CN102661733B - Front vehicle ranging method based on monocular vision - Google Patents

Abstract

The invention belongs to the technical field of intelligent transportation and relates to a front vehicle ranging method based on monocular vision. The method includes: adjusting the posture of a vehicular camera, considering the pitch angle variation range due to bumping to adjust the pitch angle of the camera, and using the obtained actual measurement data to segmentally fit the functional relation between an ordinate and an actual longitudinal distance at different pitch angles and the functional relation between the ordinate and an actual transverse distance represented by per horizontal pixel at different pitch angles; and when a vehicle runs, calibrating real-time pitch angle of the camera according to left and right lane lines detected in real time and a three-line calibration method, and then using the linear interpolation method to calculate to obtain the distance between the vehicle and a target vehicle. The front vehicle ranging method based on monocular vision has a real-time dynamic compensation effect on ranging error caused by vehicle bumping and can be used for accurately calculating the distance between the vehicle and the front vehicle.

Description

A kind of front vehicles distance-finding method based on monocular vision

Technical field

The present invention relates to a kind of method that can accurately calculate the distance of this car and front vehicles under high speed or expressway condition, the method has real-time dynamic compensation effect to the vehicle range error causing of jolting, belong to image and process and field of machine vision, can be applicable to the intelligent vehicle safety DAS (Driver Assistant System) in intelligent transportation field.

Background technology

Along with the development of world economy, Global Auto quantity is increased sharply, and the incidence of traffic accident also significantly increases thereupon, and people's lives and properties and the national economy of traffic hazard to every country has all caused huge loss.The main cause that highway, through street accident rate rise is mileage in highway open to traffic rapid growth, and vehicle flowrate increases causes spacing excessively near, and driver tired driving easily causes rear-end collision.Therefore,, for reducing the generation of this type of traffic hazard, as one of gordian technique of intelligent automobile safety assisting system, front vehicles detects and distance measurement technique has caused the concern of many national automobile industries, research institutions and government.

Comply with the difference of adopted distance measuring sensor, vehicle ranging technology mainly can be divided into ultrasonic ranging, laser ranging, infrared distance measuring and machine vision range finding at present.Ultrasonic ranging, laser ranging and infrared distance measuring, by target object, the principle of the reflections such as ultrasound wave, laser, infrared ray is realized to measurement, these three kinds of technological system equipment complex and expensive, easily be interfered, and effective detection range is less, be mainly applicable to the occasion that car speed is lower.Machine vision range finding is analyzed by the image to camera collection, and the position of positioned vehicle in image calculates actual range through range finding model, and its equipment is simple, has a extensive future.

Machine vision metrology is mainly divided into: monocular vision is measured, binocular vision is measured, structure light vision is measured.Binocular vision difficult point is the coupling of unique point, has affected precision and the efficiency measured, and the emphasis of its theoretical research concentrates on matching of feature; Structured light is due to the restriction of light source, and the occasion of application is more fixing; And single camera vision system simple structure, with low cost, computing machine only need to be processed single image, do not need to carry out complicated images match, range measurement system with respect to other based on vision, has reduced system works amount at one time, has saved the time of a large amount of computer processing datas, the real-time of vision system improves greatly, can meet better actual needs.

Monocular vision range finding is mainly divided into the measuring method based on known motion and known object according to the principle of measuring.Measuring method based on known motion refers to utilizes the mobile message of video camera and picture that video camera obtains to record depth distance, the shortcoming of this measurement is the coupling that will carry out to a width or a few width picture unique point, matching error has obvious impact to measurement result, processing time while is long, for multiple image, must need more computing time.And measuring method based on known object refers to that the Target Photo that utilizes video camera to obtain under the condition of known object information obtains depth information.These class methods are more applicable for Navigation and localization, and its shortcoming is to utilize single unique point to measure, and easily, because of the inaccuracy of feature point extraction, produce error.

Monocular ranging technology has obtained at home to be paid attention to and deep research widely, wherein the colleges and universities such as Tsing-Hua University, Jilin University all successively start to carry out the work of this respect, as: the Guo Lei research of Tsing-Hua University utilizes road Parallel Constraint in the time calculating the video camera angle of pitch, adopt geometric relationship to push over method measuring distance (Guo Lei, Xu Youchun, Li Keqiang, Lian little Min. the real time distance method research [J] based on monocular vision. Journal of Image and Graphics .2006,11 (1): 74-81).The Wang Rongben of Jilin University professor's intelligent vehicle seminar, utilize the range finding of single-frame images range finding model realization monocular (Gu Baiyuan. the safe distance between vehicles early warning system research [D] based on monocular vision. the .2006:101-105 of Jilin University).

Projection model or geometric model were derived in current monocular range finding research mostly before this, then calculated thus distance.Therefore they exist the problem of some general character: on the one hand, substantially in all research, be all the optical model of pinhole imaging system or perspective projection, and wherein the geometric relationship of coordinate transform is all take desirable light path as prerequisite, and do not consider the light path errors such as the lens distortion existing in actual imaging, cannot meet the needs in practical application.On the other hand, mostly research method has all adopted numerous hypothesis with simplification problem, and as invariable in camera horizontal positioned, camera heights, vehicle is without jolting etc.And in actual use, jolting of occurring in Vehicle Driving Cycle process can change the angle of pitch that camera is taken, experiment showed, that vehicle range finding result is comparatively responsive to the angle of pitch, the change of the angle of pitch will cause distance accuracy to reduce to a certain extent.

Summary of the invention

The object of the invention is to overcome the above-mentioned deficiency of prior art, providing a kind of can carry out real-time dynamic compensation to the vehicle range error causing of jolting, and can accurately and fast calculate the method for this car and the distance of front vehicles.For this reason, the present invention adopts following technical scheme.

A front vehicles distance-finding method based on monocular vision, comprises the following steps:

(1) camera inner parameter is demarcated;

(2) camera is arranged on automobile, with the attitude parameter of three line calibration method static demarcating cameras, if camera attitude parameter does not meet selected mount scheme, after adjusting camera attitude, again use three line calibration method static demarcating camera parameters, until the attitude parameter of its camera meets the requirement of mount scheme;

(3) establishing the scope that causes the angle of pitch in vehicle traveling process because jolting is (θ _l, θ _h), adjusting the camera angle of pitch is θ _l, on the longitudinal direction in camera front, apart from vehicle body certain distance, monumented point is set, at this some place, label is set, record the ordinate of this monumented point correspondence at image, and laterally measure again certain distance at this monumented point place, record the horizontal ordinate of its correspondence image;

(4), in the process of camera front longitudinal different distance place repeating step step (3), utilize the data sectional obtaining to simulate the angle of pitch for θ _ltime, the funtcional relationship of ordinate and actual fore-and-aft distance is designated as VD _{θ L}(y), ordinate and every pixel represent that the funtcional relationship of the actual lateral separation of level is designated as HD _{θ L}(y), wherein y represents ordinate;

(5) adjusting respectively the camera angle of pitch is θ _hbe about θ _m=(θ _l+ θ _h)/2, the process of repeating step (3) and (4), then utilizes the data that obtain, and it is θ that piecewise fitting goes out the angle of pitch _hbe about θ _m=(θ _l+ θ _h)/2 o'clock, the funtcional relationship VD of ordinate and actual fore-and-aft distance _{θ H}and VD (y) _{θ M}(y), the every pixel of ordinate and level represents the funtcional relationship HD of actual lateral separation _{θ H}and HD (y) _{θ M}(y);

(6) establish the left and right lane line equation detecting in real time and be respectively y=k ₁x+b ₁and y=k ₂x+b ₂, the Bisector of angle equation of two straight lines is

wherein construct three straight lines in fact parallel to each other: left and right lane line and track center line, the funtcional relationship of utilizing step (4) and (5) to obtain, in conjunction with three line calibration methods, calibrates the real-time luffing angle θ of camera _real-time;

(7) according to θ _real-timewith θ _l, θ _mand θ _hrelation, utilize linear interpolation method, calculate the distance of vehicle target and this car.As preferred implementation, in step (2), the requirement that the attitude parameter of camera meets mount scheme is: make level inclination

with direction deflection angle Ψ all within 0.002 radian;

In step (7), utilize linear interpolation method, the range formula that calculates vehicle target and this spacing is:

$\mathrm{Dis}=\sqrt{{\mathrm{<figure-callout\; id="2"\; label="VD"\; filenames="HSA00000723522700041.png"\; state="\{\{state\}\}">VD</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="HD"\; filenames="HSA00000723522700041.png"\; state="\{\{state\}\}">HD</figure-callout>}}^2}$

$\mathrm{VD}=\left\{\begin{array}{cc}{\mathrm{VD}}_{\mathrm{\&theta;L}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_L\\ \frac{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{VD}}_{\mathrm{\&theta;L}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_L}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{VD}}_{\mathrm{\&theta;M}}\left(y\right),& {\mathrm{\&theta;}}_L<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_M\\ \frac{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{VD}}_{\mathrm{\&theta;M}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_M}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{VD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_M<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_H\\ {\mathrm{VD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}>{\mathrm{\&theta;}}_H\end{array}\right.$

$\mathrm{HD}=\left\{\begin{array}{cc}\left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;{\mathrm{HD}}_{\mathrm{\&theta;L}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_L\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;[\frac{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{HD}}_{\mathrm{\&theta;L}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_L}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{HD}}_{\mathrm{\&theta;M}}\left(y\right)],& {\mathrm{\&theta;}}_L<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_M\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;[\frac{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{HD}}_{\mathrm{\&theta;M}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_M}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{HD}}_{\mathrm{\&theta;H}}\left(y\right)],& {\mathrm{\&theta;}}_M<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_H\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;{\mathrm{HD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}>{\mathrm{\&theta;}}_H\end{array}\right.$

In formula, VD represents the fore-and-aft distance of vehicle target and this car, and HD represents the lateral separation of vehicle target and this car, and Dis represents the distance of vehicle target and this car, P _xthe horizontal resolution that represents collected by camera image, x and y represent respectively the transverse and longitudinal coordinate of target.

Substantive distinguishing features of the present invention is, adopt the monocular vision ranging technology based on known object, the light path errors such as the lens distortion existing in consideration actual imaging, first camera is carried out to calibration of camera, and verify the accuracy of three line calibration methods by geometric calculation and total station survey method, then utilize three line calibration method static demarcating camera attitude parameters, the angle of pitch different according to camera, adopt the method for piecewise fitting function, the funtcional relationship of matching ordinate and actual fore-and-aft distance, matching ordinate and every pixel represent the funtcional relationship of actual lateral separation.Finally, utilize the result of lane detection, again utilize the actual angle of pitch of the real-time computing camera of three line calibration methods, calculate the distance of vehicle target and this car by linear interpolation, vehicle is jolted to the angle of pitch that causes changes and the range error that causes has real-time dynamic compensation effect.The present invention has following technique effect:

1. method is simple, easy to implement.The present invention utilizes data regression modeling principle, and experiment measuring obtains necessary data, just can complete front target distance measurement, and does not need high-precision instrument and equipment, method simple practical.

2. accuracy is high.In data acquisition of the present invention, consider the light path errors such as the lens distortion that exists in actual imaging, adopted the method for piecewise fitting function, guaranteed range finding accuracy.And sampling frequency can adjust data acquisition according to actual needs time and adjust fitting precision.

3. pair vehicle jolts and causes that the angle of pitch changes the range error of bringing, and calculates target and this spacing by the real-time calibration angle of pitch and linear interpolation, makes the present invention have real-time dynamic compensation effect.

Accompanying drawing explanation

Fig. 1: distance-finding method process flow diagram of the present invention.

Fig. 2: the checking schematic diagram of three line calibration methods.

Fig. 3: camera mount scheme schematic diagram.Fig. 3-1 is vertical view, and Fig. 3-2 are side view, and Fig. 3-3 are front view.

Fig. 4: vertical and horizontal range finding fitting function figure.Fig. 4-1 is three kinds of fore-and-aft distances under the different angles of pitch and the fitting function figure of ordinate, and Fig. 4-2 are the fitting function figure that the horizontal every pixel under three kinds of different angles of pitch represents actual range and ordinate.

Fig. 5: angle of pitch real-time calibration method schematic diagram.Fig. 5-1 is the distribution plan of the interior left and right lane line of image and track center line, and Fig. 5-2 are the distribution plan of actual road surface left and right lane line and track center line.

Fig. 6: the result of finding range under different distance figure.Fig. 6-1 is the range finding figure of distance between 20 to 30 meters, Fig. 6-2 are the range finding figure of distance between 30 to 40 meters, Fig. 6-3 are the range finding figure of distance between 40 to 50 meters, Fig. 6-4 are the range finding figure of distance between 50 to 60 meters, Fig. 6-5 are the range finding figure of distance between 60 to 70 meters, Fig. 6-6 are the range finding figure of distance between 80 to 90 meters, and Fig. 6-7 are the range finding figure of distance between 110 to 120 meters, and Fig. 6-8 are the range finding figure of distance between 120 to 130 meters.

Embodiment

Process flow diagram of the present invention as shown in Figure 1, first camera is carried out to calibration of camera, and verify the accuracy of three line calibration methods by geometric calculation and total station survey method, then utilize three line calibration method static demarcating camera attitude parameters, the angle of pitch different according to camera, adopt the method for piecewise fitting function, the funtcional relationship of matching ordinate and actual fore-and-aft distance, matching ordinate and every pixel represent the funtcional relationship of actual lateral separation.Finally, utilize the result of lane detection, again utilize the actual angle of pitch of the real-time computing camera of three line calibration methods, calculate the distance of vehicle target and this car by linear interpolation.Below in conjunction with accompanying drawing, the specific implementation process of technical solution of the present invention is illustrated.

1. camera calibration of camera

Adopt Zhang Zhengyou method to demarcate camera inner parameter, obtain principal point for camera coordinate (i ₀, j ₀), focal length (take pixel as the unit) f of horizontal direction and vertical direction _iand f _j.

2. three line calibration method computing method

If three straight line l parallel to each other on ground ₁, l ₂, l ₃, they are all parallel to vehicle axis Xv, are respectively a with the distance of Xv ₁, a ₂, a ₃.Their common intersecting point coordinates are (i _h, j _h), and U _h1=U _h2=U _h3=U _h=(i _h-i ₀) d _x, V _h1=V _h2=V _h3=V _h=(j _h-j ₀) d _y.D _x, d _ybe respectively photo coordinate system and deposit horizontal, the longitudinal scale-up factor of coordinate system conversion to frame.At l1, l2, any point of getting respectively non-intersection point on l3, coordinate is (i _n, j _n), n=1,2,3.Three line calibration method computing method are as follows:

ψ＝arctg{[(r ₁-r ₃)(a ₁-a ₂)-(r ₁-r ₂)(a ₁-a ₃)]/[(r ₁-r ₃)(r ₁a ₁-r ₂a ₂)-(r ₁-r ₂)(r ₁a ₁-r ₃a ₃)]}

θ＝arctg(U _h?sinψ/f _id _x+V _h?cosψ/f _jd _y)，

h＝(a ₂-a ₁)AC/(BC-AD)，d＝(B/A)(a ₂-a ₁)AC/(BC-AD)+a ₁

In formula

A＝r ₁?sinψcosθ-cosθcosψ

C＝r ₂?sinψcosθ-cosθcosψ

r _n＝-(f _i/f _j)(i _h-i _n)/(j _h-j _b)，n＝1，2，3

Wherein arctg represents arctan function.In result of calculation, Ψ is camera direction drift angle, for camera level inclination, θ is the camera angle of pitch, and h is camera heights, and d is the distance of camera to Xv.

3. the accuracy of checking three line calibration methods

3.1 as shown in Figure 2, and camera is fixed on tripod, and by tripod base furnishing level, a mark post is placed perpendicular to ground in front, determines A, B, C 3 points on label.Camera photocentre is aimed at respectively to front A, B, C 3 points, record three groups of angles of pitch and camera heights value, θ 1 and h1, θ 2 and h2, θ 3 and h3 with three line calibration methods.

3.2 use meter rulers are measured overhead height D1, D2 and D3 of A, B, C at 3, measure camera photocentre arrives camera horizontal range D to floor level h and A, B, 3 of C, the angle of pitch while can computing camera photocentre aiming at A, B, C at 3 by geometric calculation, obtain θ 1 '=arctg[(h-D1)/D], θ 2 '=arctg[(h-D2)/D], θ 3 '=arctg[(h-D3) and/D].

3.3 remove camera, and by the position that total powerstation is placed in camera is identical just now, the angle of pitch β 1 when recording respectively total powerstation telescope right-angled intersection point and aiming at front A, B, C at 3, β 2, β 3.

3.4 compare θ 1 and θ 1 ', θ 2 and θ 2 ', θ 3 and θ 3 ', h1, h2, h3 and h, (θ 1-θ 2), (θ 1 '-θ 2 ') and (β 1-β 2), (θ 1-θ 3), (θ 1 '-θ 3 ') and (β 1-β 3), the accuracy of checking three line calibration methods.Can satisfy the demand through experimental verification three line calibration method stated accuracies.

4. the funtcional relationship of piecewise fitting ordinate and actual fore-and-aft distance, the every pixel of piecewise fitting ordinate and level represents the funtcional relationship of actual lateral separation.

4.1 camera mount scheme schematic diagram as shown in Figure 3, Fig. 3-1 is vertical view, Fig. 3-2 are side view, Fig. 3-3 are front view, camera is arranged on after automotive interior front windshield, and direction is directed straight ahead, and level is put, camera optical axis overlaps with vehicle axis Xv in the projection on ground, makes the distance d=(d of camera to Xv ₂-d ₁)/2=0, level inclination

direction deflection angle Ψ=0.Three straight lines that are parallel to each other are found or draw in camera front, and all parallel with Xv, and measure the distance of Xv and three straight lines, with three line calibration method calibration for cameras attitude parameters, adjust camera attitude, until make level inclination

with all within 0.002 radian (within 0.12 degree) of direction deflection angle Ψ.

4.2 aim at respectively, 75 meters, front and 40 meters of when camera photocentre, and the angle of pitch under both of these case is about the scope (θ that causes the angle of pitch to change because jolting in vehicle traveling process _l, θ _h), aim at respectively, 50 meters, front when camera photocentre, the angle of pitch is now about θ _m=(θ _l+ θ _h)/2.

4.3 adjust the camera angles of pitch makes photocentre aim at, 75 meters, front, and meets 4.1 mount scheme requirement, demarcates now pitching angle theta with three line calibration methods _l, start along Xv from minimum sighting distance, every 5 meters, label is set one time, and record the ordinate of this correspondence image.Laterally measure certain distance with meter ruler again on edge, label place, and record the horizontal ordinate of its correspondence image.The data sectional that utilization obtains simulates the angle of pitch for θ _ltime, the funtcional relationship of ordinate and actual fore-and-aft distance is designated as VD _{θ L}(y), the every pixel of ordinate and level represents that the funtcional relationship of actual lateral separation is designated as HD _{θ L}(y), wherein y represents ordinate.

4.4 adjust the camera angle of pitch makes photocentre aim at, 50 meters, front and 40 meters of, repeats embodiment 4.3 processes, and it is θ that piecewise fitting goes out the angle of pitch _mand θ _htime, the funtcional relationship VD of ordinate and actual fore-and-aft distance _{θ M}and VD (y) _{θ H}(y), the every pixel of ordinate and level represents the funtcional relationship HD of actual lateral separation _{θ M}and HD (y) _{θ H}(y).Fore-and-aft distance under three kinds of different angles of pitch and the fitting function of ordinate are as shown in Fig. 4-1, and the fitting function that the horizontal every pixel under three kinds of different angles of pitch represents actual range and ordinate is as shown in Fig. 4-2.

5. the real-time calibration camera angle of pitch

As shown in Figure 5, the distribution of left and right lane line and track center line in the presentation video of Fig. 5-1, Fig. 5-2 represent the distribution of actual road surface left and right lane line and track center line.If the left and right lane line equation in image detecting is in real time respectively: l _left: y=k ₁x+b ₁and l _right: y=k ₂x+b ₂, the Bisector of angle equation of two straight lines is

wherein

so just construct three straight lines parallel to each other, left and right lane line and the track center line on actual road surface.If the horizontal linear intersection point of left and right lane line and track center line and image y=0 is respectively P _l, P _rand P _m, the horizontal ordinate of 3 is respectively

with the horizontal ordinate at camera position place is

wherein P _xrepresent the horizontal resolution of collected by camera image.

As shown in Figure 4, HD _{θ L}(y), HD _{θ M}and HD (y) _{θ H}(y) in the time of y=0, error is very little.So by X _l, X _r, X _mand X _ain conjunction with HD _{θ M}(y) can calculate actual road surface straight line l _left, l _right, l _middledistance a with vehicle axis Xv _left, a _rightand a _middle, by itself and P _l, P _rand P _mcoordinate bring in the formula of embodiment step 2, can calibrate real-time camera pitching angle theta _real-time.

6. linear interpolation is calculated the distance of vehicle target and this car

According to θ _real-timewith θ _l, θ _mand θ _hrelation, utilize linear interpolation method, calculate the distance of target vehicle and this car, computing method are as follows:

$\mathrm{Dis}=\sqrt{{\mathrm{<figure-callout\; id="2"\; label="VD"\; filenames="HSA00000723522700041.png"\; state="\{\{state\}\}">VD</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="HD"\; filenames="HSA00000723522700041.png"\; state="\{\{state\}\}">HD</figure-callout>}}^2}$

$\mathrm{VD}=\left\{\begin{array}{cc}{\mathrm{VD}}_{\mathrm{\&theta;L}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_L\\ \frac{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{VD}}_{\mathrm{\&theta;L}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_L}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{VD}}_{\mathrm{\&theta;M}}\left(y\right),& {\mathrm{\&theta;}}_L<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_M\\ \frac{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{VD}}_{\mathrm{\&theta;M}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_M}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{VD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_M<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_H\\ {\mathrm{VD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}>{\mathrm{\&theta;}}_H\end{array}\right.$

$\mathrm{HD}=\left\{\begin{array}{cc}\left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;{\mathrm{HD}}_{\mathrm{\&theta;L}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_L\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;[\frac{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{HD}}_{\mathrm{\&theta;L}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_L}{{\mathrm{\&theta;}}_M-{\mathrm{\&theta;}}_L}{\mathrm{HD}}_{\mathrm{\&theta;M}}\left(y\right)],& {\mathrm{\&theta;}}_L<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_M\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;[\frac{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_{\mathrm{realtime}}}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{HD}}_{\mathrm{\&theta;M}}\left(y\right)+\frac{{\mathrm{\&theta;}}_{\mathrm{realtime}}-{\mathrm{\&theta;}}_M}{{\mathrm{\&theta;}}_H-{\mathrm{\&theta;}}_M}{\mathrm{HD}}_{\mathrm{\&theta;H}}\left(y\right)],& {\mathrm{\&theta;}}_M<{\mathrm{\&theta;}}_{\mathrm{realtime}}\&le;{\mathrm{\&theta;}}_H\\ \left[\mathrm{abs}(x-\frac{P_X}{2})\right]\&CenterDot;{\mathrm{HD}}_{\mathrm{\&theta;H}}\left(y\right),& {\mathrm{\&theta;}}_{\mathrm{realtime}}>{\mathrm{\&theta;}}_H\end{array}\right.$

In formula, VD represents the fore-and-aft distance of vehicle target and this car, and HD represents the lateral separation of vehicle target and this car, and Dis represents the distance of vehicle target and this car, P _xthe horizontal resolution that represents collected by camera image, x and y represent respectively the transverse and longitudinal coordinate of target.

Distance-finding method used in the present invention, carries out emulation to the video image that uses industrial camera collection, under different distance, finds range result as shown in Figure 6.In sum, first method of the present invention has verified the accuracy of three line calibration methods, then use three line calibration method calibration for cameras attitude parameters, by actual measurement, obtain the corresponding relation between distance sample point and picture plane, then use the method fitting function relation of piecewise fitting, set up range finding model corresponding to these mapping relations, and be equipped with the real-time dynamic compensation changing for the angle of pitch, thereby realize accurate detection and calculated the distance-finding method of distance between current driving vehicle and front vehicles.

Claims (2)

1. the front vehicles distance-finding method based on monocular vision, comprises the following steps:

(1) camera inner parameter is demarcated;

(2) camera is arranged on automobile, with the attitude parameter of three line calibration method static demarcating cameras, if camera attitude parameter does not meet selected mount scheme, after adjusting camera attitude, again use three line calibration method static demarcating camera parameters, until the attitude parameter of its camera meets the requirement of mount scheme;

(3) establishing the scope that causes the angle of pitch in vehicle traveling process because jolting is (θ _l, θ _h), adjusting the camera angle of pitch is θ _l, on the longitudinal direction in camera front, apart from vehicle body certain distance, monumented point is set, at this some place, label is set, record the ordinate of this monumented point correspondence at image, and laterally measure again certain distance at this monumented point place, record the horizontal ordinate of its correspondence image;

(4), in the process of camera front longitudinal different distance place repeating step step (3), utilize the data sectional obtaining to simulate the angle of pitch for θ _ltime, the funtcional relationship of ordinate and actual fore-and-aft distance is designated as VD _{θ L}(y), ordinate and every pixel represent that the funtcional relationship of the actual lateral separation of level is designated as HD _{θ L}(y), wherein y represents ordinate;

(5) adjusting respectively the camera angle of pitch is θ _hbe about θ _m=(θ _l+ θ _h)/2, the process of repeating step (3) and (4), then utilizes the data that obtain, and it is θ that piecewise fitting goes out the angle of pitch _hbe about θ _m=(θ _l+ θ _h)/2 o'clock, the funtcional relationship VD of ordinate and actual fore-and-aft distance _{θ H}and VD (y) _{θ M}(y), the every pixel of ordinate and level represents the funtcional relationship HD of actual lateral separation _{θ H}and HD (y) _{θ M}(y);

(6) establish the left and right lane line equation detecting in real time and be respectively y=k ₁x+b ₁and y=k ₂x+b ₂, the Bisector of angle equation of two straight lines is wherein

construct three straight lines in fact parallel to each other: left and right lane line and track center line, the funtcional relationship of utilizing step (4) and (5) to obtain, in conjunction with three line calibration methods, calibrates the real-time luffing angle θ of camera _real-time;

(7) according to θ _real-timewith θ _l, θ _mand θ _hrelation, utilize linear interpolation method, calculate the distance of vehicle target and this car,

Computing formula is as follows:

In formula, VD represents the fore-and-aft distance of vehicle target and this car, and HD represents the lateral separation of vehicle target and this car, and Dis represents the distance of vehicle target and this car, P _xthe horizontal resolution that represents collected by camera image, x and y represent respectively the transverse and longitudinal coordinate of target.

2. a kind of front vehicles distance-finding method based on monocular vision according to claim 1, is characterized in that, in step (2), the requirement that the attitude parameter of camera meets mount scheme is: make level inclination with direction deflection angle Ψ all within 0.002 radian.

Images