloud data. It focuses on the relationship of point cloud data in space to ensure local consistency in classification.

[0102] The above-mentioned local feature expression enhancement processing refers to further optimizing the local feature expression ability of point cloud data on the basis of the preset spatial layer to ensure more accurate classification data.

[0103] The above-mentioned second point cloud classification data refers to the optimized classification data obtained after local feature expression enhancement processing. Compared with the first point cloud classification data, it contains more refined feature information.

[0104] The above-mentioned preset hierarchical weight refers to assigning different weights to the data of different feature layers in the final classification stage to ensure more accurate classification results.

[0105] The above-mentioned target unevenness data refers to the final output road unevenness classification result after hierarchical weight calculation and is used for road condition analysis and driving optimization.

[0106] Specific feature extraction can be achieved according to the following steps: Based on the local features extracted by the improved Pointnet++ model, inverse density features are extracted from the point cloud using a three-layer MLP transformation, inverse density factors are obtained by grouping local regions, and they are weighted with the original point cloud features to obtain point cloud features with density information; Based on the kernel density estimation method, the density feature of each point is calculated to better estimate the point cloud distribution in the local region. At the point ( , ) the kernel density calculation formula is:

[0107] where n is the number of points in the local area, is the current point and other points in the local area The square of the distance, bandwidth is the bandwidth, which determines the scope of action of the kernel function; The density distribution is obtained through kernel density estimation, the non-linear transformation is realized by using a multi-layer perceptron to obtain the inverse density factor, and the feature weight of the edge points is increased based on the neural network learning of the inverse density factor. The inverse density factor The formula is expressed as:

[0108] The smaller the point density, the larger the inverse density factor, indicating that the point is more at the edge and its contribution to the features of the local area is greater.

[0109] In a possible embodiment, the above road surface unevenness extraction system performs feature enhancement processing on the feature point cloud data, extracts key information by using a hybrid pooling layer to generate first point cloud classification data; optimizes the local feature expression through a spatial layer to obtain second point cloud classification data, and finally performs comprehensive analysis on the data of different feature layers by combining hierarchical weight calculation to output the unevenness level of the road surface where the vehicle is located, improving the classification accuracy of the unevenness data and ensuring the stability and reliability of the unevenness detection.

[0110] As Figure 4 shown, an embodiment of the present invention further provides a road surface unevenness extraction device 400. The road surface unevenness extraction device 400 includes: A first acquisition module 401, configured to acquire the original point cloud data of the current vehicle; A second acquisition module 402, configured to perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data; A third acquisition module 404, configured to perform preprocessing on the target point cloud data to obtain feature point cloud data; A fourth acquisition module 404, configured to perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located.

[0111] Optionally, the above first acquisition module 401 includes: A first determination sub-module, configured to scan the current vehicle in the horizontal and vertical angle directions through a lidar to determine the three-dimensional coordinates in the corresponding directions; A first generation sub-module, configured to generate sparse point cloud data based on the three-dimensional coordinates; A second determination sub-module, configured to detect the acceleration and angular velocity of the current vehicle in real time through an IMU sensor, and determine the attitude data of the current vehicle, where the attitude data includes the bumpiness degree, steering data, and acceleration condition.

[0112] Optionally, the above-mentioned second acquisition module 402 includes: A first acquisition sub-module, configured to perform anti-shake compensation processing on the original point cloud data through a preset anti-shake compensation algorithm to obtain a first compensation parameter; A second acquisition sub-module, configured to acquire the vibration parameters of the current vehicle through an IMU sensor; Based on the ARIMA model, establish a motion state equation and the vibration parameters, perform prediction processing on the original point cloud data, and determine a second compensation parameter; A third acquisition sub-module, configured to process the original point cloud data based on the first compensation parameter and the second compensation parameter to obtain target point cloud data.

[0113] Optionally, the above-mentioned third acquisition module 404 includes: A fourth acquisition sub-module, configured to denoise the target point cloud data through a median filtering algorithm to obtain first feature point cloud data; A third determination sub-module, configured to calculate the first feature point cloud data through an ICP algorithm and a bidirectional KD algorithm to determine a point cloud rotation matrix R and a translation amount t; A second generation sub-module, configured to register the first feature point cloud data according to the point cloud rotation matrix R and the translation amount t to generate second feature point cloud data; A fifth acquisition sub-module, configured to perform point cloud segmentation processing on the second feature point cloud data through a point cloud segmentation algorithm to obtain third feature point cloud data.

[0114] Optionally, the above-mentioned device further includes: A first determination module, configured to extract domain features from the second feature point cloud data to determine each center point and its adjacent point cloud group in the second feature point cloud data; A second determination module, configured to generate corresponding feature coordinate groups based on each center point and its adjacent point cloud group, where the feature coordinate groups include local polar coordinates, Cartesian coordinates, and position feature coordinates; A third determination module, configured to perform point feature fusion processing on each feature coordinate group according to a point cloud semantic segmentation algorithm to obtain a feature value vector corresponding to each feature coordinate group; A fourth determination module, configured to use the feature value vector as the third feature point cloud data.

[0115] Optionally, the above-mentioned fourth acquisition module 404 includes: The fifth acquisition sub-module is used to perform feature enhancement processing on the feature point cloud data through a preset mixing pool layer to obtain first point cloud classification data; The sixth acquisition sub-module is used to perform local feature expression enhancement processing on the first point cloud classification data through a preset space layer to obtain second point cloud classification data; The output sub-module is used to process the second point cloud classification data through a preset hierarchical weight and output target unevenness data of the road surface where the current vehicle is located.

[0116] As Figure 5 shown, an embodiment of the present invention further provides an electronic device 500, including a processor, and the above-mentioned processor can execute any one of the above-mentioned road surface unevenness extraction methods.

[0117] Specifically, it includes a processor 501, a memory 502, and a computer program for executing the road surface unevenness extraction method stored on the memory 502 and capable of running on the processor 501, where: The processor 501 runs the calculator program of the road surface unevenness extraction method stored in the memory 502 and executes the following steps: Obtain the original point cloud data of the current vehicle; Perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data; Perform preprocessing on the target point cloud data to obtain feature point cloud data; Perform point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain target unevenness data of the road surface where the current vehicle is located.

[0118] Optionally, the processor 501 executes the original point cloud data including the attitude data and point cloud data of the vehicle, and obtaining the original point cloud data of the current vehicle includes: Scan the current vehicle in the horizontal and vertical angle directions through a lidar to determine the three-dimensional coordinates in the corresponding directions; Generate sparse point cloud data based on the three-dimensional coordinates; Real-time detect the acceleration and angular velocity of the current vehicle through an IMU sensor to determine the attitude data of the current vehicle, and the attitude data includes the bumpiness degree, steering data, and acceleration situation.

[0119] Optionally, the processor 501 executes performing anti-shake compensation processing on the original point cloud data to obtain target point cloud data, including: Perform anti-shake compensation processing on the original point cloud data through a preset anti-shake compensation algorithm to obtain a first compensation parameter; Obtain the vibration parameters of the current vehicle through an IMU sensor; Based on the ARIMA model, establish a motion state equation and the vibration parameters, and perform prediction processing on the original point cloud data to determine the second compensation parameter; Based on the first compensation parameter and the second compensation parameter, process the original point cloud data to obtain the target point cloud data.

[0120] Optionally, the processor 501 executes the preprocessing of the target attitude data and the target point cloud data to obtain the feature point cloud data, including: Denoise the target point cloud data through a median filtering algorithm to obtain the first feature point cloud data; Calculate the first feature point cloud data through the ICP algorithm and the bidirectional KD algorithm to determine the point cloud rotation matrix R and the translation amount t; Register the first feature point cloud data according to the point cloud rotation matrix R and the translation amount t to generate the second feature point cloud data; Perform point cloud segmentation processing on the second feature point cloud data through a point cloud segmentation algorithm to obtain the third feature point cloud data.

[0121] Optionally, when the processor 501 executes the point cloud segmentation processing on the second feature point cloud data through the point cloud segmentation algorithm to obtain the third feature point cloud data, the method further includes: Extract the domain features of the second feature point cloud data to determine each center point and its adjacent point cloud group in the second feature point cloud data; Generate corresponding feature coordinate groups based on each center point and its adjacent point cloud group, and the feature coordinate groups include local polar coordinates, Cartesian coordinates, and position feature coordinates; Perform point feature fusion processing on each feature coordinate group based on a point cloud semantic segmentation algorithm to obtain an eigenvalue vector corresponding to each feature coordinate group; Use the eigenvalue vector as the third feature point cloud data.

[0122] Optionally, the processor 501 also executes the point cloud classification processing on the feature point cloud data through a preset unevenness model to obtain the target unevenness data of the road surface where the current vehicle is located, including: Perform feature enhancement processing on the feature point cloud data through a preset hybrid pooling layer to obtain the first point cloud classification data; Perform local feature expression enhancement processing on the first point cloud classification data through a preset spatial layer to obtain the second point cloud classification data; Process the second point cloud classification data through a preset hierarchical weight and output the target unevenness data of the road surface where the current vehicle is located.

[0123] An embodiment of the present invention also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, it implements each process of the road surface unevenness extraction method or the application-side road surface unevenness extraction method provided by the embodiment of the present invention, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0124] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0125] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for extracting road roughness, characterized in that: include: Get the original point cloud data of the current vehicle; Performing anti-shake compensation processing on the original point cloud data to obtain target point cloud data; Preprocessing the target point cloud data to obtain feature point cloud data; The feature point cloud data is subjected to point cloud classification processing by using a preset roughness model to obtain target roughness data of the road surface where the current vehicle is located.

2. The road surface roughness extraction method according to claim 1, characterized in that: The original point cloud data includes posture data and point cloud data of the vehicle, and the obtaining of the original point cloud data of the current vehicle includes: Scan the current vehicle in the horizontal and vertical directions through the laser radar to determine the three-dimensional coordinates in the corresponding direction; Based on the three-dimensional coordinates, generating sparse point cloud data; The IMU sensor detects the acceleration and angular velocity of the current vehicle in real time to determine the posture data of the current vehicle, which includes the degree of bumpiness, steering data, and acceleration conditions.

3. The road surface roughness extraction method according to claim 1, characterized in that: The performing anti-shake compensation processing on the original point cloud data to obtain target point cloud data includes: Performing anti-shake compensation processing on the original point cloud data by using a preset anti-shake compensation algorithm to obtain a first compensation parameter; Obtain the vibration parameters of the current vehicle through the IMU sensor; Establishing a motion state equation and the vibration parameters based on the ARIMA model, performing prediction processing on the original point cloud data, and determining a second compensation parameter; The original point cloud data is processed based on the first compensation parameter and the second compensation parameter to obtain target point cloud data.

4. The road surface roughness extraction method according to claim 1, characterized in that: The preprocessing of the target posture data and the target point cloud data to obtain feature point cloud data includes: De-noising the target point cloud data by using a median filtering algorithm to obtain first feature point cloud data; The first feature point cloud data is calculated by using the ICP algorithm and the bidirectional KD algorithm to determine the point cloud rotation matrix R and the translation t; Registering the first feature point cloud data according to the point cloud rotation matrix R and translation t to generate second feature point cloud data; The second feature point cloud data is processed by point cloud segmentation algorithm to obtain third feature point cloud data.

5. The method for extracting road roughness according to claim 4, characterized in that: The method further comprises: performing point cloud segmentation processing on the second feature point cloud data by using a point cloud segmentation algorithm to obtain third feature point cloud data. Performing domain feature extraction on the second feature point cloud data to determine each center point and its adjacent point cloud groups in the second feature point cloud data; Based on each center point and its adjacent point cloud group, a corresponding feature coordinate group is generated, wherein the feature coordinate group includes local polar coordinates, Cartesian coordinates and position feature coordinates; Based on the point cloud semantic segmentation algorithm, point feature fusion processing is performed according to each feature coordinate group to obtain the eigenvalue vector corresponding to each feature coordinate group; The eigenvalue vector is used as the third feature point cloud data.

6. The method for extracting road roughness as claimed in claim 1, characterized in that: The step of performing point cloud classification processing on the characteristic point cloud data by using a preset roughness model to obtain target roughness data of the road surface where the current vehicle is located includes: Performing feature enhancement processing on the feature point cloud data through a preset hybrid pool layer to obtain first point cloud classification data; Performing local feature expression enhancement processing on the first point cloud classification data through a preset spatial layer to obtain second point cloud classification data; The second point cloud classification data is processed by preset hierarchical weights to output target roughness data of the road surface where the current vehicle is located.

7. A road surface roughness extraction device, characterized in that: include: The first acquisition module is used to acquire the original point cloud data of the current vehicle; A second acquisition module is used to perform anti-shake compensation processing on the original point cloud data to obtain target point cloud data; A third acquisition module is used to pre-process the target point cloud data to obtain feature point cloud data; The fourth acquisition module is used to perform point cloud classification processing on the characteristic point cloud data through a preset roughness model to obtain target roughness data of the road surface where the vehicle is currently located.

8. A road roughness extraction system, characterized in that: The road surface roughness extraction system comprises: a road surface roughness extraction device; The road surface roughness extraction device implements the road surface roughness extraction method described in claim 1.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the road surface roughness extraction method as claimed in any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the road surface roughness extraction method according to any one of claims 1 to 6 are implemented.

Images