CN106370884A - Vehicle speed measurement method based on binocular camera computer vision technology - Google Patents

Abstract

The invention relates to a vehicle speed measurement technology in the field of computer vision, and discloses a vehicle speed measurement method based on a binocular camera computer vision technology, for solving the problem of vehicle speed measurement inaccuracy in the prior art. The method comprises the following steps: A, acquiring a vehicle driving video image by use of a binocular camera; B, performing detection and tracking positioning on a vehicle target in the video image; and C, obtaining coordinate information of a driving vehicle in realistic three-dimensional space, obtaining a real distance which a vehicle pass during driving, and accordingly, calculating a driving speed of the vehicle. The method is applied to real-time and accurate speed measurement of the vehicle.

Description

Vehicle speed measuring method based on binocular camera computer vision technique

Technical field

The present invention relates to the vehicle speed measuring technology in computer vision field is and in particular to a kind of calculated based on binocular camera The vehicle speed measuring method of machine vision technique.

Background technology

Existing vehicle speed measuring is based primarily upon laser, radar and image/video and monitors this few class method:

Laser velocimeter system, by repeatedly being found range to vehicle, obtains Vehicle Speed with the ratio of time of measuring.Should Method requires velocity-measuring system just to range accuracy highest during direction of vehicle movement, and when there is measured deviation angle, measurement is accurate Exactness is big and decline with the change of measured deviation angle, therefore, it is difficult to meet actual vehicle testing the speed demand.

Radar velocity measurement system application Doppler effect, when there is relative motion relative to radar velocity measurement system in vehicle, radar Echo produces Doppler frequency change, utilizes Doppler frequency and speed further according to radar system and vehicle heading angled relationships Degree formula is calculated Vehicle Speed.During vehicle speed measuring, vehicle can not be considered as simply putting mesh relative to radar velocity measurement system Mark, therefore vehicle each several part all reflects radar echo signal, and Doppler frequency has differences, impact radar velocity measurement system Accuracy.

Current image/video monitoring vehicle speed measuring is based primarily upon monocular camera image processing techniquess, by following two sides Formula carries out vehicle speed measuring:

Measuring vehicle move distance in known time section obtains Vehicle Speed: by setting in shooting area in advance Vertical gauge scope, such as demarcates the distance between certain two baseline in advance, then obtains vehicle by baseline by camera shooting Between before and after time difference, be calculated Vehicle Speed；

Measuring vehicle obtains Vehicle Speed by known distance required time: vehicle in known time section is being shot The change of image internal coordinate is compared with image internal standard length coordinate, and after being converted, true operating range is calculated car again Travel speed.

The vehicle speed measuring method based on monocular camera computer vision technique for the both the above when obtaining vehicle operating range, The former due to hardware limitation it is impossible to ensure vehicle turnover baseline moment just obtained by cameras capture, the time difference obtaining It is not the time difference truly passing through distance between baseline；The latter ensure that the accuracy of time difference, but because image is clapped Take the photograph and can there is perspective effect, the distant view of same coordinate difference and the actual distance representated by close shot have larger difference, therefore coordinate There is larger error in the vehicle operating range that the mode that change compares obtains.

Content of the invention

The technical problem to be solved is: proposes a kind of vehicle based on binocular camera computer vision technique and surveys Fast method, solves the inaccurate problem of vehicle speed measuring in conventional art.

The scheme that present invention solution above-mentioned technical problem is adopted is:

Based on the vehicle speed measuring method of binocular camera computer vision technique, comprise the following steps:

A, using binocular camera collection vehicle travel video image；

B, the vehicle target in video image is detected and track and localization；

C, acquisition coordinate information in reality three-dimensional for the driving vehicle, obtain the actual distance that vehicle drives through, Thus calculating the speed of vehicle traveling.

As optimizing further, in step a, before travelling video image using binocular camera collection vehicle, to binocular Camera parameter is demarcated.

As optimizing further, in step b, using adaboost method by way of training Weak Classifier cascade to car Detected, vehicle tracking is carried out using kernelized correlation filters (kcf) algorithm.

As optimizing further, in step c, described acquisition coordinate information in reality three-dimensional for the driving vehicle, obtain The actual distance driving through to vehicle, thus the method calculating the speed of vehicle traveling is:

First, Stereo matching is carried out to the left images of synchronization collection and obtain disparity map, the corresponding picture in disparity map The parallax d of vegetarian refreshments p is that it corresponds to abscissa x in the imaging point of the left and right plane of delineation respectively_left、x_rightDifference:

D=x_left-x_right(1)

Baseline distance b is the distance of the projection centre line of two cameras, represents the focal length of camera with f, former according to triangulation Reason, obtains triangle geometrical relationship:

$\left\{\begin{array}{c}{x}_{left}=f\frac{x}{z}\\ x_{right}=f\frac{(x-b)}{z}\\ y_{left}=y_{right}=f\frac{y}{z}\\ d=x_{left}-x_{right}\end{array}\right.---\left(2\right)$

It is calculated the D coordinates value (x, y, z) of pixel p:

$(x,y,z)=(\frac{{\mathrm{bx}}_{left}}{d},\frac{{\mathrm{by}}_{eft}}{d},\frac{bf}{d})---\left(3\right)$

Using previous moment shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate Position,

True three-dimension coordinate (the x in previous moment for the vehicle is obtained by formula (1) (2) (3)₀,y₀,z₀)；

Using later moment in time shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate Position, calculates the true three-dimension coordinate (x obtaining vehicle in later moment in time by formula (1) (2) (3)₁,y₁,z₁)

Then, the true operating range between the moment before and after calculating acquisition vehicle:

$d=\sqrt{{(x_0-x_1)}^2+{(y_0-y_1)}^2+{(z_0-z_1)}^2}---\left(4\right)$

Time point t finally according to previous moment₀, the time point t of later moment in time₁The travel speed of calculating vehicle:

$v=\frac{d}{\left|t_0-t_1\right|}---\left(5\right).$

The invention has the beneficial effects as follows:

The present invention obtains coordinate information in reality three-dimensional for the driving vehicle by binocular camera, can be calculated Driving vehicle real operating range during collected by camera, this distance is directly calculated according to coordinate, therefore more accelerates Speed, more precisely, improves real-time and the accuracy of vehicle speed measuring effectively.

Brief description

Fig. 1 is the vehicle speed measuring method general flow chart based on binocular camera computer vision technique；

Fig. 2 is binocular camera principle of triangulation figure.

Specific embodiment

Binocular camera is the embodiment that mankind's eyes function extends, and the mankind pass through eyes on the basis of obtaining external image, The three dimensional depth of external object can also be perceived simultaneously.And external image under binocular camera collection synchronization, regard through computer Feel algorithm process, equally can obtain the three-dimensional coordinate information of external object.Compare other side in order to solve monocular camera changes in coordinates The larger problem of formula computed range error, the present invention obtains the three-dimensional coordinate information of object by binocular camera, using several where Method calculates operating range, can be greatly reduced by time difference institute in the way of free access time point calculates time difference simultaneously The impact bringing.

As shown in figure 1, being included based on the vehicle speed measuring method of binocular camera computer vision technique in the present invention is following Step:

1) video acquisition:

Before this step, need for binocular camera to be arranged on the accuracy that road is other (on street lamp), measures for guarantee, Binocular camera is carried out demarcating its parameter；When using, travel video image using binocular camera collection vehicle, obtain video sequence Row image incoming moving vehicles detection and tracking module.

2) moving vehicles detection and tracking:

Moving vehicles detection and tracking module is detected and track and localization to the vehicle target in video image；

Moving vehicles detection and tracking is the basis realizing binocular range finding.Detect firstly the need of to the vehicle in scene, only Have in the case of vehicle target is detected and just can carry out the positioning to vehicle；Afterwards the vehicle detecting is tracked, with The coordinate of track to change is used for calculating the distance that vehicle was travelled within the time period followed the tracks of.The present invention is using classics Adaboost method is detected to vehicle by way of training Weak Classifier cascade, then adopts kernelized Correlation filters (kcf) algorithm carries out vehicle tracking.

Because the vehicle detection employed in the present invention and track algorithm are this area conventional algorithm, here no longer to it Implement and repeated.

3) binocular measurement and calculating:

In this step, seat in reality three-dimensional for the driving vehicle is obtained with parallax calculation method by binocular ranging Mark information, obtains the actual distance that vehicle drives through, thus calculating the speed of vehicle traveling, specific means is as follows:

First, Stereo matching is carried out to the left images of synchronization collection and obtain disparity map, as shown in Fig. 2 disparity map In corresponding pixel points p parallax d be its respectively the left and right plane of delineation imaging point correspond to abscissa x_left、x_rightDifference Value:

D=x_left-x_right(1)

Baseline distance b is the distance of the projection centre line of two cameras, represents the focal length of camera with f, former according to triangulation Reason, obtains triangle geometrical relationship:

$\left\{\begin{array}{c}{x}_{left}=f\frac{x}{z}\\ x_{right}=f\frac{(x-b)}{z}\\ y_{left}=y_{right}=f\frac{y}{z}\\ d=x_{left}-x_{right}\end{array}\right.---\left(2\right)$

It is calculated the D coordinates value (x, y, z) of pixel p:

$(x,y,z)=(\frac{{\mathrm{bx}}_{left}}{d},\frac{{\mathrm{by}}_{left}}{d},\frac{bf}{d})---\left(3\right)$

Using previous moment shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate Position, using this coordinate position as above-mentioned pixel p, obtains vehicle true the three of previous moment by formula (1) (2) (3) Dimension coordinate (x₀,y₀,z₀)；

Using later moment in time shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate Position, using this coordinate position as above-mentioned pixel p, calculates acquisition vehicle by formula (1) (2) (3) true in later moment in time Real three-dimensional coordinate (x₁,y₁,z₁)；

Then, the true operating range between the moment before and after calculating acquisition vehicle:

$d=\sqrt{{(x_0-x_1)}^2+{(y_0-y_1)}^2+{(z_0-z_1)}^2}---\left(4\right)$

Time point t finally according to previous moment₀, the time point t of later moment in time₁The travel speed of calculating vehicle:

$v=\frac{d}{\left|t_0-t_1\right|}---\left(5\right).$

According to above-mentioned implementation, the present invention can be by the three-dimensional coordinate of vehicle geometric center between different frame number images Change, calculates the distance of vehicle movement in different frame number images, and according to the corresponding time, calculates the travel speed of vehicle, from And the Vehicle Speed between random time point can be calculated, to solve to test the speed and can not protect using monocular camera in conventional art Card time of measuring difference is as differed from by the actual time of in front and back's baseline and brings the problem of measurement error.

Claims (4)

1. the vehicle speed measuring method based on binocular camera computer vision technique is it is characterised in that comprise the following steps:

A, using binocular camera collection vehicle travel video image；

B, the vehicle target in video image is detected and track and localization；

C, acquisition coordinate information in reality three-dimensional for the driving vehicle, obtain the actual distance that vehicle drives through, thus Calculate the speed of vehicle traveling.

2. the vehicle speed measuring method based on binocular camera computer vision technique as claimed in claim 1 is it is characterised in that walk In rapid a, before travelling video image using binocular camera collection vehicle, binocular camera parameter is demarcated.

3. the vehicle speed measuring method based on binocular camera computer vision technique as claimed in claim 1 is it is characterised in that walk In rapid b, vehicle is detected by way of training Weak Classifier cascade using adaboost method, using kernelized Correlation filters algorithm carries out vehicle tracking.

4. the vehicle speed measuring method based on binocular camera computer vision technique as claimed in claim 1 is it is characterised in that walk In rapid c, described acquisition coordinate information in reality three-dimensional for the driving vehicle, obtain the actual distance that vehicle drives through, Thus the method calculating the speed of vehicle traveling is:

First, Stereo matching is carried out to the left images of synchronization collection and obtain disparity map, corresponding pixel points p in disparity map Parallax d be its respectively the left and right plane of delineation imaging point correspond to abscissa x_left、x_rightDifference:

D=x_left-x_right(1)

Baseline distance b is the distance of the projection centre line of two cameras, represents the focal length of camera with f, according to principle of triangulation, obtains To triangle geometrical relationship:

$\left\{\begin{array}{c}{x}_{left}=f\frac{x}{z}\\ x_{right}=f\frac{(x-b)}{z}\\ y_{left}=y_{right}=f\frac{y}{z}\\ d=x_{left}-x_{right}\end{array}\right.---\left(2\right)$

It is calculated the D coordinates value (x, y, z) of pixel p:

$(x,y,z)=(\frac{{\mathrm{bx}}_{left}}{d},\frac{{\mathrm{by}}_{left}}{d},\frac{bf}{d})---\left(3\right)$

Using previous moment shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate position,

True three-dimension coordinate (the x in previous moment for the vehicle is obtained by formula (1) (2) (3)₀,y₀,z₀)；

Using later moment in time shoot vehicle image in driving vehicle geometric center as vehicle plane picture coordinate position, Calculate the true three-dimension coordinate (x obtaining vehicle in later moment in time by formula (1) (2) (3)₁,y₁,z₁)

Then, the true operating range between the moment before and after calculating acquisition vehicle:

$d=\sqrt{{(x_0-x_1)}^2+{(y_0-y_1)}^2+{(z_0-z_1)}^2}---\left(4\right)$

Time point t finally according to previous moment₀, the time point t of later moment in time₁The travel speed of calculating vehicle:

$v=\frac{d}{\left|t_0-t_1\right|}---\left(5\right).$