CN106250912A - Vehicle position acquisition method based on image - Google Patents

Abstract

The invention provides a kind of vehicle position acquisition method based on image, the method includes: given motor vehicles features training sample image；In features training, for the Orientation Features simple structure grader of each sample motor vehicle, and select the simple classification device of iteration error minimum；Exported final grader by multiple simple classification device linear combinations, select optimal threshold by final grader, distinguish sample motor vehicle and Fei Che sample；Calculate the symmetrical metrics of vehicle-mounted image, determine whether front motor vehicles exists.The present invention proposes a kind of vehicle position acquisition method based on image, during motor vehicles outline identification, make full use of Orientation Features information, take into account the real-time during identification and robustness, and in safe distance between vehicles monitors, the speed polytropy relatively travelled according to adjacent two cars provides comprehensively consideration, improves public transport safe early warning operating capability.

Description

Vehicle position acquisition method based on image

Technical field

The present invention relates to machine vision, particularly to a kind of vehicle position acquisition method based on image.

Background technology

While transportation fast development, traffic safety problem becomes increasingly conspicuous, the traffic emerged in an endless stream Accident causes huge economic loss and human life's loss to society, becomes new social unstable factor.Motor line Sail safe condition vision early warning system to contribute to improving traffic safety and improving road safety operating capability, and can be to accident Confirmation of responsibility provides partial visual evidence.Although motor vehicles visual security early warning technology has become the design of current automobile safety Emphasis, but existing safe early warning technology all has some limitations.Such as front truck image-recognizing method have ignored machine The Orientation Features information that motor-car profile is shown in the picture, causes the damage of topically effective authentication information under single features Lose, fail preferably to solve the contradiction between real-time and robustness during motor vehicles identification.Office in safe distance between vehicles monitors It is limited to the calculating relative to travel speed of before and after's car, does not solve the work of driver's subjective response time in safe distance between vehicles monitoring decision-making With, adjacent two cars relative to the uncertain problem such as polytropy of transport condition.

Summary of the invention

For solving the problem existing for above-mentioned prior art, the present invention proposes a kind of vehicle location based on image and obtains Method, including:

Given motor vehicles features training sample image (x₁, y₁) ..., (x_N, y_N), y_i∈ 0,1}, represent non-car sample respectively And sample motor vehicle；Initialization sample weights δ_i:

${\mathrm{\δ}}_i=\left\{\begin{array}{cc}1/2p& y_i=1\\ 1/2q& y_i=0\end{array}\right.$

In formula, p is sample motor vehicle quantity；Q is non-car sample size；

Make t=1 ..., T, T are total iterations, in t wheel features training, many for each sample motor vehicle Direction character, simple structure grader h_t∈ [0,1]；H (f (x), p, η) is: h=1, otherwise h=0 as pf (x) < p η；f(x) For the eigenvalue of pixel x, η is predetermined threshold value；P is polarity function；

Calculate training sample at h_tThe classification error rate sum of middle appearance: ε_t=∑ [h_t(x_i)-y_i]²；

Select simple classification device h minimum for iteration error ε_t；

Update training sample weights: will ε_tIt is updated to ε_t exp[-y_ih_t(x_i)], i=1,2 ..., N, regularization weights: ∑_iε_t=1；

By T the simple classification device linear combination final grader of output:

$H\left(x\right)=sign\left(\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{h}_t\right(x\left)\right)$

Select optimal threshold by final grader, utilize this threshold zone to divide sample motor vehicle and Fei Che sample；To vehicle-mounted The road image of image capture device collection detects in real time and judges, obtains in present image behind the position at front truck place, Be expert at for horizontal axis brightness data is considered as one-dimensional functions g (x) of abscissa, and axis of symmetry is taken as rectangle vertical axis x_s, wide Degree is rectangle width w, with x_sFor midpoint, g (x)=g (x_s+ u) odd component O (u, x_s, w) with even component E (u, x_s, w):

E (u, x_s, w)=g (x_s+u)/2+g(x_s-u)/2 -w/2≤u≤w/2

O (u, x_s, w)=g (x_s+u)/2-g(x_s-u)/2 -w/2≤u≤w/2

Symmetrical metrics is calculated by energy function:

$\begin{array}{c}S(x_s,w)=\left(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\right(u,x_s,w)-\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2(u,x_s,w\left)\right)/\left(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\right(u,x_s,w)+\\ \underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2\left(u,x_s,w\right))\end{array}$

In formula, x_sFor the vertical axis of rectangle；W is rectangle width；As S (x_s, time w)=-1, represent the most asymmetric；S(x_s, When w)=1, represent full symmetric；

Above-mentioned symmetrical metrics is carried out entropy regularization: SYM=S (x_s, w)/4E_m

Wherein E_mFor gray level image information entropy maximum, predefine threshold value when SYM is more thanTime, determine that front motor vehicles is deposited ?.

The present invention compared to existing technology, has the advantage that

The present invention proposes a kind of vehicle position acquisition method based on image, during motor vehicles outline identification, fills Divide and utilize Orientation Features information, take into account the real-time during identification and robustness, and in monitoring in safe distance between vehicles, according to The speed polytropy that adjacent two cars travel relatively provides comprehensively consideration, improves public transport safe early warning operating capability.

Accompanying drawing explanation

Fig. 1 is the flow chart of vehicle position acquisition method based on image according to embodiments of the present invention.

Detailed description of the invention

Hereafter provide retouching in detail one or more embodiment of the present invention together with the accompanying drawing of the diagram principle of the invention State.Describe the present invention in conjunction with such embodiment, but the invention is not restricted to any embodiment.The scope of the present invention is only by right Claim limits, and the present invention contains many replacements, amendment and equivalent.Illustrate in the following description many details with Thorough understanding of the present invention is just provided.These details are provided for exemplary purposes, and without in these details Some or all details can also realize the present invention according to claims.

An aspect of of the present present invention provides a kind of vehicle position acquisition method based on image.Fig. 1 is real according to the present invention Execute the vehicle position acquisition method flow chart based on image of example.

The present invention, in front truck image recognition, merges the performance characteristic in the priori to front truck and image thereof, limits Object detecting and tracking region, proposes front truck existence it is assumed that extractor according to the detection of road image luminance mean value Characteristics of Mutation Motor-car sample Orientation Features, to the feature samples training classification extracted, it is achieved being accurately positioned of front truck, by entropy regularization Symmetrical metrics checking front truck existence is it is assumed that achieve the detection identification of motor vehicles.In conjunction with the forecast model improved to motor vehicles Testing result carries out dynamic realtime tracking.In safe distance between vehicles calculates, by setting up distance survey based on track plane restriction Model, it is achieved that the accurate measurement of car spacing before and after adjacent, uses multisensor from driver's response feature, motor vehicles reaction spy Property and road environment set out, and establish the safe distance between vehicles model of Multi-source Information Fusion.

According to the Prior Knowledge Constraints hunting zone of vehicle-mounted image capture device installation site, by Foregut fermenters tracing area Being narrowed down in current lane by entire image, centered by the accumulation of lane line, arranging a width is 70 pixels, and height is 50 The hough transform region of pixel, motor vehicle detecting region is the inside that arranged on left and right sides lane line limits jointly with rectangular area Region.

In the hough transform region determined, calculate the flat of every one-row pixels the most respectively from the bottom of road image All brightness values, search out the region that in image, luminance mean value sudden change is maximum, it becomes possible to tentatively conclude in road image and whether may There is motor vehicles.

$G\left(r\right)=\frac{\underset{i=b{\left(r\right)}_L}{\overset{b{\left(r\right)}_R}{\mathrm{\&Sigma;}}}g(r,i)}{b{\left(r\right)}_R+b{\left(r\right)}_L+1}$

In formula, G (r) is luminance mean value；b(r)_LLeft end pixel coordinate for the r row in region of search；b(r)_RFor search The right-hand member pixel coordinate of the r row in region；(r i) is pixel (r, brightness value i) to g.

In region of search, luminance mean value G (r) produces the position that Spline smoothing is expert at, and may correspond to motor vehicles Edge bottom, temporary transient setting has detected front truck.If there is no the saltus step of gray average in whole region of search, then it is assumed that when Motor vehicles is there is not in front detection region.

Select following wave filter:

$\begin{array}{c}\mathrm{\&Phi;}(x,y)=\frac{{\mathrm{\&omega;}}_0}{\sqrt{\mathrm{\&pi;}}k}\exp \&lsqb;-\frac{{\mathrm{\&omega;}}_0}{2k^2}({\left(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;}\right)}^2+{\left(-x\sin \mathrm{\&theta;}+y\cos \mathrm{\&theta;}\right)}^2)\&rsqb;*\sin \left({\mathrm{\&omega;}}_0\right(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})\\ )\end{array}$

Direction θ is selected by expression:

θ=2k π/n, k={0,1,2...n-1}

In formula, the sum of the angle that n comprises by the wave filter used；ω₀For predetermined radial center frequency.

By image I, (x y) is carried out convolution with above-mentioned wave filter, is obtained on different scale by the change of the yardstick of wave filter Orienting response, thus form 8 directions.With wavelet conversion coefficient W, (x y) reflects the characteristics of image of motor vehicles.

W (x, y)=∫ I (x₁, y₁)Φ(x,y)*(x-x₁, y-y₁)dxdy

Given motor vehicles features training sample image (x₁, y₁) ..., (x_N, y_N), y_i∈ 0,1}, represent non-car sample respectively And sample motor vehicle.Then the weight initialization process of sample is as follows:

${\mathrm{\&delta;}}_i=\left\{\begin{array}{cc}1/2p& y_i=1\\ 1/2q& y_i=0\end{array}\right.$

In formula, p is sample motor vehicle quantity；Q is non-car sample size.

In t wheel features training, (t=1 ..., T, T are total iterations), for 8 sides of each sample motor vehicle To feature, simple structure grader h_t∈ [0,1]；H (f (x), p, η) is: h=1, otherwise h=0 as pf (x) < p η；F (x) is The eigenvalue of pixel x, η is predetermined threshold value；P is polarity function.

Calculate training sample at h_tThe classification error rate sum of middle appearance: ε_t=∑ [h_t(x_i)-y_i]²；

Select simple classification device h minimum for iteration error ε_t；

Update training sample weights: will ε_tIt is updated to ε_t exp[-y_ih_t(x_i)], i=1,2 ..., N, regularization weights: ∑_iε_t=1；

By T the simple classification device linear combination final grader of output:

$H\left(x\right)=sign\left(\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{h}_t\right(x\left)\right)$

Select simple classification device, and training sample is classified by feature based value.Optimal threshold is selected by final grader, This threshold zone is utilized to divide sample motor vehicle and Fei Che sample.In off-line training module, by big to collecting under circumstances The sample motor vehicle of amount and the study of Fei Che sample, for 8 direction characters of sample motor vehicle, by grader features training shape Become a series of simple classification device, according to weight, these simple classification devices are combined into final grader.ONLINE RECOGNITION module is then The final grader obtained according to off-line training module, the road image gathering vehicle-mounted image capture device detects in real time With judgement, obtain the position at front truck place in present image.

When carrying out symmetrical metrics and calculating, be expert at for horizontal axis brightness data is considered as the one-dimensional functions g of abscissa X (), axis of symmetry is taken as rectangle vertical axis x_s, width is rectangle width w, with x_sFor midpoint, g (x)=g (x_s+ u) odd component O (u, x_s, w) with even component E (u, x_s, w) it is shown below:

E (u, x_s, w)=g (x_s+u)/2+g(x_s-u)/2 -w/2≤u≤w/2

O (u, x_s, w)=g (x_s+u)/2-g(x_s-u)/2 -w/2≤u≤w/2

Calculated symmetrical metrics by energy function to be shown below:

$\begin{array}{c}S(x_s,w)=\left(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\right(u,x_s,w)-\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2(u,x_s,w\left)\right)/\left(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\right(u,x_s,w)+\\ \underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2\left(u,x_s,w\right))\end{array}$

In formula, x_sFor the vertical axis of rectangle；W is rectangle width.

As S (x_s, time w)=-1, represent the most asymmetric；S(x_s, when w)=1, represent full symmetric.

The symmetrical metrics quoting entropy regularization gets rid of false motor vehicles target, and symmetrical metrics is defined as: SYM=S (x_s, w)/4E_m

Wherein E_mFor gray level image information entropy maximum, predefine threshold value when SYM is more thanTime, determine that front motor vehicles is deposited ?.

In a further embodiment, on the basis of single-frame static images range finding model, the present invention is according to road geometry Morphological characteristic establishes a kind of spacing vision measurement model.Calculate target characteristic point P₀Aperture with vehicle-mounted image capture device C projected position in the plane of track in the center of circle i.e. distance of initial point O | OP₀| for:

$\left|{\mathrm{OP}}_0\right|=\sqrt{{\left|OP\right|}^2+{\left|{\mathrm{PP}}_0\right|}^2}$

Its midpoint P is calculated as | CP |=h/cos β, and β is the angle of pitch of vehicle-mounted image capture device, and h is vehicle-mounted image capturing The terrain clearance of equipment.

| OP |=h × tan (pi/2-β+tan (C₀-P’₀)/f)

C₀Optical axis and plane of delineation intersection point, P ' for vehicle-mounted image capture device₀For target characteristic point P₀In the plane of delineation Corresponding pixel, f is focal length.

$\left|{\mathrm{PP}}_0\right|=h/cos\mathrm{\&beta;}\&CenterDot;\frac{\frac{k_L+{\mathrm{\&mu;k}}_R}{1+\mathrm{\&mu;}}{P^{\&prime;}}_0}{f\sqrt{{\left(\frac{k_L+{\mathrm{\&mu;k}}_R}{1+\mathrm{\&mu;}}\right)}^2+1}}$

k_L、k_RThe slope of arranged on left and right sides lane line in expression pixel planes respectively；Coefficient

By calculating some O and distance | OP | of some P and some P₀Distance with a P | PP₀|, obtain | OP₀| i.e. as this car And the spacing between front truck.

The safe distance between vehicles model of Multi-source Information Fusion that the present invention sets up, by being the vision of status information to garage of front and back Cognition Understanding, the relative velocity of car and spacing before and after estimation.

In the process of moving, before and after driver is adjacent when implementing brake, the relative position relation of car is expressed as follows motor vehicles: L_sFor front and back's car safe distance between vehicles, L_hFor the operating range in this car braking time, L_fFor front truck operating range, D₀It is that two cars are relative Fixing safe distance between vehicles time static.

Then safe distance between vehicles L_s=L_h-L_f+D₀

1. when current vehicle is static, L_fBeing 0, front and back the safe distance between vehicles of car is:

L_s=v_h(t_e+t_c+t_z/2)+v_h ²/2a_hmax+D₀

Wherein v_hFor this vehicle speed, t_eFor time of driver's reaction, t_cTime, t is coordinated for brakes_zFor deceleration time, a_hmaxFor this car maximum brake acceleration.

If 2. front truck is at the uniform velocity to travel, L_f=v_f(t_e+t_c+t_z/2+(v_h-v_f)a_bh)；

a_bhFor this car brake acceleration.

Before and after then, the safe distance between vehicles of car is:

L_s=v_r(t_e+t_c+t_z/2)+v_r ²/2a_bh+D₀

Wherein v_rBeing two car relative velocities, its acquisition methods is:

v_r=(L_i-L_i+1)/Δt；

L_iWith L_i+1Obtained by said process measurement by vehicle-mounted image capture device before and after being respectively time interval Δ t This car and the spacing of front truck.

3., if front truck is in even Reduced Speed Now, it is divided into this car speed to be higher than, equal to two kinds of situations of front truck speed.

1., when this car speed is higher than front truck speed, the operating range of front truck is:

L_f=(v_f ²-v’_f ²)2a_bf

In formula, v '_fFor moment instantaneous velocity a certain in front truck moderating process；a_bfFor front truck brake acceleration, its calculating side Method is:

a_bf=(vⁱ⁺¹ _f-vⁱ _f)/Δt；

vⁱ⁺¹ _f=(Lⁱ⁺²-Lⁱ⁺¹)/Δt+vⁱ⁺¹ _h

vⁱ _f=(Lⁱ⁺¹-Lⁱ)/Δt+vⁱ _h

Lⁱ,Lⁱ⁺¹,Lⁱ⁺²It is respectively three the continuous spacing numerical value recorded at interval of Δ t.vⁱ _h, vⁱ⁺¹ _hBefore and after interval of delta t This car speed.

This car driver reaction and brake coordination time come into effect deceleration, and its operating range is:

L_h=v_h(t_e+t_c+t_z/2)+(v_h ²-v’_f ²)/2a_bh

Now, if two car maximum brake acceleration are identical, then:

L_s=L_h-L_f+D₀=v_h(t_e+t_c+t_z/2)+v_r(2v_h-v_r)/2a_hmax+D₀

2., when this car speed is equal to front truck speed, this car adjusts brake acceleration to maintain the following state with front truck, then Safe distance between vehicles is:

L_s=v_h(t_e+t_c+t_z/2)+(v_f ²/2)(1/a_bh-1/a_bf)+v_f ²/2a_bf+D₀

In sum, the present invention proposes a kind of vehicle position acquisition method based on image, at motor vehicles outline identification During, make full use of Orientation Features information, take into account the real-time during identification and robustness, and supervising in safe distance between vehicles In control, the speed polytropy relatively travelled according to adjacent two cars provides comprehensively consideration, improves public transport safe early warning pipe Reason ability.

Obviously, it should be appreciated by those skilled in the art, each module of the above-mentioned present invention or each step can be with general Calculating system realize, they can concentrate in single calculating system, or be distributed in multiple calculating system and formed Network on, alternatively, they can realize with the executable program code of calculating system, it is thus possible to by they store Performed by calculating system within the storage system.So, the present invention is not restricted to the combination of any specific hardware and software.

It should be appreciated that the above-mentioned detailed description of the invention of the present invention is used only for exemplary illustration or explains the present invention's Principle, and be not construed as limiting the invention.Therefore, that is done in the case of without departing from the spirit and scope of the present invention is any Amendment, equivalent, improvement etc., should be included within the scope of the present invention.Additionally, claims purport of the present invention Whole within containing the equivalents falling into scope and border or this scope and border change and repair Change example.

Claims (1)

1. a vehicle position acquisition method based on image, it is characterised in that including:

Given motor vehicles features training sample image (x₁, y₁) ..., (x_N, y_N), y_i∈ 0,1}, represent non-car sample and machine respectively Motor-car sample；Initialization sample weights δ_i:

${\mathrm{\&delta;}}_i=\left\{\begin{array}{cc}1/2p& y_i=1\\ 1/2q& y_i=0\end{array}\right.$

In formula, p is sample motor vehicle quantity；Q is non-car sample size；

Make t=1 ..., T, T are total iterations, in t wheel features training, multi-direction for each sample motor vehicle Feature, simple structure grader h_t∈ [0,1]；H (f (x), p, η) is: h=1, otherwise h=0 as pf (x) < p η；F (x) is picture The eigenvalue of element x, η is predetermined threshold value；P is polarity function；

Calculate training sample at h_tThe classification error rate sum of middle appearance: ε_t=∑ [h_t(x_i)-y_i]²；

Select simple classification device h minimum for iteration error ε_t；

Update training sample weights: will ε_tIt is updated to ε_t exp[-y_ih_t(x_i)], i=1,2 ..., N, regularization weights: ∑_iε_t =1；

By T the simple classification device linear combination final grader of output:

$H\left(x\right)=sign\left(\underset{t=1}{\overset{T}{\&Sigma;}}{h}_t\right(x\left)\right)$

Select optimal threshold by final grader, utilize this threshold zone to divide sample motor vehicle and Fei Che sample；To vehicle-mounted image The road image of capture device collection detects in real time and judges, obtains in present image behind the position at front truck place, by water The be expert at brightness data of flat axis is considered as one-dimensional functions g (x) of abscissa, and axis of symmetry is taken as rectangle vertical axis x_s, width is Rectangle width w, with x_sFor midpoint, g (x)=g (x_s+ u) odd component O (u, x_s, w) with even component E (u, x_s, w):

E (u, x_s, w)=g (x_s+u)/2+g(x_s-u)/2 -w/2≤u≤w/2

O (u, x_s, w)=g (x_s+u)/2-g(x_s-u)/2 -w/2≤u≤w/2

Symmetrical metrics is calculated by energy function:

$\begin{array}{c}S\left(x_s,w\right)=\left(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\left(u,x_s,w\right)-\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2\left(u,x_s,w\right)\right)/(\underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{E}^2\left(u,x_s,w\right)+\\ \underset{u=-w/2}{\overset{w/2}{\mathrm{\&Sigma;}}}{O}^2\left(u,x_s,w\right))\end{array}$

In formula, x_sFor the vertical axis of rectangle；W is rectangle width；As S (x_s, time w)=-1, represent the most asymmetric；S(x_s, w)= When 1, represent full symmetric；

Above-mentioned symmetrical metrics is carried out entropy regularization: SYM=S (x_s, w)/4 E_m

Wherein E_mFor gray level image information entropy maximum, predefine threshold value when SYM is more thanTime, determine that front motor vehicles exists.