CN106846328A - A kind of tunnel brightness detection method based on video - Google Patents

Abstract

The invention discloses a kind of tunnel brightness detection method based on video, its step includes：Propose for ensureing that used video camera has the video image stabilization detection algorithm of good stability, the method demarcated by gridiron pattern carries out distortion correction treatment to the video image that video camera shoots, realize the detection and correction to the uniformity of camera image sensor, extract the starting color background image of video, the extraction of pictorial element is carried out using the computational methods of the row pixel of two row of constant interval two after accurate setting detection zone, determine the brightness actually measured and the pictorial element R for detecting, G, the brightness detection algorithm of relation between B.The scheme of the tunnel brightness detection system based on video proposed by the present invention can carry out continuous brightness detection, and can provide real-time brightness value for illumination monitoring computer using the continuous videos picture frame of camera acquisition to surface layer.

Description

A kind of tunnel brightness detection method based on video

Technical field

The present invention relates to Computer Vision, visibility detection technique field, and in particular to a kind of tunnel based on video Brightness detection method.

Background technology

In tunnel system design and engineering construction, illuminate in occupation of considerable status.Because tunnel environment phase To obturation, the same irradiation that can directly receive sunshine of common road is not as, the brightness in tunnel is relative to ambient light For can be relatively low.Therefore, tunnel needs to provide suitable vision light conditions for driver by means of artificial light, and this will Ask illuminator to disclosure satisfy that the optimal bright demand that driver is adapted to, to ensure traffic safety, reduce traffic accident and occur Rate.In tunnel Mechatronic Systems, illuminator is energy consumption highest part, and energy is reduced on the premise of vehicle safe driving is ensured Consumption is an important goal of energy-conservation.A kind of effective method that real-time photocontrol is reducing energy consumption is carried out to illumination, Photocontrol depends on the real-time measurement values of brightness inside and outside Tunnel.Therefore, tunnel internal brightness of illumination is detected to reducing Energy consumption and ensure that traffic safety plays an important role.

For tunnel illumination brightness test problems, many research is carried out both at home and abroad and has been made certain gains.The patent No. CN202396058U Chen Rong, woods write the highway tunnel illumination brightness adaptive control systems such as favour, in various software algorithm logic point Under analysis treatment can lamp brightness in real time or in timing controlled tunnel, realize tunnel illumination brightness Based Intelligent Control, meet tunnel row The Self Adaptive Control of the maximum energy-conservation under car safety requirements, but, do not consider how to detect tunnel internal brightness problem.

The content of the invention

For above-mentioned problems of the prior art, it is an object of the present invention to propose a kind of tunnel based on video Brightness detection method, can simultaneously carry out tunnel brightness detection and the detection of tunnel luminance uniformity, while extending tunnel brightness inspection The scope of application of method of determining and calculating.

In order to realize above-mentioned task, the present invention uses following technical scheme：

The present invention provides a kind of stability judging method of video image, concretely comprises the following steps：

The stability of the video image of collection is prejudged according to video image stabilization detection algorithm, using following public affairs Formula,

Wherein, R, G, B represent three color components of red, green, blue of each pixel of video image, and MSE represents collection Three total mean square deviations of color component of video image, MSE_RRepresent the mean square deviation of each pixel color component of video image R, MSE_G Represent the mean square deviation of each pixel color component of video image G, MSE_BRepresent that each pixel color component of video image B's is square Difference, i represents the i-th two field picture in the video image of collection, i=1,2,3 ..., n, n >=1, R_i、G_i、B_iRespectively the i-th two field picture The value of R, G, B of all pixels point, The R of all images in the video image for respectively gathering_i、G_i、B_iIt is equal Value；

If MSE, MSE_R、MSE_GAnd MSE_BValue is respectively less than 0.5, then the video image stabilization sexual satisfaction requirement for gathering, is steady Fixed video image, otherwise resurveys video image.

The present invention also provides a kind of tunnel brightness detection method based on video, including：

Collection video image, the stabilization of the video image of collection is prejudged according to video image stabilization detection algorithm Property, using equation below,

Wherein, R, G, B represent three color components of red, green, blue of each pixel of video image, and MSE represents collection Three total mean square deviations of color component of video image, MSE_RRepresent the mean square deviation of each pixel color component of video image R, MSE_G Represent the mean square deviation of each pixel color component of video image G, MSE_BRepresent that each pixel color component of video image B's is square Difference, i represents the i-th two field picture in the video image of collection, i=1,2,3 ..., n, n >=1, R_i、G_i、B_iRespectively the i-th two field picture The value of R, G, B of all pixels point, The R of all images in the video image for respectively gathering_i、G_i、B_iIt is equal Value；

If MSE, MSE_R、MSE_GAnd MSE_BValue is respectively less than 0.5, then the video image stabilization sexual satisfaction requirement for gathering, is stabilization Video image, otherwise resurvey video image.

The video image for choosing multiple stabilizations carries out distortion correction treatment to video camera, obtains camera intrinsic parameter and outer ginseng Number；

Carry out the homogeneity correction of camera sensor；

Extract starting color background image；

The row pixel calculating method of two row of constant interval two is used to starting color background image, starting color background image is drawn It is divided into multiple detection zones, the arbitrary image pixel in any detection zone is with (x_p, y_q) represent, carry out image brightness values T meters Calculate, the calculating of T uses equation below,

Wherein, T (x_p, y_q) represent detection zone image slices vegetarian refreshments (x_p, y_q) image brightness values, y_tRepresent per sub-regions The image pixel average of top edge, y_bRepresent the image pixel average per sub-regions lower edge, x_lRepresent per the sub-regions left side The image pixel average of edge, x_rThe image pixel average per sub-regions right hand edge is represented, p and q is natural number, and INT [] is represented Formula inside bracket is rounded；

Calculate tunnel brightness value L according to video image brightness detection algorithm, the computation model for using for,

It is described that distortion correction treatment is carried out to video camera refers to 50 in the video image for select using gridiron pattern standardization stabilization Two field picture carries out camera calibration.

The uniformity of the correcting camera sensor, concretely comprises the following steps：The video image of stabilization is divided intoIt is individual Area identical subregion, (j, k) represent bySub-regions formed arbitrary region, j ∈ ζ andζ andFor Natural number, on the basis of the video image center of stabilization, carries out the homogeneity correction of camera sensor, and the formula of use is such as Under,

Wherein, K_R(j, k), K_G(j, k), K_B(j, k) is respectively the correction coefficient of arbitrary region (j, k) corresponding R, G, B, R (j, k), G (j, k), B (j, k) represent R, G, B value of arbitrary region (j, k) respectively.

The starting color background image of the video image of stabilization is extracted using medium filtering statistic law.

The starting color background image is obtained using median filter method after being trained to 100 two field pictures.

Obtain always equal to the brightness of starting color background image after light-source brightness value L according to video image brightness detection algorithm Evenness U₀With center line brightness longitudinal uniformity U_mDetected, the formula of use is as follows：

Wherein, L_minIt is the average value of the minimum luminance value of all subregions, L_avIt is the average brightness value to all subregions The average value asked for, L '_minIt is the average value of the center line minimum luminance value of all subregions, L '_maxIt is the center line of all subregions The average value of maximum brightness value.

The beneficial effects of the invention are as follows：The present invention can be detected to tunnel internal monochrome information, it is possible to while entering The brightness detection of row tunnel and luminance uniformity detect, to reducing tunnel illuminating system energy consumption and to ensure that vehicle safe driving has very heavy Big effect.Simultaneously video image brightness detection algorithm of the present invention by different exposure time and different image element values with it is bright An one-to-one relation is established between angle value, rather than being limited only to certain specific time for exposure or specific figure As element value, the scope of application of brightness detection algorithm is extended.

Brief description of the drawings

Fig. 1 is the Technology Roadmap of tunnel brightness detection method of the present invention based on video；

The subregion that Fig. 2 is counted by the camera sensor uniformity of the present invention detection；

Fig. 3 is the horizontal and vertical variation tendency of R values when the camera sensor uniformity of the present invention is detected, the horizontal stroke of (a) R values To the longitudinally varying trend of variation tendency (b) R values；

Fig. 4 is that the present invention is trained using the frame of background (a) medium filtering statistic law 25 that medium filtering statistic law is trained Background, the background that the frame of (b) medium filtering statistic law 50 is trained, the background that the frame of (c) medium filtering statistic law 100 is trained, D background that the frame of () medium filtering statistic law 150 is trained；

Fig. 5 is the comparison diagram and relative error of light source intrinsic brilliance value and the calculated value of brightness detection model.

Specific embodiment

Specific embodiment of the invention given below is, it is necessary to explanation is the invention is not limited in implementing in detail below Example, all equivalents done on the basis of technical scheme each fall within protection scope of the present invention.

Embodiment 1

The video camera configuration that the present invention is used is PAL colour television standards, and it is taking into full account the visual characteristic of human eye After the nonlinear characteristic of cathode-ray tube (CRT), according to the importance of component value in coloured image, by three color component R, G, B are weighted averagely with different weights.Y therein is brightness (Luminance), reflects brightness degree, brightness-formula For：

Y=0.299R+0.587G+0.114B

The coloured image that collection is come is gray level image using YUV model conversations in pretreatment stage, that is, Rgb color space is converted to YUV color spaces, and treatment is only just analyzed to Y afterwards, and monochrome information Y can show complete Luminance picture, so can both improve the speed of subsequent algorithm, and can save memory headroom, reach more preferably Ask.

Tunnel brightness detection method of the present invention based on video, its overall flow figure is as shown in figure 1, detailed process includes：

Collection video image, the stabilization of the video image of collection is prejudged according to video image stabilization detection algorithm Property, specific formula is：

Wherein, R, G, B represent three color components of red, green, blue of each pixel of video image, and MSE represents collection Three total mean square deviations of color component of video image, MSE_RRepresent the mean square deviation of each pixel color component of video image R, MSE_G Represent the mean square deviation of each pixel color component of video image G, MSE_BRepresent that each pixel color component of video image B's is square Difference, i represents the i-th two field picture in the video image of collection, i=1,2,3 ..., n, n >=1, R_i、G_i、B_iRespectively the i-th two field picture The value of R, G, B of all pixels point, The R of all images in the video image for respectively gathering_i、G_i、B_iIt is equal Value；If MSE, MSE_R、MSE_G、MSE_BValue be respectively less than 0.5, then the requirement of the video image stabilization sexual satisfaction that gathers is regarding for stabilization Frequency image；

Table 1 R, G, B Data-Statistics result

When video image stabilization test is carried out, constant detection zone R, G, the value of B are counted every 1 minute, Namely every 1500 frame video images, 20 groups of data (being shown in Table 1) are asked for altogether, journey is disperseed come analyze data using mean square deviation Degree, mean square deviation can well weigh the stable sex expression of camera chain.

The corresponding stability numerical value of table 2 R, G, B

Can be seen that the single mean square deviation of whether R, G, B from the result of calculation in table 2, or totality mean square deviation, The value of variance is both less than 0.5, is almost gathered in around average, is in close proximity to average, embodies camera chain not It is carved with identical shooting effect simultaneously, it may be determined that it has good stability.Therefore, the camera chain can be used into The research of tunnel brightness detection system of the row based on video.

The video image for choosing multiple stabilizations carries out distortion correction treatment to video camera, obtains camera intrinsic parameter and outer ginseng Number；

The inner parameter of video camera is primarily used to determine relation between image coordinate system and camera coordinate system, specific public Formula：

In above formula, μ, v are the pixel coordinate of two dimensional image, and the summit with the plane of delineation upper left corner is as origin；μ₀、v₀It is figure As coordinate of the physical coordinates system origin in image pixel coordinates system；x_c、y_c、z_cIt is the coordinate value in camera coordinate system；x、y It is the physical coordinates of two dimensional image；1/dx and 1/dy represent the number (pixels/ of the pixel of unit length in image coordinate system Mm), referred to as proportionality coefficient；

The parameter of the camera model of table 3

As the coefficient of radial distortion k in view of camera lens₁、k₂With tangential distortion coefficient p₁、p₂When, it is necessary to use Distortion model；

In above formula, μ, v are the pixel coordinate of two dimensional image, μ₀、v₀It is image physical coordinates system origin in image pixel coordinates Coordinate in system；x_c、y_c、z_cIt is the coordinate value in camera coordinate system；X, y are the physical coordinates of two dimensional image；f_x、f_yIt is with picture Element is the focal length of unit, x_w、y_w、z_wIt is the coordinate value in world coordinate system；

The world coordinate system of video camera and the transformational relation of image pixel coordinates system can be according to inner parameter and outside ginsengs It is several to determine jointly, expression：

In above formula, f_x、f_yIt is the focal length in units of pixel, actually the physics focal length of lens is sensed with imaging Each unit of device x, y direction size product, and f_x=f/dx, f_y=f/dy represents scale factor of the image on transverse and longitudinal axle；

It is 6mm that the present invention uses the focal length of video camera, and sensor type is 1/3 " Progressive Scan COMS, 1/ 3 " it is the sensor diagonal size that is represented with foot.By taking the video camera as an example, the solution of parameter matrix is carried out, its parameter is such as Under：Focal length f=6mm, the resolution ratio for using is 1280*720, and sensor correspondingly-sized is horizontal 4.8mm, vertical 3.6mm, diagonal 6mm；

It can be seen from definition according to its nominal parameter and each video camera internal reference：In f_x=f/dx, f_yF in=f/dy_x、f_y It is respectively the normalization focal length on u, v axle, f is the focal length of video camera, and dx, dy represent unit picture of the sensor in u, v axle respectively The size of element, u₀、v₀The intersection point of optical centre, i.e. camera optical axis and the plane of delineation is represented, picture centre is usually located at Place, therefore its value often takes the half of resolution ratio, the result of the geometric parameter for calculating is as follows：

Wherein DPI_wIt is horizontal resolution, DPI_hIt is vertical resolution, Sen_wIt is the width of sensor, Sen_hIt is sensor Height.The quantity of gridiron pattern picture for participating in demarcating is different, the result for calibrating also can difference, therefore, to different chesses The calibrating parameters value obtained in the case of disk lattice picture number has carried out contrast statistics with theoretical value；

As shown in Table 4, the precision of camera calibration has direct relation with the quantity for demarcating picture.With demarcation picture Quantity increases, error rate constantly reducing, but, after image is more than 50, the perspective of error rate and video camera The variation tendency of transformation matrix and distortion factor is not obvious, and with increasing for picture is demarcated, it is time-consuming also to increase, therefore, It is very rational to carry out camera calibration from 50.Low error rate shows the video camera mathematics used in the present invention simultaneously Model is rational, and can find that the tangential distortion of camera lens is than radial distortion according to video camera geometric correction result It is much smaller.

The result of the geometric parameter that the video camera of table 4 passes through after demarcating

The video image of stabilization is divided intoIndividual area identical subregion, (j, k) represent byIndividual sub-district Domain formed arbitrary region, j ∈ ζ andζ andIt is natural number, on the basis of the video image center of stabilization, is taken the photograph The homogeneity correction of camera sensor, the formula of use is as follows,

Wherein, K_R(j, k), K_G(j, k), K_B(j, k) is respectively the correction coefficient of arbitrary region (j, k) corresponding R, G, B, R (j, k), G (j, k), B (j, k) represent R, G, B value of arbitrary region (j, k) respectively.

Video image is divided into many sub-regions, gray value is asked for every sub-regions, from video image element laterally and Longitudinal distributing law carries out the detection of camera sensor uniformity and correction；

Camera chain uniformity be primarily referred to as cmos image sensor on the whole to the uniformity of Intensity response with it is consistent Property.Here the hawk image that video camera shoots is divided into many sub-regions, gray value then is asked for every sub-regions, so that from The horizontal and vertical upper regularity of distribution of each pictorial element judges whether uniformity is good.

The resolution ratio unification of camera acquisition video image is set to 1280*720 in the present invention, in order to image is divided into The subregion of formed objects, by the length of subregion and it is wide be chosen for 80 pixels, therefore whole image can be divided into 16*9 it is sub Region, as shown in Fig. 2 each image element value that statistics is drawn per sub-regions in figure is counted, because being standard Uniform luminance source, tri- image element value variation tendencies of R, G, the B for counting are essentially identical, therefore, only list pass through here The statistic (being shown in Table 5) of the R values that detection is obtained.

The regularity of distribution can not be intuitively found from the statistics of R values, can be by means of being capable of performance statistics number directly perceived According to the chart of rule, the data variation of R values is represented with the form of broken line graph from horizontal, longitudinal both direction respectively here.

From Fig. 3 it can be found that the uniformity of imageing sensor and bad, when carrying out brightness and detecting, it is necessary first to uniform Property is modified.Can be learnt from Fig. 3, performance of the R values on horizontal and vertical is all：From picture centre more close to, then be worth bigger； From picture centre more away from, then be worth smaller, and change from picture centre to image border is presented good symmetry.Picture centre Can be good at react Intensity response, therefore timing choose picture centre on the basis of.

The array of cmos image sensor is inconsistent to Intensity response to cause system inhomogeneity poor, accordingly, it would be desirable to This inconsistency is corrected, the method that at this moment piecemeal homogeneity correction can be taken for entire image.The think of of the method Think：The quantity of block is 16*9 (Fig. 2), takes the average R values of R (4,8), R (4,9), R (5,8), R (5,9), R (6,8), R (6,9) and isThe R values of each block are R (j, k), and the corresponding correction coefficient of each block is K (j, k), specific such as formula：

Correction coefficient result such as table 6 per sub-regions, after correction, in the case of uniform luminance source, camera chain The R values of each block of output are 130.2364, and this value exactly corrects the average R values of six blocks of preceding picture centre.Entirely The Intensity response of camera image sensor array is uniform, has reached preferable uniformity.Need in later stage software development simultaneously When by using the coefficient in table to camera acquisition to image be modified, just can be accurately worth, so elimination take the photograph Camera system inhomogeneities influences on brightness detection system.

The statistic of the R values that the detection of table 5 16*9 sub-regions are obtained

The homogeneity correction coefficient of table 6

The starting color background image of the video image of stabilization is extracted using medium filtering statistic law；

The static scene image sequence that brightness detection in the present invention is fixed suitable for video camera, initial static background is big Mostly it is to be asked for by the sequence of video images of fixed frame number.The method for obtaining initial background is medium filtering statistic law.

Bg (x, y)=median (fr_i ^R(x, y), fr_i ^G(x, y), fr_i ^B(x, y))

In above formula, i is for seeking the sequence of video images of initial background from the 1st width to the n-th width；fr_i ^R(x, y), fr_i ^G(x, y)、fr_i ^B(x, y) is that correspondence transverse and longitudinal coordinate is R, G, B of the pixel of (x, y) respectively by from small to large or from big in each image To small sorted value；Median is to seek intermediate value to sorted sequence；X, y are horizontal and vertical pixel coordinate.

In the case of moving object is less, medium filtering statistic law at 25,50,100 and 150 frame it is time-consuming be respectively 1.048s, 1.722s, 3.678s and 5.994s, take in the case of moving object is more and are respectively 1.361s, 2.129s, 4.355s With 6.799, it is seen then that median filter method can obtain preferable and accurate initial background after being trained to 100 two field pictures.

Using constant interval pixel computational methods starting color background image is carried out detection zone brightness of image element R, G, B numerical value and image brightness values T are calculated；

Because the weight of the brightness of image element of each pixel in detection zone is identical, mainly it is larger with it is less Brightness of image element value all revert to intermediate mean values level, therefore uses arithmetic mean of instantaneous value, by detection zone select The brightness of image element of pixel seeks arithmetic mean of instantaneous value, the area image luminance elements value for as detecting.Using constant interval The row pixel computational methods speed of two row two is fast, takes 0.011s.

In above formula, T (x_p, y_q) represent detection zone image slices vegetarian refreshments (x_p, y_q) image brightness values, y_tRepresent that each is detected The image pixel average of region top edge, y_bRepresent the image pixel average of each detection zone lower edge, x_lRepresent that each is detected The image pixel average of region left hand edge, x_rThe image pixel average of each detection zone right hand edge is represented, p and q is natural number, INT [] is represented and the formula inside bracket is rounded；T is the arithmetic mean of instantaneous value of the brightness of image element of detection zone, is calculated Obtain R values, G values, B values and gray value and be respectively 83.455260,82.378638,97.273623 and 87.058411.Calculate To brightness of image element T value be the area image luminance elements value 87.058411 that detects.

Light-source brightness value L is calculated according to video image brightness detection algorithm；

Determine that optimal Intensity response function model is based on logarithm with the data of image element value and time for exposure Polynomial form, is analyzed using the model to the situation of time for exposure and light-source brightness, so that it is determined that going out based on light intensity The brightness detection algorithm of receptance function model.

The light exposure H of video camera and the relation of lens parameters：

In above formula, T is the time for exposure, and F is the aperture size of camera lens, and τ is the transmittance of camera optics camera lens, L is the brightness on testee surface.

Intensity response function model formula：

In above formula, (x, y) represents measured point, and H represents the light exposure of pixel, f_i(i=1,2,3) is respectively three images The corresponding Intensity response function of element.

When video camera is shot, light-source brightness is fixed as 31.06cd/m², while exposure mode is set as into hand Dynamic model formula, 43 time for exposure of selection carry out video image acquisition, and the time for exposure is respectively：1/25、1/30、1/35、1/40、1/ 45、1/50、1/60、1/70、1/80、1/90、1/100、1/125、1/150、1/175、1/200、1/250、1/300、1/350、 1/400、1/450、1/500、1/550、1/600、1/650、1/700、1/750、1/800、1/850、1/900、1/950、1/ 1000、1/1100、1/1200、1/1300、1/1400、1/1500、1/1600、1/1700、1/1800、1/1900、1/2000、1/ 4000th, 1/10000, the unit of gear is all the second, computation complexity can be greatly reduced using 1 order polynomial, while phase relation All more than 0.99, its scope meets the requirements number R values.

Pictorial element is to pictorial element and time for exposure with the Intensity response function model of time for exposure and light-source brightness The extension of Intensity response function model, it will be appreciated that for (light-source brightness is fixed under the multigroup constant light source brightness case of measure 31.06、50.36、101.33、195.67、256.86、361.14、522.95、744.51、963.77、1207.66、1565.79、 2118.13、2640.87、4016.45cd/m²), video image acquisition is carried out to 43 time for exposure respectively, so as to pass through 602 Group data are fitted assessment to the Intensity response function model of pictorial element and time for exposure T and light-source brightness L, as a result use The good function of fitting effect

Coefficient R value is 0.999792, therefore using the function as brightness detection algorithm model, during by pictorial element and exposure Between draw the calculated value of light-source brightness, the comparison diagram and relative error result of 602 groups of actual values of light-source brightness and calculated value As shown in Figure 5.As shown in Figure 5, the relative error of most data is less than 10%, and the degree of accuracy is high, and the algorithm model is by difference An one-to-one relation is established between time for exposure and different image element values and brightness value, rather than only limiting to In certain specific time for exposure or specific image element value, the scope of application of brightness detection algorithm is extended.

Tunnel brightness detection system will not only be detected to the brightness of designated area, in addition it is also necessary to the brightness to monitor area Total uniformity and center line brightness longitudinal uniformity detected, the computing formula of both uniformitys.

In above formula, L_minIt is the average value of the minimum luminance value of all subregions, L_avIt is the mean flow rate to all subregions The average value that value is asked for, L '_minIt is the average value of the center line minimum luminance value of all subregions, L '_maxFor in all subregions The average value of line maximum brightness value.

When carrying out the total uniformity of brightness and calculating, because the pixel of whole image is 1280*720, therefore will can entirely scheme Uniform detection block as being divided into the rows of 16 row * 9,32 * 18 rows of row, row these three types of 64 row * 36, each detection block is right respectively The pixel answered is 80*80,40*40,20*20.And the detection of corresponding center line brightness longitudinal uniformity, it is by figure The transversal centerline of picture carries out detecting realization that the 5th row all 16 to 16 * 9 row types of row is arranged respectively, 32 arrange * 18 row types All 32 row of 9th and 10 rows, all 64 row of the 18th and 19 rows of 64 * 36 row types of row carry out the inspection of center line brightness longitudinal uniformity Survey.

It is as follows testing result to be carried out with the three types block video image different to six sections：

The brightness detection system of table 7 is detected to the three types block of the total uniformity of brightness and center line brightness longitudinal uniformity

As shown in Table 7, the block for dividing is smaller, and the value for detecting gets over refinement, and the gap of the maximum of brightness and minimum value It is bigger, also resulted in that brightness longitudinal uniformity is smaller, conversely, point block it is bigger, then the value of brightness longitudinal uniformity is bigger. But the block for being not point is the smaller the better, and the smaller amount of calculation of block is bigger, therefore, it is proposed with 32*18 blocks in the present invention Computational methods carry out the detection of the total uniformity of brightness and center line brightness longitudinal uniformity.

Claims (9)

1. a kind of stability judging method of video image, it is characterised in that the judgement of stability of the video image is using such as Lower formula,

${\mathrm{MSE}}_R=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(R_i-\stackrel{\‾}{R_t})}^2}$

${\mathrm{MSE}}_G=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(G_i-\stackrel{\‾}{G_t})}^2}$

${\mathrm{MSE}}_B=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(B_i-{\stackrel{\‾}{B}}_t)}^2}$

$MSE=\sqrt{\frac{1}{3}({\mathrm{MSE}}_R^2+{\mathrm{MSE}}_G^2+{\mathrm{MSE}}_B^2)}$

Wherein, R, G, B represent three color components of red, green, blue of each pixel of video image, and MSE represents the video of collection Three total mean square deviations of color component of image, MSE_RRepresent the mean square deviation of each pixel color component of video image R, MSE_GRepresent The mean square deviation of each pixel color component of video image G, MSE_BRepresent the mean square deviation of each pixel color component of video image B, i Represent the i-th two field picture, i=1,2,3 ..., n, n >=1, R in the video image of collection_i、G_i、B_iThe all pictures of respectively the i-th two field picture The value of R, G, B of vegetarian refreshments,The R of all images in the video image for respectively gathering_i、G_i、B_iAverage；

If MSE, MSE_R、MSE_GAnd MSE_BValue is respectively less than 0.5, then the video image stabilization sexual satisfaction requirement for gathering, is regarding for stabilization Frequency image, otherwise resurveys video image.

2. a kind of tunnel brightness detection method based on video, it is characterised in that the detection method includes：

Step one, gathers video image, and the stability of video image is judged using claim 1 methods described, is stablized Video image；

Step 2, extracts the starting color background image of the video image of stabilization；

Step 3, calculates light-source brightness value L, specifically includes：

Starting color background image to step 2 uses the row pixel calculating method of two row of constant interval two, by starting color Background As being divided into multiple detection zones, the arbitrary image pixel in any detection zone is with (x_p,y_q) represent, carry out starting color Background image brightness value T is calculated, and the calculating of T uses equation below,

$T=\frac{\underset{p=p+2}{\overset{x_r}{\underset{p=x_l}{\Σ}}}\underset{q=q+2}{\overset{y_t}{\underset{q=y_b}{\Σ}}}T(x_p,y_q)}{INT\[(y_t-y_b+1)/2\]*-INT\[(x_r-x_l+1)/2\]}$

Wherein, T (x_p,y_q) represent detection zone image slices vegetarian refreshments (x_p,y_q) image brightness values, y_tRepresent in each detection zone The image pixel average at edge, y_bRepresent the image pixel average of each detection zone lower edge, x_lRepresent that each detection zone is left The image pixel average at edge, x_rThe image pixel average of each detection zone right hand edge is represented, p and q values are natural number, INT [] is represented and the formula inside bracket is rounded；

Calculate tunnel brightness value L according to video image brightness detection algorithm, the computation model for using for,

3. the tunnel brightness detection method of video is based on as claimed in claim 2, it is characterised in that obtain the video image of stabilization Afterwards, it is necessary to the video image for passing through stabilization carries out distortion correction treatment to video camera.

4. as claimed in claim 2 based on video tunnel brightness detection method, it is characterised in that it is described video camera is carried out it is abnormal It is that, using gridiron pattern standardization, 50 two field pictures carry out camera calibration in selecting stable video image to become correction process.

5. the tunnel brightness detection method of video is based on as claimed in claim 3, it is characterised in that enter line distortion school to video camera Need to carry out the homogeneity correction of camera sensor after positive treatment.

6. the tunnel brightness detection method of video is based on as claimed in claim 5, it is characterised in that the camera sensor Homogeneity correction, concretely comprises the following steps：The video image of stabilization is divided intoIndividual area identical subregion, (j, k) is represented BySub-regions formed arbitrary region, j ∈ ζ andζ andIt is natural number, with the video image of stabilization On the basis of the heart, the homogeneity correction of camera sensor is carried out, the formula of use is as follows,

Wherein, K_R(j,k)、K_G(j,k)、K_B(j, k) is respectively the correction coefficient of arbitrary region (j, k) corresponding R, G, B, R (j, K), G (j, k), B (j, k) represent R, G, B value of arbitrary region (j, k) respectively.

7. the tunnel brightness detection method of video is based on as claimed in claim 2, it is characterised in that use medium filtering statistic law Extract the starting color background image of the video image of stabilization.

8. the tunnel brightness detection method of video is based on as claimed in claim 7, it is characterised in that the starting color Background Seem to be obtained after being trained to 100 two field pictures using median filter method.

9. the tunnel brightness detection method of video is based on as claimed in claim 2, it is characterised in that according to video image brightness inspection Method of determining and calculating obtains uniformity U total to tunnel brightness after light-source brightness value L₀With center line brightness longitudinal uniformity U_mDetected, used Formula it is as follows：

Wherein, L_minIt is the average value of the minimum luminance value of all subregions, L_avIt is that the average brightness value of all subregions is asked for Average value, L^’ _minIt is the average value of the center line minimum luminance value of all subregions, L '_maxFor the center line of all subregions is maximum The average value of brightness value.