CN106034195A - Mobile detecting method based on grey correlation analysis - Google Patents

Abstract

A mobile detecting method based on grey correlation analysis comprises the following steps of receiving a video frame including plural input pixels; through a grey correlation analysis technology, determining a bit rate change of the input pixels so as to establish a plurality of quality background models; through two phase detecting steps based on a block and the pixels, detecting a mobile object so as to generate a binary moving mask; through entropy calculation, detecting a brightness change of the video frame so as to update the background models at a proper time; and providing a setting interface so that a user sets detecting sensitivity, and estimating error detection of the moving mask based on the sensitivity. Through the plurality of quality background models, the mobile object can be correctly detected in a video string flow possessing a variable bit rate; through the two phase detecting steps based on the block and the pixels, complete and accurate moving detecting can be realized; through detecting the brightness change, the background models are updated at the proper time so that an influence caused by the brightness change can be eliminated; and through the adjustable detecting sensitivity, accuracy can be increased and the error detection can be reduced.

Description

Mobile detection method based on grey correlation analysis

Technical field

The present invention relates to technical field of image processing, particularly relate to extract about shifting from video streaming (video stream) Mobile detection (motion detection) method of animal body information.

Background technology

Mobile detection is a crucial technology in automated video monitoring system, and it is for extracting about shifting from video streaming The information of animal body.Mobile detection method is mainly segmented into three types: time differencing method (temporal Differencing), optical flow method (optical flow) and background subtracting method (background subtraction).Time difference Method is easily achieved, and amount of calculation is low, but mobile object is easily caused internal fracture, it is not easy to extract complete movement Object information.Optical flow method can extract complete mobile object information, and can be used for the mobile detection of mobile camera, But computationally intensive, it is not easy to the application of real-time, very sensitive to noise in addition.Background subtracting method is easily achieved, By using background model and more complete mobile object letter can be extracted with moderate amount of calculation under the background of relative quiescent Breath, but the change to light is the most sensitive.Therefore, background subtracting method is commonly used in mobile detection application, and develops Go out various background subtracting method, such as: gauss hybrid models method (Gaussian mixtures model, GMM) and difference are estimated Survey method (sigma difference estimation, SDE), multiple SDE method (multiple SDE, MSDE), multiple Time differencing method (multiple temporal difference, MTD), simple statistics calculus of finite differences (simple statistical Difference, SSD).

Along with the development of video communication technology, radio communication becomes to be practically applicable to mobile detection application, and it can strengthen greatly The detecting ability of the mobile object of scope.Unfortunately, radio communication is limited by the bandwidth of reality network, is particularly easy to send out Raw network congestion or server are delayed machine, to this end, video streaming introduces video code rate controls (video rate control) technology, The most H.264/AVC video format, it uses variable bit rate (variable bitrate, VBR) to adapt to reality net Network condition.Above-mentioned existing background subtracting method, in the video streaming with fixed bit rate, can detect motive objects Body, this is because in the environment of the most preferable, stable, it is easy to distinguish mobile object by background model.But, Owing to reality network seldom provides such ideal, stable environment, above-mentioned existing background subtracting method having variable bit In the video streaming of rate, easily change because of the unexpected of bit rate of video streaming, cause the erroneous judgement of mobile object, it is impossible to have Effect detects mobile object.

For example, referring to Fig. 6, it is to have plural number video pictures (video in the video streaming of variable bit rate Frame) the measurement figure of the brightness value (luminance value) of upper same pixel.At the beginning, video streaming is the 150th～240 Being the video streaming of the high bit rate (or high-quality) of 200kbps during individual video pictures, it has the back of the body that relatively high-amplitude wave is dynamic Scape signal B1.Existing background subtracting method produces background model according to so having the dynamic background signal B1 of relatively high-amplitude wave.When When video communication is hindered by network congestion, video code rate controls technology and distributes the remaining network bandwidth, therefore with backsight Frequently crossfire becomes the video streaming of low bit rate (or low quality) of 5kbps during the 240th～280 video pictures, It has a background signal B2 relatively smoothed out, and in this example because of have mobile object by and have the dynamic movement of a relatively high-amplitude wave Signal P1.If when movable signal P1 occurs, background model not yet update and still according to there is the back of the body that relatively high-amplitude wave is dynamic Scape signal B1 is produced, and it is background signal that existing background subtracting method may judge movable signal P1 by accident.And if when one section After between, existing background subtracting method may update background model according to having the background signal B2 relatively smoothed out.But, When video communication is no longer influenced by hindering, video streaming reverts to 200kbps's during the 280th～325 video pictures The video streaming of high bit rate (or high-quality), it has the background signal B3 that relatively high-amplitude wave is dynamic.If at background signal When the fluctuation of B3 occurs, background model not yet update and still according to there is the background signal B2 relatively smoothed out produced, show It is movable signal that some background subtracting methods may judge the fluctuation of background signal B3 by accident.So, at the bit of video streaming When rate is by high step-down or by low uprising, existing background subtracting method is all it may happen that judge by accident.

Summary of the invention

It is an object of the invention to provide a kind of mobile detection method, correctly can detect in the video streaming have variable bit rate Survey mobile object, and the impact more completely caused with mobile detection accurately and elimination brightness flop can be realized.

For achieving the above object, the present invention provides a kind of shifting based on grey correlation analysis (grey relational analysis) Dynamic method for detecting, comprising:

S1) receiving video pictures (video frame), video pictures includes plural number input pixel；

S2) multimass (multi-quality) background model is set up, comprising: S21) calculate the picture of each input pixel Element value and the Euclidean distance (Euclidean distance) between the pixel value of corresponding plural candidate background pixel； S22) according to Euclidean distance, the pixel value of each input pixel and the picture of corresponding plural candidate background pixel are calculated Grey incidence coefficient (grey relational coefficient) between element value；S23) judge in grey incidence coefficient minimum Whether value, less than or equal to first threshold, if then judging that the bit rate of input pixel changes, and adds input pixel As candidate background pixel, if otherwise judging, the bit rate of input pixel does not changes；

S3) detecting mobile object, comprising: S31) video pictures is divided into plurality of blocks, for each block, By the grey correlation between pixel value and the pixel value of corresponding plural candidate background pixel of input pixel each in block In coefficient, maximum adds up, to produce the grey incidence coefficient summation about block；S32) judge that grey incidence coefficient is total Whether more than or equal to Second Threshold, if then judging that block is background block, if otherwise judging, block is as mobile block； S33) for each mobile block, it is judged that in mobile block, the pixel value of each input pixel is carried on the back with corresponding plural candidate In grey incidence coefficient between the pixel value of scene element, whether maximum is more than or equal to the 3rd threshold value, if then judging defeated Entering pixel is background pixel, if otherwise judging to input pixel as mobile pixel；S34) produce binaryzation and move shade (binary motion mask)；

S4) brightness flop of video pictures is detected, comprising: S41) calculate the grey incidence coefficient summation of each block Entropy (entropy), and add up to produce about the entropy summation of video pictures；S42) video pictures and previous video are judged Whether the difference of the entropy summation of picture is more than or equal to the 4th threshold value, if then judging that video pictures has brightness flop, and root Updating, according to video pictures, the candidate background pixel that each input pixel is corresponding, if otherwise judging, video pictures does not has brightness to become Change；

S5) error detection of the mobile shade of assessment, comprising: S51) provide setting interface for users setting detecting sensitive Degree；S52) by the total quantity of pixel mobile in mobile shade and background pixel divided by the area of mobile shade, comment to produce Valuation；S53) judging whether assessed value is more than the product of the 5th threshold value and detecting sensitivity, the 5th threshold value is for being just predicted as The discreet of positive pixel (true positive pixel) divided by the area of mobile shade, if then judging that there is mistake detects Survey, if otherwise judging to there is not error detection.

In one embodiment of this invention, candidate background pixel corresponding to the input pixel that is updated is selected from inputting pixel Time corresponding to maximum in Euclidean distance between pixel value with the pixel value of corresponding plural candidate background pixel Select background pixel.

In one embodiment of this invention, input pixel can all include brightness value with the pixel value of candidate background pixel (luminance value), blue color difference value (blue-difference chroma value) and red color value (red-difference chroma value)。

In one embodiment of this invention, brightness value, blue color difference value can all use 8 bits to represent with red color value.

In one embodiment of this invention, first threshold can be set to 0.6.

In one embodiment of this invention, Second Threshold can be set to 245.

In one embodiment of this invention, the 3rd threshold value can be set to 0.6.

In one embodiment of this invention, the 4th threshold value can be set to 0.05.

In one embodiment of this invention, it was predicted that the discreet for positive positive pixel can be set to 30 × 30, detects sensitivity Setting according to user can be 0～10.

In one embodiment of this invention, plurality of blocks can be 16 × 16 blocks.

Technological means described in said one embodiment can be applicable in another embodiment above-mentioned, new to obtain one Embodiment, as long as these technological means are the most conflicting.

The present invention is by grey correlation analysis technology, it is judged that the bit rate variation of input pixel, to set up multimass background mould Type, therefore can correctly detect mobile object in the video streaming have variable bit rate；Further by based on block with The two benches detecting step of pixel, detects mobile object, moves shade producing binaryzation, therefore can realize more complete with Mobile detection accurately；By the calculating of entropy, the brightness flop of detecting video pictures, with timely replacement background model, because of This can eliminate the impact that brightness flop causes；Sensitive by providing setting interface for users to set detecting according to environmental demand Degree, utilizes the error detection of the mobile shade of adjustable detecting sensitivity assessment, therefore can increase accuracy and detect with reducing mistake Survey.

Accompanying drawing explanation

The present invention is further detailed explanation with detailed description of the invention below in conjunction with the accompanying drawings.

Fig. 1 is the flow chart of the mobile detection method based on grey correlation analysis according to one embodiment of the invention.

Fig. 2 is the flow chart setting up multimass background model step according to one embodiment of the invention.

Fig. 3 is the flow chart based on block with the two benches detecting step of pixel according to one embodiment of the invention.

Fig. 4 is the flow chart of the brightness flop step of the detecting video pictures according to one embodiment of the invention.

Fig. 5 is the flow chart of the error detection step of the mobile shade of the assessment according to one embodiment of the invention.

Fig. 6 is to have in the video streaming of variable bit rate the measurement figure of the brightness value of same pixel on plural number video pictures.

Description of symbols:

B1, B2, B3 background signal

P1 movable signal

S1 receives the video pictures including plural number input pixel

S2 is by grey correlation analysis technology, it is judged that the bit rate variation of input pixel, to set up multimass background model

S3, by two benches detecting step based on block and pixel, detects mobile object, moves screening producing binaryzation Cover

S4 is by the calculating of entropy, and the brightness flop of detecting video pictures, with timely replacement background model

S5 provides and sets interface for users setting detecting sensitivity, and the error detection of the mobile shade of assessment according to this

The Europe that S21 calculates between pixel value and the pixel value of corresponding plural candidate background pixel of each input pixel is several Reed distance

S22, according to Euclidean distance, calculates the pixel value of each input pixel and corresponding plural candidate background pixel Grey incidence coefficient between pixel value

In S23 grey incidence coefficient, whether minima is less than or equal to first threshold

The bit rate of S231 input pixel changes, adds input pixel as candidate background pixel

The bit rate of S232 input pixel does not changes

Video pictures is divided into plurality of blocks by S31, for each block, by the pixel value of input pixel each in block Add up with maximum in the grey incidence coefficient between the pixel value of corresponding plural candidate background pixel, to produce about district The grey incidence coefficient summation of block

Whether S32 grey incidence coefficient summation is more than or equal to Second Threshold

S321 block is background block

S322 block is mobile block

The pixel value of the pixel value of each input pixel and corresponding plural candidate background pixel in each mobile block of S33 Between grey incidence coefficient in maximum whether more than or equal to the 3rd threshold value

S331 input pixel is background pixel

S332 input pixel is mobile pixel

S34 produces binaryzation and moves shade

S41 calculates the entropy of the grey incidence coefficient summation of each block, and adds up to produce the entropy summation about video pictures

Whether S42 video pictures is more than or equal to the 4th threshold value with the difference of the entropy summation of previous video pictures

S421 video pictures has brightness flop, updates, according to video pictures, the candidate background pixel that each input pixel is corresponding

S422 video pictures does not has brightness flop

S51 provides and sets interface for users setting detecting sensitivity

S52 will move the total quantity area divided by mobile shade of pixel and background pixel, to produce assessment in mobile shade Value

Whether S53 assessed value is more than the 5th threshold value and the product detecting sensitivity, and the 5th threshold value is to be predicted as positive positive pixel Discreet divided by the area of mobile shade

There is error detection in S531

There is not error detection in S532.

Detailed description of the invention

Refer to the stream that Fig. 1, Fig. 1 are the mobile detection method based on grey correlation analysis according to one embodiment of the invention Cheng Tu.In step S1, receiving video streaming, wherein, video streaming includes plural number video pictures, each video pictures Including plural number input pixel.Distinguishing for convenience and illustrate, the t video pictures can be denoted as I_t, at the t video In picture, (x, input pixel y) can be denoted as p in position_t(x, y).In the present embodiment, YC is used_bC_rColor space, Each input pixel p_t(x, color or pixel value y) can be by brightness value (Y), blue color difference value (C_b) and red color Value (C_r) represented by three variablees, it addition, brightness value (Y), blue color difference value (C_b) and red color value (C_r) All use 8 bits to represent, i.e. its value is 0～255.

In step S2, by grey correlation analysis technology, it is judged that the bit rate variation of input pixel, to set up multimass Background model.In the present embodiment, as in figure 2 it is shown, Fig. 2 is to set up multimass background according to one embodiment of the invention The flow chart of model step, in step S21, calculates each input pixel p_t(x, pixel value y) and corresponding plural candidate Background pixel B (x, y)₁～B (x, y)_MPixel value between Euclidean distance Δ (1)～Δ (M), be expressed as follows with formula:

Δ (k)=‖ p_t(x, y)-B (x, y)_k‖

Wherein, k is 1～M any integer.Additionally it is noted that owing to using YC_bC_rColor space, the most each candidate Background pixel B (x, y)_kColor or pixel value equally by brightness value (Y), blue color difference value (C_b) and red color value (C_r) represented by three variablees.When Euclidean distance Δ (k) is the biggest, represent input pixel p_t(x, y) with candidate background picture Element B (x, y)_kBetween difference the biggest.Below this difference is quantified.

In step S22, according to Euclidean distance Δ (k), calculate each input pixel p_t(x, pixel value y) is with corresponding Plural candidate background pixel B (x, y)₁～B (x, y)_MPixel value between grey incidence coefficient γ(p_t(x, y), B (x, y)₁)～γ (p_t(x, y), B (x, y)_M), it is expressed as follows with formula:

$\mathrm{\γ}(p_t(x,y),B{(x,y)}_k)=\frac{{\mathrm{\Δ}}_{\min }+\mathrm{\ξ}{\mathrm{\Δ}}_{\max }}{\mathrm{\Δ}\left(k\right)+\mathrm{\ξ}{\mathrm{\Δ}}_{\max }}$

Wherein, k is 1～M any integer.It addition, Δ_minRepresent minimal difference, Δ_maxRepresent maximum difference.Due to brightness value (Y), blue color difference value (C_b) and red color value (C_r) all use 8 bits to represent, i.e. its value is 0～255, Therefore Δ_minWith Δ_max0 and 255 can be set to.It addition, ξ represents resolution ratio (distinguishing coefficient), It is between 0～1, and in the present embodiment, ξ is set to 0.2.As grey incidence coefficient γ (p_t(x, y), B (x, y)_k) the least, represent Input pixel p_t(x, y) with candidate background pixel B (x, y)_kBetween relatedness the least, i.e. difference is the biggest.

In step S23, it is judged that grey incidence coefficient γ (p_t(x, y), B (x, y)₁)～γ (p_t(x, y), B (x, y)_MIn), minima is No less than or equal to first threshold ε.If so, input pixel p is represented_t(x, y) with candidate background pixel B (x, y)₁～B (x, y)_M Between all do not have relatedness, then arrive step S231, it is judged that input pixel p_t(x, bit rate y) changes, and by defeated Enter pixel p_t(x, bit rate variation judgment value V y)_kIt is set to 1, furthermore, it is possible to pixel p will be inputted_t(x y) adds as waiting Select background pixel, thereby can set up multimass background model.If it is not, represent input pixel p_t(x y) carries on the back with certain candidate Between scene element, there is great relevance, then arrive step S232, it is judged that input pixel p_t(x, bit rate y) does not changes, And pixel p will be inputted_t(x, bit rate variation judgment value V y)_kIt is set to 0.Narration above can be expressed as follows with formula:

$V_k=\left\{\begin{array}{cc}1,& \mathrm{if}\min \left(\mathrm{\γ}(p_t(x,y),B{(x,y)}_k)\right)\≤\mathrm{\ϵ}\\ 0,& \mathrm{otherwise}\end{array}\right.$

Wherein, k is 1～M any integer, and first threshold ε can be set to 0.6.

In step S3, by two benches detecting step based on block with pixel, detect mobile object, to produce two-value Change mobile shade.In the present embodiment, as it is shown on figure 3, Fig. 3 be according to one embodiment of the invention based on block and picture The flow chart of the two benches detecting step of element, reconnaissance phase based on block includes step S31, S32, S321 and S322. In step S31, by video pictures I_t(i, j), in the present embodiment, according to H.264 Video coding to be divided into plurality of blocks M Macro zone block (MacroBlock) type supported is by video pictures I_tSplit 16 × 16 blocks, therefore i is 1～16 One integer, j is 1～16 any integer.For each block M (i, j), by block M (i, j) in each input pixel p_t(x, y) Pixel value and corresponding plural candidate background pixel B (x, y)₁～B (x, y)_MPixel value between grey incidence coefficient γ(p_t(x, y), B (x, y)₁)～γ (p_t(x, y), B (x, y)_MIn), maximum adds up, to produce the grey incidence coefficient about block Summation S (i, j), is expressed as follows with formula:

$S(i,j)=\underset{p_t\∈M(i,j)}{\mathrm{\Σ}}\max \left(\mathrm{\γ}(p_t(x,y),B{(x,y)}_k)\right)$

Wherein, k is 1～M any integer.

In step S32, it is judged that (whether i, j) more than or equal to Second Threshold α for grey incidence coefficient summation S.If so, represent Block M (i, is not j) mobile block, then arrives step S321, it is judged that block M (i, j) is background block, and by block M (i, j) Mobile block judgment value R (i j) is set to 0.If it is not, represent block M (i, j) in have many input pixel p_t(x is y) mobile A part for object, then arrive step S322, it is judged that (i, j) for mobile block, and by block M (i, turnover zone j) for block M (i j) is set to 1 to block judgment value R.Narration above can be expressed as follows with formula:

$R(i,j)=\left\{\begin{array}{cc}0,& \mathrm{if\; S}(i,j)\≥\mathrm{\α}\\ 1,& \mathrm{otherwise}\end{array}\right.$

Wherein, Second Threshold α can be set to 245.

Reconnaissance phase based on pixel includes step S33, S331, S332 and S34.In step S33, for each shifting Dynamic block, it is judged that each input pixel p in mobile block_t(x, pixel value y) and corresponding plural candidate background pixel B (x, y)₁～B (x, y)_MPixel value between grey incidence coefficient γ (p_t(x, y), B (x, y)₁)～γ (p_t(x, y), B (x, y)_M) Whether middle maximum is more than or equal to the 3rd threshold value beta.If so, represent that this inputs pixel p_t(x, y) with certain candidate background Pixel has great relevance, then arrive step S331, it is judged that input pixel p_t(x y) is background pixel.If it is not, represent This inputs pixel p_t(x, y) with candidate background pixel B (x, y)₁～B (x, y)_MAll do not have relatedness, then arrive step S332, Judge input pixel p_t(x, y) for mobile pixel.Then, in step S34, produce binaryzation and move shade BM, with public affairs Formula is expressed as follows:

$\mathrm{BM}(x,y)=\left\{\begin{array}{cc}0,& \mathrm{if}\max \left(\mathrm{\γ}(p_t(x,y),B{(x,y)}_k)\right)\≥\mathrm{\β}\\ 1,& \mathrm{otherwise}\end{array}\right.$

Wherein, k is 1～M any integer, and the 3rd threshold value beta can be set to 0.6.

In step S4, by the calculating of entropy, the brightness flop of detecting video pictures, with timely replacement background model.? In the present embodiment, as shown in Figure 4, Fig. 4 is the brightness flop step of the detecting video pictures according to one embodiment of the invention Flow chart, in step S41, calculate each block M (i, grey incidence coefficient summation S j) (i, entropy j), i.e. -S (i, j) log (S (i, j)), and add up with produce about video pictures I_tEntropy summation E_t, it is expressed as follows with formula:

$E_t=-\underset{(i,j)\∈I_t}{\mathrm{\Σ}}S(i,j)\log \left(S(i,j)\right)$

In step S42, it is judged that video pictures I_tEntropy summation E_tWith previous video pictures I_t-1Entropy summation E_t-1Difference be No it is more than or equal to the 4th threshold value μ.If so, video pictures I is represented_tThere is unexpected brightness flop, then arrive step S421, Judge video pictures I_tThere is brightness flop, and by brightness flop judgment value L_tIt is set to 1.If it is not, then arrive step S422, sentence Disconnected video pictures I_tThere is no brightness flop, and by brightness flop judgment value L_tIt is set to 0.Narration above can be expressed as follows with formula:

$L_t=\left\{\begin{array}{cc}1,& \mathrm{if}|E_t-E_{t-1}|\≥\mathrm{\μ}\\ 0,& \mathrm{otherwise}\end{array}\right.$

Wherein, the 4th threshold value μ can be set to 0.05.Additionally, in step S421, owing to judging video pictures I_tThere is brightness flop, Therefore can be according to video pictures I_tUpdate each input pixel p_t(x, and y) corresponding candidate background pixel B (x, y)_s, with formula It is expressed as follows:

B (x, y)_s=B (x, y)_s+ρ(p_t(x, y)-B (x, y)_s)

Wherein, and B (x, y)_sFor the candidate background pixel being updated, and B (x, y)_sFor the candidate background pixel after updating, ρ is for presetting ginseng Number.And the candidate background pixel B being updated (x, y)_sIt is selected from inputting pixel p_t(x, pixel value y) and corresponding plural candidate Background pixel B (x, y)₁～B (x, y)_MPixel value between Euclidean distance Δ (1)～Δ (M) in corresponding to maximum Candidate background pixel, is expressed as follows with formula:

$B{(x,y)}_s=\arg \underset{k=1~M}{\max }\mathrm{\Δ}\left(k\right)$

In step S5, it is provided that set interface for users and set detecting sensitivity, and the mistake of the mobile shade of assessment according to this is detectd Survey.In the present embodiment, as it is shown in figure 5, the mistake that Fig. 5 is the mobile shade of the assessment according to one embodiment of the invention is detectd Survey the flow chart of step, in step S51, it is provided that set interface for users and set detecting sensitivity ds, set interface such as It is graphical slide bar, and detects sensitivity ds and such as can be adjusted between 0～10.

In step S52, by total quantity n of pixel mobile in mobile shade BM with background pixel_dpDivided by mobile shade BM Area, to produce assessed value E_BM, it is expressed as follows with formula:

$E_{\mathrm{BM}}=\frac{n_{\mathrm{dp}}}{\mathrm{dimXdimY}}$

Wherein, n_dp=p_p+p_n, p_pFor the quantity of pixel mobile in mobile shade BM, p_nFor mobile shade BM carries on the back The quantity of scene element, dimX Yu dimY is respectively the width of mobile shade BM with high.

In step S53, it is judged that assessed value E_BMWhether more than the 5th threshold value δ and the product detecting sensitivity ds, wherein, 5th threshold value is the discreet Ω being predicted as positive positive pixel (the true positive pixel) face divided by mobile shade BM Long-pending, it is expressed as follows with formula:

$\mathrm{\δ}=\frac{\mathrm{\Ω}}{\mathrm{dimXdimY}}$

If so, it is assessed value E_BMMore than product δ × ds, then arrive step S531, it is judged that mobile shade BM exists error detection, And warn judgment value F to be set to 1 mistake.If it is not, i.e. assessed value E_BMLess than or equal to product δ × ds, then arrive step S532, it is judged that mobile shade BM does not exist error detection, and warns judgment value F to be set to 0 mistake.Narration can above It is expressed as follows with formula:

$F=\left\{\begin{array}{cc}1,& \mathrm{if}{E}_{\mathrm{BM}}>\mathrm{\δ}\×\mathrm{ds}\\ 0,& \mathrm{otherwise}\end{array}\right.$

Wherein, it was predicted that the discreet Ω for positive positive pixel can be set to 30 × 30.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all spirit in the present invention and former Within then, any modification, equivalent substitution and improvement etc. made, should be included within the scope of the present invention.

Claims (10)

1. mobile detection method based on grey correlation analysis, it is characterised in that comprising:

S1) receiving video pictures, described video pictures includes plural number input pixel；

S2) multimass background model is set up, comprising:

S21) Euclidean distance between the pixel value of each described input pixel and the pixel value of corresponding plural candidate background pixel is calculated,

S22) according to described Euclidean distance, the grey incidence coefficient between the pixel value of each described input pixel and the pixel value of described corresponding plural candidate background pixel is calculated,

S23) judge that in described grey incidence coefficient, whether minima is less than or equal to first threshold, if then judging that the bit rate of described input pixel changes, and described input pixel is added as candidate background pixel, if otherwise judging, the bit rate of described input pixel does not changes；

S3) detecting mobile object, comprising:

S31) described video pictures is divided into plurality of blocks, for each block, maximum in grey incidence coefficient between pixel value and the pixel value of described corresponding plural candidate background pixel of described input pixel each in described block is added up, to produce the grey incidence coefficient summation about described block

S32) judge described grey incidence coefficient summation whether more than or equal to Second Threshold, if then judging that described block is background block, if otherwise judge described block for movement block,

S33) for each mobile block, judge in described mobile block that in the grey incidence coefficient between the pixel value of each described input pixel and the pixel value of described corresponding plural candidate background pixel, whether maximum is more than or equal to the 3rd threshold value, if then judging that described input pixel is background pixel, if otherwise judging, described input pixel is as mobile pixel

S34) produce binaryzation and move shade；

S4) brightness flop of described video pictures is detected, comprising:

S41) calculate the entropy of the described grey incidence coefficient summation of each block, and add up to produce the entropy summation about described video pictures,

S42) judge that whether the described video pictures difference with the entropy summation of previous video pictures is more than or equal to the 4th threshold value, if then judging that described video pictures has brightness flop, and update, according to described video pictures, the candidate background pixel that each described input pixel is corresponding, if otherwise judging, described video pictures does not has brightness flop；

S5) error detection of described mobile shade is assessed, comprising:

S51) setting interface for users is provided to set detecting sensitivity,

S52) by described in described mobile shade, the total quantity of mobile pixel and described background pixel is divided by the area of described mobile shade, to produce assessed value,

S53) judge that whether described assessed value is more than the 5th threshold value and the product of described detecting sensitivity, described 5th threshold value is the discreet the being predicted as positive positive pixel area divided by described mobile shade, if then judging to there is error detection, if otherwise judging to there is not error detection.

Mobile detection method based on grey correlation analysis the most according to claim 1, wherein, candidate background pixel corresponding to maximum in candidate background pixel corresponding to the described input pixel that is updated Euclidean distance between pixel value and the pixel value of described corresponding plural candidate background pixel of described input pixel.

Mobile detection method based on grey correlation analysis the most according to claim 1, wherein, described input pixel all includes brightness value, blue color difference value and red color value with the pixel value of described candidate background pixel.

Mobile detection method based on grey correlation analysis the most according to claim 3, wherein, described brightness value, blue color difference value and red color value all use 8 bits to represent.

Mobile detection method based on grey correlation analysis the most according to claim 4, wherein, described first threshold is set to 0.6.

Mobile detection method based on grey correlation analysis the most according to claim 4, wherein, described Second Threshold is set to 245.

Mobile detection method based on grey correlation analysis the most according to claim 4, wherein, described 3rd threshold value is set to 0.6.

Mobile detection method based on grey correlation analysis the most according to claim 4, wherein, described 4th threshold value is set to 0.05.

Mobile detection method based on grey correlation analysis the most according to claim 1, wherein, described in be predicted as the discreet of positive positive pixel and be set to 30 × 30, described detecting sensitivity is set as 0 ~ 10 according to user.

Mobile detection method based on grey correlation analysis the most according to claim 1, wherein, described plurality of blocks is 16 × 16 blocks.