CN111222418B - Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line - Google Patents
Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line Download PDFInfo
- Publication number
- CN111222418B CN111222418B CN201911350314.3A CN201911350314A CN111222418B CN 111222418 B CN111222418 B CN 111222418B CN 201911350314 A CN201911350314 A CN 201911350314A CN 111222418 B CN111222418 B CN 111222418B
- Authority
- CN
- China
- Prior art keywords
- point
- lane line
- points
- shape
- axis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a method for rapidly fusing and optimizing crowdsourcing data of a plurality of road segments of a lane line, which comprises the following steps: establishing a lane line local piecewise function model, and performing local fitting optimization on each shape point to form a section of lane line; establishing a global lane line model according to the local piecewise function model to generate a plurality of sections of the whole lane line; and carrying out error fracture and error connection processing on the lane lines to generate a lane line set. The lane linear points with high precision, integrity and continuity can be obtained from the disordered original input data points, and the linear equation corresponding to the lane line can be obtained, so that vectorized lane line data is obtained; the method is simple, convenient and quick to execute, has low requirements on input data and high robustness, can adapt to 3D lane line point data and complex urban road data, and has obvious advantages compared with the conventional lane line clustering algorithm.
Description
Technical Field
The invention relates to the field of high-precision maps, in particular to a method for quickly fusing and optimizing crowdsourcing data of multiple road segments of a lane line.
Background
In the field of automatic driving, in order to accurately control the driving of a vehicle, high-precision map drawing is often involved, and lane line-shaped point data of a road surface is needed in the process of high-precision map drawing, so that lane-level driving guidance is provided for the automatic driving vehicle. The high-precision map can be drawn by using a surveying and mapping vehicle with high price through long-time data acquisition, but the high freshness requirement of the high-precision map is difficult to meet due to high cost, long acquisition period and slow updating.
Compared with a high-precision mapping vehicle, the crowdsourcing collecting vehicle is low in cost and suitable for being widely arranged to collect high-freshness data and improve the updating frequency of a high-precision map, the crowdsourcing collecting vehicle is low in precision, errors of collected data points are large and often wrong data points exist, and therefore high-precision lane line data are expected to be obtained through large data volume fusion optimization of multiple frequent collection, but the large data volume means the increase of calculation complexity and calculation time. Therefore, under the condition of large data volume, how to perform fusion optimization of crowdsourcing data on 2D or 3D data of the lane lines acquired by crowdsourcing acquisition vehicles becomes a difficult point.
Disclosure of Invention
The invention provides a rapid crowd-sourced data fusion optimization method for multiple road segments of a lane line, aiming at the technical problems in the prior art, and solves the problem that 3D data of the lane line collected by a crowd-sourced collection vehicle cannot be rapidly and effectively fused in the prior art.
The technical scheme for solving the technical problems is as follows: a method for fast fusion and optimization of crowdsourcing data of multiple lane segments of a lane line, the method comprising:
step 1, establishing a lane line local piecewise function model, and performing local fitting optimization on each shape point to form a section of lane line;
step 2, establishing a global lane line model according to the local piecewise function model to generate a plurality of sections of the whole lane line;
and 3, carrying out error fracture and error connection processing on the lane line to generate a lane line set.
The invention has the beneficial effects that: the lane linear points with high precision, integrity and continuity can be obtained from the disordered original input data points, and the linear equation corresponding to the lane line can be obtained, so that vectorized lane line data is obtained; the method is simple, convenient and quick to execute, has low requirements on input data and high robustness, can adapt to 2D lane line point data and complex urban road data, and has obvious advantages compared with the conventional lane line clustering algorithm.
On the basis of the technical scheme, the invention can be further improved as follows.
Further, the step 1 comprises:
step 101, establishing a set omega of shape points including crowd-sourced data of all to-be-processed lane road segments0;
Step 102, constructing a cuboid retrieval frame<l,w,h>Selecting a point of the shape p0(x0,y0,z0) Is the starting point of the search frame;
step 103, will form point p with0Does not exceed L in absolute value, and is in contact with the point p0Does not exceed W in absolute value, and is in contact with the point p0All shape point generation sets omega with Z-axis coordinate values having absolute values of differences not exceeding H1;
Step 104, from the set Ω1Selecting the shape points in the retrieval frame to generate a shape point set omega 2;
step 105, for the set Ω2Performing linear regression on the figure points to generate a section of lane line.
In the step 104, the set Ω is determined1Any point of (1) pk(xk,yk,zk) The method for judging whether the search frame is in the search frame is as follows: determining eight corner points of the retrieval frame according to the size and the direction of the retrieval frame and the coordinates and the directions of the shape points, and determining eight corner points of the retrieval frame according to the corner points and the shape points pkDetermining the shape point pkWhether it is within the search box;
the process of determining the corner point comprises:
the initial coordinates of the eight corner points are:
and the coordinates of each angular point are multiplied by the rotation angular points of the rotation matrix to obtain:
after translational rotation, the angular points are obtained:
wherein the rotation matrixAngle of rotationI is an identity matrix and is a matrix of the identity,representing a vectorAndcross product of (1), unit rotation vectorIs a vectorThe corresponding antisymmetric matrix:
further, the step 104 includes:
step 10401, taking the corner point as c31,c32,c33,c34,c35,c36,c37,c38Four vertices of three edges of a given rectangular parallelepiped, which are perpendicular to each other and intersect at a point, are denoted asWherein the content of the first and second substances,three mutually perpendicular edgesThe intersection point of (a);
step 10402, according to the intersection point and the shape point p selected in the step 10401kCoordinates of (2), calculating vectors
step 10404, when the inner product results of the six vector inner products are not less than 0, determining that the shape point p iskWithin or on the boundary of the search box; when at least one of the inner product results of the six vector inner products is less than 0, the shape point p is judgedkOutside the search box.
Further, the step 105 comprises:
step 10501, set an angle difference threshold and a distance difference threshold from the set Ω2Is selected from the said shape point p0The angle difference of is not more than the shape point generation set omega of the angle difference threshold value3From said set Ω2Is selected from the said shape point p0The set omega of shape point generation whose distance difference is not greater than the distance difference threshold4Generating said set Ω3And said set omega4Of (a) intersection omega5;
10502, according to said set Ω2The set omega3The set omega4And the intersection Ω5The number of interior shape points determines whether to perform a linear regression, yes, go to step 10503, no, go to step 10504;
step 10503, aligning the intersection Ω5The shape points in (1) are subjected to linear regression.
Further, in the step 10502, the set Ω2When the number of the shape points is less than a set threshold value, or the intersection omega5Is less than 2, or the set omega3Or the set Ω4The number of the shaped points of (C) accounts for the set omega2When the ratio of the number of the dots is smaller than a set threshold value, it is determined that linear regression cannot be performed.
Further, the step 10503 comprises:
step 1050301, selecting the set Ω5The coordinate value of X axis and the coordinate value of Z axis are used as characteristic independent variables, and the coordinate value of Y axis is subjected to linear regression to obtain a regression coefficient coef1=(cf3,cf4)TAnd intercept ic3;
Step 1050302, selecting the set Ω5The coordinate value of the Y axis and the coordinate value of the Z axis are used as characteristic independent variables, and the coordinate value of the X axis is subjected to linear regression to obtain a regression coefficient coef2=(cf5,cf6)TAnd intercept ic4;
Step 1050303, the equation for establishing the shape point to perform linear regression is:
parameter A1,B1,C1,D1,A2,B2,C2,D2Are respectively cf3,-1,cf4,ic3,-1,cf5,cf6,ic4。
Further, the method for establishing the global lane line model according to the local piecewise function model in the step 2 comprises the following steps:
after the step 102 and 105 generate a segment of lane line, the starting point of the search frame and the set Ω are determined2Form point in (b) in the set omega0Marked as a processed shape point; determining the starting point and direction of the next search box, and performing the step 102-105 on the set omega0The unprocessed shape points are processed until the set omega is obtained0Marking the points in the lane line as processed points, and forming a plurality of sections of the whole lane line after each section of the lane line;
the process for determining the starting point and the direction of the next retrieval frame comprises the following steps:
and selecting a point which is farthest away from the starting point of the search frame on the independent variable as a tail point of the search frame, wherein the tail point of the search frame is used as the starting point of the next search frame, and the direction vector of a straight line obtained by linear regression is used as the direction vector of the starting point of the next search frame.
Further, the process of performing error connection processing on the lane line in step 3 includes: performing linear regression on the shape points to generate each section of lane line, and comparing the variation ranges of the X axis, the Y axis and the Z axis of the coordinates of all the shape points in a single lane line;
when the variation range of the X axis is larger than the variation ranges of the Y axis and the Z axis, the shape points are sorted from small to large according to the X axis coordinate;
when the variation range of the Y axis is larger than the variation ranges of the X axis and the Z axis, the shape points are sorted from small to large according to the Y axis coordinate;
when the variation range of the Z axis is larger than the variation ranges of the X axis and the Y axis, the shape points are sorted from small to large according to the Z axis coordinate;
the process of performing the error fracture treatment on the lane line comprises the following steps:
for any two lane lines L0And L1If said lane line L is0And the tail point of the road line L1The distance of the first point is less than the set Euclidean distance, and the midpoint PLTo the lane line L0And the lane line L1The driving direction vertical distance of the head point of the vehicle is smaller than a set vertical distance threshold value;
merge the lane lines L0And L1Is a lane line L0Deleting the lane line L from the lane line set1;
Wherein the midpoint PLIs the lane line L0And the tail point of the road line L1Is the midpoint of the line connecting the first points of (a).
Further, the step 3 is followed by:
step 4, storing the lane lines in a graph structure by taking the figure points or straight lines as vertexes;
when the lane lines are stored in the graph structure in the form of shape points, the shape points of the lane lines searched in sequence are used as vertexes, all shape points in the one-way retrieval process are added into the graph in sequence, and edges are used for connecting two vertexes in an adjacent sequence; each vertex comprises an ID, a coordinate position, a parameter of a linear equation obtained by the belonged linear regression and an ID of the belonged straight line;
when the lane lines are stored in the graph structure in a straight line mode, the coordinates of the head point, the coordinates of the tail point and the straight line parameters of each straight line are stored to form a graph with straight line data as the top point by taking the straight line obtained by linear fitting optimization as a unit.
The beneficial effect of adopting the further scheme is that: firstly, judging similar lane linear points based on the lane linear point coordinates and direction angles or direction vectors thereof, and forming a plurality of sections of whole lane lines through fitting optimization of a plurality of sections of continuous similar points; carrying out processing optimization on the error fracture and the error connection of the multiple sections of lane lines after fitting optimization; the lane linear points with high precision, integrity and continuity can be obtained from the disordered original input data points, and the linear equation corresponding to the lane line can be obtained, so that vectorized lane line data is obtained; the method is simple, convenient and quick to execute, has low requirements on input data and high robustness, can adapt to 3D lane line point data and complex urban road data, and has obvious advantages compared with the conventional lane line clustering algorithm.
Drawings
Fig. 1 is a flowchart of a method for fast fusing and optimizing crowdsourcing data of multiple road segments of a lane line according to the present invention;
fig. 2 is a flowchart of an embodiment of a method for fast fusing and optimizing crowdsourcing data of multiple road segments of a lane line according to the present invention;
fig. 3 is a flowchart of an embodiment of local fitting optimization for shape points according to the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a method for fast fusing and optimizing crowdsourcing data of multiple road segments of a lane line, which is provided by the present invention, and as shown in fig. 1, the method includes:
step 1, establishing a lane line local piecewise function model, and performing local fitting optimization on each shape point to form a section of lane line.
And 2, establishing a global lane line model according to the local piecewise function model to generate a plurality of sections of the whole lane line.
And 3, carrying out error fracture and error connection processing on the lane lines to generate a lane line set. The lane linear points with high precision, integrity and continuity can be obtained from the disordered original input data points, and the linear equation corresponding to the lane line can be obtained, so that vectorized lane line data is obtained; the method is simple, convenient and quick to execute, has low requirements on input data and high robustness, can adapt to 3D lane line point data and complex urban road data, and has obvious advantages compared with the conventional lane line clustering algorithm.
Example 1
Embodiment 1 provided by the present invention is a preferred embodiment of a method for fast fusing and optimizing crowdsourcing data of a multi-channel segment of a lane line provided by the present invention, and as shown in fig. 2, is a flowchart of an embodiment of a method for fast fusing and optimizing crowdsourcing data of a multi-channel segment of a lane line provided by the present invention, and as shown in fig. 3, is a flowchart of an embodiment of local fitting and optimizing a shape point provided by the present invention, as can be seen from fig. 2 and 3, the embodiment includes:
and converting the coordinates of the multi-channel segment data into plane coordinates by adopting a Gaussian-Kruger projection method based on longitude and latitude, altitude height coordinates and a projection zone range given by the segment data.
Step 1, establishing a lane line local piecewise function model, and performing local fitting optimization on each shape point to form a section of lane line.
Preferably, step 1 comprises:
step 101, establishing a set omega of shape points including crowd-sourced data of all to-be-processed lane road segments0。
Step 102, constructing a cuboid retrieval frame<l,w,h>Selecting a point of the shape p0(x0,y0,z0) Is the starting point of the search box.
Step 103, mixing the shape point p0Does not exceed L in absolute value of the difference between the X-axis coordinate values and the shape point p0The absolute value of the difference between the Y-axis coordinate values of (a) and (b) does not exceed W and is equal to the shape point p0All shape point generation sets omega with Z-axis coordinate values having absolute values of differences not exceeding H1。
Given a range of regions<L,W,H>Generating a set omega1Set omega1The shape point of the inner is the shape point p0Candidate points located on the same lane line.
d is a parameter of the direction of the search frame, sgn (d) represents the symbol of d, when d>When d is less than or equal to 0, sgn (d) is1,the included angle formed by the positive direction of the three axes of X, Y and Z is alpha, beta and gamma.
Step 104, from the set Ω1And selecting the shape points in the retrieval frame to generate a shape point set omega 2.
In step 104, the set Ω is judged1Any point of (1) pk(xk,yk,zk) The method for judging whether the search frame is in the search frame is as follows: determining eight corner points of the retrieval frame according to the size and direction of the retrieval frame and the coordinates and direction of the shape points, and determining eight corner points of the retrieval frame according to the corner points and the shape points pkCoordinate judging point p ofkWhether it is in the search box;
specifically, the process of determining the corner point includes:
the initial coordinates of the eight corner points are:
and (3) multiplying the coordinates of each angular point by the rotation angular point of the rotation matrix to obtain:
obtaining the angular point after translation and rotation:
wherein the rotation matrixAngle of rotationI is an identity matrix and is a matrix of the identity,representing a vectorAndcross product of (1), unit rotation vectorIs a vectorThe corresponding antisymmetric matrix:
preferably, step 104 includes:
step 10401, taking the corner point as c31,c32,c33,c34,c35,c36,c37,c38Four vertices of three edges of a given rectangular parallelepiped, which are perpendicular to each other and intersect at a point, are denoted asWherein the content of the first and second substances,three mutually perpendicular edgesOf (2)And (4) point.
Step 10402, according to the intersection point and the shape point p selected in step 10401kCoordinates of (2), calculating vectors
wherein the content of the first and second substances,representing a calculated vectorAndthe vector inner product of (a).
Step 10404, when the inner product results of the six vector inner products are not less than 0, determining the shape point pkInside or on the boundary of the search box; when at least one of the inner product results of the six vector inner products is less than 0, the shape point p is judgedkOutside the search box.
Step 105, for the set omega2The figure points in the road are subjected to linear regression to generate a section of lane line.
Preferably, step 105 comprises:
step 10501, set an angle difference threshold and a distance difference threshold from the set Ω2In selecting and forming point p0The shape point generating set omega with the angle difference not larger than the angle difference threshold value3From the set Ω2In selecting and forming point p0The set omega of shape point generation for which the distance difference is not greater than the distance difference threshold4Generating a set omega3And set omega4Of (a) intersection omega5。
In particular, the parameter Δ d is given1,Δd2,Δd1Represents a point p0And set omega2Any point p inkA threshold value of an angle or an angle difference between direction vectors; Δ d2Represents the set omega2Any point p inkAnd a passing point p0And the direction vector is a point p0A threshold value for the distance between the straight lines of the direction vector. Condition p0And pkIs not more than delta d1Or the difference between the two angles is not more than deltad1Defined as the "angular condition". Condition pkTo a point of passing p0And the direction vector is a point p0The distance of the straight line of the direction vector is not more than delta d2Defined as the "distance condition". Set omega2The set of points satisfying the angle condition is omega3Set omega2The set of points satisfying the distance condition is omega4The set of points satisfying both the angle condition and the distance condition is Ω5。Ω5Is determined as the point p0Points on the same lane line.
10502, according to the set Ω2Set omega3Set omega4And the intersection Ω5The number of shape points in determines whether to perform a linear regression, yes, step 10503 is performed, no, step 10504 is performed.
In step 10502, set Ω2When the number of the shape points is less than the set threshold value, or the intersection omega5Is less than 2, or set omega3Or set omega4The number of the figure points of (1) accounts for the set omega2When the ratio of the number of the dots is smaller than a set threshold value, it is determined that linear regression cannot be performed.
Specifically, for the point set Ω satisfying the angle condition3,Ω3In the number of elements ofPoint set omega satisfying distance condition4,Ω4In the number of elements ofPoint set omega satisfying retrieval condition2,Ω2In the number of elements ofTo obtain the ratioGiven a parameter n1Representing the minimum number of points, parameter ratio, that satisfy the search criteria1Represents rt1Minimum, parameter ratio2Represents rt2A minimum value. When in useWhen this is the case, linear regression is not possible. When rt is1<ratio1Or rt2<ratio2When this is the case, linear regression is not possible. Select set omega3And set omega4Of (a) intersection omega5If Ω is5If the number of the medium elements is less than 2, linear regression cannot be performed.
Step 10503, for the intersection Ω5The shape points in (1) are subjected to linear regression.
The linear Regression problem may use a general linear least square method, or a random consensus sampling (RANSAC) algorithm, or a teerson Regression (Theil-Sen Regression) algorithm, or a robust Regression algorithm such as a Huber Regression (Huber Regression).
Preferably, step 10503 comprises:
step 1050301, select set Ω5The coordinate value of X axis and the coordinate value of Z axis are used as characteristic independent variables, and the coordinate value of Y axis is subjected to linear regression to obtain a regression coefficient coef1=(cf3,cf4)TAnd intercept ic3。
Step 1050302, select set Ω5The coordinate value of the Y axis and the coordinate value of the Z axis are used as characteristic independent variables, and the coordinate value of the X axis is subjected to linear regression to obtain a regression coefficient coef2=(cf5,cf6)TAnd intercept ic4。
Step 1050303, the equation for linear regression of the shape points is established as:
parameter A1,B1,C1,D1,A2,B2,C2,D2Are respectively cf3,-1,cf4,ic3,-1,cf5,cf6,ic4。
The model of the lane line in the retrieval frame is a straight line formed by the intersection of two planes, so that the equation for performing linear regression is an equation of two straight lines, and the linear regression can be used for solving twice.
Set of points omega5After the points in the step (3) are subjected to linear regression according to the method from the step 1050301 to the step 1050303, a regression equation of the determined parameters is obtained, the parameter pnum is given, and the point set omega is subjected to linear regression6And comparing the difference between the maximum value and the minimum value of the X coordinate and the difference between the maximum value and the minimum value of the Y coordinate in the point set, selecting the coordinate value of the dimension with a larger difference value as an independent variable, taking points at intervals between the maximum value and the minimum value of the dimension according to the parameter pnum, arranging the independent variables from large to small or from small to large according to the given search frame direction, substituting the independent variable values into a regression equation, obtaining the coordinate value of the dependent variable Y, Z or X, Z, and obtaining the shape point coordinate output by the fused lane line.
And 2, establishing a global lane line model according to the local piecewise function model to generate a plurality of sections of the whole lane line.
Preferably, the method for establishing the global lane line model according to the local piecewise function model in the step 2 comprises:
after generating a segment of lane line in step 102-105, the starting point of the search frame and the set omega are set2Form point in (1) is in the set omega0Marked as a processed shape point; the loop execution step 102 and 105 for the set omega for determining the starting point and direction of the next search box0Treating the untreated shape pointsUp to the set omega0The points in the middle are marked as processed, and all sections of lane lines are combined into a whole lane line.
Once the set Ω is obtained5And if the linear regression is successfully carried out, the set omega meeting the retrieval condition at the starting point of the retrieval frame and in the direction of the retrieval frame is used2Marked as processed, the remaining points are unprocessed points. If a point in the unprocessed source data set is selected, the corresponding set Ω can be found5And linear regression is successfully carried out, the point is considered to be a feasible starting point of local optimization.
If the starting point of the search box can not find the point set omega capable of linear regression5The processing strategy of (1); when the point set omega capable of linear regression can not be found from the starting point of the search box5In this case, the parameter L may be optionally increased as appropriate to increase the length of the search box in the X direction until the point set Ω that can be linearly regressed is found5Or the stopping strategy of the lane line search given in the third part is satisfied, or alternatively, an unprocessed point set is selected, is closest to the starting point of the current search frame, and is searched again with another point in the search direction selected at the starting point of the current search frame as the starting point of the search frame until a point set omega capable of linear regression is found5Or a stopping strategy satisfying the lane line finding given in the third section.
The process of determining the starting point and the direction of the next retrieval frame comprises the following steps:
and selecting a point which is farthest away from the starting point of the search frame on the independent variable as a search frame tail point, taking the search frame tail point as the starting point of the next search frame, and taking the direction vector of a straight line obtained by linear regression as the direction vector of the starting point of the next search frame.
Specifically, in the process of selecting the starting point and the direction vector of the next search frame, after selecting the tail point of the search frame, judging whether the direction needs to be changed, if the direction vector of the straight line advancing is different from the direction guided by the direction parameter of the search frame, changing the direction of the search frame to enable the direction to be changed. This process continues until three search stop conditions are met. First search stop condition, according to claimIn the first part of 5, given the parameter scnt, if the appropriate search box still cannot be found at consecutive scnt points, it is considered that the lane line search in the direction should be stopped at this time. The second search stop condition, the selection parameter, cutdis1, represents the maximum value of the distance between the start point of the current search box and the end point of the previous search box, and if the distance is greater than cutdis1, the search is considered to be stopped. The third search stop condition, selection parameter cutdis2, represents the starting point of the current search box and the previous successful set of points Ω6The maximum value of the distance between the straight lines of the above-performed linear regression result, if the distance is greater than cutdis2, it is considered that the search should be stopped. After the continuous search in one direction meets any search stop condition, if the search process is only to the direction d of the search frame>0 or d<And 0, if the search is carried out in one direction, changing the direction of the search frame, searching from the other direction, and repeating the search process in the searched direction. After the retrieval in the two directions is finished, the lane line is considered to be searched once, and a complete lane line is found. And selecting a starting point for searching the lane line from the rest unprocessed point sets, and repeating the process of searching the lane line. And until the number of the remaining unprocessed points is less than 2, all data is considered to be processed, and all lane lines are found.
And 3, carrying out error fracture and error connection processing on the lane lines to generate a lane line set.
The process of performing error connection processing on the lane line in the step 3 comprises the following steps: performing linear regression on the shape points to generate each section of lane line, and comparing the variation ranges of the X axis, the Y axis and the Z axis of the coordinates of all the shape points in a single lane line:
when the variation range of the X axis is larger than that of the Y axis and the Z axis, the shape points are sorted from small to large according to the X axis coordinate; when the variation range of the Y axis is larger than that of the X axis and the Z axis, the shape points are sorted from small to large according to the Y axis coordinate; and when the variation range of the Z axis is larger than that of the X axis and the Y axis, sorting the shape points from small to large according to the Z axis coordinate.
In X-axis coordinates of all shape points in a single lane line, max (X), min (X) respectively represent the maximum value and the minimum value in the X-axis coordinates. Similarly, max (Y), min (Y) respectively represent the maximum value and the minimum value in the Y-axis coordinate, and similarly max (z), min (z) respectively represent the maximum value and the minimum value in the Y-axis coordinate. If max (X) -min (X) > max (Y) -min (Y) and max (X) -min (X) > max (Z) -min (Z), all the shape points in the lane line are connected into a line according to the descending order of the X value; if max (Y) -min (Y) > max (X) -min (X) and max (Y) -min (Y) > max (Z) -min (Z), all the shape points in the lane line are connected into a line according to the descending order of the Y value; if max (Z) -min (Z) > max (X) -min (X) and max (Z) -min (Z) > max (Y) -min (Y), all the shape points in the lane line are connected into a line according to the Z value from small to large. All lane lines are reordered and continuous according to the rule, namely the problem of wrong connection of the lane lines caused by wrong shape point sequencing is solved.
The process of performing the error fracture treatment on the lane line comprises the following steps:
for any two lane lines L0And L1If the lane line L is0Tail point of and lane line L1The distance of the first point is less than the set Euclidean distance, and the midpoint PLLane line L0Tail point and lane line L of1The driving direction vertical distance of the head point of (2) is smaller than the set vertical distance threshold value.
Merge lane line L0And L1Is a lane line L0Deleting the lane line L from the lane line set1。
Wherein, the middle point PLIs a lane line L0Tail point of and lane line L1Is the midpoint of the line connecting the first points of (a).
The Euclidean distance may be set to 10 meters, and the sag threshold may be set to 0.8 meters. And (4) performing the operation on all the lane lines, wherein the finally remained lane lines are the lane line set after the completion of the fault fracture.
Step 3 is followed by: step 4, storing the lane lines in a graph structure by taking the figure points or straight lines as vertexes;
when the lane lines are stored in the graph structure in the form of shape points, the shape points of the lane lines searched in sequence are used as vertexes, all shape points in the one-way retrieval process are added into the graph in sequence, and edges are used for connecting two vertexes in an adjacent sequence; each vertex comprises an ID, a coordinate position, a parameter of a linear equation obtained by the belonged linear regression, an ID of the belonged linear equation and other attributes.
After the unidirectional retrieval is finished, all lane linear points found from the other direction are also sequentially added into the graph, edges are used for connecting two points in adjacent sequence, and the ID of each vertex is used as the unique identifier of each vertex in the graph. All the lane lines are searched in sequence, all the shape points on each lane line are stored in the graph structure, and the ID of each lane line shape point and the ID of each straight line are guaranteed to be different.
When the lane lines are stored in the graph structure in a straight line mode, the straight line obtained by linear fitting optimization is used as a unit, and the coordinates of the head point, the coordinates of the tail point and the straight line parameters of the straight line are stored to form a graph with straight line data as the top point.
By utilizing the graph storage data structure of the lane line and combining with the algorithm for analyzing the graph, partial topological features on the map can be conveniently judged.
The invention provides a rapid fusion optimization method of multi-channel segment data of lane line crowdsourcing data, which comprises the steps of firstly judging similar lane line points based on lane line point coordinates and direction angles or direction vectors thereof, and forming a plurality of sections of whole lane lines through fitting optimization of a plurality of sections of continuous similar points; carrying out processing optimization on the error fracture and the error connection of the multiple sections of lane lines after fitting optimization; the lane linear points with high precision, integrity and continuity can be obtained from the disordered original input data points, and the linear equation corresponding to the lane line can be obtained, so that vectorized lane line data is obtained; the method is simple, convenient and quick to execute, has low requirements on input data and high robustness, can adapt to 3D lane line point data and complex urban road data, and has obvious advantages compared with the conventional lane line clustering algorithm.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. A method for rapidly fusing and optimizing crowdsourcing data of multiple road segments of a lane line is characterized by comprising the following steps:
step 1, establishing a lane line local piecewise function model, and performing local fitting optimization on each shape point to form a section of lane line;
step 2, establishing a global lane line model according to the local piecewise function model to generate a plurality of sections of the whole lane line;
step 3, carrying out error fracture and error connection processing on the lane line to generate a lane line set;
the step 1 comprises the following steps:
step 101, establishing a set omega of shape points including crowd-sourced data of all to-be-processed lane road segments0;
Step 102, constructing a cuboid retrieval frame<l,w,h>Selecting a point of the shape p0(x0,y0,z0) Is the starting point of the search frame;
step 103, will form point p with0Does not exceed L in absolute value, and is in contact with the point p0Does not exceed W in absolute value, and is in contact with the point p0All shape point generation sets omega with Z-axis coordinate values having absolute values of differences not exceeding H1;
Step 104, from the set Ω1Selecting the shape points in the search frame to generate a shape point set omega2;
Step 105, for the set Ω2Performing linear regression on the shape points to generate a section of lane line;
In the step 104, the set Ω is determined1Any point of (1) pk(xk,yk,zk) The method for judging whether the search frame is in the search frame is as follows: determining eight corner points of the retrieval frame according to the size and the direction of the retrieval frame and the coordinates and the directions of the shape points, and determining eight corner points of the retrieval frame according to the corner points and the shape points pkDetermining the shape point pkWhether it is within the search box;
the process of determining the corner point comprises:
the initial coordinates of the eight corner points are:
and the coordinates of each angular point are multiplied by the rotation angular points of the rotation matrix to obtain:
after translational rotation, the angular points are obtained:
wherein the rotation matrixAngle of rotation I is an identity matrix and is a matrix of the identity,representing a vectorAndcross product of (1), unit rotation vector Is a vectorThe corresponding antisymmetric matrix:
2. the method of claim 1, wherein the step 104 comprises:
step 10401, taking the corner point as c31,c32,c33,c34,c35,c36,c37,c38Four vertices of three edges of a given rectangular parallelepiped, which are perpendicular to each other and intersect at a point, are denoted asWherein the content of the first and second substances,three mutually perpendicular edgesThe intersection point of (a);
step 10402, according to the intersection point and the shape point p selected in the step 10401kCoordinates of (2), calculating vectors
step 10404, when the inner product results of the six vector inner products are not less than 0, determining that the shape point p iskWithin or on the boundary of the search box; when at least one of the inner product results of the six vector inner products is less than 0, the shape point p is judgedkOutside the search box.
3. The method of claim 1, wherein the step 105 comprises:
step 10501, set an angle difference threshold and a distance difference threshold from the set Ω2Is selected from the said shape point p0The angle difference of is not more than the shape point generation set omega of the angle difference threshold value3From said set Ω2Is selected from the said shape point p0The set omega of shape point generation whose distance difference is not greater than the distance difference threshold4Generating said set Ω3And said set omega4Of (a) intersection omega5;
10502, according to said set Ω2The set omega3The set omega4And the intersection Ω5The number of interior shape points determines whether to perform a linear regression, yes, go to step 10503, no, go to step 10504;
step 10503, aligning the intersection Ω5The shape points in (1) are subjected to linear regression.
4. The method according to claim 3, wherein in step 10502 said set Ω2When the number of the shape points is less than a set threshold value, or the intersection omega5Is less than 2, or the set omega3Or the set Ω4The number of the shaped points of (C) accounts for the set omega2When the ratio of the number of the dots is smaller than a set threshold value, it is determined that linear regression cannot be performed.
5. The method of claim 3, wherein the step 10503 comprises:
step 1050301, selecting the set Ω5The coordinate value of X axis and the coordinate value of Z axis are used as characteristic independent variables, and the coordinate value of Y axis is subjected to linear regression to obtain a regression coefficient coef1=(cf3,cf4)TAnd intercept ic3;
Step 1050302, selecting the set Ω5The coordinate value of the Y axis and the coordinate value of the Z axis are used as characteristic independent variables, and the coordinate value of the X axis is subjected to linear regression to obtain a regression coefficient coef2=(cf5,cf6)TAnd intercept ic4;
Step 1050303, the equation for establishing the shape point to perform linear regression is:
parameter A1,B1,C1,D1,A2,B2,C2,D2Are respectively cf3,-1,cf4,ic3,-1,cf5,cf6,ic4。
6. The method according to claim 1, wherein the step 2 of building a global lane line model according to the local piecewise function model comprises:
after the step 102 and 105 generate a segment of lane line, the starting point of the search frame and the set Ω are determined2Form point in (b) in the set omega0Marked as a processed shape point; determining the starting point and direction of the next search box, and performing the step 102-105 on the set omega0The unprocessed shape points are processed until the set omega is obtained0Marking the points in the lane line as processed points, and connecting all the lane lines to form a plurality of sections of whole lane lines;
the process for determining the starting point and the direction of the next retrieval frame comprises the following steps:
and selecting a point which is farthest away from the starting point of the search frame on the independent variable as a tail point of the search frame, wherein the tail point of the search frame is used as the starting point of the next search frame, and the direction vector of a straight line obtained by linear regression is used as the direction vector of the starting point of the next search frame.
7. The method according to claim 1, wherein the step 3 of performing the error connection processing on the lane line comprises: performing linear regression on the shape points to generate each section of lane line, and comparing the variation ranges of the X axis, the Y axis and the Z axis of the coordinates of all the shape points in a single lane line;
when the variation range of the X axis is larger than the variation ranges of the Y axis and the Z axis, the shape points are sorted from small to large according to the X axis coordinate;
when the variation range of the Y axis is larger than the variation ranges of the X axis and the Z axis, the shape points are sorted from small to large according to the Y axis coordinate;
when the variation range of the Z axis is larger than the variation ranges of the X axis and the Y axis, the shape points are sorted from small to large according to the Z axis coordinate;
the process of performing the error fracture treatment on the lane line comprises the following steps:
for any two lane lines L0And L1If said lane line L is0And the tail point of the road line L1The distance of the first point is less than the set Euclidean distance, and the midpoint PLTo the lane line L0And the lane line L1The driving direction vertical distance of the head point of the vehicle is smaller than a set vertical distance threshold value;
merge the lane lines L0And L1Is a lane line L0Deleting the lane line L from the lane line set1;
Wherein the midpoint PLIs the lane line L0And the tail point of the road line L1Is the midpoint of the line connecting the first points of (a).
8. The method of claim 1, wherein step 3 is further followed by:
step 4, storing the lane lines in a graph structure by taking the figure points or straight lines as vertexes;
when the lane lines are stored in the graph structure in the form of shape points, the shape points of the lane lines searched in sequence are used as vertexes, all shape points in the one-way retrieval process are added into the graph in sequence, and edges are used for connecting two vertexes in an adjacent sequence; each vertex comprises an ID, a coordinate position, a parameter of a linear equation obtained by the belonged linear regression and an ID of the belonged straight line;
when the lane lines are stored in the graph structure in a straight line mode, the coordinates of the head point, the coordinates of the tail point and the straight line parameters of each straight line are stored to form a graph with straight line data as the top point by taking the straight line obtained by linear fitting optimization as a unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911350314.3A CN111222418B (en) | 2019-12-24 | 2019-12-24 | Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911350314.3A CN111222418B (en) | 2019-12-24 | 2019-12-24 | Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111222418A CN111222418A (en) | 2020-06-02 |
CN111222418B true CN111222418B (en) | 2022-04-29 |
Family
ID=70829164
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911350314.3A Active CN111222418B (en) | 2019-12-24 | 2019-12-24 | Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111222418B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111708856B (en) * | 2020-06-03 | 2022-07-29 | 武汉中海庭数据技术有限公司 | Crowdsourcing data segmented fusion method of lane line based on reinforcement learning |
CN111932506B (en) * | 2020-07-22 | 2023-07-14 | 四川大学 | Method for extracting discontinuous straight line in image |
CN112835363B (en) * | 2020-12-29 | 2023-08-01 | 武汉中海庭数据技术有限公司 | Method and device for controlling flow of large-scale crowdsourcing map lane line data fusion |
CN112733078B (en) * | 2020-12-29 | 2023-10-10 | 武汉中海庭数据技术有限公司 | Method and device for smooth connection among multiple paths of fragments of crowdsourcing data |
CN112580744B (en) * | 2020-12-29 | 2022-07-29 | 武汉中海庭数据技术有限公司 | Optimized fusion method for measuring data of same lane line in crowdsourced data road segment |
CN112418192B (en) * | 2021-01-21 | 2021-06-04 | 武汉中海庭数据技术有限公司 | Multi-line direct connection method and device among multi-channel segments of crowdsourcing data |
CN113435427B (en) * | 2021-08-27 | 2021-11-09 | 湖北亿咖通科技有限公司 | Method and device for aggregating lane lines |
CN114067562B (en) * | 2021-10-28 | 2022-08-19 | 武汉中海庭数据技术有限公司 | Lane effectiveness prediction method and prediction system based on multi-dimensional features |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104008387A (en) * | 2014-05-19 | 2014-08-27 | 山东科技大学 | Lane line detection method based on feature point piecewise linear fitting |
CN109737976A (en) * | 2019-01-07 | 2019-05-10 | 上海极奥网络科技有限公司 | Map road section and lane line automatic Generation |
CN110287904A (en) * | 2019-06-27 | 2019-09-27 | 武汉中海庭数据技术有限公司 | A kind of lane line extracting method, device and storage medium based on crowdsourcing data |
CN110345952A (en) * | 2019-07-09 | 2019-10-18 | 同济人工智能研究院(苏州)有限公司 | A kind of serializing lane line map constructing method and building system |
CN110414387A (en) * | 2019-07-12 | 2019-11-05 | 武汉理工大学 | A kind of lane line multi-task learning detection method based on lane segmentation |
CN110569704A (en) * | 2019-05-11 | 2019-12-13 | 北京工业大学 | Multi-strategy self-adaptive lane line detection method based on stereoscopic vision |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190325349A1 (en) * | 2018-04-19 | 2019-10-24 | Here Global B.V. | Method, apparatus, and system for traffic sign learning based on road network connectivity |
-
2019
- 2019-12-24 CN CN201911350314.3A patent/CN111222418B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104008387A (en) * | 2014-05-19 | 2014-08-27 | 山东科技大学 | Lane line detection method based on feature point piecewise linear fitting |
CN109737976A (en) * | 2019-01-07 | 2019-05-10 | 上海极奥网络科技有限公司 | Map road section and lane line automatic Generation |
CN110569704A (en) * | 2019-05-11 | 2019-12-13 | 北京工业大学 | Multi-strategy self-adaptive lane line detection method based on stereoscopic vision |
CN110287904A (en) * | 2019-06-27 | 2019-09-27 | 武汉中海庭数据技术有限公司 | A kind of lane line extracting method, device and storage medium based on crowdsourcing data |
CN110345952A (en) * | 2019-07-09 | 2019-10-18 | 同济人工智能研究院(苏州)有限公司 | A kind of serializing lane line map constructing method and building system |
CN110414387A (en) * | 2019-07-12 | 2019-11-05 | 武汉理工大学 | A kind of lane line multi-task learning detection method based on lane segmentation |
Also Published As
Publication number | Publication date |
---|---|
CN111222418A (en) | 2020-06-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111222418B (en) | Crowdsourcing data rapid fusion optimization method for multiple road segments of lane line | |
CN109541997B (en) | Spraying robot rapid intelligent programming method for plane/approximate plane workpiece | |
Carsten et al. | 3d field d: Improved path planning and replanning in three dimensions | |
CN107749079B (en) | Point cloud quality evaluation and track planning method for unmanned aerial vehicle scanning reconstruction | |
KR900003123B1 (en) | Method for evaluating free surface and n.c. system | |
CN108519094B (en) | Local path planning method and cloud processing terminal | |
US10288434B2 (en) | Map-centric map matching method and apparatus | |
CN110599506B (en) | Point cloud segmentation method for three-dimensional measurement of complex special-shaped curved surface robot | |
CN104200212B (en) | A kind of building external boundary line drawing method based on airborne LiDAR data | |
CN103247041B (en) | A kind of dividing method of the cloud data of the many geometric properties based on local sampling | |
CN111209805B (en) | Rapid fusion optimization method for multi-channel segment data of lane line crowdsourcing data | |
CN114526739B (en) | Mobile robot indoor repositioning method, computer device and product | |
Kim et al. | GP-ICP: Ground plane ICP for mobile robots | |
CN107818598B (en) | Three-dimensional point cloud map fusion method based on visual correction | |
CN105654483A (en) | Three-dimensional point cloud full-automatic registration method | |
CN113610917A (en) | Circular array target center image point positioning method based on blanking points | |
CN103744886B (en) | Directly extracted k nearest neighbor searching algorithm | |
CN111192364A (en) | Low-cost mobile multi-robot vision simultaneous positioning and map creating method | |
CN114332291A (en) | Oblique photography model building outer contour rule extraction method | |
CN109919955A (en) | The tunnel axis of ground formula laser radar point cloud extracts and dividing method | |
CN111141296B (en) | Preprocessing method and system for multi-channel fragment data of lane line crowdsourcing data | |
CN106931978B (en) | Indoor map generation method for automatically constructing road network | |
Zhu et al. | Vdb-edt: An efficient euclidean distance transform algorithm based on vdb data structure | |
CN107562779B (en) | Spatial topology processing method for two-dimensional vector polygon self-intersection | |
CN112269848A (en) | Crowdsourcing trajectory data fusion method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |