CN101294801A - Vehicle distance measuring method based on binocular vision - Google Patents

Abstract

The invention provides a method of measuring the vehicle distance based on binocular vision. Required parameters are obtained by adopting a self-adapting front vehicle distance measurement system of stereoscopic vision, which comprises a camera unit used for shooting traffic information ahead. The camera unit comprises a first camera C1 and a second camera C2; the coordinate system of either camera in the camera unit is taken as the world coordinate system of the whole measurement process, the photo-center of the camera is taken as the origin of the world coordinate system and is noted as 0, the positional relation between the two cameras in the camera unit is calibrated, and the size of the driveway area and the traffic lane line information of the vehicle are obtained and noted as S1, thus improving the measurement accuracy and ensuring the driving safety.

Description

Vehicle distance measurement method based on binocular vision

Technical field

The present invention relates to the vehicle distance measurement method between a kind of front vehicles and the front vehicle.

Background technology

As the important component part of intelligent transportation system (ITS), the security system of vehicle always is important field of research, and the front vehicles distance measurement technique then is one of technology the most key in the Vehicle security system.Front vehicles distance survey technology is not only forward position and the focus of studying in the current Vehicle security system field, also is the hi-tech that computer vision technique and electronic technology combine.The precision of front vehicles range observation has determined the Vehicle security system performance to a great extent.

In present vehicle range measurement system, distance-finding method based on vision is the most promising a kind of method by generally acknowledging, the method of the front vehicles distance survey that all is based on the monocular vision principle that most of vision range measurement system adopts in the present prior art, this method is placed in camera on the vehicle by certain fixed form, in advance height and luffing angle that the back camera is installed are measured, and calculated the measurement that realizes the front vehicles distance by a series of formula.Low and the spacing computation complexity reduction of simplicity of design, cost of system also exist some problems though adopt these class methods to make.At first, because the measurement result of this method depends on the camera setting height(from bottom) and the luffing angle of prior measurement strongly, therefore along with vehicle ', camera position can be because the shake effect, original relatively inevitably installation site changes, thereby changed the camera parameter, luffing angle especially, this can make measurement result generation deviation.Secondly it is comparatively accurate for the measurement that is near the measured point the ground that this monocular is measured spacing method, if the measured point on the front vehicles has certain height apart from ground, then this moment, measurement result existed than mistake, when especially two spacings are nearer, error is more obvious, and this will have a strong impact on the security of vehicle.

On the other hand, binocular stereo vision is widely used as a kind of advanced person's vision measuring method, and stereoscopic vision can directly be recovered the three-dimensional coordinate of measured point, thereby can accurately obtain space length information.Therefore adopt stereoscopic vision to measure the front vehicles distance and have better accuracy and reliability.

Summary of the invention

The object of the present invention is to provide a kind of vehicle distance measurement method based on binocular vision, solved the big problem of measuring error in the prior art, the present invention can be automatically according to vehicle ' s contour information and road information, accurately measure the distance between this car and the front vehicles, and compensation improves the accuracy and the stability of systematic survey because the camera parameters that the vehicle shake causes changes automatically.

For reaching above-mentioned technical purpose, the technical solution used in the present invention is:

Adopt the self-adaptation front vehicles distance survey system of stereoscopic vision to obtain the required parameter of measurement, the self-adaptation front vehicles distance survey system of described stereoscopic vision comprises the camera unit that is used to take information on the road ahead, and described camera unit comprises the first camera C1 and the second camera C2; The described first camera C1 and the second camera C2 are installed in the vehicle windscreen back, choose the world coordinate system of the coordinate system of arbitrary camera in the described camera unit as whole measuring process, the photocentre of this camera is as the initial point of described world coordinate system, be designated as O, demarcate in the described camera unit position relation between the camera, obtain this car place track area size and lane line information, the area of the rectangular area that described car place track area size obtaining and lane line information are constituted is designated as S1;

Step 1: survey and the segment rectangle zone of the contour of the vehicle profile of checking front vehicles, the area in the segment rectangle zone of the contour of the vehicle profile of the front vehicles that obtains is designated as S2;

Step 2: the segment rectangle region area S2 of the contour of the vehicle profile of the rectangular area area S1 that described car place track area size will obtaining and lane line information constitute and the described front vehicles of acquisition compares, if S1/S2＜=threshold value T, execution in step 3, if S1/S2＞threshold value T, execution in step 4, threshold value T=2;

Step 3: calculate the centre of form of rectangle in segment rectangle zone of the contour of the vehicle profile of described front vehicles, be designated as P1, execution in step 5;

Step 4: extract the point at arbitrary angle of the rectangular area of described front vehicles car plate profile, be designated as P2;

Step 5: the three-dimensional coordinate P that recovers described P1 or P2;

Step 6: calculate the distance between the OP, the distance before and after obtaining between two cars.

In a preferred embodiment of the invention, in the process of calculating front-and-rear vehicle distance, can dynamically adjust the outer parameter of the camera in the described camera unit simultaneously, method is as follows:

Step 11: calculate the three-dimensional coordinate of putting on two lane lines in track, described front vehicles place, be designated as Pi;

Step 12: according to the three-dimensional coordinate Pi match space plane V that puts on described two lane lines; The method of match space plane is as follows: establish three-dimensional point P _iCoordinate is (X _i, Y _i, Z _i), plane equation is AX+BY+CZ+1=0, and then the purpose of match is obtained the plane equation parameter A exactly, and B and C are with three-dimensional point P _iAmong the coordinate substitution AX+BY+CZ+1=0, be constructed as follows system of equations:

$\left[\begin{array}{cccc}{x}_1& y_1& z_1& 1\\ x_2& y_2& z_2& 1\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ x_n& y_n& z_n& 1\end{array}\right]\left[\begin{array}{c}A\\ B\\ C\\ 1\end{array}\right]=0$

Find the solution system of equations as implied above, can obtain the plane equation of space plane V;

Step 13: the three-dimensional coordinate Pi that puts on described two lane lines to described space plane V projection, is obtained the planar coordinate Qi of three-dimensional coordinate Pi on space plane V;

Step 14: described planar coordinate Qi is carried out fitting a straight line, obtain two straight line vector a and b; The method of fitting a straight line is as follows: the coordinate of establishing two-dimensional points Qi is (x _i, y _i), total m point, the place straight line is y=a ₀+ a ₁X then utilizes the least square method formula can obtain the parameter a of straight-line equation ₀And a ₁:

$\left[\begin{array}{cc}m& \mathrm{\Σ}{x}_i\\ \mathrm{\Σ}{x}_i& \mathrm{\Σ}{x}_i^2\end{array}\right]\left[\begin{array}{c}{a}_0\\ a1\end{array}\right]=\left[\begin{array}{c}\mathrm{\Σ}{y}_i\\ \mathrm{\Σ}{x}_i{y}_i\end{array}\right];$

Step 15: judge whether described straight line vector a is parallel with b,, then demarcate in the described camera unit position relation between the camera again, obtain new camera calibration parameter if not parallel;

In a preferred embodiment of the invention, the segment rectangle region method step of the contour of the vehicle profile of the front vehicles of acquisition is as follows:

1) on the standard road surface, can think in the zone, track of above-mentioned acquisition, the grey scale change of road in image is mild between we and the front vehicles, but in road surface and vehicle intersection, owing to there is shade, therefore form the edge of gradation of image from bright to dark, think that this edge is exactly the lower limb of vehicle.In this car road ahead, according to the profile of horizon scan line front vehicles, obtain the image of the profile of described front vehicles, calculate every capable average gray RowAvr (r) in the described image, be shown below:

$\mathrm{RowAvr}\left(r\right)=\frac{1}{\mathrm{Right}\left(r\right)-\mathrm{Left}\left(r\right)}\underset{c=\mathrm{Left}\left(r\right)}{\overset{\mathrm{Right}\left(r\right)}{\mathrm{\Σ}}}I(r,c),$ In the following formula, zone r capable rightmost coordinate in track in Right (r) presentation video; Zone r capable Far Left coordinate in track in Left (r) presentation video; (r c) is (r, gray scale c) in the image to I; The capable average gray of r in RowAvr (r) the expression zone, track changes the lower limb of maximum r as front vehicles with RowAvr (r) and RowAvr (r-1), is designated as LowerPos;

2) in the zone, track, carry out the edge image that Suo Beier (SOBEL) edge extracting algorithm obtains front vehicles in the zone, track, [LowerPos-Tr at edge image, LowerPos] row in, judge leftmost edge point position and rightmost edge point position, and, be designated as LeftPos and RightPos respectively with its left hand edge and right hand edge of thinking front vehicles; Wherein Tr gets 2 ~ 6, and LeftPos and RightPos represent the Far Left and the rightmost of vehicle region respectively;

3) length breadth ratio of supposing the vehicle rectangle is Ta, then according to step 2) as can be known the broadside in the segment rectangle zone of contour of the vehicle profile be RightPos-LeftPos, the long limit that then obtains the rectangular area is (RightPos-LeftPos) * Ta, and the Ta value is generally at 0.5-1.0;

In a preferred embodiment of the invention, in the described step 3, comprise the steps:

Step 31: extract segment rectangle region S 2 interior all marginal point Pi of the contour of the vehicle profile of described front vehicles, and put into the edge sequence;

Step 32: the mean value of the coordinate of described edge sequence Pi as centre of form coordinate, is designated as P1, with P1 as measurement point.

In a preferred embodiment of the invention, in the described step 4, comprise the steps:

Step 41: in the rectangular area of described front vehicles car plate, utilize the Suo Beier boundary operator to extract edge image;

Step 42: the described edge image that extracts is carried out binary conversion treatment according to image segmentation algorithm;

Step 43: the described edge image that will extract carries out Gray Projection, obtains the right limit value d of higher limit a, the lower limit b of car plate, left limit value c;

Step 44: obtain representing the upper left of license plate area rectangle according to higher limit a, the lower limit b of the described car plate that obtains, left limit value c and right limit value d, the lower-left, the upper right and bottom right coordinate of four angle points in image be (c, a), (c, b), (d, a) with (d, b);

Step 45: the point of location on arbitrary angle in described license plate area, be designated as P2, with P2 as measurement point.

In a preferred embodiment of the invention, when step 1 is carried out, follow the tracks of front vehicles in conjunction with Kalman filtering algorithm.

In a preferred embodiment of the invention, described Camera extrinsic number comprises rotation matrix R and translation matrix T.

In a preferred embodiment of the invention, in the step 15, whether parallel employing is listed as civilian Burger-Ma Kuaerte algorithm (Levenberg-Marquardt) with vectorial b to vectorial a;

In a preferred embodiment of the invention, the described edge image that will extract described in the described step 43 carries out Gray Projection, comprise the described edge image that extracts is carried out horizontal Gray Projection and vertical Gray Projection simultaneously, and horizontal Gray Projection figure sought two maximal values from top to bottom, as the higher limit and the lower limit of described front truck car plate; Vertical Gray Projection figure is sought two maximal values from top to bottom, as the left limit value and the right limit value of described front truck vehicle license plate.

In a preferred embodiment of the invention, described projecting method is: the described edge image that will obtain is for being designated as I, I (x, y) the presentation video mid point (x, the gray-scale value of y) locating, horizontal Gray Projection promptly, fixedly the y coordinate carries out gray scale and adds up, formula is as follows:

$H_j=\underset{i=0}{\overset{\mathrm{width}}{\mathrm{\Σ}}}I(x_i,y_j),$ Wherein width presentation video width, fixedly y _jCoordinate;

In like manner, vertically Gray Projection promptly, fixedly the x coordinate carries out gray scale and adds up, formula is as follows:

$V_j=\underset{i=0}{\overset{\mathrm{height}}{\mathrm{\Σ}}}I(x_j,y_i),$ Wherein height presentation video height, fixedly x _jCoordinate; For license plate area, horizontal Gray Projection H _jMiddle two maximum value, the higher limit of corresponding car plate and lower limit, vertically the Gray Projection V of existing _jMiddle two maximum value, the left limit value of corresponding car plate and the right limit value of existing.

Description of drawings

The present invention will be further described below in conjunction with the drawings and specific embodiments.

Fig. 1 is the constructional device synoptic diagram that two cameras of the present invention are installed;

Fig. 2 is the process flow diagram of the vehicle distance measurement method based on binocular vision provided by the invention;

Fig. 3 is the process flow diagram that the outer parameter of the camera in the camera unit is carried out dynamic adjusting method provided by the invention;

Fig. 4 is the segment rectangle region method process flow diagram of contour of the vehicle profile of the front vehicles of acquisition provided by the invention;

Fig. 5 is the method flow diagram of the centre of form of rectangle in segment rectangle zone of the contour of the vehicle profile of the described front vehicles of calculating provided by the invention;

Fig. 6 is the process flow diagram of point methods at arbitrary angle of the rectangular area of the described front vehicles car plate of extraction provided by the invention profile.

Embodiment

As shown in Figure 1, two cameras are installed in the vehicle windscreen back, adjust the disposing way of two cameras, make two cameras that the public visual field be arranged, and the public visual field can comprise the most information on the road ahead.

The high precision that stereoscopic vision is measured depends on two prerequisites, the one, and the front truck measurement point is correct, and the 2nd, the camera calibration parameter is correct.

Because vehicle is as a kind of artificial rigid body, its afterbody image has following feature:

(1) vehicle edge is compared than the natural scene in the road, and it is obvious to have an edge, continuously, and the characteristics of direction near vertical or level;

(2) the contour of the vehicle profile is observed from behind, is roughly to meet rectangle;

(3) there are bigger difference in vehicle back texture and road texture;

(4) there is shade in the vehicle bottom, has tangible gray feature, is generally place the darkest in the road.

So can be according to the feature of vehicle afterbody image, determine the measurement point of front truck according to different situations, as shown in Figure 2, a kind of vehicle distance measurement method based on binocular vision, adopt the self-adaptation front vehicles distance survey system of stereoscopic vision to obtain the required parameter of measurement, the self-adaptation front vehicles distance survey system of described stereoscopic vision comprises the camera unit that is used to take information on the road ahead, and described camera unit comprises the first camera C1 and the second camera C2; The described first camera C1 and the second camera C2 are installed in the vehicle windscreen back, choose the world coordinate system of the coordinate system of arbitrary camera in the described camera unit as whole measuring process, the photocentre of this camera is as the initial point of described world coordinate system, be designated as 0, demarcate in the described camera unit position relation between the camera, (the present invention demarcates the high precision camera calibration algorithm based on plane template (" A Flexible New Technique for Camera Calibration " that the position relation between the camera utilizes Zhang Zhengyou to propose, PatternAnalysis and Machine Intelligence, IEEE Transactions on, 2000,11 (22): 1330-1334)), obtain this car place track area size and lane line information, (zone, track and lane line acquisition algorithm (Zhu Wennan that the present invention utilizes following this piece article to propose; Chen Qiang; Wang Hong " lane detection in some complex conditions ", Intelligent Robots and Systems, 2006IEEE/RSJ International Conference on, 2006,117-122), the area of the rectangular area that described car place track area size obtaining and lane line information are constituted is designated as S1;

Step 1: survey and the segment rectangle zone of the contour of the vehicle profile of checking front vehicles, the segment rectangle region area of the contour of the vehicle profile of the front vehicles that obtains is designated as S2;

The segment rectangle region method step of the contour of the vehicle profile of the front vehicles that obtains is as follows:

1) on the standard road surface, can think in the zone, track of above-mentioned acquisition, the grey scale change of road in image is mild between we and the front vehicles, but in road surface and vehicle intersection, owing to there is shade, therefore form the edge of gradation of image from bright to dark, think that this edge is exactly the lower limb of vehicle.In this car road ahead, according to the profile of horizon scan line front vehicles, obtain the image of the profile of described front vehicles, calculate every capable average gray RowAvr (r) in the described image, be shown below:

$\mathrm{RowAvr}\left(r\right)=\frac{1}{\mathrm{Right}\left(r\right)-\mathrm{Left}\left(r\right)}\underset{c=\mathrm{Left}\left(r\right)}{\overset{\mathrm{Right}\left(r\right)}{\mathrm{\Σ}}}I(r,c),$ In the following formula, zone r capable rightmost coordinate in track in Right (r) presentation video; Zone r capable Far Left coordinate in track in Left (r) presentation video; (r c) is (r, gray scale c) in the image to I; The capable average gray of r in RowAvr (r) the expression zone, track changes the lower limb of maximum r as front vehicles with RowAvr (r) and RowAvr (r-1), is designated as LowerPos;

2) in the zone, track, carry out the edge image that Suo Beier (SOBEL) edge extracting algorithm obtains front vehicles in the zone, track, [LowerPos-Tr at edge image, LowerPos] row in, judge leftmost edge point position and rightmost edge point position, and, be designated as LeftPos and RightPos respectively with its left hand edge and right hand edge of thinking front vehicles; Wherein Tr gets 2 ~ 6, and LeftPos and RightPos represent the Far Left and the rightmost of vehicle region respectively;

3) length breadth ratio of supposing the vehicle rectangle is Ta, then according to step 2) as can be known the broadside in the segment rectangle zone of contour of the vehicle profile be RightPos-LeftPos, the long limit that then obtains the rectangular area is (RightPos-LeftPos) * Ta, and the Ta value is generally at 0.5-1.0; Greater than 1 situation, the vehicle rectangular area that obtain this moment just truly comprises the part of the rectangle of vehicle for length breadth ratios such as lorries, does not still influence follow-up result.

Step 2: the segment rectangle region area S2 of the contour of the vehicle profile of the rectangular area area S1 that described car place track area size will obtaining and lane line information constitute and the described front vehicles of acquisition compares, if S1/S2＜=threshold value T, execution in step 3, if S1/S2＞threshold value T, execution in step 4, threshold value T=2;

Step 3: calculate the centre of form of rectangle in segment rectangle zone of the contour of the vehicle profile of described front vehicles, be designated as P1, execution in step 5;

Calculate the centre of form of rectangle in segment rectangle zone of the contour of the vehicle profile of described front vehicles, method is as follows:

Step 31: extract segment rectangle region S 2 interior all marginal point Pi of the contour of the vehicle profile of described front vehicles, and put into the edge sequence;

Step 32: the mean value of the coordinate of described edge sequence Pi as centre of form coordinate, is designated as P1, with P1 as measurement point.

Step 4: extract the point at arbitrary angle of the rectangular area of described front vehicles car plate profile, be designated as P2;

The method of point at arbitrary angle of extracting the rectangular area of described front vehicles car plate profile is:

Step 41: in the rectangular area of described front vehicles car plate, utilize the Suo Beier boundary operator to extract edge image;

Step 42: the described edge image that extracts is carried out binary conversion treatment according to image segmentation algorithm; (image segmentation algorithm is referring to Kong Ming, Sun Xiping, Wang Yongji, " a kind of improved thresholding method based on inter-class variance ", Central China University of Science and Technology's journal, 2007,7];

Step 43: the described edge image that will extract carries out Gray Projection, obtains the right limit value d of higher limit a, the lower limit b of car plate, left limit value c;

Step 44: obtain representing the upper left of license plate area rectangle according to higher limit a, the lower limit b of the described car plate that obtains, left limit value c and right limit value d, the lower-left, the upper right and bottom right coordinate of four angle points in image be (c, a), (c, b), (d, a) with (d, b);

Step 45: the point of location on arbitrary angle is designated as P2 in described license plate area, with P2 as measurement point.Because license plate area is thought the rectangular area, any angle point (intersection point) of therefore getting the rectangular area can.But must be noted that corner location must be unified for the image of taking for left and right sides camera,, then also must select upper left angle point at right figure license plate area if promptly select upper left angle point at left figure license plate area.

Step 5: the three-dimensional coordinate P that recovers described P1 or P2;

Step 6: calculate the distance between the OP, the distance before and after obtaining between two cars.

In the process of calculating front-and-rear vehicle distance, can dynamically adjust the outer parameter of the camera in the described camera unit simultaneously, method is as follows:

Step 11: calculate the three-dimensional coordinate of putting on two lane lines in track, described front vehicles place, be designated as Pi; According to (Zhu Wennan; Chen Qiang; Wang Hong " lane detection in some complexconditions ", Intelligent Robots and Systems, 2006 IEEE/RSJ InternationalConference on, 2006,117-122) algorithm obtains the front vehicles place and gets the near linear equation that the track gets two lane lines, be labeled as L1 and L2 respectively, (Stereo Matching Algorithm and three-dimensional reconstruction algorithm see Ma Songde for details to utilize the Stereo Matching Algorithm of straight line and the three-dimensional reconstruction algorithm of point, Zhang Zhengyou. computer vision---the theory of computation and algorithm basis [C]. Beijing: Science Press, 1997), obtain the three-dimensional coordinate that two lane line L1 and L2 go up point, the unified Pi that is designated as; For the binocular camera of having demarcated, if obtained the corresponding point in the image of the left and right sides, can utilize the three-dimensional reconstruction algorithm to recover the three-dimensional coordinate of corresponding point, specific algorithm is referring to Ma Songde, Zhang Zhengyou. computer vision---the theory of computation and algorithm basis [C]. Beijing: Science Press, 1997

Step 12: according to the three-dimensional coordinate Pi match space plane V that puts on described two lane lines; The method of match space plane is as follows: establish three-dimensional point P _iCoordinate is (X _i, Y _i, Z _i), plane equation is AX+BY+CZ+1=0, and then the purpose of match is obtained the plane equation parameter A exactly, and B and C are with three-dimensional point P _iAmong the coordinate substitution AX+BY+CZ+1=0, be constructed as follows system of equations:

$\left[\begin{array}{cccc}{x}_1& y_1& z_1& 1\\ x_2& y_2& z_2& 1\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ x_n& y_n& z_n& 1\end{array}\right]\left[\begin{array}{c}A\\ B\\ C\\ 1\end{array}\right]=0$

Adopt singular value decomposition method to find the solution system of equations as implied above, can obtain the plane equation of space plane V;

Step 13: the three-dimensional coordinate Pi that puts on described two lane lines to described space plane V projection, is obtained the planar coordinate Qi of three-dimensional coordinate Pi on space plane V;

Step 14: described planar coordinate Qi is carried out fitting a straight line, obtain two straight line vector a and b; The method of fitting a straight line is as follows: the coordinate of establishing two-dimensional points Qi is (x _i, y _i), total m point, the place straight line is y=a ₀+ a ₁X then utilizes the least square method formula can obtain the parameter a of straight-line equation ₀And a ₁:

$\left[\begin{array}{cc}m& \mathrm{\Σ}{x}_i\\ \mathrm{\Σ}{x}_i& \mathrm{\Σ}{x}_i^2\end{array}\right]\left[\begin{array}{c}{a}_0\\ a1\end{array}\right]=\left[\begin{array}{c}\mathrm{\Σ}{y}_i\\ \mathrm{\Σ}{x}_i{y}_i\end{array}\right];$

Step 15: judge whether described straight line vector a is parallel with b,, then demarcate in the described camera unit position relation between the camera again, obtain new camera calibration parameter if not parallel; If the outer parameter of not parallel expression camera this moment changes, again take the target image this moment, the high precision camera calibration algorithm based on plane template (" A Flexible New Technique for CameraCalibration " that utilizes Zhang Zhengyou to propose, Pattern Analysis and Machine Intelligence, IEEE Transactions on, 2000,11 (22): 1330-1334) again camera is demarcated.

When step 1 is carried out, follow the tracks of front vehicles in conjunction with Kalman filtering algorithm.Kalman filtering algorithm is existing method, and list of references is: Zhao Li, Chen Quanlin, " based on the realization of the vehicle detection and the tracker of Kalman wave filter ", electronic measurement technique, 2007,30 (2): 165-168.

The Camera extrinsic number comprises rotation matrix R and translation matrix T.

Whether parallel employing is listed as civilian Burger-Ma Kuaerte algorithm (Levenberg-Marquardt) with vectorial b to vectorial a;

The described edge image that will extract described in the step 43 carries out Gray Projection, comprise the described edge image that extracts is carried out horizontal Gray Projection and vertical Gray Projection simultaneously, and horizontal Gray Projection figure sought two maximal values from top to bottom, as the higher limit and the lower limit of described front truck car plate; Vertical Gray Projection figure is sought two maximal values from top to bottom, as the left limit value and the right limit value of described front truck vehicle license plate.

Described projecting method is: the described edge image that will obtain is for being designated as I, I (x, y) the presentation video mid point (x, the gray-scale value of y) locating, horizontal Gray Projection promptly, fixedly the y coordinate carries out gray scale and adds up, formula is as follows:

$H_j=\underset{i=0}{\overset{\mathrm{width}}{\mathrm{\Σ}}}I(x_i,y_j),$ Wherein width presentation video width, fixedly y _jCoordinate;

In like manner, vertically Gray Projection promptly, fixedly the x coordinate carries out gray scale and adds up, formula is as follows:

$V_j=\underset{i=0}{\overset{\mathrm{height}}{\mathrm{\Σ}}}I(x_j,y_i),$ Wherein height presentation video height, fixedly x _jCoordinate; For license plate area, horizontal Gray Projection H _jMiddle two maximum value, the higher limit of corresponding car plate and lower limit, vertically the Gray Projection V of existing _jMiddle two maximum value, the left limit value of corresponding car plate and the right limit value of existing.

Claims (10)

1, a kind of vehicle distance measurement method based on binocular vision, it is characterized in that: adopt the self-adaptation front vehicles distance survey system of stereoscopic vision to obtain the required parameter of measurement, the self-adaptation front vehicles distance survey system of described stereoscopic vision comprises the camera unit that is used to take information on the road ahead, and described camera unit comprises the first camera C1 and the second camera C2; The described first camera C1 and the second camera C2 are installed in the vehicle windscreen back, choose the world coordinate system of the coordinate system of arbitrary camera in the described camera unit as whole measuring process, the photocentre of this camera is as the initial point of described world coordinate system, be designated as 0, demarcate in the described camera unit position relation between the camera, obtain this car place track area size and lane line information, the area of the rectangular area that described car place track area size obtaining and lane line information are constituted is designated as S1;

Step 1: survey and the segment rectangle zone of the contour of the vehicle profile of checking front vehicles, the area in the segment rectangle zone of the contour of the vehicle profile of the front vehicles that obtains is designated as S2;

Step 2: the segment rectangle region area S2 of the contour of the vehicle profile of the area S1 of the rectangular area that described car place track area size will obtaining and lane line information constitute and the described front vehicles of acquisition compares, if S1/S2＜=threshold value T, execution in step 3, if S1/S2＞threshold value T, execution in step 4, threshold value T=2;

Step 3: calculate the centre of form of rectangle in segment rectangle zone of the contour of the vehicle profile of described front vehicles, be designated as P1, execution in step 5;

Step 4: extract the point at arbitrary angle of the rectangular area of described front vehicles car plate profile, be designated as P2;

Step 5: the three-dimensional coordinate P that recovers described P1 or P2;

Step 6: calculate the distance between the OP, the distance before and after obtaining between two cars.

2, the vehicle distance measurement method based on binocular vision as claimed in claim 1 is characterized in that, in the process of calculating front-and-rear vehicle distance, can dynamically adjust the outer parameter of the camera in the described camera unit simultaneously, and method is as follows:

Step 11: calculate the three-dimensional coordinate of putting on two lane lines in track, described front vehicles place, be designated as Pi;

Step 12: according to the three-dimensional coordinate Pi match space plane V that puts on described two lane lines; The method of match space plane is as follows: establish three-dimensional point P _iCoordinate is (X _i, Y _i, Z _i), plane equation is AX+BY+CZ+1=0, and then the purpose of match is obtained the plane equation parameter A exactly, and B and C are with three-dimensional point P _iAmong the coordinate substitution AX+BY+CZ+1=0, be constructed as follows system of equations:

$\left[\begin{array}{cccc}{x}_1& y_1& z_1& 1\\ x_2& y_2& z_2& 1\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ x_n& y_n& z_n& 1\end{array}\right]\left[\begin{array}{c}A\\ B\\ C\\ 1\end{array}\right]=0$

Find the solution system of equations as implied above, can obtain the plane equation of space plane V;

Step 13: the three-dimensional coordinate Pi that puts on described two lane lines to described space plane V projection, is obtained the planar coordinate Qi of three-dimensional coordinate Pi on space plane V;

Step 14: described planar coordinate Qi is carried out fitting a straight line, obtain two straight line vector a and b; The method of fitting a straight line is as follows: the coordinate of establishing two-dimensional points Qi is (x _i, y _i), total m point, the place straight line is y=a ₀+ a ₁X then utilizes the least square method formula can obtain the parameter a of straight-line equation ₀And a ₁:

$\left[\begin{array}{cc}m& {\mathrm{\Σx}}_i\\ {\mathrm{\Σx}}_i& {\mathrm{\Σx}}_i^2\end{array}\right]\left[\begin{array}{c}{a}_0\\ a1\end{array}\right]=\left[\begin{array}{c}{\mathrm{\Σy}}_i\\ {\mathrm{\Σx}}_i{y}_i\end{array}\right];$

Step 15: if described straight line vector a and b are not parallel, then demarcate in the described camera unit position relation between the camera again, obtain new camera calibration parameter.

3, the vehicle distance measurement method based on binocular vision as claimed in claim 1 is characterized in that, in the described step 1, the segment rectangle region method step of the contour of the vehicle profile of the front vehicles of acquisition is as follows:

1) in this car road ahead, according to the profile of horizon scan line front vehicles, obtain the image of the profile of described front vehicles, calculate every capable average gray RowAvr (r) in the described image, be shown below:

$\mathrm{RowAvr}\left(r\right)=\frac{1}{\mathrm{Right}\left(r\right)-\mathrm{Left}\left(r\right)}\underset{c=\mathrm{Left}\left(r\right)}{\overset{\mathrm{Right}\left(r\right)}{\mathrm{\Σ}}}I(r,c),$ In the following formula, zone r capable rightmost coordinate in track in Right (r) presentation video; Zone r capable Far Left coordinate in track in Left (r) presentation video; (r c) is (r, gray scale c) in the image to I; The capable average gray of r in RowAvr (r) the expression zone, track changes the lower limb of maximum r as front vehicles with RowAvr (r) and RowAvr (r-1), is designated as LowerPos;

2) in the zone, track, carry out the edge image that Suo Beier (SOBEL) edge extracting algorithm obtains front vehicles in the zone, track, [LowerPos-Tr at edge image, LowerPos] row in, judge leftmost edge point position and rightmost edge point position, and, be designated as LeftPos and RightPos respectively with its left hand edge and right hand edge of thinking front vehicles; Wherein Tr gets 2 ~ 6, and LeftPos and RightPos represent the Far Left and the rightmost of vehicle region respectively;

3) length breadth ratio of supposing the vehicle rectangle is Ta, then according to step 2) as can be known the broadside in the segment rectangle zone of contour of the vehicle profile be RightPos-LeftPos, the long limit that then obtains the rectangular area is (RightPos-LeftPos) * Ta, and the Ta value is generally at 0.5-1.0.

4, the vehicle distance measurement method based on binocular vision as claimed in claim 1 is characterized in that, in the described step 3, comprises the steps:

Step 31: extract segment rectangle region S 2 interior all marginal point Pi of the contour of the vehicle profile of described front vehicles, and put into the edge sequence;

Step 32: the mean value of the coordinate of described edge sequence Pi as centre of form coordinate, is designated as P1, with P1 as measurement point.

5, the vehicle distance measurement method based on binocular vision as claimed in claim 1 is characterized in that, in the described step 4, comprises the steps:

Step 41: in the rectangular area of described front vehicles car plate, utilize the Suo Beier boundary operator to extract edge image;

Step 42: the described edge image that extracts is carried out binary conversion treatment according to image segmentation algorithm;

Step 43: the described edge image that will extract carries out Gray Projection, obtains the right limit value d of higher limit a, the lower limit b of car plate, left limit value c;

Step 44: according to higher limit a, the lower limit b of the described car plate that obtains, left limit value c and right limit value d, obtain representing the upper left of license plate area rectangle, the lower-left, the upper right and bottom right coordinate of four angle points in image be (c, a), (c, b), (d, a) with (d, b);

Step 45: the point of location on arbitrary angle in described license plate area, be designated as P2, with P2 as measurement point.

6, the vehicle distance measurement method based on binocular vision as claimed in claim 1 is characterized in that, when step 1 is carried out, follows the tracks of front vehicles in conjunction with Kalman filtering algorithm.

7, the vehicle distance measurement method based on binocular vision as claimed in claim 2 is characterized in that, described Camera extrinsic number comprises rotation matrix R and translation matrix T.

8, the vehicle distance measurement method based on binocular vision as claimed in claim 2 is characterized in that, in the step 15, whether parallel employing is listed as civilian Burger-Ma Kuaerte algorithm (Levenberg-Marquardt) with vectorial b to vectorial a.

9, the vehicle distance measurement method based on binocular vision as claimed in claim 5, it is characterized in that, the described edge image that will extract described in the described step 43 carries out Gray Projection, comprise the described edge image that extracts is carried out horizontal Gray Projection and vertical Gray Projection simultaneously, and horizontal Gray Projection figure sought two maximal values from top to bottom, as the higher limit and the lower limit of described front truck car plate; Vertical Gray Projection figure is sought two maximal values from top to bottom, as the left limit value and the right limit value of described front truck vehicle license plate.

10, the vehicle distance measurement method based on binocular vision as claimed in claim 9, it is characterized in that, described projecting method is: the described edge image that will obtain is for being designated as I, I (x, y) (horizontal Gray Projection promptly for x, the gray-scale value of y) locating for the presentation video mid point, fixedly the y coordinate carries out gray scale and adds up, and formula is as follows:

$H_j=\underset{i=0}{\overset{\mathrm{width}}{\mathrm{\Σ}}}I(x_i,y_j),$ Wherein width presentation video width, fixedly y _jCoordinate;

In like manner, vertically Gray Projection promptly, fixedly the x coordinate carries out gray scale and adds up, formula is as follows:

$V_j=\underset{i=0}{\overset{\mathrm{height}}{\mathrm{\Σ}}}I(x_j,y_i),$ Wherein height presentation video height, fixedly x _jCoordinate; For license plate area, horizontal Gray Projection H _jMiddle two maximum value, the higher limit of corresponding car plate and lower limit, vertically the Gray Projection V of existing _jMiddle two maximum value, the left limit value of corresponding car plate and the right limit value of existing.

Images