CN105182320A - Depth measurement-based vehicle distance detection method - Google Patents

Abstract

The invention provides a depth measurement-based vehicle distance detection method. The method includes the following steps that: relevant parameters of a license plate detection classifier and two vehicle-mounted cameras are acquired; a front target vehicle in images acquired by the vehicle-mounted cameras is positioned; the license plate of the front target vehicle in the images acquired by the vehicle-mounted cameras is tracked respectively; the position of the central point of the license plate in the current frames of images acquired by the two vehicle-mounted cameras is obtained; the three-dimensional coordinates of the position of the central point of the license plate in a world coordinate system are obtained; three-dimensional coordinate transformation is carried out, so that the distance from the target vehicle to the vehicle-mounted cameras can be obtained; and the distance from the vehicle-mounted cameras to the head of a local vehicle is subtracted from the distance from the target vehicle to the vehicle-mounted cameras, so that the distance from the target vehicle to the head of the local vehicle can be obtained. According to the depth measurement-based vehicle distance detection method of the invention, a machine vision learning algorithm is adopted, and the position of the front vehicle can be precisely positioned; repeat positioning of the license plate is carried out based on target tracking technologies; the distance between the two vehicles can be calculated precisely based on binocular stereo vision. The depth measurement-based vehicle distance detection method has the advantages of higher speed and more accurate vehicle distance calculation.

Description

A kind of vehicle distance detecting method based on depth survey

Technical field

The present invention relates to safe driving technical field, specifically a kind of vehicle distance detecting method based on depth survey.

Background technology

In daily motor vehicle driving, enough spacing is kept to be the most effective means avoiding rear-end collision with front truck.And for the judgement of spacing, mainly obtained by the experience range estimation of driver, there is wretched insufficiency in this mode: first, the sitting posture of driver and the difference at visual angle, can relatively large deviation be there is in the result of range estimation, especially on a highway, because the speed of a motor vehicle is too fast, acquisition spacing more accurately cannot be estimated at all; Secondly, drive for a long time often to make driver attention not concentrate, easily ignore and judge to there is comparatively big error with the spacing of front truck or spacing, and then cause traffic hazard.

In recent years, there are some spacing detection techniques, mainly contain following a few class:

(1) physically based deformation ranging technology, such technology, mainly through transmitting and receiving ultrasound wave or infrared laser line, obtains the distance with preceding vehicle.There is more deficiency in this technology: equipment cost is high, and telemeasurement error is comparatively large, is subject to the impact of preceding object thing and causes flase drop, can exist interfering with each other simultaneously when multiple vehicle uses.

(2) based on video processing technique, as Chinese patent application CN104392629A discloses a kind of method and apparatus of inspection vehicle distance, Chinese patent CN101941438B discloses a kind of safe distance between vehicles intelligence measuring and controlling device and method, such technology is mainly through video image processing technology, on each two field picture of video, obtain the feature relevant to front vehicles, according to depth of field mapping table or three-dimensional measurement technology, obtain the spacing with front truck.The advantage of this method is that cost is low, Active measuring, wide adaptability, and its shortcoming is that algorithm is more complicated, and location front vehicles out of true, spacing error calculated is larger.

Summary of the invention

The object of the present invention is to provide that a kind of algorithm speed is faster, spacing calculates the more accurate vehicle distance detecting method based on depth survey.

Technical scheme of the present invention is:

Based on a vehicle distance detecting method for depth survey, comprise the following steps:

(1) correlation parameter that car plate detects sorter and two vehicle-mounted vidicons is obtained;

(2) judge whether to need detection and location objects ahead vehicle again, if so, then perform step (3), if not, then perform step (4);

(3) detect car plate position in sorter and image based on the car plate obtained and, apart from the distance of image lower boundary, locate the objects ahead vehicle in the image of a wherein vehicle-mounted vidicon collection; Detect car plate location similarity in sorter and image based on car plate again, locate the objects ahead vehicle in the image of another vehicle-mounted vidicon collection;

(4) car plate of objects ahead vehicle in the image of two vehicle-mounted vidicons collections is followed the tracks of respectively, namely according to the car plate position of target vehicle in previous frame image, the car plate position of target vehicle in prediction current frame image;

(5) the car plate center position of target vehicle in the current frame image of two vehicle-mounted vidicons collections is obtained respectively;

(6) according to the correlation parameter of vehicle-mounted vidicon, obtain the three-dimensional coordinate of car plate center position in world coordinate system of target vehicle in the current frame image of two vehicle-mounted vidicons collections, be the three-dimensional coordinate of coordinate system with vehicle-mounted vidicon by described three-dimensional coordinate transformation one-tenth in world coordinate system, obtain the distance of target vehicle apart from vehicle-mounted vidicon;

(7) distance of described target vehicle distance vehicle-mounted vidicon is deducted the distance of vehicle-mounted vidicon apart from this car headstock, namely obtain the distance of target vehicle apart from this car headstock.

The described vehicle distance detecting method based on depth survey, in step (1), described acquisition car plate detects sorter, comprising:

In a, positive and negative 30 degree of collection front, the license plate image of different distance is as the positive sample of training;

B, detect sorter file based on haar characteristic sum adaboost learning algorithm training car plate.

The described vehicle distance detecting method based on depth survey, in step (1), the correlation parameter of described acquisition two vehicle-mounted vidicons, comprising:

A, in the common visual field of two vehicle-mounted vidicons, place the scaling board of a width gridiron pattern pattern, use two vehicle-mounted vidicons to gather scaling board image respectively;

The angle position of b, conversion scaling board, until two vehicle-mounted vidicons acquire five width scaling board images respectively;

C, on each width scaling board image, obtain all gridiron patterns two dimension angular coordinate;

D, based on the classical calibration algorithm of Zhang Zhengyou, obtain the correlation parameter formula of vehicle-mounted vidicon:

$K=s\left[\begin{array}{ccc}{f}_x\×k_x& 0& u_0\\ 0& f_y\×k_y& v_0\\ 0& 0& 1\end{array}\right]$

$R=\left[\begin{array}{ccc}{r}_1& r_2& r_3\\ r_4& r_5& r_6\\ r_7& r_8& r_9\end{array}\right]$

$T=\left[\begin{array}{c}{t}_1\\ t_2\\ t_3\end{array}\right]$

Wherein, K represents the Intrinsic Matrix of vehicle-mounted vidicon, f _x, f _yrepresent the effective focal length of camera lens, k _x, k _yrepresent distortion factor, (u ₀, v ₀) representing picture centre, s represents the uncertain graphical rule factor, and R, T are all outer parameter matrixs of vehicle-mounted vidicon, represents respectively from world coordinates and is tied to rotational transformation matrix between vehicle-mounted vidicon coordinate system and translation transformation matrix.

The described vehicle distance detecting method based on depth survey, step (3), comprising:

A, based on obtain car plate detect sorter, detect first vehicle-mounted vidicon collection image in track, front whether there is car plate;

B, judge whether vehicle corresponding to car plate obtained is positioned at current lane, is if so, considered as effective car plate, if not, is considered as invalid car plate;

If c only exists an effective car plate, then it can be used as target license plate; If there is multiple effective car plate, then the target license plate in the image selecting car plate position in image to gather as first vehicle-mounted vidicon apart from minimum effective car plate apart from image lower boundary, vehicle corresponding to target license plate is target vehicle;

D, detect sorter based on the car plate obtained, obtains all candidate license plate in the image of second vehicle-mounted vidicon collection;

Target license plate in e, the image that gathers with first vehicle-mounted vidicon is for standard, all candidate license plate in the image of second vehicle-mounted vidicon collection are carried out Similarity Measure with it respectively, select candidate license plate corresponding to maximum similarity to be target license plate in the image that gathers of second vehicle-mounted vidicon, vehicle corresponding to target license plate is target vehicle.

The described vehicle distance detecting method based on depth survey, in step (4), the described car plate position according to target vehicle in previous frame image, the car plate position of target vehicle in prediction current frame image, comprising:

A, in previous frame image, the car plate central point of target vehicle is saved as impact point, and the car plate band of position of target vehicle is saved as the plate template of target vehicle in current frame image after surrounding expansion;

B, impact point to be followed the tracks of, based on kalman Filter Principle, the position of car plate central point in current frame image of target of prediction vehicle;

C, car plate size based on target vehicle in position in current frame image of the car plate central point of target vehicle of prediction and previous frame image, according to following formula, obtain the Plate searching rectangular area of target vehicle in current frame image:

$\left\{\begin{array}{c}rect.x=center.x-car\_width*0.7\\ rect.y=center.y-car\_height\\ rect.width=car\_width*1.4\\ rect.height=car\_height*2\end{array}\right.$

Wherein, rect.x, rect.y represent upper left corner horizontal ordinate and the ordinate of the Plate searching rectangular area rect of target vehicle in current frame image respectively, center.x, center.y represent the position horizontal ordinate of the car plate central point of the target vehicle of prediction in current frame image and ordinate respectively, rect.width, rect.height represent width and the height of the Plate searching rectangular area rect of target vehicle in current frame image respectively, and car_width, car_height represent car plate width and the height of target vehicle in previous frame image respectively;

D, in current frame image target vehicle Plate searching rectangular area in, plate template is used to carry out search spread, to each traversal position, according to following formula, calculate the degree of confidence that this traversal position belongs to target vehicle car plate position, the traversal position selecting maximum confidence corresponding is as the car plate position of target vehicle in current frame image:

${\mathrm{conf}}_{ij}=\frac{1}{255*N}\Σ\Σf(i+x,j+y)-M(x,y)$

Wherein, conf _ijrepresent that current traversal position belongs to the degree of confidence of target vehicle car plate position, N represents the pixel quantity of the plate template of target vehicle in current frame image, M (x, y) represent that the plate template of target vehicle in current frame image is at (x, y) grey scale pixel value at place, f (i+x, j+y) represent in the Plate searching rectangular area of target vehicle in current frame image, with current traversal position top left co-ordinate (i, j) place is benchmark, the grey scale pixel value at skew (x, y) place.

The described vehicle distance detecting method based on depth survey, step (5), comprising:

A, according to following formula, obtain the binary image bin (x, y) of the car plate band of position of target vehicle in current frame image:

$bin(x,y)=\left\{\begin{array}{cc}255& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\\ 0& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\end{array}\right.$

Wherein, f (x, y) represents the grey scale pixel value at the car plate band of position (x, the y) place of target vehicle in current frame image, f (x _i, y _i) representing grey scale pixel value in N neighborhood centered by (x, y), n represents the number of pixels in the N neighborhood centered by (x, y), and T represents binary-state threshold;

B, to obtain binary image carry out morphology operations, remove interference;

C, obtain the car plate center position of target vehicle in current frame image according to following formula:

$\left\{\begin{array}{c}center.x=\frac{\Σx_i*bin(x_i,y_i)}{\Σbin(x_i,y_i)}\\ center.y=\frac{\Σy_i*bin(x_i,y_i)}{\Σbin(x_i,y_i)}\end{array}\right.$

Wherein, center.x, center.y represent horizontal ordinate and the ordinate of the car plate center position of target vehicle in current frame image respectively, bin (x _i, y _i) represent (x in binary image _i, y _i) grey scale pixel value at place.

The described vehicle distance detecting method based on depth survey, step (6), comprising:

A, based on following Binocular vision photogrammetry principle type, obtain the three-dimensional coordinate of car plate center position in world coordinate system (X, Y, Z) of target vehicle in the current frame image that two vehicle-mounted vidicons gather:

$\left\{\begin{array}{c}{s}_1\left[\begin{array}{c}{u}_1\\ v_1\\ 1\end{array}\right]=K_1\×\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\\ s_r\left[\begin{array}{c}{u}_r\\ v_r\\ 1\end{array}\right]=K_r\×\[R_r|T_r\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\end{array}\right.$

Wherein, s _l, K _l, R _l, T _lbe the correlation parameter of first vehicle-mounted vidicon, (u _l, v _l) be the car plate center position coordinate of target vehicle in the current frame image that gathers of first vehicle-mounted vidicon, s _r, K _r, R _r, T _rbe the correlation parameter of second vehicle-mounted vidicon, (u _r, v _r) be the car plate center position coordinate of target vehicle in the current frame image that gathers of second vehicle-mounted vidicon;

B, according to following formula, described three-dimensional coordinate transformation in world coordinate system is become the three-dimensional coordinate being coordinate system with first vehicle-mounted vidicon, obtains the distance of target vehicle apart from first vehicle-mounted vidicon:

$\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\\ 1\end{array}\right]=\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]$

Wherein, (X, Y, Z) represents the three-dimensional coordinate of car plate center position in world coordinate system of target vehicle in the current frame image that above-mentioned steps a obtains, (X _c, Y _c, Z _c) represent the three-dimensional coordinate being coordinate system with first vehicle-mounted vidicon, Z _crepresent the distance of car plate center position apart from first vehicle-mounted vidicon of target vehicle in current frame image, as the distance of objects ahead vehicle apart from first vehicle-mounted vidicon.

The described vehicle distance detecting method based on car plate position, also comprises: when described target vehicle is less than default safe distance apart from the distance of this car headstock time, carry out audio alert.

As shown from the above technical solution, the present invention adopts machine vision learning algorithm, accurately locates front vehicle position, and based target tracking technique carries out the resetting of car plate, based on binocular stereo vision accurate Calculation vehicle distances; Compared with the conventional method, algorithm speed is faster, and spacing calculates more accurate.

Accompanying drawing explanation

Fig. 1 is binocular stereo vision schematic diagram, and wherein equipment 1 is first vehicle-mounted vidicon being positioned at left side, and equipment 2 is second vehicle-mounted vidicon being positioned at right side;

Fig. 2 is method flow diagram of the present invention;

Fig. 3 is the gridiron pattern scaling board image gathered;

Fig. 4 is that positive sample image trained by car plate detection sorter;

Fig. 5 is car plate Detection results figure;

Fig. 6 is the binary image of the car plate band of position;

Fig. 7 is car plate center position figure.

Embodiment

The present invention is further illustrated below in conjunction with the drawings and specific embodiments.

As shown in Figure 1, what the three-dimensional measurement algorithm in the present embodiment adopted is technique of binocular stereoscopic vision, and two vehicle-mounted vidicons adopt parallel placement, and the parallax range between them is L.

As shown in Figure 2, a kind of vehicle distance detecting method based on depth survey, comprises the following steps:

S1, judge whether to need initializes system parameters, if needed, enter step S2, otherwise, enter step S3.

S2, initializes system parameters, systematic parameter comprises the correlation parameter that car plate detects sorter and vehicle-mounted vidicon, and concrete demarcating steps is as follows:

S21, obtain the correlation parameter of vehicle-mounted vidicon, specifically parameter matrix outside the Intrinsic Matrix of vehicle-mounted vidicon and position.Before system life's work, obtain the correlation parameter of two vehicle-mounted vidicons, this is the requisite program of binocular stereo vision, the classical calibration algorithm that what the scaling method in the present embodiment adopted is based on Zhang Zhengyou, and concrete demarcating steps is as follows:

S211, in the common visual field of two vehicle-mounted vidicons, place the scaling board of a width gridiron pattern pattern, use two vehicle-mounted vidicons to gather a width scaling board image respectively, as shown in Figure 3.

S212, convert the angle position of scaling board, again gather a width scaling board image according to step S211.

S213, repetition step S212, until two vehicle-mounted vidicons acquire five width scaling board images respectively.

S214, on each width scaling board image, obtain all gridiron patterns two dimension angular coordinate.

S215, based on classical theory of calibration, obtain the correlation parameter formula of vehicle-mounted vidicon:

$K=s\left[\begin{array}{ccc}{f}_x\×k_x& 0& u_0\\ 0& f_y\×k_y& v_0\\ 0& 0& 1\end{array}\right]$

$R=\left[\begin{array}{ccc}{r}_1& r_2& r_3\\ r_4& r_5& r_6\\ r_7& r_8& r_9\end{array}\right]$

$T=\left[\begin{array}{c}{t}_1\\ t_2\\ t_3\end{array}\right]$

Wherein, K is the Intrinsic Matrix of vehicle-mounted vidicon, and intrinsic parameter refers to that those can describe the physical parameter of vehicle-mounted vidicon image-forming principle and imaging characteristic, comprises the effective focal length f of camera lens _x, f _ywith distortion factor k _x, k _yand picture centre (u ₀, v ₀) and uncertain graphical rule factor s.R, T are the outer parameter matrixs of vehicle-mounted vidicon, outer parameter refers to those geometric parameters that can represent the particular location of vehicle-mounted vidicon in selected world coordinate system, comprises and is tied to rotational transformation matrix between vehicle-mounted vidicon coordinate system and translation transformation matrix from world coordinates.

S22, acquisition car plate detect sorter, and concrete steps are as follows:

In S221, positive and negative 30 degree of collection vehicle front, the license plate image of different distance is as the positive sample of training, as shown in Figure 4;

S222, based on haar characteristic sum adaboost learning algorithm, training car plate detects sorter file.

S3, judge whether to need to detect front vehicles, if needed, enter step S4, otherwise, enter step S5.

S4, detection front vehicles, judge whether the dead ahead of current lane exists target vehicle, as shown in Figure 5, concrete steps are as follows for effect:

S41, the car plate detection sorter obtained based on step S2, detect track, front in the image of first vehicle-mounted vidicon collection and whether there is car plate;

S42, judge whether vehicle corresponding to car plate obtained is positioned at current lane, if not, be considered as invalid car plate, if, be considered as effective car plate;

If S43 only exists an effective car plate, then it can be used as target license plate; If there is multiple effective car plate, the target license plate in the image selecting its position in the picture to gather as first vehicle-mounted vidicon apart from minimum effective car plate apart from image lower boundary, vehicle corresponding to target license plate is target vehicle;

S44, the car plate detection sorter obtained based on step S2, obtain all candidate license plate in the image of second vehicle-mounted vidicon collection;

S45, with step S43 obtain target license plate for standard, all candidate license plate that step S44 obtains are carried out Similarity Measure with it, select candidate license plate corresponding to maximum similarity to be target license plate in the image that gathers of second vehicle-mounted vidicon, vehicle corresponding to target license plate is target vehicle.

S5, tracking objects ahead vehicle, according to the car plate position in previous frame image, the prediction position of car plate in current frame image, each two field picture is avoided all to carry out the car plate detection and location of repetition, in the image of two vehicle-mounted vidicon collections, perform identical tracking step respectively, concrete steps are as follows:

S51, renewal tracking data, mainly in previous frame image, preserve car plate central point as tracking target point, and the car plate band of position is saved as the plate template of current frame image after surrounding expansion;

S52, impact point to be followed the tracks of, mainly based on kalman Filter Principle, the prediction position of car plate central point in current frame image;

S53, selection Plate searching rectangular area, mainly based on the car plate size in the car plate center position predicted and previous frame image, according to following formula, obtain the approximate region scope rect of car plate in current frame image:

$\left\{\begin{array}{c}rect.x=center.x-car\_width*0.7\\ rect.y=center.y-car\_height\\ rect.width=car\_width*1.4\\ rect.height=car\_height*2\end{array}\right.$

Wherein, rect.x, rect.y represent upper left corner horizontal ordinate and the ordinate of Plate searching rectangular area rect in current frame image respectively, center.x, center.y represent the position horizontal ordinate of the car plate central point of prediction in current frame image and ordinate respectively, rect.width, rect.height represent width and the height of rect respectively, and car_width, car_height represent width and the height of the car plate obtained in previous frame image respectively.

S54, accurately positioning licence plate position, main method is: in the Plate searching rectangular area of current frame image, plate template is used to carry out search spread, each travels through position, according to following formula, calculate the degree of confidence that this traversal position belongs to car plate position, the traversal position selecting maximum confidence corresponding is as the optimal location of car plate in current frame image:

${\mathrm{conf}}_{ij}=\frac{1}{255*N}\Σ\Σf(i+x,j+y)-M(x,y)$

Wherein, conf _ijrepresent that current traversal position belongs to the degree of confidence of car plate position, N represents the pixel quantity of the plate template of current frame image, M (x, y) represents the gray-scale value of plate template at pixel (x, y) place, f (i+x, j+y) represent in the Plate searching rectangular area of current frame image, with top left co-ordinate (i, j) place, current traversal position for benchmark, the grey scale pixel value at skew (x, y) place.

S6, the car plate center position obtained in current frame image, concrete steps are as follows:

S61, according to following formula, obtain the binary image bin (x, y) of the car plate band of position in current frame image, effect as shown in Figure 6:

$bin(x,y)=\left\{\begin{array}{cc}255& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\\ 0& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\end{array}\right.$

Wherein, f (x, y) is the grey scale pixel value at the car plate band of position (x, y) place in current frame image, f (x _i, y _i) be grey scale pixel value in N neighborhood centered by (x, y), n is the number of pixels in the N neighborhood centered by (x, y), and T is binary-state threshold.

S62, morphology operations, remove the interference that area is less.

S63, according to following formula, obtain the car plate center position center in current frame image, effect as shown in Figure 7:

$\left\{\begin{array}{c}center.x=\frac{\Σx_i*bin(x_i,y_i)}{\Σbin(x_i,y_i)}\\ center.y=\frac{\Σy_i*bin(x_i,y_i)}{\Σbin(x_i,y_i)}\end{array}\right.$

Wherein, center.x, center.y are horizontal ordinate and the ordinate of car plate center position in current frame image respectively, bin (x _i, y _i) be (x in binary image _i, y _i) grey scale pixel value at place.

The distance of S7, acquisition target vehicle, concrete steps are as follows:

S71, based on following Binocular vision photogrammetry principle type, obtain the three-dimensional coordinate of car plate center position in world coordinate system (X, Y, Z) in current frame image:

$\left\{\begin{array}{c}{s}_1\left[\begin{array}{c}{u}_1\\ v_1\\ 1\end{array}\right]=K_1\×\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\\ s_r\left[\begin{array}{c}{u}_r\\ v_r\\ 1\end{array}\right]=K_r\×\[R_r|T_r\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\end{array}\right.$

Wherein, s _l, K _l, R _l, T _lthe correlation parameter of first vehicle-mounted vidicon, (u _l, v _l) be car plate center position coordinate in the image that gathers of first vehicle-mounted vidicon, s _r, K _r, R _r, T _rthe correlation parameter of second vehicle-mounted vidicon, (u _r, v _r) be car plate center position coordinate in the image that gathers of second vehicle-mounted vidicon.

S72, carry out three-dimensional system of coordinate conversion according to following formula:

$\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\\ 1\end{array}\right]=\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]$

Wherein, (X, Y, Z) is the three-dimensional coordinate of car plate center position in world coordinate system that step S71 obtains, (X _c, Y _c, Z _c) be coordinate system with first vehicle-mounted vidicon three-dimensional coordinate, Z _crepresent the distance of car plate center position apart from first vehicle-mounted vidicon, as the distance of objects ahead vehicle apart from first vehicle-mounted vidicon.

S73, correction target vehicle distances, because vehicle-mounted vidicon is all generally be placed on vehicle interior, the target vehicle distance that step S72 obtains contains the distance of this car headstock apart from vehicle-mounted vidicon, therefore, according to following formula, obtain the distance Z of final this car headstock apart from the front truck tailstock:

Z＝Z _c-Z _car

Wherein, Z _cthe distance of vehicle-mounted vidicon apart from front truck, Z _carthe distance of vehicle-mounted vidicon apart from this car headstock.

S8, audio alert, according to the safe distance arranged in advance, if the spacing of reality is less than safe distance, audio alert, reminds driver to adjust spacing in time, safe driving.

The above embodiment is only be described the preferred embodiment of the present invention; not scope of the present invention is limited; under not departing from the present invention and designing the prerequisite of spirit; the various distortion that those of ordinary skill in the art make technical scheme of the present invention and improvement, all should fall in protection domain that claims of the present invention determine.

Claims (8)

1. based on a vehicle distance detecting method for depth survey, it is characterized in that, comprise the following steps:

(1) correlation parameter that car plate detects sorter and two vehicle-mounted vidicons is obtained;

(2) judge whether to need detection and location objects ahead vehicle again, if so, then perform step (3), if not, then perform step (4);

(3) detect car plate position in sorter and image based on the car plate obtained and, apart from the distance of image lower boundary, locate the objects ahead vehicle in the image of a wherein vehicle-mounted vidicon collection; Detect car plate location similarity in sorter and image based on car plate again, locate the objects ahead vehicle in the image of another vehicle-mounted vidicon collection;

(4) car plate of objects ahead vehicle in the image of two vehicle-mounted vidicons collections is followed the tracks of respectively, namely according to the car plate position of target vehicle in previous frame image, the car plate position of target vehicle in prediction current frame image;

(5) the car plate center position of target vehicle in the current frame image of two vehicle-mounted vidicons collections is obtained respectively;

(6) according to the correlation parameter of vehicle-mounted vidicon, obtain the three-dimensional coordinate of car plate center position in world coordinate system of target vehicle in the current frame image of two vehicle-mounted vidicons collections, be the three-dimensional coordinate of coordinate system with vehicle-mounted vidicon by described three-dimensional coordinate transformation one-tenth in world coordinate system, obtain the distance of target vehicle apart from vehicle-mounted vidicon;

(7) distance of described target vehicle distance vehicle-mounted vidicon is deducted the distance of vehicle-mounted vidicon apart from this car headstock, namely obtain the distance of target vehicle apart from this car headstock.

2. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, in step (1), described acquisition car plate detects sorter, comprising:

In a, positive and negative 30 degree of collection front, the license plate image of different distance is as the positive sample of training;

B, detect sorter file based on haar characteristic sum adaboost learning algorithm training car plate.

3. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, in step (1), the correlation parameter of described acquisition two vehicle-mounted vidicons, comprising:

A, in the common visual field of two vehicle-mounted vidicons, place the scaling board of a width gridiron pattern pattern, use two vehicle-mounted vidicons to gather scaling board image respectively;

The angle position of b, conversion scaling board, until two vehicle-mounted vidicons acquire five width scaling board images respectively;

C, on each width scaling board image, obtain all gridiron patterns two dimension angular coordinate;

D, based on the classical calibration algorithm of Zhang Zhengyou, obtain the correlation parameter formula of vehicle-mounted vidicon:

$K=s\left[\begin{array}{ccc}{f}_x\×k_x& 0& u_0\\ 0& f_y\×k_y& v_0\\ 0& 0& 1\end{array}\right]$

$R=\left[\begin{array}{ccc}{r}_1& r_2& r_3\\ r_4& r_5& r_6\\ r_7& r_8& r_9\end{array}\right]$

$T=\left[\begin{array}{c}{t}_1\\ t_2\\ t_3\end{array}\right]$

Wherein, K represents the Intrinsic Matrix of vehicle-mounted vidicon, f _x, f _yrepresent the effective focal length of camera lens, k _x, k _yrepresent distortion factor, (u ₀, v ₀) representing picture centre, s represents the uncertain graphical rule factor, and R, T are all outer parameter matrixs of vehicle-mounted vidicon, represents respectively from world coordinates and is tied to rotational transformation matrix between vehicle-mounted vidicon coordinate system and translation transformation matrix.

4. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, step (3), comprising:

A, based on obtain car plate detect sorter, detect first vehicle-mounted vidicon collection image in track, front whether there is car plate;

B, judge whether vehicle corresponding to car plate obtained is positioned at current lane, is if so, considered as effective car plate, if not, is considered as invalid car plate;

If c only exists an effective car plate, then it can be used as target license plate; If there is multiple effective car plate, then the target license plate in the image selecting car plate position in image to gather as first vehicle-mounted vidicon apart from minimum effective car plate apart from image lower boundary, vehicle corresponding to target license plate is target vehicle;

D, detect sorter based on the car plate obtained, obtains all candidate license plate in the image of second vehicle-mounted vidicon collection;

Target license plate in e, the image that gathers with first vehicle-mounted vidicon is for standard, all candidate license plate in the image of second vehicle-mounted vidicon collection are carried out Similarity Measure with it respectively, select candidate license plate corresponding to maximum similarity to be target license plate in the image that gathers of second vehicle-mounted vidicon, vehicle corresponding to target license plate is target vehicle.

5. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, in step (4), and the described car plate position according to target vehicle in previous frame image, the car plate position of target vehicle in prediction current frame image, comprising:

A, in previous frame image, the car plate central point of target vehicle is saved as impact point, and the car plate band of position of target vehicle is saved as the plate template of target vehicle in current frame image after surrounding expansion;

B, impact point to be followed the tracks of, based on kalman Filter Principle, the position of car plate central point in current frame image of target of prediction vehicle;

C, car plate size based on target vehicle in position in current frame image of the car plate central point of target vehicle of prediction and previous frame image, according to following formula, obtain the Plate searching rectangular area of target vehicle in current frame image:

$\left\{\begin{array}{c}rect.x=center.x-car\_width*0.7\\ rect.y=center.y-car\_height\\ rect.width=car\_width*1.4\\ rect.height=car\_height*2\end{array}\right.$

Wherein, rect.x, rect.y represent upper left corner horizontal ordinate and the ordinate of the Plate searching rectangular area rect of target vehicle in current frame image respectively, center.x, center.y represent the position horizontal ordinate of the car plate central point of the target vehicle of prediction in current frame image and ordinate respectively, rect.width, rect.height represent width and the height of the Plate searching rectangular area rect of target vehicle in current frame image respectively, and car_width, car_height represent car plate width and the height of target vehicle in previous frame image respectively;

D, in current frame image target vehicle Plate searching rectangular area in, plate template is used to carry out search spread, to each traversal position, according to following formula, calculate the degree of confidence that this traversal position belongs to target vehicle car plate position, the traversal position selecting maximum confidence corresponding is as the car plate position of target vehicle in current frame image:

${\mathrm{conf}}_{ij}=\frac{1}{255*N}\mathrm{\Σ}\mathrm{\Σ}f(i+x,j+y)-M(x,y)$

Wherein, conf _ijrepresent that current traversal position belongs to the degree of confidence of target vehicle car plate position, N represents the pixel quantity of the plate template of target vehicle in current frame image, M (x, y) represent that the plate template of target vehicle in current frame image is at (x, y) grey scale pixel value at place, f (i+x, j+y) represent in the Plate searching rectangular area of target vehicle in current frame image, with current traversal position top left co-ordinate (i, j) place is benchmark, the grey scale pixel value at skew (x, y) place.

6. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, step (5), comprising:

A, according to following formula, obtain the binary image bin (x, y) of the car plate band of position of target vehicle in current frame image:

$bin(x,y)=\left\{\begin{array}{cc}255& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\\ 0& f(x,y)-\frac{1}{n}{\Σ}_{i=1}^{N}f(x_i,y_i)\≥T\end{array}\right.$

Wherein, f (x, y) represents the grey scale pixel value at the car plate band of position (x, the y) place of target vehicle in current frame image, f (x _i, y _i) representing grey scale pixel value in N neighborhood centered by (x, y), n represents the number of pixels in the N neighborhood centered by (x, y), and T represents binary-state threshold;

B, to obtain binary image carry out morphology operations, remove interference;

C, obtain the car plate center position of target vehicle in current frame image according to following formula:

$\left\{\begin{array}{c}center.x=\frac{{\mathrm{\Σx}}_i*bin(x_i.y_i)}{\mathrm{\Σ}bin(x_i.y_i)}\\ center.y=\frac{{\mathrm{\Σy}}_i*bin(x_i,y_i)}{\mathrm{\Σ}bin(x_i.y_i)}\end{array}\right.$

Wherein, center.x, center.y represent horizontal ordinate and the ordinate of the car plate center position of target vehicle in current frame image respectively, bin (x _i, y _i) represent (x in binary image _i, y _i) grey scale pixel value at place.

7. the vehicle distance detecting method based on depth survey according to claim 1, is characterized in that, step (6), comprising:

A, based on following Binocular vision photogrammetry principle type, obtain the three-dimensional coordinate of car plate center position in world coordinate system (X, Y, Z) of target vehicle in the current frame image that two vehicle-mounted vidicons gather:

$\left\{\begin{array}{c}{s}_1\left[\begin{array}{c}{u}_1\\ v_1\\ 1\end{array}\right]=K_1\×\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\\ s_r\left[\begin{array}{c}{u}_r\\ v_r\\ 1\end{array}\right]=K_r\×\[R_r|T_r\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]\end{array}\right.$

Wherein, s ₁, K ₁, R ₁, T ₁be the correlation parameter of first vehicle-mounted vidicon, (u ₁, v ₁) be the car plate center position coordinate of target vehicle in the current frame image that gathers of first vehicle-mounted vidicon, s _r, K _r, R _r, T _rbe the correlation parameter of second vehicle-mounted vidicon, (u _r, v _r) be the car plate center position coordinate of target vehicle in the current frame image that gathers of second vehicle-mounted vidicon;

B, according to following formula, described three-dimensional coordinate transformation in world coordinate system is become the three-dimensional coordinate being coordinate system with first vehicle-mounted vidicon, obtains the distance of target vehicle apart from first vehicle-mounted vidicon:

$\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\\ 1\end{array}\right]=\[R_1|T_1\]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]$

Wherein, (X, Y, Z) represents the three-dimensional coordinate of car plate center position in world coordinate system of target vehicle in the current frame image that above-mentioned steps a obtains, (X _c, Y _c, Z _c) represent the three-dimensional coordinate being coordinate system with first vehicle-mounted vidicon, Z _crepresent the distance of car plate center position apart from first vehicle-mounted vidicon of target vehicle in current frame image, as the distance of objects ahead vehicle apart from first vehicle-mounted vidicon.

8. the vehicle distance detecting method based on car plate position according to claim 1, is characterized in that, also comprise: when described target vehicle is less than default safe distance apart from the distance of this car headstock time, carry out audio alert.