CN104988818B - Intersection multi-lane calibration method based on perspective transformation - Google Patents

Abstract

Disclosed is an intersection multi-lane calibration method based on perspective transformation. The intersection multi-lane calibration method comprises the following steps that 1, an image with an obvious lane line is selected from a monitor video of an intersection; 2, edge extraction is performed through a Canny operator on the basis of the selected image, and then n straight line segments are extracted by means of Hough transformation; 3, horizontal line segment sets H and vertical line segment sets S are obtained according to slopes; 4, the sets H and S obtained in the step 3 are screened preliminarily; 5, the H and S finally obtained from the step 4 are marked as H' and S', and the line segments are further filtered; 6, a red light line is determined in H'' and S'' is further filtered so as to determine a lane line; 7, the line segment representing the lane line is found out from the set S''' obtained from the step 6.

Description

A kind of intersection multilane scaling method based on perspective transform

Technical field

The present invention relates to intelligent transportation field, specifically a kind of to analyze intersection image to demarcate respectively by perspective transform The method of lane position.

Background technology

With developing rapidly for China's communication, the automobile quantity in city increases year by year, and urban highway traffic is blocked, gathered around Crowded problem seems more and more prominent.The construction of modern transportation control system, can but be greatly improved the utilization of existing road Rate, and reliable peccancy detection is the guarantee of intelligent transportation application.

It is to differentiate that violating the regulations one is crucial that the multilane of intersection is demarcated.At present in the embedded induction coil in intersection Not only high cost, construction are difficult for mode, are also inconvenient for safeguarding in some crowded section of highway.Computer vision technique as an alternative is because of which Light the features such as, has obtained extensive concern in intelligent transportation field.

Different from the single, double lane identification currently used for DAS (Driver Assistant System), the intersection in city generally requires together When recognize more tracks and calibrate red light stop line position.

The content of the invention

The present invention to be overcome the disadvantages mentioned above of prior art, with perspective transform as foundation, there is provided a kind of intersection it is many Track scaling method.The present invention comprises the steps：

Step 1：The obvious image of lane line is chosen from the monitor video of intersection, it is to avoid occlusion etc. is disturbed, Wherein the height of image is height, and width is width, and unit is pixel；

Step 2：Edge extracting is carried out using Canny operators in the image basis chosen, is then carried using Hough transform Take out n bar straightway l_i(i=1,2 ..., n), wherein l_iStart-stop extreme coordinates be denoted as A respectively_i(x_Ai,y_Ai) and B_i(x_Bi, y_Bi), zero is the image upper left corner, and the corresponding slope of every line segment is k_i, computing formula is as follows：

Step 3：Horizontal line section set H and vertical line segment aggregate S is obtained according to slope, wherein：

H={ l_i|k_i∈(-0.1,0.1),i∈{1,2,…,n}} (2)

S={ l_i|k_i∈(-∞,-0.65)∪(0.65,∞),i∈{1,2,…,n}} (3)

Step 4：The set H obtained to step 3 and S carries out preliminary screening, specially：

Step 4.1：All extreme coordinates are metAndStraightway l_iFrom Reject in horizontal line section set H and vertical line segment aggregate S, obtain new H and S；

Step 4.2 (optional)：If video image includes that timestamp is stabbed with place, further extreme coordinates are located at The image set time stabs and stabs the straightway l in region with place_iReject from horizontal line section set H and vertical line segment aggregate S, again Obtain H and the S for updating；

Step 5：Make the final H that step 4 is obtained H ' and S ' is designated as with S, further line segment is filtered：

Step 5.1：Calculated by below equation and horizontal line section set H ' is clustered and r subset h is obtained_t：

AndAnd h_t={ l_tk| t=1,2 ..., r；K=1,2 ... w_t} (4)

Wherein p=1,2 ..., r；Q=1,2 ..., r；p≠q (5)

Wherein, w_tRepresent subset h_tIn line segment number；l_tkRepresent the kth bar line segment of t-th subset；In formula (6), l_psRepresent Subset h_pIn the s article line segment, l_qzRepresent subset h_qIn the z article line segment, DistY (l_ps,l_qz) represent line segment l_psWith line segment l_qz Between vertical distance, y_ApsRepresent subset h_pIn s article of line segment starting endpoint vertical coordinate, y_BpsRepresent subset h_pIn the s article The vertical coordinate of the termination end points of line segment, y_AqzRepresent subset h_qIn z article of line segment starting endpoint vertical coordinate, y_BqzRepresent subset h_qIn z article of line segment termination end points vertical coordinate；Formula (7) represents line segment l_psWith line segment l_qzBetween vertical distance should meet Condition；Formula (8) represents subset h_pInternal i-th line section l_piWith j-th strip line segment l_pjBetween vertical distance should meet Condition, w_pRepresent subset h_pLine segment number；

Step 5.2：Extract each subset h_t(t=1,2 ..., r) in most long line segment constitute new horizontal line section set H″；

Step 5.3：Cluster is carried out to vertical line segment aggregate S ' by below equation calculating and obtains m subset s_t：

AndAnd s_t={ l_tv| t=1,2 ..., m；V=1,2 ... u_t} (9)

Wherein p=1,2 ..., m；Q=1,2 ..., m；p≠q (10)

Wherein, u_tRepresent subset s_tIn line segment number, l_tvRepresent the v article line segment of t-th subset；In formula (11), l_psTable Show subset s_pIn the s article line segment, l_qzRepresent subset s_qIn the z article line segment, DistX (l_ps,l_qz) represent line segment l_psWith line segment l_qzBetween horizontal range, x_ApsRepresent subset s_pIn s article of line segment starting endpoint abscissa, x_BpsRepresent subset s_pIn The abscissa of the termination end points of s bar line segments, x_AqzRepresent subset s_qIn z article of line segment starting endpoint abscissa, x_BqzRepresent Subset s_qIn z article of line segment termination end points abscissa；Formula (12) represents line segment l_psWith line segment l_qzBetween horizontal range should The condition of the satisfaction；Formula (13) represents subset s_pInternal i-th line section l_piL is estimated with j-th strip line_pjBetween horizontal range should The condition of satisfaction, u_pRepresent subset s_pLine segment number；

Step 5.4：Extract each subset s_t(t=1,2 ..., m) in most long line segment constitute new vertical line segment aggregate S″；

Step 6：Further filter in H " middle to determine red light line and to S ", specially：

Step 6.1：Horizontal line section set H " is sorted according to the following rules：Make i-th line section l_iWith j-th strip line segment l_jEnd Point coordinates meetsWherein l_i∈ H ", l_j∈ H ", i ＜ j；

Step 6.2：Retain sequence after H " in the 1st article of line segment as red light stop line；

Step 6.3：Calculate vertical line segment aggregate S " in any two line segments l_iWith l_jExtending line intersection point C_ij(x_ij,y_ij), i ≠ J, builds intersection point collection P：

P={ C_ij(x_ij,y_ij) | i=1,2 ..., n "；J=1,2 ..., n " } (14)

y_ij=k_i(x_ij-x_Ai)+y_Ai (16)

Wherein k_iAnd k_j" the middle conductor l corresponding to S_iAnd l_jSlope, the quantity of n " represent S " middle conductor, (x_Ai,y_Ai) represent Line segment l_iStarting endpoint coordinate, (x_Aj,y_Aj) represent line segment l_jStarting endpoint coordinate；

Step 6.4：Vanishing point V (x are found out from intersection point collection P_o,y_o), wherein vanishing point represents friendship of the parallel lines in perspective projection Point, vanishing point V (x_o,y_o) feature be：And any point G (x in antinode collection P_g,y_g) meetG is subscript, and other arbitrfary points of vanishing point V are different from representing intersection point collection P；

Step 6.5：Obtain vanishing point V (x_o,y_o) to vertical line segment aggregate S " middle arbitrary line sections l_iApart from d_i, whereinReject vertical line segment aggregate S " inAll line segments obtain new vertical Line segment aggregate S " '；

Step 7：Find out the set S that step 6 is obtained " ' in represent the line segment of lane line, specially：

Step 7.1：To vertical line segment aggregate S " ' be ranked up so that i-th line section l after sequence_iWith j-th strip line segment l_j's Extreme coordinates meetl_i∈ S " ', l_j∈ S " ', i ＜ j, i=1,2 ..., n " ', j=1,2 ..., N " ', wherein n " ' represent S " ' in line segment quantity, x_AiAnd x_BiRepresent line segment l_iStarting and termination end points abscissa, x_AjWith x_BjRepresent line segment l_jStarting and termination end points abscissa；

Step 7.2：Using dynamic programming algorithm to vertical line segment aggregate S " ' screen, reservation as much as possible is wherein Line segment, and make any of which line segment l_iTo adjacent segments l_i+1Horizontal range meet formula (17) and (18)：

Formula (11) is shown in the wherein definition of DistX；

Step 7.3：Vanishing point V (x are drawn one by one_o,y_o) to through step 7.2 screening after vertical line segment aggregate S " ' in per bar Line segment l_iThe ray at midpoint can mark the multilane position at each crossing.

It is an advantage of the invention that：The present invention can effectively exclude the interference of intersection complex background, accurate using perspective transform Intersection multilane position is calibrated really.

Description of the drawings

Fig. 1 is four intersection images that embodiments of the invention are chosen.

Fig. 2 is four intersection images through step 3 process of the present invention.

Fig. 3 is four intersection images through step 4 process of the present invention.

Fig. 4 is four intersection images through step 5 process of the present invention.

Fig. 5 is four intersection images through step 6.3 process of the present invention.

Fig. 6 is four intersection images through step 6.4 process of the present invention.

Fig. 7 is four intersection images through step 6.5 process of the present invention.

Fig. 8 is four intersection images through step 7.2 process of the present invention.

Fig. 9 is four intersection images through step 7.3 process of the present invention.

Specific embodiment

The intersection multilane scaling method based on perspective transform of the present invention is elaborated with reference to embodiment Specific embodiment.

Step 1：The obvious image of lane line is chosen from the monitor video of intersection, it is to avoid occlusion etc. is disturbed, Wherein the height of image is height, and width is width, and unit is pixel.The four intersection images such as Fig. 1 institutes for choosing Show.

Step 2：Edge extracting is carried out using Canny operators in the image basis chosen, is then carried using Hough transform Take out n bar straightway l_i(i=1,2 ..., n), wherein l_iStart-stop extreme coordinates be denoted as A respectively_i(x_Ai,y_Ai) and B_i(x_Bi, y_Bi), zero is the image upper left corner.The corresponding slope of every line segment is k_i, computing formula is as follows：

Step 3：Horizontal line section set H and vertical line segment aggregate S is obtained according to slope, wherein：

H={ l_i|k_i∈(-0.1,0.1),i∈{1,2,…,n}} (2)

S={ l_i|k_i∈(-∞,-0.65)∪(0.65,∞),i∈{1,2,…,n}} (3)

As shown in Fig. 2 wherein, the straightway in horizontal line section set H is marked the straightway for extracting with a, vertical line segment Straightway in set S is marked with b.

Step 4：The set H obtained to step 3 and S carries out simple screening, specially：

Step 4.1：All extreme coordinates are metAndStraightway l_iFrom Reject in horizontal line section set H and vertical line segment aggregate S, obtain new H and S.

Step 4.2 (optional)：If video image includes that timestamp is stabbed with place, further extreme coordinates are located at The image set time stabs and stabs the straightway l in region with place_iReject from horizontal line section set H and vertical line segment aggregate S, again Obtain H and the S for updating.

As shown in figure 3, the straightway in horizontal line section set H this moment is marked with a ＇, the straight line in vertical line segment aggregate S Section is marked with b ＇.

Step 5：Make the final H that step 4 is obtained H ' and S ' is designated as with S, carry out line segment and filter roughly：

Step 5.1：Cluster is carried out to horizontal line section set H ' and obtains r subset h_t, it is desired to meet following condition：

AndAnd h_t={ l_tk| t=1,2 ..., r；K=1,2 ... w_t} (4)

Wherein p=1,2 ..., r；Q=1,2 ..., r；p≠q (5)

Wherein, w_tRepresent subset h_tIn line segment number；l_tkRepresent the kth bar line segment of t-th subset；In formula (6), l_psRepresent Subset h_pIn the s article line segment, l_qzRepresent subset h_qIn the z article line segment, DistY (l_ps,l_qz) represent line segment l_psWith line segment l_qz Between vertical distance, y_ApsRepresent subset h_pIn s article of line segment starting endpoint vertical coordinate, y_BpsRepresent subset h_pIn the s article The vertical coordinate of the termination end points of line segment, y_AqzRepresent subset h_qIn z article of line segment starting endpoint vertical coordinate, y_BqzRepresent subset h_qIn z article of line segment termination end points vertical coordinate；Formula (7) represents line segment l_psWith line segment l_qzBetween vertical distance should meet Condition；Formula (8) represents subset h_pInternal i-th line section l_piWith j-th strip line segment l_pjBetween vertical distance should meet Condition, w_pRepresent subset h_pLine segment number；

Step 5.2：Extract each subset h_t(t=1,2 ..., r) in most long line segment constitute new horizontal line section set H″。

Step 5.3：Cluster is carried out to vertical line segment aggregate S ' and obtains m subset s_t, it is desirable to meet following condition：

AndAnd s_t={ l_tv| t=1,2 ..., m；V=1,2 ... u_t} (9)

Wherein p=1,2 ..., m；Q=1,2 ..., m；p≠q (10)

Wherein, u_tRepresent subset s_tIn line segment number, l_tvRepresent the v article line segment of t-th subset；In formula (11), l_psTable Show subset s_pIn the s article line segment, l_qzRepresent subset s_qIn the z article line segment, DistX (l_ps,l_qz) represent line segment l_psWith line segment l_qzBetween horizontal range, x_ApsRepresent subset s_pIn s article of line segment starting endpoint abscissa, x_BpsRepresent subset s_pIn The abscissa of the termination end points of s bar line segments, x_AqzRepresent subset s_qIn z article of line segment starting endpoint abscissa, x_BqzRepresent Subset s_qIn z article of line segment termination end points abscissa；Formula (12) represents line segment l_psWith line segment l_qzBetween horizontal range should The condition of the satisfaction；Formula (13) represents subset s_pInternal i-th line section l_piL is estimated with j-th strip line_pjBetween horizontal range should The condition of satisfaction, u_pRepresent subset s_pLine segment number；

Step 5.4：Extract each subset s_t(t=1,2 ..., m) in most long line segment constitute new vertical line segment aggregate S″。

As shown in figure 4, horizontal line section set H this moment " in straightway first straight line (1) mark, vertical line-segment sets Straightway second straight line (2) in conjunction S " is marked.

Step 6：Further filter in H " middle to determine red light line and to S ", specially：

Step 6.1：Horizontal line section set H " is sorted according to the following rules：Make i-th line section l_iWith j-th strip line segment l_jEnd Point coordinates meetsWherein l_i∈ H ", l_j∈ H ", i ＜ j.

Step 6.2：Retain sequence after H " in the 1st article of line segment as red light stop line.

Step 6.3：Calculate vertical line segment aggregate S " in any two line segments l_iWith l_jExtending line intersection point C_ij(x_ij,y_ij), i ≠ J, builds intersection point collection P：

P={ C_ij(x_ij,y_ij) | i=1,2 ..., n "；J=1,2 ..., n " } (14)

y_ij=k_i(x_ij-x_Ai)+y_Ai (16)

Wherein k_iAnd k_j" the middle conductor l corresponding to S_iAnd l_jSlope, the quantity of n " represent S " middle conductor, (x_Ai,y_Ai) represent Line segment l_iStarting endpoint coordinate, (x_Aj,y_Aj) represent line segment l_jStarting endpoint coordinate.As shown in figure 5, red light this moment stops Line is marked with the 5th straight line (5), vertical line segment aggregate S " in straightway marked with the 4th straight line (4), the prolongation of all line segments Line and its intersection point all use the 3rd straight line (3) labelling.

Step 6.4：Vanishing point V (x are found out from intersection point collection P_o,y_o), wherein vanishing point represents friendship of the parallel lines in perspective projection Point, vanishing point V (x_o,y_o) feature be：And any point G (x in antinode collection P_g,y_g) meetG is subscript, and other arbitrfary points of vanishing point V are different from representing intersection point collection P.As shown in fig. 6, Red light stop line this moment is marked with the 8th straight line (8), vertical line segment aggregate S " in straightway marked with the 7th straight line (7), The extended line of all line segments and its intersection point all use the 6th straight line (6) labelling.Vanishing point position O labellings.

Step 6.5：Obtain vanishing point V (x_o,y_o) to vertical line segment aggregate S " middle arbitrary line sections l_iApart from d_i, whereinReject vertical line segment aggregate S " inAll line segments obtain new vertical Line segment aggregate S " '.It is straight in as shown in fig. 7, red light stop line this moment is marked with the 9th straight line (9), vertical line segment aggregate S " ' Line segment is marked with the tenth straight line (10), vanishing point position O labellings.

Step 7：Find out the set S that step 6 is obtained " ' in represent the line segment of lane line, specially：

Step 7.1：To vertical line segment aggregate S " ' be ranked up so that wherein i-th line section l_iWith j-th strip line segment l_jEnd Point coordinates meetsl_i∈ S " ', l_j∈ S " ', i ＜ j, i=1,2 ..., n " ', j=1,2 ..., n " ', Line segment quantity in wherein n " ' expression S " ', x_AiAnd x_BiRepresent line segment l_iStarting and termination end points abscissa, x_AjAnd x_BjTable Timberline section l_jStarting and termination end points abscissa.

Step 7.2：Using dynamic programming algorithm to vertical line segment aggregate S " ' screen, reservation as much as possible is wherein Line segment, and make any of which line segment l_iTo adjacent segments l_i+1Horizontal range meet formula (17) and (18)：

Formula (11) is shown in the wherein definition of DistX.As shown in figure 8, red light stop line this moment is marked with the 11st straight line (11) Go out, vertical line segment aggregate S " ' in straightway marked with the 12nd straight line (12), vanishing point position O labellings.

Step 7.3：Vanishing point V (x are drawn one by one_o,y_o) to through step 7.2 screening after vertical line segment aggregate S " ' in per bar Line segment l_iThe ray at midpoint can mark the multilane position at each crossing.As shown in figure 9, red light stop line this moment is used 14th straight line (14) is marked, and the lane position in single intersection direction is marked with the tenth trilete rays (13), vanishing point position O labellings.

Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, the protection of the present invention Being not construed as of scope is only limitted to the concrete form stated by embodiment, and protection scope of the present invention is also and in this area skill Art personnel according to present inventive concept it is conceivable that equivalent technologies mean.

Claims (1)

1. a kind of intersection multilane scaling method based on perspective transform, comprises the following steps：

Step 1：The obvious image of lane line is chosen from the monitor video of intersection, it is to avoid the interference of occlusion, wherein The height of image is height, and width is width, and unit is pixel；

Step 2：Edge extracting is carried out using Canny operators in the image basis chosen, is then extracted using Hough transform N bar straightway l_i(i=1,2 ..., n), wherein l_iStart-stop extreme coordinates be denoted as A respectively_i(x_Ai,y_Ai) and B_i(x_Bi,y_Bi), sit Mark origin is the image upper left corner, and the corresponding slope of every line segment is k_i, computing formula is as follows：

Step 3：Horizontal line section set H and vertical line segment aggregate S is obtained according to slope, wherein：

H={ l_i|k_i∈(-0.1,0.1),i∈{1,2,…,n}} (2)

S={ l_i|k_i∈(-∞,-0.65)∪(0.65,∞),i∈{1,2,…,n}} (3)

Step 4：The set H obtained to step 3 and S carries out preliminary screening, specially：

Step 4.1：All extreme coordinates are metAndStraightway l_iFrom horizontal line Reject in Duan Jihe H and vertical line segment aggregate S, obtain new H and S；

Step 4.2：If video image includes that timestamp is stabbed with place, when extreme coordinates being located at image fixation further Between stamp with place stab region straightway l_iReject from horizontal line section set H and vertical line segment aggregate S, obtain what is updated again H and S；

Step 5：Make the final H that step 4 is obtained H ' and S ' is designated as with S, further line segment is filtered：

Step 5.1：Calculated by below equation and horizontal line section set H ' is clustered and r subset h is obtained_t：

AndAnd h_t={ l_tk| t=1,2 ..., r；K=1,2 ... w_t} (4)

Wherein p=1,2 ..., r；Q=1,2 ..., r；p≠q (5)

$DistY(l_{ps},l_{qz})=|\frac{y_{Aps}+y_{Bps}}{2}-\frac{y_{Aqz}+y_{Bqz}}{2}|---\left(6\right)$

$DistY(l_{ps},l_{qz})>\frac{height}{10}---\left(7\right)$

$DistY(l_{pi},l_{pj})<\frac{height}{10},i=1,2,...,w_p,j=1,2,...,w_p,i\&NotEqual;j---\left(8\right)$

Wherein, w_tRepresent subset h_tIn line segment number；l_tkRepresent the kth bar line segment of t-th subset；In formula (6), l_psRepresent subset h_pIn the s article line segment, l_qzRepresent subset h_qIn the z article line segment, DistY (l_ps,l_qz) represent line segment l_psWith line segment l_qzBetween Vertical distance, y_ApsRepresent subset h_pIn s article of line segment starting endpoint vertical coordinate, y_BpsRepresent subset h_pIn the s article line segment Termination end points vertical coordinate, y_AqzRepresent subset h_qIn z article of line segment starting endpoint vertical coordinate, y_BqzRepresent subset h_qIn The vertical coordinate of the termination end points of z article of line segment；Formula (7) represents line segment l_psWith line segment l_qzBetween vertical distance should meet Condition；Formula (8) represents subset h_pInternal i-th line section l_piWith j-th strip line segment l_pjBetween the bar that should meet of vertical distance Part, w_pRepresent subset h_pLine segment number；

Step 5.2：Extract each subset h_t(t=1,2 ..., r) in most long line segment constitute new horizontal line section set H "；

Step 5.3：Cluster is carried out to vertical line segment aggregate S ' by below equation calculating and obtains m subset s_t：

AndAnd s_t={ l_tv| t=1,2 ..., m；V=1,2 ... u_t} (9)

Wherein p=1,2 ..., m；Q=1,2 ..., m；p≠q (10)

$DistX(l_{ps},l_{qz})=|\frac{x_{Aps}+x_{Bps}}{2}-\frac{x_{Aqz}+x_{Bqz}}{2}|---\left(11\right)$

$DistX(l_{ps},l_{qz})>\frac{width}{10}---\left(12\right)$

$DistX(l_{pi},l_{pj})<\frac{width}{10},i=1,2,...,u_p,j=1,2,...,u_p,i\&NotEqual;j---\left(13\right)$

Wherein, u_tRepresent subset s_tIn line segment number, l_tvRepresent the v article line segment of t-th subset；In formula (11), l_psRepresent son Collection s_pIn the s article line segment, l_qzRepresent subset s_qIn the z article line segment, DistX (l_ps,l_qz) represent line segment l_psWith line segment l_qzIt Between horizontal range, x_ApsRepresent subset s_pIn s article of line segment starting endpoint abscissa, x_BpsRepresent subset s_pIn the s bar line The abscissa of the termination end points of section, x_AqzRepresent subset s_qIn z article of line segment starting endpoint abscissa, x_BqzRepresent subset s_q In z article of line segment termination end points abscissa；Formula (12) represents line segment l_psWith line segment l_qzBetween horizontal range should meet Condition；Formula (13) represents subset s_pInternal i-th line section l_piL is estimated with j-th strip line_pjBetween horizontal range should meet Condition, u_pRepresent subset s_pLine segment number；

Step 5.4：Extract each subset s_t(t=1,2 ..., m) in most long line segment constitute new vertical line segment aggregate S "；

Step 6：Further filter in H " middle to determine red light line and to S ", specially：

Step 6.1：Horizontal line section set H " is sorted according to the following rules：Make i-th line section l_iWith j-th strip line segment l_jEnd points sit Mark meetsWherein l_i∈ H ", l_j∈ H ", i ＜ j；

Step 6.2：Retain sequence after H " in the 1st article of line segment as red light stop line；

Step 6.3：Calculate vertical line segment aggregate S " in any two line segments l_iWith l_jExtending line intersection point C_ij(x_ij,y_ij), i ≠ j, structure Establish diplomatic relations point set P：

P={ C_ij(x_ij,y_ij) | i=1,2 ..., n "；J=1,2 ..., n " } (14)

$x_{ij}=\frac{k_i\&CenterDot;x_{Ai}-k_j\&CenterDot;x_{Aj}+y_{Aj}-y_{Ai}}{k_i-k_j}---\left(15\right)$

y_ij=k_i(x_ij-x_Ai)+y_Ai (16)

Wherein k_iAnd k_j" the middle conductor l corresponding to S_iAnd l_jSlope, the quantity of n " represent S " middle conductor, (x_Ai,y_Ai) represent line segment l_iStarting endpoint coordinate, (x_Aj,y_Aj) represent line segment l_jStarting endpoint coordinate；

Step 6.4：Vanishing point V (x are found out from intersection point collection P_o,y_o), wherein vanishing point represents intersection point of the parallel lines in perspective projection, Vanishing point V (x_o,y_o) feature be：And any point G (x in antinode collection P_g,y_g) meetG is subscript, and other arbitrfary points of vanishing point V are different from representing intersection point collection P；

Step 6.5：Obtain vanishing point V (x_o,y_o) to vertical line segment aggregate S " middle arbitrary line sections l_iApart from d_i, whereinReject vertical line segment aggregate S " inAll line segments obtain new vertical Line segment aggregate S " '；

Step 7：Find out the set S that step 6 is obtained " ' in represent the line segment of lane line, specially：

Step 7.1：To vertical line segment aggregate S " ' be ranked up so that i-th line section l after sequence_iWith j-th strip line segment l_jEnd points Coordinate meetsl_i∈ S " ', l_j∈ S " ', i ＜ j, i=1,2 ..., n " ', j=1,2 ..., n " ', wherein Line segment quantity in n " ' expression S " ', x_AiAnd x_BiRepresent line segment l_iStarting and termination end points abscissa, x_AjAnd x_BjRepresent line Section l_jStarting and termination end points abscissa；

Step 7.2：Using dynamic programming algorithm to vertical line segment aggregate S " ' screen, retain line segment therein, and make wherein Arbitrary line segment l_iTo adjacent segments l_i+1Horizontal range meet formula (17) and (18)：

$DistX\left(l_i,l_{i+1}\right)\&Element;\&lsqb;\frac{width}{5},\frac{width}{2}\&rsqb;---\left(17\right)$

$|DistX(l_i,l_{i+1})-DistX(l_{i+1},l_{i+2})|<\frac{width}{5}---\left(18\right)$

Formula (11) is shown in the wherein definition of DistX；

Step 7.3：Vanishing point V (x are drawn one by one_o,y_o) to through step 7.2 screening after vertical line segment aggregate S " ' in every line segment l_iThe ray at midpoint can mark the multilane position at each crossing.