CN-121982517-A - Road alignment safety evaluation method based on three-dimensional point cloud space modeling

CN121982517ACN 121982517 ACN121982517 ACN 121982517ACN-121982517-A

Abstract

The invention relates to a road line shape safety evaluation method based on three-dimensional point cloud space modeling, which aims at the defects of poor adaptability of a road point cloud segmentation algorithm, strong dependence on road marking due to flat line shape fitting and the like in a complex scene, and utilizes the acquired high-density point cloud data characteristic to provide a road key information collaborative extraction method fused with multi-scale characteristics, wherein firstly, the influence of environmental noise on the road characteristics is reduced through normalized gradient filtering and statistical outlier filtering algorithm; and then extracting road edge lines to obtain road center line data, and generating a safety evaluation result list based on road plane, longitudinal section and cross section line parameters of the road center line. The research result provides technical support for application scenes such as road maintenance detection, automatic driving high-precision map construction and the like.

Inventors

WANG XIAOFEI
CAO YUANSHENG
YANG HUI
LIN YONGJIE

Assignees

华南理工大学

Dates

Publication Date: 20260505
Application Date: 20251216

Claims (10)

1. A road line shape safety evaluation method based on three-dimensional point cloud space modeling is characterized by comprising the following steps, S1, acquiring original three-dimensional point cloud data of a target road; S2, constructing an octree space index structure for the original three-dimensional point cloud data, and sequentially performing gradient filtering and statistical outlier removal processing to obtain de-noised and thinned point cloud data; S3, extracting a road surface point cloud data set by adopting an improved region growing algorithm based on the denoised and thinned point cloud data, wherein the improved region growing algorithm starts a growing process by taking an interactively specified road starting and ending point, a curve center and a shielding region edge as seed points, and carries out growing judgment by adopting a composite growing criterion function comprising plane consistency errors, elevation deviations, intensity differences and dynamic space inhibition items so as to improve the segmentation precision of the road surface point cloud data set; S4, projecting the road point cloud data set to a two-dimensional plane, extracting road boundary points by adopting ALPHA SHAPE algorithm, calculating geometric midlines of left and right boundary lines after B spline curve smoothing is carried out on the road boundary points, and mapping the geometric midlines back to a three-dimensional space to obtain a three-dimensional road centerline; S5, calculating plane line shape parameters comprising curvature radius through cubic spline interpolation based on a three-dimensional road center line, calculating longitudinal section line shape parameters comprising longitudinal slope gradient and vertical curve radius through resampling and derivative analysis, and calculating cross section line shape parameters comprising lane width and transverse slope parameters through cross section sampling and RANSAC straight line fitting in a direction perpendicular to a tangent line of the center line; and S6, automatically comparing the obtained plane linear parameters, the longitudinal section linear parameters and the cross section linear parameters with safety thresholds respectively, and generating a safety evaluation result list containing risk levels, hidden danger types and optimization suggestions based on single-parameter compliance judgment and multiparameter coupling effect evaluation.
2. The method of claim 1, wherein the raw three-dimensional point cloud data in S1 comprises three-dimensional coordinates, intensity, echo times, and color information.
3. The method according to claim 1, wherein S2 comprises in particular: S21, constructing an octree data structure based on original three-dimensional point cloud data, recursively dividing a three-dimensional space into a plurality of subspaces, and realizing quick neighborhood search of the point cloud data; S22, searching neighborhood points of each point based on subspace division results of the octree spatial index structure, calculating elevation gradient, carrying out gradient filtering by adopting a normalized elevation gradient algorithm, setting a normalized elevation gradient threshold value range to be 0-0.15, and filtering out vertical structure object points; S23, rapidly searching K neighbors based on the octree spatial index structure, executing a statistical outlier removal algorithm on the point cloud data after gradient filtering, calculating the mean value and standard deviation of the K neighbors distance of each point, and setting a global threshold value tau=mu global + n⋅σ global , wherein mu global and sigma global are the mean value and standard deviation of the neighborhood distances of all points respectively, n takes a value of 1.0-3.0, and the outlier with the distance mean value larger than tau is removed.
4. The method of claim 1, wherein the composite growth criterion function in S3 is defined as Wherein: Is a weight coefficient, e plane is a plane consistency error term, the terrain continuity is guaranteed by combining a normal vector included angle constraint theta <10 degrees and OcTree radius search r=0.2m, d elevation is an elevation difference term, a threshold h=0.15m is set to filter abrupt change characteristics such as a curb, I intensity is an intensity difference term, a threshold I=0.8Imax is set, and phi (p) is a dynamic inhibition term.
5. The method of claim 1, wherein the interactive seed point optimization strategy in S3 is to assign a plurality of seed points to initialize corresponding growing areas in a road starting end point, a curve curvature center area and a tree shielding hole edge area in a visual interface, assign a unique area identifier to each seed point and record a growth threshold parameter of each seed point, and realize multi-area automatic fusion by judging the elevation continuity of boundary points of adjacent areas when the growing sub-areas of different seed points are expanded to the adjacent boundaries.
6. The method according to claim 1, wherein in S4, a ALPHA SHAPE algorithm is adopted to extract road boundary points from the projected road surface point cloud data set with a rolling circle radius α=2.0-3.0 meters, a cubic B-spline basis function is applied to the road boundary points to perform smooth fitting to eliminate jagged boundaries, the smoothed boundary lines are divided into left and right side boundaries according to the road traveling direction, and a geometric center line is obtained by calculating an arithmetic average of corresponding point coordinates of the left and right side boundary lines at the same parameter position.
7. The method of claim 1, wherein the calculation of the planar linear parameters in S5 comprises converting discrete centerline points to continuous parameters t using cumulative chord length parameterization, constructing cubic spline functions x (t) and y (t), and calculating curvature by analytical derivative Wherein the first derivative in the formula And second derivative Obtained by analytical differentiation of spline function, the corresponding curvature radius is R (t) is And setting a curvature threshold value to automatically identify the straight line segment and the curve segment and outputting the radius and the length of each segment.
8. The method of claim 1, wherein the calculating of the longitudinal section linear parameter in S5 specifically comprises the steps of carrying out high Cheng Chong sampling along the center line interval of the three-dimensional road, smoothing the slope sequence of adjacent points by adopting a Hampel filter, obtaining a first derivative of the smoothed slope curve, classifying the road sections with the absolute values of the first derivatives continuously approaching zero as straight line slope sections and outputting slope values, classifying the continuous road sections with the first derivatives being obviously nonzero as vertical curve sections and carrying out secondary parabolic fitting by adopting a least square method, and calculating the radius and the length of the vertical curve.
9. The method of claim 1, wherein the calculating of the cross-section linear parameters in S5 specifically comprises the steps of calculating tangential directions of resampled center line points, taking vertical directions of the tangential directions as cross-section directions, extracting two-side road point clouds along the cross-section directions, fitting straight lines to transverse positions of the point clouds and elevation data by adopting a weighted RANSAC algorithm after the point clouds are divided into boxes, wherein the slope of the straight lines is a transverse slope value, and calculating lane width and lateral width by the distance from the center line points to two side boundaries.
10. The method of claim 1, wherein the single parameter-based compliance determination in S6 includes comparing the planar line shape parameter, the vertical line shape parameter, and the cross-sectional line shape parameter to a road design specification database, one by one, with the parameter value being lower than a specification limit, with the parameter value being marked as high risk, the parameter value being between the limit and a general value, and the parameter value being marked as low risk, the parameter value being higher than the general value; The multi-parameter coupling effect evaluation is carried out based on a preset rule base, corresponding circle curve radius and longitudinal slope threshold values are set according to different design speeds, a sharp-curve and steep-slope high-risk combined road section is judged, when the top of a vertical curve is connected with a small-radius curve, the combination is judged to be a poor-vision high-risk combination, a visual map file is output through a safety evaluation result list, the high-risk road section is marked red, the middle risk is marked yellow, the low risk is marked green, and an evaluation report comprising risk positions, types, grades and improvement measures is synchronously generated.

Description

Road alignment safety evaluation method based on three-dimensional point cloud space modeling Technical Field The invention relates to the technical field of traffic road engineering, in particular to a road alignment safety evaluation method based on three-dimensional point cloud space modeling. Background With the increasing perfection of traffic infrastructure networks, the industry center of gravity is gradually changing from large-scale construction to the refined operation, maintenance and management of full life cycle, and under the background, the high-efficiency and accurate evaluation and maintenance of the safety condition of the existing road become urgent demands of traffic management departments. However, the root of the road safety condition is often deeply planted in the design parameters of the internal geometric shape, namely the plane, the vertical section and the cross section, which directly determine the stability, the comfort and the safety of the driving. The road alignment safety evaluation method in the prior art has the following technical problems: The limitations of the existing road geometric data deletion and the traditional measurement means are that for a large number of roads operated for many years, particularly for low-grade roads built at early stage or mountainous roads with complex topography, the current situations of complete drawing deletion and incomplete archives exist, so that objective data base for accurately and safely evaluating the road geometric conditions is lacking. Meanwhile, the traditional road geometric information acquisition means has obvious bottlenecks that manual investigation and conventional instrument measurement are low in efficiency and high in operation risk, discrete point data are acquired, linear continuity is difficult to comprehensively reflect, efficiency, precision, safety and cost are also difficult to consider, and the requirements of large-scale and normalized road safety screening cannot be met. The adaptability of the road surface point cloud segmentation in the complex scene is insufficient, the road environment presents terrain, environment and structural complexity, the terrain dimension relates to a return curve, a high slope and the like, the environment dimension comprises high-density vegetation shielding and dynamic vehicle pedestrian interference, and the structural dimension has unstructured characteristics such as marking wear, road surface damage and the like. The existing segmentation method based on single elevation threshold or deep learning has low segmentation accuracy and integrity under the pseudo-ground point interference of tree branch and leaf multi-mode distribution, roof squares and the like, and severely restricts the follow-up parameter extraction precision. Disclosure of Invention Aiming at the problems in the prior art, the invention aims to provide the road line shape safety evaluation method based on three-dimensional point cloud space modeling, which can improve the road data acquisition efficiency and safety, effectively treat vegetation shielding and dynamic target interference problems, clear hidden danger positions and optimization suggestions and improve the scientificity and the accuracy of maintenance decisions. In order to achieve the above purpose, the invention adopts the following technical scheme: A road line shape safety evaluation method based on three-dimensional point cloud space modeling, which comprises the following steps, S1, acquiring original three-dimensional point cloud data of a target road; S2, constructing an octree space index structure for the original three-dimensional point cloud data, and sequentially performing gradient filtering and statistical outlier removal processing to obtain de-noised and thinned point cloud data; S3, extracting a road surface point cloud data set by adopting an improved region growing algorithm based on the denoised and thinned point cloud data, wherein the improved region growing algorithm starts a growing process by taking an interactively specified road starting and ending point, a curve center and a shielding region edge as seed points, and carries out growing judgment by adopting a composite growing criterion function comprising plane consistency errors, elevation deviations, intensity differences and dynamic space inhibition items so as to improve the segmentation precision of the road surface point cloud data set; S4, projecting the road point cloud data set to a two-dimensional plane, extracting road boundary points by adopting ALPHA SHAPE algorithm, calculating geometric midlines of left and right boundary lines after B spline curve smoothing is carried out on the road boundary points, and mapping the geometric midlines back to a three-dimensional space to obtain a three-dimensional road centerline; S5, calculating plane line shape parameters comprising curvature radius through cubic spline interpolation based on a three-dimensional