CN105678287A - Ridge-measure-based lane line detection method - Google Patents

Abstract

The invention, which belongs to the technical field of active safety of the automobile, discloses a ridge-measure-based lane line detection method. The method comprises: step one, an original image of a road in front of a vehicle is collected; step two, a region of interest is determined; step three, a ridge measure numerical value of an image in the region of interest is calculated; step four, an average value p and a variance sigma of ridge measure values of all pixel points are calculated, all pixel points with the ridge measure values larger than p-3sigma are viewed as potential lane line feature points and are set as 1, and non-potential feature points in the image are set as 0, so that a binary image only including the potential lane line feature points is obtained; step five, the extracted lane line feature points are screened to obtain a binary image Img_B only including a lane line segment; and step six, linear hough transform is carried out on the binary image Img_B obtained by the step five and a linear model parameter is obtained. According to the invention, with ridge measure feature point extraction and hough transform, lane line edge features are extracted and the linear model is established. The method has advantages of high stability and high robustness.

Description

A kind of method for detecting lane lines based on ridge tolerance

Technical field

The invention belongs to technical field of image processing, relate to image segmentation and the detection of image curve geometrical property, be specifically related to a kind of through street method for detecting lane lines based on ridge tolerance.

Background technology

Traffic safety problem has become international big problem, and therefore the impact of human life's property is self-evident by the safety of automobile. Raising along with the development of highway and automotive performance, automobile driving speed is also accelerated accordingly, in addition the increase of automobile quantity and transportation are day by day busy, motor-vehicle accident increases caused casualties and property loss, having become a social problem that can not be ignored, the traffic safety of automobile more seems extremely important. And the accident that traditional passive security has far from been avoided that traffic occurs, the therefore concept of active safety formation slowly constantly perfect. Visual sensing due to have contain much information, with low cost, have a wide range of applications in field of automotive active safety.

Lane detection technology refers to that means such as utilizing image sensing detect the technology of road track deficiency and excess graticule, and it is one of key technology of field of automotive active safety. In the Lane Keeping System of view-based access control model, the detection of lane line and tracking are basic, necessary functions, and it can prevent vehicle lane departure, can also give the road environment information that the offer such as other active safety system that includes anti-collision warning is important simultaneously. From middle nineteen nineties in last century, include the state such as the U.S., German America and Europe and carried out the research of a large amount of related direction, and successfully developed some Lane Departure Warning System differed from one another. These systems when there is skew or there is movement tendency in vehicle to driver with information warning, even actively get involved wagon control, to reach the purpose of Accident prevention generation.

Current existing through street method for detecting lane lines often adopts relatively simple method and loose constraint to obtain lane line Edge Feature Points, and adopts complex model in the lane line parameter estimation stage, such as optimum Bayesian Estimation and maximal possibility estimation etc. Such method has good Detection results under major part scene, but some road is due to by trees, light intensity, hollow, road surface material inequality, other reason such as pavement markers and shadow effect, make the method when lane line marked situation is complicated and road surface uniformity is poor, often a large amount of non-lane line characteristic points are judged as lane line characteristic point, cause lane line parameter estimation deviation.

In order to convenient, present disclosure is described, it is necessary to some concepts are illustrated.

Concept 1. camera parameters and camera calibration: what camera parameter described is the imaging geometry model of video camera itself. It characterizes object and is mapped to the transformational relation under two dimensional image coordinate system from three-dimensional world coordinate system. The process being obtained these parameters by some experiment is then referred to as camera calibration (or calibration). The parameter of video camera includes inner parameter and external parameter, and inner parameter includes principal point coordinate, focal length etc., and external parameter includes camera position, attitude etc.

Concept 2. region of interest (ROI:ReignofInterest): in image processing field, area-of-interest (ROI) refers to the local image region selected from image, and this region is the emphasis that graphical analysis is paid close attention to. Determine this region to be further processed. Use ROI often can reduce the process time, increase precision.

Concept 3. Hough transformation: i.e. Hough transform, its basic thought is the duality utilizing point-line, it may be assumed that the straight line intersected in the corresponding parameter space of the point of conllinear in image space; Otherwise, intersect at all straight lines of same point in parameter space has the point of conllinear corresponding with it in image space. Utilize the maximum value search problem that linear feature search problem can be converted in parameter space by Hough transformation.

Summary of the invention

For the problems referred to above, the present invention proposes a kind of method for detecting lane lines based on ridge tolerance, and compared with congenic method, ridge metric calculation has been applied to traffic lane line neighborhood territory pixel, and conventional edge extracts the difference having taken into consideration only between adjacent two pixels. Therefore, it is strong that institute's extracting method has stability, the advantages such as applicable working condition is relatively broad.

The definition that ridge is measured by the present invention: a kind of metric to image intensity value and ridge form degree of approximation. Such as, ridge present middle high, both sides are low, and have certain symmetric form, if certain area grayscale value is big in image, both sides gray value is low, and has good symmetry, then the ridge metric of this point is relatively big, otherwise less.

Realize technical scheme as follows:

A kind of method for detecting lane lines based on ridge tolerance, comprises the steps:

Step 1: collection vehicle road ahead original image Img;

Step 2: determine region of interest I;

Step 3: calculate the ridge metric values of image in region of interest;

Step 4: calculate meansigma methods p and the variances sigma of the ridge metric of all pixels, all ridge metrics pixel more than p-3 σ is considered as potential lane line characteristic point and is set to 1, potential characteristic point non-in image is then set to 0, obtains containing only the binary image of potential lane line characteristic point;

Step 5: the lane line characteristic point extracted through step 4 is screened, it is thus achieved that contain only the binary image Img_B of lane line line segment;

Step 6: the binary image Img_B obtained for step 5, carries out traditional straight line Hough transformation, and obtains straight line model parameter.

It is preferred that scheme, implementing of described step 1 includes: by being arranged on the camera acquisition vehicle front road conditions original image Img of vehicle interior or outside.

It is preferred that scheme, described in described step 2, the defining method of region of interest I is: according to the inner parameter of video camera and external parameter, in acquisition camera view, the region below ground level vanishing line, within image right boundary is region of interest.

It is preferred that scheme, implementing of described step 3 includes:

Step 3-1: by gray level image g (x) in original ROI and 2-d gaussian filters device G_σdDo convolution algorithm:

L_σd(x)=G_σd(x) * g (x);

Wherein, G_σdBeing an anisotropic Gaussian core, its covariance matrix is ∑=diag (σ_dx,σ_dy), wherein, σ_dyIt is constant H_R, σ_dxFor variable quantity, its numerical value is the half of track live width corresponding to each row of image;

Step 3-2: calculate the image each pixel x gradient vector field along u row and v column direction:

$w_{\mathrm{\σ}d}\left(x\right)={\left({\∂}_u{L}_{\mathrm{\σ}d}\right(x),{\∂}_v{L}_{\mathrm{\σ}d}(x\left)\right)}^T;$

Additionally, calculate new matrix:

s_σd(x)=w_σd(x)·(w_σd(x))^T;

Step 3-3: computation structure tensor field:

$s_{{\mathrm{\σd\σ}}_i}\left(x\right)=G_{{\mathrm{\σ}}_i}\left(x\right)*w_{\mathrm{\σ}d}\left(x\right);$

Wherein,For another gaussian kernel;

Step 3-4: setForCharacteristic vector corresponding to eigenvalue of maximum, then corresponding for certain pixel x ridge metricCalculating formula is as follows:

$R_{{\mathrm{\σd\σ}}_i}\left(x\right)=\left|div\left(w_{{\mathrm{\σd\σ}}_i}\right(x\left)\right)\right|;$

Wherein, div is divergence;

Step 3-5: pixels all in area-of-interest in image are carried out the calculating of ridge tolerance.

It is preferred that scheme, implementing of described step 5 comprises the steps:

Step 5-1: line segment is added up: in binary image, all continuous line segments are added up, and each line segment is considered as a unit U_i;

Step 5-2: line segment parameter calculates: to each line segment unit U_i, calculate its length l_i, G-bar a_iWith line segment slope concordance δ_i;

A certain line segment unit U_iG-bar a_iComputing formula is as follows:

$a_i=\frac{1}{n-1}\underset{k=1:n}{\Σ}{a}_k;$

Wherein, a_kFor the slope of point-to-point transmission in line segment, the sub-slope of called after,u_k,v_kFor the coordinate of certain point x in line segment;

A certain line segment unit U_iLine segment slope concordance δ_iComputational methods are the mean square deviation adding up sub-slope:

δ_i=E (a_k)

Step 5-3: carry out line segment screening, it is thus achieved that remove the binary image Img_B containing only lane line line segment after interference line segment.

It is preferred that scheme, the rule of the screening of line segment described in described step 5-3 is:

1) length is removed less than 0.07H_RLine segment, i.e. short-term section;

2) G-bar line segment outside [π/8,3 π/8] and [5 π/8,7 π/8] scope is removed, namely bigger with dream car road line slope difference line segment;

3) line segment slope concordance δ is removed_i> 6.73 line segment, i.e. line segment in irregular shape.

It is preferred that scheme, implementing of described step 6 comprises the steps:

Step 6-1: traversing graph is as each pixel (x in Img_B, y), calculate ρ=xcos (θ)+ysin (θ): θ ∈ [0 °～180 °], obtain all through pixel (x, straight line group y) (ρ, θ) | θ ∈ [0 °～180 °] };

Wherein: (x, y) represents the position of pixel in image Img_B; ρ represents through pixel (x, air line distance zero y), the i.e. citing of image Img_B lower-left angle point; θ represents angle and θ ∈ [0 °～180 °];

Step 6-2: by pixels all in image Img_B (x, straight line group y) { (ρ, θ) | θ ∈ [0 °～180 °] } is mapped to H (ρ, θ) space, obtains ρ-θ parameter space accumulated image Img_H;

Step 6-3: search out two maximum M in the upper half images of ρ-θ parameter space accumulated image Img_H, lower half images respectively₁、M₂, then its corresponding ρ-θ parameter is to (ρ_l,θ_l)、(ρ_r,θ_r) the straight line model parameter that is in region of interest ROI-I under left and right lane line polar form;

Further conversion, obtains the left-lane line under pixel coordinate system, right lane line straight line model is:

$\left\{\begin{array}{c}xsin{\mathrm{\θ}}_l+ycos{\mathrm{\θ}}_l={\mathrm{\ρ}}_l\\ {\mathrm{xsin\θ}}_r+{\mathrm{ycos\θ}}_r={\mathrm{\ρ}}_r\end{array}\right.;$

Wherein, l represents left-lane line, and r represents right lane line.

Beneficial effects of the present invention:

The present invention takes full advantage of lane line feature of distribution in far and near visual field in image, by feature point extraction and the Hough transformation of ridge tolerance, effectively lane line edge feature is extracted and sets up straight line model.Plurality of committed step (ridge tolerance, Eigenvector screening strategy, Hough transformation) all have employed the algorithm with relatively strong adaptability and certain fault-tolerance, substantially increases stability and the robustness of the present invention.

Accompanying drawing explanation

Fig. 1 is the method flow diagram of the lane detection based on ridge tolerance that the present invention proposes.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the invention will be further described.

As it is shown in figure 1, the flow chart of the method for detecting lane lines based on ridge tolerance proposed for the present invention, specifically include following steps:

Step 1: collection vehicle road ahead original image Img.

In vehicle forward process, by being arranged on the camera acquisition vehicle front road conditions original image Img of vehicle interior or outside.

Step 2: determine region of interest I (ROI).

Inner parameter according to video camera and external parameter, obtaining in camera view the region below ground level vanishing line, within image right boundary is region of interest ROI, if the high width respectively H that ROI is in the picture_RAnd W_R。

Step 3: in region of interest ROI, the ridge metric values of image calculates. Concrete calculating comprises the steps:

Step 3-1: by gray level image g (x) in original ROI and 2-d gaussian filters device G_σdDo convolution algorithm:

L_σd(x)=G_σd(x)*g(x)(1)

In formula (1), G_σdBeing an anisotropic Gaussian core, its covariance matrix is ∑=diag (σ_dx,σ_dy) (Σ is exactly a symbol representing covariance matrix here, is not summation symbol), wherein, σ_dyIt is constant H_R, σ_dxFor variable quantity, its numerical value is the half of track live width corresponding to each row of image (calculating in pixels).

Step 3-2: calculate the image each pixel x gradient vector field along u row and v column direction, as follows:

$w_{\mathrm{\σ}d}\left(x\right)={\left({\∂}_u{L}_{\mathrm{\σ}d}\right(x),{\∂}_v{L}_{\mathrm{\σ}d}(x\left)\right)}^T---\left(2\right)$

Additionally, calculate new matrix:

s_σd(x)=w_σd(x)·(w_σd(x))^T(3)

Step 3-3: computation structure tensor field:

$s_{{\mathrm{\σd\σ}}_i}\left(x\right)=G_{{\mathrm{\σ}}_i}\left(x\right)*w_{\mathrm{\σ}d}\left(x\right)---\left(4\right)$

In formula (4),For another gaussian kernel.

Step 3-4: setForCharacteristic vector corresponding to eigenvalue of maximum, then corresponding for certain pixel x ridge metricCalculating formula is as follows:

$R_{{\mathrm{\σd\σ}}_i}\left(x\right)=\left|div\left(w_{{\mathrm{\σd\σ}}_i}\right(x\left)\right)\right|---\left(5\right)$

In formula (5), div is divergence.

Step 3-5: pixels all in region of interest ROI in image are carried out the calculating of ridge tolerance.

Step 4: calculate meansigma methods p and the variances sigma of the ridge metric of all pixels in this step, is considered as potential lane line characteristic point by all ridge metrics pixel more than p-3 σ and is set to 1, potential characteristic point non-in image is then set to 0. So far, obtain containing only the binary image of potential lane line characteristic point.

Step 5: the lane line characteristic point extracted through step 4 is screened, it is thus achieved that contain only the binary image Img_B of lane line line segment. Specifically include following steps:

Step 5-1: line segment is added up. In binary image, there is a large amount of line segment. In this step, all continuous line segments are added up, and each line segment is considered as a unit U_i。

Step 5-2: line segment parameter calculates. To each line segment unit U_i, calculate its length l_i(namely line segment is containing number n a little), G-bar a_iWith line segment slope concordance δ_i。

A certain line segment unit U_iG-bar a_iComputing formula is as follows:

$a_i=\frac{1}{n-1}\underset{k=1:n}{\Σ}{a}_k---\left(6\right)$

In formula (6), a_kFor the slope of point-to-point transmission in line segment, the sub-slope of called after,u_k,v_kFor the coordinate of certain point x in line segment.

A certain line segment unit U_iLine segment slope concordance δ_iComputational methods are the mean square deviation adding up sub-slope:

δ_i=E (a_k)(7)

Step 5-3: line segment screens. Screening rule has following three:

1) length is removed less than 0.07H_RLine segment, i.e. short-term section;

2) G-bar line segment outside [π/8,3 π/8] and [5 π/8,7 π/8] scope is removed, namely bigger with dream car road line slope difference line segment;

3) line segment slope concordance δ is removed_i> 6.73 line segment, i.e. line segment in irregular shape.

So far, it is thus achieved that remove the binary image Img_B containing only lane line line segment after interference line segment.

Step 6: because through street curvature is relatively big, therefore can be similar to and be considered as straight line model, for the binary image Img_B that step 5 obtains, carry out traditional straight line Hough transformation, and obtain straight line model parameter. Specifically include following steps:

Step 6-1: traversing graph is as each pixel (x in Img_B, y), calculate ρ=xcos (θ)+ysin (θ): θ ∈ [0 °～180 °], obtain all through pixel (x, straight line group y) (ρ, θ) | θ ∈ [0 °～180 °] }; Wherein: (x, y) represents the position of pixel in image Img_B; ρ represents through pixel (x, air line distance zero y), the i.e. citing of image Img_B lower-left angle point; θ represents angle and θ ∈ [0 °～180 °].

Step 6-2: by pixels all in image Img_B (x, straight line group y) { (ρ, θ) | θ ∈ [0 °～180 °] } is mapped to H (ρ, θ) space, obtains ρ-θ parameter space accumulated image Img_H;

Step 6-3: search out two maximum M in the upper half images of ρ-θ parameter space accumulated image Img_H, lower half images respectively₁、M₂. Then its corresponding ρ-θ parameter is to (ρ_l,θ_l)、(ρ_r,θ_r) the straight line model parameter that is in region of interest ROI-I under left and right lane line polar form.

Further, the left and right lane line straight line model obtained under pixel coordinate system is:

$\left\{\begin{array}{c}{\mathrm{xsin\θ}}_l+{\mathrm{ycos\θ}}_l={\mathrm{\ρ}}_l\\ {\mathrm{xsin\θ}}_r+{\mathrm{ycos\θ}}_r={\mathrm{\ρ}}_r\end{array}\right.---\left(8\right)$

Wherein, l represents left-lane line, and r represents right lane line.

The specific embodiment of the invention

The method adopting the present invention, writes lane detection software first by C Plus Plus; Then video camera is arranged on automobile (inside and outside all can). Then the internal and external parameter of video camera is demarcated, and in vehicle travel process, forward image is acquired; Subsequently, the original image (720x480) photographed is input in lane detection software processes; Experiment gathers the video of about 90 hours under various working altogether, and when fair weather, the lane detection algorithm success rate of the present invention is about 98%; Also the success rate of about 96% is had under the adverse weather such as night, sleet. Average every frame process time is about 50ms, and running environment is Win7, CPU is four core 2.4GHz.

In sum, the present invention makes full use of far and near visual field lane line feature, adopts a kind of lane detection strategy based on ridge tolerance, it is achieved thereby that the method that lane line is provided accurately from the input source image provided.

The above is only used for describing technical scheme and specific embodiment; the protection domain being not intended to limit the present invention; under the premise without prejudice to flesh and blood of the present invention and spirit, changed and retouching etc. falls within protection scope of the present invention.

Claims (7)

1. the method for detecting lane lines based on ridge tolerance, it is characterised in that comprise the steps:

Step 1: collection vehicle road ahead original image Img;

Step 2: determine region of interest I;

Step 3: calculate the ridge metric values of image in region of interest;

Step 4: calculate meansigma methods p and the variances sigma of the ridge metric of all pixels, all ridge metrics pixel more than p-3 σ is considered as potential lane line characteristic point and is set to 1, potential characteristic point non-in image is then set to 0, obtains containing only the binary image of potential lane line characteristic point;

Step 5: the lane line characteristic point extracted through step 4 is screened, it is thus achieved that contain only the binary image Img_B of lane line line segment;

Step 6: the binary image Img_B obtained for step 5, carries out traditional straight line Hough transformation, and obtains straight line model parameter.

2. a kind of method for detecting lane lines based on ridge tolerance according to claim 1, it is characterised in that implementing of described step 1 includes: by being arranged on the camera acquisition vehicle front road conditions original image Img of vehicle interior or outside.

3. a kind of method for detecting lane lines based on ridge tolerance according to claim 2, it is characterized in that, described in described step 2, the defining method of region of interest I is: according to the inner parameter of video camera and external parameter, in acquisition camera view, the region below ground level vanishing line, within image right boundary is region of interest.

4. a kind of method for detecting lane lines based on ridge tolerance according to claim 1, it is characterised in that implementing of described step 3 includes:

Step 3-1: by gray level image g (x) in original ROI and 2-d gaussian filters device G_σdDo convolution algorithm:

L_σd(x)=G_σd(x) * g (x);

Wherein, G_σdBeing an anisotropic Gaussian core, its covariance matrix is ∑=diag (σ_dx,σ_dy), wherein, σ_dyIt is constant H_R, σ_dxFor variable quantity, its numerical value is the half of track live width corresponding to each row of image;

Step 3-2: calculate the image each pixel x gradient vector field along u row and v column direction:

$w_{\mathrm{\σ}d}\left(x\right)={\left({\∂}_u{L}_{\mathrm{\σ}d}\right(x),{\∂}_v{L}_{\mathrm{\σ}d}(x\left)\right)}^T;$

Additionally, calculate new matrix:

s_σd(x)=w_σd(x)·(w_σd(x))^T;

Step 3-3: computation structure tensor field:

$s_{{\mathrm{\σd\σ}}_i}\left(x\right)=G_{{\mathrm{\σ}}_i}\left(x\right)*w_{\mathrm{\σ}d}\left(x\right);$

Wherein,For another gaussian kernel;

Step 3-4: setForCharacteristic vector corresponding to eigenvalue of maximum, then corresponding for certain pixel x ridge metricCalculating formula is as follows:

$R_{{\mathrm{\σd\σ}}_i}\left(x\right)=\left|div\left(w_{{\mathrm{\σd\σ}}_i}\right(x\left)\right)\right|;$

Wherein, div is divergence;

Step 3-5: pixels all in area-of-interest in image are carried out the calculating of ridge tolerance.

5. a kind of method for detecting lane lines based on ridge tolerance according to claim 1, it is characterised in that implementing of described step 5 comprises the steps:

Step 5-1: line segment is added up: in binary image, all continuous line segments are added up, and each line segment is considered as a unit U_i;

Step 5-2: line segment parameter calculates: to each line segment unit U_i, calculate its length l_i, G-bar a_iWith line segment slope concordance δ_i;

A certain line segment unit U_iG-bar a_iComputing formula is as follows:

$a_i=\frac{1}{n-1}\underset{k=1:n}{\Σ}{a}_k;$

Wherein, a_kFor the slope of point-to-point transmission in line segment, the sub-slope of called after,u_k,v_kFor the coordinate of certain point x in line segment;

A certain line segment unit U_iLine segment slope concordance δ_iComputational methods are the mean square deviation adding up sub-slope:

δ_i=E (a_k)

Step 5-3: carry out line segment screening, it is thus achieved that remove the binary image Img_B containing only lane line line segment after interference line segment.

6. a kind of method for detecting lane lines based on ridge tolerance according to claim 5, it is characterised in that the rule of the screening of line segment described in described step 5-3 is:

1) length is removed less than 0.07H_RLine segment, i.e. short-term section;

2) G-bar line segment outside [π/8,3 π/8] and [5 π/8,7 π/8] scope is removed, namely bigger with dream car road line slope difference line segment;

3) line segment slope concordance δ is removed_i> 6.73 line segment, i.e. line segment in irregular shape.

7. a kind of method for detecting lane lines based on ridge tolerance according to claim 1, it is characterised in that implementing of described step 6 comprises the steps:

Step 6-1: traversing graph is as each pixel (x in Img_B, y), calculate ρ=xcos (θ)+ysin (θ): θ ∈ [0 °～180 °], obtain all through pixel (x, straight line group y) (ρ, θ) | θ ∈ [0 °～180 °] };

Wherein: (x, y) represents the position of pixel in image Img_B; ρ represents through pixel (x, air line distance zero y), the i.e. citing of image Img_B lower-left angle point; θ represents angle and θ ∈ [0 °～180 °];

Step 6-2: by pixels all in image Img_B (x, straight line group y) { (ρ, θ) | θ ∈ [0 °～180 °] } is mapped to H (ρ, θ) space, obtains ρ-θ parameter space accumulated image Img_H;

Step 6-3: search out two maximum M in the upper half images of ρ-θ parameter space accumulated image Img_H, lower half images respectively₁、M₂, then its corresponding ρ-θ parameter is to (ρ_l,θ_l)、(ρ_r,θ_r) the straight line model parameter that is in region of interest ROI-I under left and right lane line polar form;

Further conversion, obtains the left-lane line under pixel coordinate system, right lane line straight line model is:

$\left\{\begin{array}{c}xsin{\mathrm{\θ}}_l+ycos{\mathrm{\θ}}_l={\mathrm{\ρ}}_l\\ {\mathrm{xsin\θ}}_r+{\mathrm{ycos\θ}}_r={\mathrm{\ρ}}_r\end{array}\right.;$

Wherein, l represents left-lane line, and r represents right lane line.