CN103456171B - A kind of based on fish-eye vehicle flow detection system, method and method for correcting image - Google Patents

Abstract

The invention discloses a kind of based on fish-eye vehicle flow detection system, method and method for correcting image, comprise image collecting device, vehicle Flow Detection module and data storage device, vehicle Flow Detection module connects image collecting device and data storage device respectively, image collecting device comprises integrated fish eye lens and sensor, and sensor connects vehicle Flow Detection module by image rectification chip.The present invention's design based on fish-eye vehicle flow detection system and method for correcting image, the scene solving the monitoring of existing vehicle flowrate monitoring system is abundant not, and construction costs high in cost of production problem.

Description

A kind of based on fish-eye vehicle flow detection system, method and method for correcting image

Technical field

The invention belongs to a kind of vehicle flow detection system, be specifically related to based on fish-eye vehicle flow detection system, method and method for correcting image.

Background technology

Existing monitoring system adopts single common view angle camera lens, monitors vehicle flowrate; Adopt the system of common lens, the scene of monitoring is abundant not, and current flux monitoring system mainly operates on industrial computer, so not only adds construction costs cost, and needs larger industrial computer equipment to process.

Summary of the invention

In order to the scene solving the monitoring of existing vehicle flowrate monitoring system is abundant not, and construction costs high in cost of production problem, the present invention devises a kind of based on fish-eye vehicle flow detection system and method for correcting image.

A kind of based on fish-eye vehicle flow detection system, comprise image collecting device, vehicle Flow Detection module and data storage device, vehicle Flow Detection module connects image collecting device and data storage device respectively, image collecting device comprises integrated fish eye lens and sensor, and sensor connects vehicle Flow Detection module by image rectification chip.Its advantage is: system of the present invention comprises three part compositions, Part I image collecting device, and this part is primarily of fish eye lens and sensor imaging moiety composition; Part II vehicle Flow Detection module realizes checking vehicle flowrate in this part; Part III data memory device, stores the data after Part II process in this part.There is the vehicle flowrate that fish-eye system has monitoring road surface, better visual angle; The scope checked is more wide, and native system does not need the checkout facility added, and reduces cost.

Described based on fish-eye vehicle flow detection system, image rectification chip is fpga chip, and vehicle Flow Detection module is DSP data processing module.Its advantage is: after fish eye lens collected by camera to view data, through correction algorithm process, the image restoring of distortion is become normal picture, sends to vehicle Flow Detection module.Fish eye lens correct process in fpga chip, need to open up one piece of internal memory cache region in fpga chip inside, this region be used for store correct after view data.First, calculate the corresponding relation of the coordinate points on the position of coordinate points of two field picture after correcting and former two field picture, then the numerical value assignment on the position of this on former two field picture in the corresponding point of frame after the rectification of inside, buffer zone, then dispose this frame internal storage data and carry out next frame process.Image after data processing module receives video acquisition module rectification carries out vehicle flowrate analysis.A panel region in this module intercepts image is analyzed, and vehicle sails this region into and is not more than general vehicle length in the picture with the distance leaving this region, and this region needs to demarcate according to actual scene.

A kind of based on fish-eye traffic flow detecting method, comprise the following steps:

Step one, fish eye lens acquisition of image data, and be sent to FPGA image rectification chip by sensor;

Step 2, FPGA image rectification chip corrects fault image;

Step 3, the self-adaption binaryzation module in DSP data processing module is by the yuv data image binaryzation process after correction;

Step 4, DSP data processing module adopts gradient vehicle local feature to detect and whether HOG vehicle characteristics matching process joint-detection has vehicle to pass through;

Step 5, by the image information containing vehicle stored in data storage device.

Described based on fish-eye traffic flow detecting method, in step 3, the gray-scale value in image is set to 0 or 1 by self-adaption binaryzation module, and the setting of threshold value adopts the method for adaptive threshold.

Described based on fish-eye traffic flow detecting method, in step 4, gradient information process adopts the pattern of one dimension or two dimension.

Described based on fish-eye traffic flow detecting method, HOG vehicle characteristics matching process in step 4, adopts the HOG feature of headstock or tailstock model, carries out off-line training and classification with SVM support vector machine linear classifier.

Its advantage is: after fish eye lens collected by camera to view data, corrects correcting algorithm process, the image restoring of distortion is become normal picture, sends to vehicle Flow Detection module through fisheye image.Self-adaption binaryzation module is by the yuv data image binaryzation process after correction.Gray-scale value in image is set to 0 or 1, and the setting of threshold value adopts the method for adaptive threshold.Whether the detection of gradient vehicle local feature and HOG vehicle characteristics matching process join together to detect surveyed area in this two field picture has vehicle to pass through.There is abundant gradient information vehicle local, the information of vehicles in the analysis surveyed area that employing the method can be rough.Gradient information process can adopt the pattern of one dimension or two dimension, and the higher need of the dimension time to be processed is longer.HOG vehicle characteristics matching process adopts the HOG feature of headstock or tailstock model, carries out off-line training and classification with SVM support vector machine linear classifier.Which image pattern contains vehicle to utilize this feature to judge, those image patterns do not have vehicle, thus carries out accurate vehicle flowrate.Image data information containing vehicle is filed, returns and start to carry out data analysis next time.

A kind of fish-eye method for correcting image, comprises the following steps:

(1) according to the spherical co-ordinate model in panorama picture of fisheye lens principle, set up XYZ space rectangular coordinate system, wherein, P is for inciding fish-eye light; F1 is fish-eye focal length; The angle of light P and Z axis is θ; A (x1, y1) for light P through common lens perfect condition fall x-axis y-axis composition plane on point, form angle β with Z axis, form angle ψ with X-axis, the distance of A point and central point is R; K is projective invariant coefficient; B (x2, y2) falls the point in the plane of x-axis y-axis composition after fish eye lens refraction for P, form angle α with Z axis, the distance being ψ, B point and central point with the angle of X-axis is r;

(2) by the conversion relation of β and θ and θ and α, α and β angle is determined; The conversion relation formula of β and θ is:

sinθ＝sinβ①

The refraction law formula of fish eye lens in plane limited range is as follows:

sinθ＝k*sinα②

1. 2. obtained with formula by formula

sinβ＝k*sinα③

α and β angle is

$a={\tan }^{-1}\frac{r}{f_1}$ ④

$\mathrm{\β}={\tan }^{-1}\frac{R}{f_1}$ ⑤

(3) relation of R and r 3., 4. and is 5. obtained by the formula in step (2):

$R=f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{r}{f_1}\right)\right)*k$ ⑥

(4) by B point from two-dimensional assemblage polar coordinates formula:

$r=\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}$ ⑦

$\mathrm{\ψ}={\tan }^{-1}\left(\frac{y_2}{x_2}\right)$ ⑧

(5) coordinate of the A (x1, y1) that light P obtains under conventional lens shooting is determined:

x ₁＝R*cos(ψ)⑨

y ₁＝R*sin(ψ)⑩

(6) 6. the range formula of R is 7. obtained with formula by formula:

$R=f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k$

(7) according to formula 8., 9., 10., obtain the coordinate transformation relation of A point and B point:

$x_1=(f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k)*\cos \left({\tan }^{-1}\left(\frac{y_2}{x_2}\right)\right)$

$y_1=(f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k)*\sin \left({\tan }^{-1}\left(\frac{y_2}{x_2}\right)\right)$

Utilize this relation preferably the image restoring of the distortion of fisheye camera shooting can be become normal picture.Then normal picture after conversion is delivered to DSP data processing module and is carried out vehicle flowrate calculating.Fish eye lens correction algorithm is exactly solve the pixel on former frame and the coordinate points corresponding relation after correcting on image.

Accompanying drawing explanation

Fig. 1 is fish-eye vehicle flow detection system principle;

Fig. 2 is fish-eye vehicle flow detection system processing procedure;

Fig. 3 is the correction process schematic diagram of flake video in FPGA inside;

Fig. 4 is video mode vehicle Flow Detection schematic diagram;

Fig. 5 is vehicle flow detection algorithm process flow diagram;

Fig. 6 is fish eye lens correcting algorithm three-dimensional imaging figure;

Fig. 7 is fish eye lens correcting algorithm planar imaging figure.

Embodiment

Below in conjunction with accompanying drawing, explanation is explained in detail to structure of the present invention, if Fig. 1 is fish-eye vehicle flow detection system principle; A kind of based on fish-eye vehicle flow detection system, comprise image collecting device, vehicle Flow Detection module and data storage device, vehicle Flow Detection module connects image collecting device and data storage device respectively, image collecting device comprises integrated fish eye lens and sensor, and sensor connects vehicle Flow Detection module by image rectification chip.If Fig. 2 is fish-eye vehicle flow detection system processing procedure; Described based on fish-eye vehicle flow detection system, image rectification chip is fpga chip, and vehicle Flow Detection module is DSP data processing module.

A kind of based on fish-eye traffic flow detecting method, comprise the following steps:

Step one, fish eye lens acquisition of image data, and be sent to FPGA image rectification chip by sensor;

Step 2, FPGA image rectification chip corrects fault image;

Step 3, the self-adaption binaryzation module in DSP data processing module is by the yuv data image binaryzation process after correction;

Step 4, DSP data processing module adopts gradient vehicle local feature to detect and whether HOG vehicle characteristics matching process joint-detection has vehicle to pass through;

Step 5, by the image information containing vehicle stored in data storage device.

Described based on fish-eye traffic flow detecting method, in step 3, the gray-scale value in image is set to 0 or 1 by self-adaption binaryzation module, and the setting of threshold value adopts the method for adaptive threshold.

Described based on fish-eye traffic flow detecting method, in step 4, gradient information process adopts the pattern of one dimension or two dimension.

Described based on fish-eye traffic flow detecting method, HOG vehicle characteristics matching process in step 4, adopts the HOG feature of headstock or tailstock model, carries out off-line training and classification with SVM support vector machine linear classifier.

A kind of fish-eye method for correcting image, comprises the following steps:

(1) according to the spherical co-ordinate model in panorama picture of fisheye lens principle, set up XYZ space rectangular coordinate system, wherein, P is for inciding fish-eye light; F1 is fish-eye focal length; The angle of light P and Z axis is θ; A (x1, y1) for light P through common lens perfect condition fall x-axis y-axis composition plane on point, form angle β with Z axis, form angle ψ with X-axis, the distance of A point and central point is R; K is projective invariant coefficient; B (x2, y2) falls the point in the plane of x-axis y-axis composition after fish eye lens refraction for P, form angle α with Z axis, the distance being ψ, B point and central point with the angle of X-axis is r;

(2) by the conversion relation of β and θ and θ and α, α and β angle is determined; The conversion relation formula of β and θ is:

sinθ＝sinβ①

The refraction law formula of fish eye lens in plane limited range is as follows:

sinθ＝k*sinα②

1. 2. obtained with formula by formula

sinβ＝k*sinα③

α and β angle is

$a={\tan }^{-1}\frac{r}{f_1}$ ④

$\mathrm{\β}={\tan }^{-1}\frac{R}{f_1}$ ⑤

(3) relation of R and r 3., 4. and is 5. obtained by the formula in step (2): if Fig. 7 is fish eye lens correcting algorithm planar imaging figure;

$R=f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{r}{f_1}\right)\right)*k$ ⑥

(4) by B point from two-dimensional assemblage polar coordinates formula:

$r=\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}$ ⑦

$\mathrm{\ψ}={\tan }^{-1}\left(\frac{y_2}{x_2}\right)$ ⑧

(5) coordinate of the A (x1, y1) that light P obtains under conventional lens shooting is determined:

x ₁＝R*cos(ψ)⑨

y ₁＝R*sin(ψ)⑩

(6) 6. the range formula of R is 7. obtained with formula by formula:

$R=f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k$

(7) according to formula 8., 9., 10., obtain the coordinate transformation relation of A point and B point:

$x_1=(f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k)*\cos \left({\tan }^{-1}\left(\frac{y_2}{x_2}\right)\right)$

$y_1=(f_1*\tan ({\sin }^{-1}(\sin \left({\tan }^{-1}\left(\frac{\sqrt{{\left(x_2\right)}^2+{\left(y_2\right)}^2}}{f_1}\right)\right)*k)*\sin \left({\tan }^{-1}\left(\frac{y_2}{x_2}\right)\right)$

After fish eye lens collected by camera to view data, first, calculate the corresponding relation of the coordinate points on the position of coordinate points of two field picture after correcting and former two field picture, then the numerical value assignment on the position of this on former two field picture in the corresponding point of frame after the rectification of inside, buffer zone, then dispose this frame internal storage data and carry out next frame process.Fish eye lens correction algorithm is exactly solve the pixel on former frame and the coordinate points corresponding relation after correcting on image, and correcting principle figure as shown in Figure 3; Correct correcting algorithm process through fisheye image, the image restoring of distortion is become normal picture, sends to vehicle Flow Detection module.Image after data processing module receives video acquisition module rectification carries out vehicle flowrate analysis.A panel region in this module intercepts image is analyzed, if Fig. 4 is video mode vehicle Flow Detection schematic diagram; Vehicle sails this region into and is not more than general vehicle length in the picture with the distance H leaving this region, and this region needs to demarcate according to actual scene.Self-adaption binaryzation module is by the yuv data image binaryzation process after correction.Gray-scale value in image is set to 0 or 1, and the setting of threshold value adopts the method for adaptive threshold.Whether the detection of gradient vehicle local feature and HOG vehicle characteristics matching process join together to detect surveyed area in this two field picture has vehicle to pass through.There is abundant gradient information vehicle local, the information of vehicles in the analysis surveyed area that employing the method can be rough.Gradient information process can adopt the pattern of one dimension or two dimension, and the higher need of the dimension time to be processed is longer.HOG vehicle characteristics matching process adopts the HOG feature of headstock or tailstock model, carries out off-line training and classification with SVM support vector machine linear classifier.Which image pattern contains vehicle to utilize this feature to judge, those image patterns do not have vehicle, thus carries out accurate vehicle flowrate.Image data information containing vehicle is filed, returns and start to carry out data analysis next time.If Fig. 5 is vehicle flow detection algorithm process flow diagram.

Technique scheme only embodies the optimal technical scheme of technical solution of the present invention, and those skilled in the art all embody principle of the present invention to some variations that wherein some part may be made, and belong within protection scope of the present invention.

Claims (4)

1. based on a fish-eye traffic flow detecting method, it is characterized in that, comprise the following steps:

Step one, fish eye lens acquisition of image data, and be sent to FPGA image rectification chip by sensor;

Step 2, FPGA image rectification chip corrects fault image;

Step 3, the self-adaption binaryzation module in DSP data processing module is by the yuv data image binaryzation process after correction;

Step 4, DSP data processing module adopts gradient vehicle local feature to detect and whether HOG vehicle characteristics matching process joint-detection has vehicle to pass through;

Step 5, by the image information containing vehicle stored in data storage device.

2. according to claim 1ly it is characterized in that based on fish-eye traffic flow detecting method, in step 3, the gray-scale value in image is set to 0 or 1 by self-adaption binaryzation module, and the setting of threshold value adopts the method for adaptive threshold.

3. according to claim 1ly it is characterized in that based on fish-eye traffic flow detecting method, in step 4, gradient information process adopts the pattern of one dimension or two dimension.

4. according to claim 1 based on fish-eye traffic flow detecting method, it is characterized in that, HOG vehicle characteristics matching process in step 4, adopts the HOG feature of headstock or tailstock model, carries out off-line training and classification with SVM support vector machine linear classifier.