Road edge feature extraction and state switching method based on laser radar point cloud data

Abstract

The invention discloses a road edge feature extraction and state switching method based on laser radar point cloud data. The method comprises the steps of obtaining target point cloud data collected by a laser radar; dividing the point cloud area into ground point cloud and non-ground point cloud; extracting road edge feature points of multi-dimensional spatial feature fusion; establishing a beam model, and classifying road feature points based on road segmentation lines; carrying out iterative fitting to remove data noise points, and smoothing fitting of road edge feature points by adopting amplitude limiting filtering and Kalman filtering; the method comprises the following steps: monitoring road edge states, carrying out a stable switching strategy between a road edge state and a road edge-free state, and identifying and smoothly processing a state switching process by analyzing a continuous multi-frame detection result. The road edge detection method has the advantages that high precision and stability of road edge detection are realized, and high efficiency and intelligence of detection are improved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a roadside feature extraction and state switching method based on laser radar point cloud data. Background Art

[0002] With the rapid development of autonomous driving technology and the increasing number of cars in cities, the ability of autonomous vehicles to perceive the environment has become a key factor in achieving safe and efficient driving. Among them, curb detection, as an important means to determine the boundaries of the vehicle's drivable area, is of great significance to ensuring the safe driving of autonomous vehicles. LiDAR, as a sensor that can obtain three-dimensional spatial information in real time, plays a vital role in curb detection.

[0003] However, the current LiDAR-based curb detection algorithm still faces many challenges in practical applications. First, traditional methods rely on the height or angle mutation characteristics of the laser scan line at the curb for feature extraction. This method has high requirements on the performance of the LiDAR and the quality of the scan line, and is easily disturbed by factors such as point cloud density and noise, making it difficult to ensure detection accuracy and stability. Especially in complex and changeable urban environments, the shapes, materials, and occlusion of curbs vary, further increasing the difficulty of detection.

[0004] In addition, existing curb detection algorithms often have problems with large computational complexity and slow processing speed when processing large-scale point cloud data, making it difficult to meet the real-time requirements of autonomous driving systems. At the same time, since curbs are usually represented as small targets distributed along a linear line in point cloud data, their feature extraction and classification processes are complex, and missed detections and false detections are prone to occur, further affecting the accuracy of the detection results.

[0005] This is an area where this application needs to be improved. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a roadside feature extraction and state switching method based on laser radar point cloud data, thereby improving the efficiency and intelligence of detection.

[0007] In order to solve the above technical problems, the present invention provides a method for extracting roadside features and switching states based on laser radar point cloud data, comprising the following steps: Step S1: Obtain target point cloud data collected by the laser radar; Use laser radar equipment to perform three-dimensional scanning of the surrounding environment and obtain point cloud data of the target scene; The point cloud data contains the three-dimensional coordinate information of the surface of each object in the scene, providing a basic data source for subsequent curb detection; Step S2: Divide the point cloud area, perform plane fitting in sections, and divide the ground point cloud and non-ground point cloud; In order to accurately divide the ground and non-ground point clouds, a piecewise plane fitting strategy is adopted: 1) Divide the original point cloud P into multiple continuous segments along the x-axis to reduce the impact of ground slope changes on fitting; 2) Apply the RANSAC algorithm to each point cloud for plane fitting. The RANSAC algorithm effectively handles noise and outliers and accurately identifies points that conform to the ground plane model, i.e., ground points; 3) According to the fitting results, the point cloud is divided into ground point cloud and non-ground point cloud; Step S3: Extracting curb feature points by fusion of multi-dimensional spatial features; Based on the height difference feature, smoothness feature and horizontal distance feature, the roadside feature points are initially extracted; 1) Height difference feature extraction: Use height difference features to capture significant changes in the vertical dimension of the curb by calculating the difference between the maximum and minimum values ​​of adjacent point cloud areas; 2) Smoothness feature evaluation: Smoothness features are introduced to evaluate the smoothness of local areas of the point cloud. By carefully analyzing the area around the candidate points, the smoothness feature distinguishes the roadside feature points from the ground or non-roadside objects; 3) Horizontal distance feature determination: Based on the height difference and smoothness analysis, the horizontal distance feature is used to measure the distance between two adjacent points on the same horizontal plane. By setting a reasonable threshold, the key points that are closely connected in the horizontal direction and constitute the roadside contour can be accurately identified; Step S4: Establish a beam model, use a peak-finding algorithm to determine the road segmentation line, and classify the road feature points based on the road segmentation line; In order to solve the problem of accurate classification of boundary points in complex road environments, the present invention constructs an off-point beam model, which quantifies the length of each beam and simulates the interaction between the beam and the ground and potential road boundaries to accurately determine the road direction and classify the feature points along the road; Specifically, the longest light beam in the area in front and behind the vehicle is studied as the baseline for road segmentation, which accurately reflects the main extension direction of the road. However, the sparsity and noise interference of point cloud data may cause the directly calculated distance function to contain multiple pseudo extreme points. Therefore, the median filtering technology is used to preprocess the distance function to filter out noise and highlight the true extreme values. Each dominant extreme value is screened according to the extreme value width to ensure that only two dominant extreme values ​​are retained, representing the left and right edges of the road respectively. By connecting the midline of the area where these two extreme points are located, the accurate definition of the road segmentation line is achieved; Based on the road segmentation line, the road feature points are classified: the feature points on the left side of the segmentation line are classified into the feature point set on the left side of the roadside, and the feature points on the right side are classified into the feature point set on the right side of the roadside. This detailed classification process enhances the semantic information of the feature points; Step S5: Iterative fitting to remove data noise points, and use limit filtering and Kalman filtering to smooth the fitting of the roadside feature points; It aims to integrate fitting and filtering optimization techniques to achieve accurate extraction of curb feature points and construction of boundary equations; Step S51, using iterative fitting technology to remove data noise, selecting the straight line equation as the fitting model, using the gradient descent algorithm to perform preliminary fitting and estimate parameters; through multiple iterations, fitting the feature points based on the current parameters and calculating the errors, continuously optimizing the algorithm to minimize the errors until the preset number of iterations is reached, and obtaining preliminary fitting parameters; calculating the accuracy of the fitting error evaluation, and using the statistical filtering method to eliminate false detection points or outliers that exceed the set threshold; refitting the remaining feature points to obtain a more accurate and stable fitting result; Step S52, applying a limiting filtering strategy to eliminate interference detection; for interference situations encountered by the laser radar during the detection process, such as a moving vehicle being mistakenly identified as a curb, by continuously storing the latest boundary equation slope and calculating its average value and variance, only when both meet the preset threshold conditions, the current linear equation is considered valid, thereby effectively eliminating interference factors; Step S53, using Kalman filtering technology to implement smoothing optimization on the boundary equation; Specifically, the slope k and intercept b of the boundary equation obtained after limiting filtering are used as observation data and input into the Kalman filter equation. In a short period of time, the changes in the slope and intercept are approximately regarded as a uniform process, so the uniform model is used to perform Kalman filtering smoothing on the boundary equation, which aims to obtain more stable and accurate boundary estimation results, thereby improving the accuracy and reliability of the overall analysis. Step S6: roadside status monitoring, smooth switching strategy between roadside and no roadside status; The aim is to achieve accurate identification and seamless connection of the curb state (with or without curb) and its dynamic switching (such as from with curb to without curb, and from without curb to with curb). Given the complex and ever-changing road environment, vehicles need to flexibly respond to frequent lane changes, turns, and road type conversions, all of which require the curb state detection to be highly sensitive and accurate; An advanced state machine model is introduced to define four curb states, each of which closely matches the road conditions that autonomous vehicles may encounter in actual driving. On this basis, an intelligent curb state assessment system is constructed by comprehensively analyzing the detected feature points, the curb results after fitting processing, and the time series of multi-frame detection results. In the case of a curb, the detection accuracy of feature points in a single frame is not only relied on, but also the comparison and analysis of multi-frame fitting results is used to effectively distinguish temporary detection failures caused by special circumstances such as occlusion, ensuring that the output curb status is accurate. On the contrary, in the scene without a curb, the multi-frame comparison strategy is also used to identify and eliminate false detections and avoid the generation of false curb information. When processing curb state switching, attention is paid to the smooth transition of the state; by analyzing the results of continuous multi-frame detection, the state switching process is identified and smoothly processed, effectively avoiding incorrect curb state output caused by abnormal detection of individual frames.

[0008] The superior effect of the present invention lies in: by integrating multi-dimensional features such as height difference and smoothness and combining with filtering algorithms, it not only achieves high precision and stability of curb detection, but also further improves the efficiency and intelligence of detection by optimizing technical processes and introducing intelligent switching mechanisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 is a flow chart of a specific embodiment of the present invention; Figure 2 is a schematic diagram of a piecewise fitting plane of a specific embodiment of the present invention; Figure 3 It is a schematic diagram of a curb state switching model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0011] like Figure 1 As shown, the present invention provides a method for extracting roadside features and switching states based on laser radar point cloud data, comprising the following steps: Step S100: Acquire target point cloud data collected by the laser radar; By deploying a 3D LiDAR system, a 360° precise scan of the surrounding environment can be achieved, thereby capturing and generating a detailed 3D point cloud data set of the surrounding environment in real time. This data set not only covers the precise 3D coordinate information of the surfaces of each object in the scene, but also includes the intensity value of the laser beam when it is reflected on the surface of these objects, as well as the timestamp record of each data point. This series of information constitutes the basic data support for subsequent tasks such as curb detection.

[0012] Step S200: Divide the point cloud area, perform plane fitting in sections, and divide the ground point cloud and non-ground point cloud; the adopted coordinate system is defined as follows: The axis points forward along the centerline of the vehicle, representing the vehicle's forward direction; The axis is located at On the same horizontal plane as the axis, perpendicular to axis and points to the left; Axis perpendicular to The positive direction of the plane points upward. In view of the specific application scenarios and roadside detection requirements, the roadside axis, Axis and The detection area of ​​the axis is , which is used as the main area for roadside detection; In order to accurately distinguish ground point cloud from non-ground point cloud data, a piecewise plane fitting strategy is introduced. Specifically, firstly The original point cloud data is divided into multiple continuous sub-segments by the axis. Considering the actual needs of curb detection and the influence of ground slope changes on fitting accuracy, The axial direction is divided into sections, every 4 Set up a testing area such as Figure 2 As shown in the figure, the detection plane is fitted on this basis, which effectively improves the accuracy of distinguishing ground and non-ground point clouds.

[0013] Step S300: Preliminarily extracting roadside feature points based on height difference, smoothness and horizontal distance spatial features; In order to maximize the capture of all potential roadside feature points, a method of integrating multi-dimensional spatial features is used to achieve accurate extraction of roadside feature points. This process integrates key features such as height difference, smoothness, and horizontal distance. The specific feature extraction method is as follows: 1) Height difference feature extraction: Use height difference features to capture significant changes in the vertical dimension of the curb; and Represents a point cloud area Point cloud in The maximum and minimum values ​​of the direction define the height difference feature as: ; ; In the formula, is the number of points in the point cloud area, is the value of each point cloud in the point cloud area, , , is the set threshold; 2) Smoothness feature extraction: The smoothness feature evaluates the smoothness of the local area of ​​the point cloud; for any point ,use represents the adjacent area of ​​the point, and the smoothness feature is defined as: ; ; In the formula, Representing point clouds The smoothness value of is the smoothness threshold; 3) Horizontal distance feature extraction: Based on the height difference and smoothness analysis, the horizontal distance feature is used to measure the interval between two adjacent points on the same horizontal plane. The horizontal distance threshold is set to determine whether a point is a feature point. Its definition is: ; In the formula, For point The absolute value of the height, is the vertical azimuth of the scanning layer, is the horizontal angular resolution of the lidar; Feature Points The horizontal distance between adjacent points should be less than .

[0014] Step S400: establishing a beam model based on the departure point, using a peak-finding algorithm to determine the road dividing line, and dividing the left and right feature points of the roadside; A ground-to-ground beam model was constructed to accurately determine the road direction by quantifying the length of each beam and simulating the interaction of the beam with the ground and potential road boundaries. In the beam model, the beam length is the distance from the emission point to the nearest point: ; Where: is the origin of the laser radar, and the laser radar scans in a clockwise direction. axis is zero degrees; Indicates the location. ; The beam length of each band is defined as the shortest distance from a point within the band to the transmitting point, that is: ; In the formula, , the beam length of each band Normalized to get the distance equation : ; The peak-finding algorithm is used to identify the local maximum in the point cloud data to determine the boundary of the road segmentation. Each point in the data set is traversed and its value is compared with that of its neighboring points; if the value of a point is greater than the values ​​of all its neighboring points, the point is considered a peak, that is, the boundary point of the road segmentation. Based on the road cutting line, the feature points on the left side of the road cutting line are divided into one category, and the feature points on the right side of the road cutting line are divided into another category.

[0015] Step S500: removing noise points by iterative fitting, and smoothing the road edge fitting result by using limiting filtering and Kalman filtering; The straight line fitting method based on the RANSAC algorithm effectively eliminates the influence of noise points. The road edge line model in the vehicle coordinate system is as follows: ; In the formula, is the slope of the roadside line in the vehicle coordinate system, is the straight line intercept in the vehicle coordinate system; Iterative fitting technology is used to remove data noise, and the gradient descent algorithm is used for preliminary fitting and parameter estimation. Through multiple iterations, the feature points are fitted based on the current parameters and the errors are calculated. Statistical filtering methods are used to eliminate false detection points or outliers that exceed the set threshold. The remaining feature points are refitted to obtain more accurate and stable fitting results. Apply a limiting filtering strategy to eliminate interference detection; for interference situations encountered by the lidar during detection, such as vehicles being mistakenly identified as curbs, the latest boundary equation slope is continuously stored and its average value is calculated and variance ,Only when both meet the preset threshold conditions, the current linear equation is considered valid, thus effectively eliminating interference factors; ; ; The slope k and intercept b of the boundary equation after limiting filtering are used as observation values ​​in the Kalman filter equation; in a short period of time, the changes in the slope and intercept are regarded as a uniform speed model, so the boundary equation is smoothed by Kalman filtering using the uniform speed model. This is to obtain more stable and accurate boundary estimation results, thereby improving the accuracy and reliability of the overall analysis.

[0016] Step S600: Processing the curb detection result to achieve efficient and accurate smooth switching between the curb state and the curb-free state; The core goal of this step is to achieve accurate recognition and seamless connection of the curb state (including curb and no curb) and its dynamic changes (such as the transition from curb to no curb, and from no curb to curb). Given the complex and changeable nature of the road environment, autonomous vehicles need to flexibly respond to frequent lane changes, turning behaviors, and changes in road surface types, all of which place stringent requirements on the sensitivity and accuracy of curb state detection.

[0017] The state machine model is introduced, which defines four curb states, each of which reflects the complex road conditions that autonomous vehicles may encounter in actual road driving. On this basis, an intelligent curb state evaluation system is constructed by analyzing the detected feature point data, the accurately fitted curb results, and the time series information across multiple frames of detection results. The curb state switching model is as follows: Figure 3 shown.

[0018] In the case of a curb, the system not only relies on the high accuracy of single-frame feature point detection, but also effectively identifies and overcomes temporary detection errors caused by special circumstances such as occlusion by comparing fitting results across multiple frames, ensuring that the output curb status information is accurate. In addition, in the case of a temporary failure to detect the curb, the vehicle coordinate system information provided by the inertial measurement unit is used to perform a short-term continuation of the previously detected curb information through coordinate conversion technology. Even in the case of a temporary interruption in detection, relatively accurate curb detection results can be continuously output, thereby maintaining the vehicle's continuous perception and accurate judgment of the road environment.

[0019] In scenarios without curbs, the multi-frame comparison strategy is also used to accurately identify and eliminate potential false detections and prevent the generation of erroneous curb information.

[0020] It is particularly important to ensure a smooth transition of the state when handling the curb state switching. Through in-depth analysis of the continuous multi-frame detection results, it is possible to intelligently identify and smoothly handle the transition phase of the state switching, effectively preventing the curb state error output caused by single-frame detection anomalies, thereby further improving the curb recognition and adaptability of autonomous vehicles.

[0021] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting roadside features and switching states based on laser radar point cloud data, comprising the following steps: Step S1: Obtain target point cloud data collected by the laser radar; Step S2: Divide the point cloud area, perform plane fitting in sections, and divide the ground point cloud and non-ground point cloud; Step S3: Extracting curb feature points by fusion of multi-dimensional spatial features; Based on the height difference feature, smoothness feature and horizontal distance feature, the roadside feature points are initially extracted; Step S4: Establish a beam model, use a peak-finding algorithm to determine the road segmentation line, and classify the road feature points based on the road segmentation line; Based on the road dividing line, the road feature points are classified: the feature points on the left side of the dividing line are classified into the feature point set on the left side of the roadside, and the feature points on the right side are classified into the feature point set on the right side of the roadside; Step S5: Iterative fitting to remove data noise points, and use limit filtering and Kalman filtering to smooth the fitting of the roadside feature points; Step S6: roadside status monitoring, smooth switching strategy between roadside and no roadside status; By analyzing the detection results of multiple consecutive frames, the state switching process is identified and smoothly processed.

2. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: The step S2 divides the ground point cloud and the non-ground point cloud by plane fitting in sections, including the following steps: Step S21, dividing the original point cloud P into a plurality of continuous segments along the x-axis; Step S22, applying the RANSAC algorithm to each segment of the point cloud to perform plane fitting and identify points that conform to the ground plane model; Step S23: according to the fitting result, the point cloud is divided into ground point cloud and non-ground point cloud.

3. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: The height difference feature extraction in step S3 is to capture the significant changes of the curb in the vertical dimension by using the height difference feature, by calculating the difference between the maximum and minimum values ​​of the adjacent point cloud areas.

4. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: The smoothness feature evaluation in step S3 is to introduce the smoothness feature to evaluate the smoothness of the local area of ​​the point cloud. By analyzing the area around the candidate point, the smoothness feature distinguishes the curb feature point from the ground or non-curb objects.

5. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: The horizontal distance feature determination in step S3 is based on the height difference and smoothness analysis, and uses the horizontal distance feature to measure the interval between two adjacent points on the same horizontal plane. By setting a threshold, the key points that are closely connected in the horizontal direction and constitute the curb contour are identified.

6. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: In step S4, a peak-finding algorithm is used to determine the boundary of the road segmentation by identifying the local maximum value in the point cloud data, traversing each point in the data set and comparing its value with that of its adjacent points; if the value of a point is greater than the values ​​of all its adjacent points, the point is regarded as a peak, i.e., a boundary point of the road segmentation.

7. The method for extracting roadside features and switching states based on laser radar point cloud data according to claim 1, characterized in that: The step S5 comprises the following steps: Step S51, using iterative fitting technology to remove data noise, selecting the straight line equation as the fitting model, using the gradient descent algorithm to perform preliminary fitting and estimate parameters; through multiple iterations, fitting the feature points based on the current parameters and calculating the error, continuously optimizing the algorithm to minimize the error, until the preset number of iterations is reached, and obtaining preliminary fitting parameters; Calculate the accuracy of fitting error assessment, and use statistical filtering methods to remove false detection points or outliers that exceed the set threshold; refit the remaining feature points to obtain more accurate and stable fitting results; Step S52, applying a limiting filtering strategy to eliminate interference detection; by continuously storing the latest boundary equation slope and calculating its average value and variance, the current linear equation is considered valid only when both meet the preset threshold conditions; Step S53: Kalman filtering technology is used to perform smooth optimization on the boundary equation.

8. A computer system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the curb feature extraction and state switching method described in any one of claims 1-7.

Images