CN107220949A - The self adaptive elimination method of moving vehicle shade in highway monitoring video - Google Patents

The self adaptive elimination method of moving vehicle shade in highway monitoring video Download PDF

Info

Publication number
CN107220949A
CN107220949A CN201710390395.4A CN201710390395A CN107220949A CN 107220949 A CN107220949 A CN 107220949A CN 201710390395 A CN201710390395 A CN 201710390395A CN 107220949 A CN107220949 A CN 107220949A
Authority
CN
China
Prior art keywords
mrow
msub
image
gradient
background
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201710390395.4A
Other languages
Chinese (zh)
Inventor
孙战里
郑超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Anhui University
Original Assignee
Anhui University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Anhui University filed Critical Anhui University
Priority to CN201710390395.4A priority Critical patent/CN107220949A/en
Publication of CN107220949A publication Critical patent/CN107220949A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/155—Segmentation; Edge detection involving morphological operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G06T5/30—Erosion or dilatation, e.g. thinning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G06T5/94—Dynamic range modification of images or parts thereof based on local image properties, e.g. for local contrast enhancement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

A kind of self adaptive elimination method of moving vehicle shade in highway monitoring video, this method gathers some two field pictures and carries out Gaussian Background modeling, obtains background model；Using current desired video image to be processed as input, moving target is detected with gauss hybrid models, and determine motion target area；The gradient map of moving target cromogram is obtained, and filling is merged with binary map progress, target area is obtained, and realize that picture is compensated；Color motion object and correspondence Background respectively to picture after compensation asks for Quadratic Pressure Gradient figure and does difference；Contours extract is carried out to differential chart and image segmentation obtains complete shadow region；Further shadow Detection is done by color space and projection properties, and eliminates shadow region；After being updated with current input picture to background model, video image input step is returned.The advantage of the invention is that：The interference of the unfavorable factors such as noise, illumination variation is significantly reduced, can realize that moving vehicle shade is rapidly and effectively adaptively eliminated.

Description

The self adaptive elimination method of moving vehicle shade in highway monitoring video
Technical field
The technology of the present invention is related to a kind of self adaptive elimination method of moving vehicle shade in highway monitoring video.
Technical background
Moving target analysis segmentation based on video sequence is widely applied to intelligent video monitoring, terrain match, image The fields such as navigational guidance, human motion detail analysis, artificial intelligence.Wherein moving object detection is moving target analytical technology The bottom, belongs to basic work, and it is subsequent video images fortune that complete moving target is accurately extracted from video image The key that can dynamic analysis be smoothed out.But in moving object detection and cutting procedure, due to light irradiation, motion can be made Target produces corresponding shade.Shade can be moved with the motion of moving target, if not in detection process of moving target Shadow Detection is come out into processing, interference may be produced to the processing of anaphase movement target.
The detection of current pair and shade is broadly divided into feature based and based on model two types, is mainly based on model logical Cross illumination invariant, scene, the available object such as foreground target sets up model, and the method is typically only applicable to special circumstances, limitation Property is larger, and amount of calculation is larger, is not suitable for practical application.The method of another feature based, by color, texture, gradient Processing etc. feature reaches the detection for shade, such as based on hsv color space, based on RGB color and is based on The method of C1C2C3 color spaces.These methods are by extracting shadow color feature, and by saturation degree, pixel change, gray scale becomes Change the detection that the features such as contrast realize shade, but these methods usually only consider moving target and the difference of shade list feature, Cause detection accuracy not high, flase drop easily occur, be vulnerable to the interference of environmental factor, be generally only applicable to more simple Single scene, it is computationally intensive, influence real-time.
The content of the invention
The present invention is a kind of self adaptive elimination method of moving vehicle shade in highway monitoring video, and the method can be applicable In most of complex scene, such as road monitoring, shadow Detection accuracy is improved, with certain real-time.
The technology of the present invention solution is as follows：
The self adaptive elimination method of moving vehicle shade in a kind of highway monitoring video.It is characterised in that it includes following step Suddenly：
Step 1：N two field pictures are gathered, background modeling is realized by gauss hybrid models, background picture is extracted；(wherein n's Span is depending on video size, and 3000) n values is in experiment
Step 2：Using current desired video image to be processed as input, moving target is detected with gauss hybrid models, And determine motion target area；
Step 3：The gradient map of moving target cromogram is obtained, and filling is merged with binary map progress, complete mesh is obtained Region is marked, image compensation is realized；
Step 4：Color motion object and correspondence Background respectively to picture after compensation is asked for Quadratic Pressure Gradient figure and made the difference Value, removes shade；
Step 5：Contours extract is carried out to differential chart, image is split and draws more complete shadow region；
Step 6：Shade is further detected by color space and projection properties, and removes shadow region again；
Step 7：Background model is updated with current input picture, return to step 2.
The step 1 is：
Gauss hybrid models background modeling is used using current t frames input picture.
For multimodal Gaussian distribution model, each pixel of image presses the superposition of multiple Gaussian Profiles of different weights To model, there may be the state that color is presented in pixel, the weights of each Gaussian Profile for every kind of Gaussian Profile correspondence one Updated with distributed constant with the time.When handling coloured image, it is assumed that tri- chrominance channels of image slices vegetarian refreshments R, G, B are separate and have There is identical variance, for stochastic variable X observation data set { x1x2x3……xn, single sampled point xt=(rt,gt,bt) clothes From Gaussian mixtures probability density function be：
W in formulai,tFor the weight of i-th of Gaussian Profile of t, k is distribution pattern sum, μi,tFor its average, τi,tFor it Covariance matrix, η (xt,μi.t,τi,t) it is i-th of Gaussian Profile of t, η is calculated by following formula：
In formula, δi,tFor variance, I is three-dimensional unit matrix, wi,tFor the weight of i-th of Gaussian Profile of t.
The step 2 is：
The first step：Each new pixel value xtIt is compared with current k model, mould of the search mean bias in 2.5 σ Type, i.e., | xt-μi,t-1|≤2.5σi,t-1, it is used as the distributed model with new pixel matching.
Second step：If the pattern matched meets context request, the pixel is background, is otherwise prospect.
3rd step：Gauss weight wk,tIt is updated according to the following formula, and the weight after renewal is normalized, it is public Formula is as follows：
wk,t=(1 α) × wk,t-1+α×mk,t
4th step：The mean μ of Gaussian ProfiletAnd variances sigmatIt is updated according to the following formula respectively：
ρ=α × η (xt,μi,t,τi,t)
μt=(1- ρ) × μt-1+ρ×xt
For single pass greyscale video image, above formula covariance τi,t=σt, α is self-defined learning rate, and α takes [0,1] Between, α determines background with new speed, wherein μtT Gaussian Profile average is represented,Represent variance.
4th step：If the first step does not have any pattern match, the minimum pattern of weight is replaced, i.e., the pattern is equal It is worth for current pixel value, standard deviation is initial higher value, and weight is smaller value.
5th step：Each pattern is according to w/ α2Arrange in descending order, weight is big, the small pattern arrangement of standard deviation is forward.
6th step：B pattern is chosen as background, B meets following equation：
The step 3 is：Image compensation, because Gauss model is poor to the robustness of illumination, directly asks for gradient map Shadow region can not can accurately be detected by making the difference, so first gradient map is sought to the cromogram of moving object, then by the ladder drawn Degree figure is merged with moving object binary map and cavity is filled using opencv unrestrained water filling, then to carry out morphology follow-up Processing, obtains than more complete, the less target object binary picture of interference, realizes the compensation to picture, comprise the following steps that：
The first step：The moving object RGB triple channels figure drawn using Scharr gradient operators to step 2 carries out edge inspection Survey.
Second step：The moving object binary picture fusion that RGB triple channels gradient map and step 2 are drawn.
3rd step：By unrestrained water filling algorithm, the connected domain in fusion picture is filled, the effect of picture compensation is reached Really.
The step 4 is：To RGB triple channels figure Background difference corresponding with its corresponding to the binary map after compensation Quadratic Pressure Gradient rim detection is carried out using Scharr gradient operators.Scharr operators are a kind of special case operators of Sobel operators, When handling less core, Scharr accuracys rate are more.Obtained using Scharr after accurate complete Gradient vector chart, to prospect and the back of the body The Gradient vector chart of scape is compared, so as to distinguish the shadow region of moving target and its generation, specific algorithm is as follows：
The gradient vector of background image is calculated, is carried out along the Scharr operators and background image of image level and vertical direction Convolution, calculates 0 ° and 90 ° of direction g of each pixel (x, y)f,1(x,y),gf,2(x, y), obtains scaling vector:
gf(x, y)={ gf,1,gf,2|x,y},gf(x,y)
Scaling vector reflects the trend that background image changes in the pixel gradient, is also the form of expression of background gradient, Formula is as follows：
gf,i(x, y)=f (x, y) × Hi(x,y)
Wherein f (x, y) is the pixel value on coordinate (x, y), Hi(x, y) is the Scharr operators on 0 ° and 90 ° of directions, respectively For：
Second step：The normalization of background image gradient vector.
3rd step：The gradient vector of current frame image is calculated, and is normalized.
Convolution both horizontally and vertically is carried out with Scharr operators and sequence image along current frame image, each picture is calculated Two yardstick g of vegetarian refreshments (x, y)c,1(x, y), gc,2(x, y), obtains scaling vector gc(x, y)={ gc,1,gc,2|x,y},gc(x, Y), current frame image variation tendency is reflected, formula is：
It is normalized, obtains vector
4th step：Calculate the difference d. of corresponding pixel points normalized vector in current frame image gradient map and its context vault
The step 5 is：
Contours extract is carried out to differential chart, is filtered.
Then mean filter is carried out to image, corrodes the Morphological scale-spaces such as expansion, determine shadow region S1。
Doing difference due to gradient can be therefore empty by using YUV colors so that region of a part of moving object itself is eliminated Between and projection properties vehicle shadow region is further detected.
Doing difference due to gradient can be therefore empty by using YUV colors so that region of a part of moving object itself is eliminated Between, projection properties further detect that step 6 is specific as follows with the mode that Gradient Features are combined to vehicle shadow region：
The first step：Shadow Detection is carried out to foreground moving object using YUV color spaces and draws mask M1, YUV colors sky Between be a kind of method of color coding, wherein Y represents the gray value of lightness and image, and U and V represent aberration, and U and V are to constitute Two colored components, YUV color spaces are defined as：
Y=0.299R+0.587G+0.114B
U=-0.1687R-0.3313G+0.5B=128
V=0.5R-0.4187G-0.0813B+128
Second step：Mask is determined according to the RGB triple channel pixels that the RGB triple channels pixel of moving object is less than correspondence background M2:
Wherein i is correspondence image frame, and t is location of pixels, μ For correspondence background model；
3rd step：Because view field does not have a light source direct projection, only air scattering, thus moving object view field relative to Color shadow region has bluenessization change, and the rate of change of channel B is less than R, and G passage rates of change draw the shade of final optimization pass Region S2:
Wherein i is correspondence image frame, and t is Location of pixels, μ is correspondence background model；
By S1And S2Region carries out determining final shadow region with computing, and contrasts original image frame removal shadow region.
Brief description of the drawings
Fig. 1 is the flow chart of the self adaptive elimination method of moving vehicle shade in highway monitoring video of the present invention；
Fig. 2 be embodiment in be based on Gradient Features shadow Detection flow chart.
Specific implementation method
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing to embodiment party of the present invention Formula is described in further detail.
Embodiment 1：
The self adaptive elimination method of moving vehicle shade is as shown in figure 1, comprise the following steps in highway monitoring video：
1. set up background model
The color characteristic information of the pixel in the preceding n two field pictures of current input image is obtained, each pixel institute The color of presentation represents that usual k is taken between 3-5 with the superposition of k Gaussian Profile.The color x that pixel is presented thinks It is stochastic variable, then the pixel value of video frame images is stochastic variable x sampled value obtained by per moment t=1 ..., T.
Before foreground detection is carried out, first background is trained, to each background in each two field picture using a mixing Gauss model is simulated, and background is once extracted, and the detection of prospect is just simple, check pixel whether the Gauss with background Model Matching, matching is background, and mismatch is exactly prospect.
Specific steps：
Parameter initialization：In the first two field picture, corresponding first Gaussian Profile of each pixel is initialized, average The value of current pixel is assigned to, weights are assigned to 1, the average of the gauss of distribution function in addition to first, weights and all initializes zero.
Parameter updates：In each pixel xs of the moment t to picture frametGauss model corresponding with it is matched.
Update rule：
A. for unmatched Gaussian Profile, their mean μ and covariance matrix keeps constant；
B. the Gaussian Profile G matchediMean μ and covariance matrix update as the following formula:
μi,t=(1- ρ) μi,t-1+ρxt
∑i,t=(1- ρ) ∑si,t-1+ρ×diag[(xt-μi,t)T(xt-μi,t)]
α is the learning rate of parameter Estimation.
If there is no Gaussian Profile and pixel value x in the corresponding mixed Gauss model of the pixeltMatching, then most will can not The Gaussian Profile G of context process can be representedjAgain assignment, i.e.,：
J=argimin{Wk,t-1}
Wj,i-1=W0,μj,t=xt,∑j,t=v0I,
In formula, W0And V0Be it is previously given on the occasion of；I is 3 × 3 unit matrixs.
Then weight coefficient w of all k Gaussian Profiles in moment t is updated as the following formulai,t:
wi,t=(1- α) wi,t-1+α(mi,t)；
In formula, if Gaussian Profile GiWith t pixel value xtMatching, then take mi,tValue 1, otherwise value is 0.
2. foreground target is extracted
Each new pixel value xtIt is compared with current k model, model of the search mean bias in 2.5 σ, formula For：
|xt-μi,t-1|≤2.5σi,t-1；
Pixel w is updated according to formulai,t=(1- α) wi,t-1+α(mi,t)。
3. image compensation
The first step：The moving object RGB triple channels figure drawn using Scharr gradient operators to step 2 carries out edge inspection Survey.
Second step：The moving object binary picture fusion that RGB triple channels gradient map and step 2 are drawn.
3rd step：By unrestrained water filling algorithm, the connected domain in fusion picture is filled, the effect of picture compensation is reached Really.
4. image Quadratic Pressure Gradient, is referred to shown in Fig. 2：
The first step:The gradient vector of background image is calculated, along image level and the Scharr operators and background of vertical direction Image carries out convolution, calculates 0 ° and 90 ° of direction g of each pixel (x, y)f,1(x,y),gf,2(x, y), obtain yardstick to Amount：
gf(x, y)={ gf,1,gf,2|x,y},gf(x,y)
The gradient form of expression；gf,i(x, y)=f (x, y) × Hi(x,y)
Gradient operator Hi(x, y) is respectively：
Present frame and correspondence background image normalization
gc(x, y)={ gc,1,gc,2|x,y}
Gradient makes the difference d:
5. subsequent motion object processing, morphological operation, image is perfect；
Contours extract is carried out to differential chart first, is filtered；
Then mean filter is carried out to image, corrodes the Morphological scale-spaces such as expansion, determine shadow region S1。
Doing difference due to gradient can be therefore empty by using YUV colors so that region of a part of moving object itself is eliminated Between, projection properties are further detected with the mode that Gradient Features are combined to vehicle shadow region.
The step is:
The first step：Shadow Detection is carried out to foreground moving object using YUV color spaces and draws mask M1
YUV color space conversion formula：
Y=0.299R+0.587G+0.114B
U=-0.1687R-0.3313G+0.5B=128
V=0.5R-0.4187G-0.0813B+128
Second step：Mask is determined according to the RGB triple channel pixels that the RGB triple channels pixel of moving object is less than correspondence background M2
3rd step：Because view field does not have a light source direct projection, only air scattering, thus moving object view field relative to Color shadow region has bluenessization change, and the rate of change of channel B is less than R, and G passage rates of change draw the shade of final optimization pass Region S2:
By S1And S2Region carries out determining final shadow region with computing, and contrasts original image frame removal shadow region.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.

Claims (7)

1. the self adaptive elimination method of moving vehicle shade in a kind of highway monitoring video, it is characterised in that：Comprise the following steps：
Step 1：N two field pictures before collection, background modeling is realized by gauss hybrid models, extracts background picture；
Step 2：Using current desired video image to be processed as input, moving target is detected with gauss hybrid models, and really Determine motion target area；
Step 3：The gradient map of moving target cromogram is obtained, and filling cavity is merged with binary map progress, complete mesh is obtained Region is marked, image compensation is realized；
Step 4：Color motion object and correspondence Background respectively to picture after compensation asks for Quadratic Pressure Gradient figure and does difference, goes Except shade；
Step 5：Contours extract is carried out to differential chart, image is split and draws more complete shadow region；
Step 6：Shade is further detected by color space and projection properties, and removes shadow region again；
Step 7：Background model is updated with current input picture, return to step 2.
2. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 1, its feature exists In：The step 1 is specially：
Current n two field pictures are gathered, background model is set up, obtains background picture；
For multimodal Gaussian distribution model, each pixel of image is built by the superposition of multiple Gaussian Profiles of different weights Mould, every kind of Gaussian Profile correspondence one weights of each Gaussian Profile and is divided there may be the state that color is presented in pixel Cloth parameter updates with the time, when handling coloured image, it is assumed that tri- chrominance channels of image slices vegetarian refreshments R, G, B are separate and with phase Same variance, for stochastic variable X observation data set { x1x2x3……xn, single sampled point xt=(rt,gt,bt) obey Gaussian mixtures probability density function is：
<mrow> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>k</mi> </munderover> <msub> <mi>w</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>&amp;times;</mo> <mi>&amp;eta;</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>,</mo> <msub> <mi>&amp;mu;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>&amp;tau;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow>
W in formulai,tFor the weight of i-th of Gaussian Profile of t, k is distribution pattern sum, μi,tFor average, τi,tFor variance square Battle array, η (xt,μi.t,τi,t) it is i-th of Gaussian Profile of t, η is calculated by following formula：
<mrow> <mi>&amp;eta;</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>,</mo> <msub> <mi>&amp;mu;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>&amp;tau;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mrow> <mo>|</mo> <msub> <mi>&amp;tau;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>|</mo> </mrow> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> </mfrac> <msup> <mi>e</mi> <mrow> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <msubsup> <mi>&amp;tau;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </msup> </mrow>
<mrow> <msub> <mi>&amp;tau;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> </msub> <mo>=</mo> <msubsup> <mi>&amp;delta;</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>t</mi> </mrow> <mn>2</mn> </msubsup> <mi>I</mi> </mrow>
In formula, δi,tFor variance, I is three-dimensional unit matrix, wi,tFor the weight of i-th of Gaussian Profile of t.
3. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 2, its feature exists In：The step 2 is specifically included：
The first step：Each new pixel value xtBe compared with current k model, model of the search mean bias in 2.5 σ, i.e., | xt-μi,t-1|≤2.5σi,t-1, it is used as the distributed model with new pixel matching；
Second step：If the pattern matched meets context request, the pixel is background, is otherwise prospect；
3rd step：Gauss weight wk,tIt is updated according to the following formula, and the weight after renewal is normalized, formula is such as Under：
wk,t=(1- α) × wk,t-1+α×mk,t
4th step：The mean μ of Gaussian ProfiletAnd variances sigmatIt is updated according to the following formula respectively：
ρ=α × η (xt,μi,t,τi,t)
μt=(1- ρ) × μt-1+ρ×xt
<mrow> <msubsup> <mi>&amp;sigma;</mi> <mi>t</mi> <mn>2</mn> </msubsup> <mo>=</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>&amp;rho;</mi> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <msubsup> <mi>&amp;sigma;</mi> <mrow> <mi>t</mi> <mo>-</mo> <mn>1</mn> </mrow> <mn>2</mn> </msubsup> <mo>+</mo> <mi>&amp;rho;</mi> <mo>&amp;times;</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>t</mi> </msub> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mo>;</mo> </mrow>
For single pass greyscale video image, covariance τi,t=σt, α is self-defined learning rate, and α is taken between [0,1], and α determines Determine context update speed.
5th step：If the first step does not have any pattern match, the minimum pattern of weight is replaced, i.e. the average of the pattern is Current pixel value, standard deviation is initial higher value, and weight is smaller value；
6th step：Each pattern is according to w/ α2Arrange in descending order, weight is big, the small pattern arrangement of standard deviation is forward；
7th step：B Gaussian Profile is chosen as background, B meets following equation
4. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 3, its feature exists In：The step 3 is specially：
Because Gauss model is poor to the robustness of illumination, directly asks for gradient map and make the difference and can not can accurately detect shadow region Domain, so first seeking gradient map to the cromogram of moving object, then merges the gradient map drawn simultaneously with moving object binary map Cavity is filled using opencv unrestrained water filling, then carries out morphology subsequent treatment, obtains smaller than more complete, interference Target object binary picture, realize compensation to picture, comprise the following steps that：
The first step：The moving object RGB triple channels figure drawn using Scharr gradient operators to step 2 carries out rim detection；
Second step：The moving object binary picture fusion that RGB triple channels gradient map and step 2 are drawn；
3rd step：By unrestrained water filling algorithm, the connected domain in fusion picture is filled, the effect of picture compensation is reached.
5. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 4, its feature exists In：The step 4 is specially：
RGB triple channels figure Background corresponding with its corresponding to binary map after compensation is calculated using Scharr gradients respectively Son carries out Quadratic Pressure Gradient rim detection, and Scharr operators are a kind of special case operators of Sobel operators, when handling less core, Scharr accuracys rate are higher.Obtained using Scharr after accurate complete Gradient vector chart, to the gradient vector of prospect and background Figure is compared, so as to distinguish the shadow region of moving target and its generation, specific algorithm is as follows：
The first step:The gradient vector of background image is calculated, along image level and the Scharr operators and background image of vertical direction Convolution is carried out, 0 ° and 90 ° of direction yardstick g of each pixel (x, y) is calculatedf,1(x,y),gf,2(x, y), obtains scaling vector gf(x, y)={ gf,1,gf,2|x,y},gf(x,y).Scaling vector reflects the trend that background image changes in the pixel gradient, It is the form of expression of background gradient, formula is as follows：
gf,i(x, y)=f (x, y) × Hi(x, y), i=1,2；
F (x, y) is the pixel value on coordinate (x, y), H in formulai(x, y) is the Scharr operators on 0 ° and 90 ° of directions, is respectively：
Second step：The normalization of background image gradient vector：
According to formulaThe gradient vector of background each pixel is normalized, normalized Gradient vector afterwardsAnd store normalized gradient vector, it is used as context vault.According to ladder The indeformable principle in direction is spent, the gradient vector after normalization is a build-in attribute of background；
3rd step：The gradient vector of current frame image is calculated, and is normalized：Both horizontally and vertically used along current frame image Scharr operators carry out convolution derivation with sequence image, calculate the yardstick g in the both direction of each pixel (x, y)c,1 (x, y), gc,2(x, y), obtains scaling vector gc(x, y)={ gc,1,gc,2|x,y},gc(x, y) reflection current frame image change becomes Gesture, formula is：It is normalized, obtains vector
4th step：Calculate the difference d of corresponding pixel points normalized vector in current frame image gradient map and its context vault：
<mrow> <mi>d</mi> <mo>=</mo> <mo>|</mo> <mo>|</mo> <msubsup> <mi>g</mi> <mi>f</mi> <mo>*</mo> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>g</mi> <mi>c</mi> <mo>*</mo> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>|</mo> <mo>|</mo> <mo>=</mo> <msubsup> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>2</mn> </msubsup> <mo>|</mo> <mrow> <msubsup> <mi>g</mi> <mrow> <mi>c</mi> <mo>,</mo> <mi>i</mi> </mrow> <mo>*</mo> </msubsup> <mo>-</mo> <msubsup> <mi>g</mi> <mrow> <mi>f</mi> <mo>,</mo> <mi>i</mi> </mrow> <mo>*</mo> </msubsup> </mrow> <mo>|</mo> </mrow>
Effectively moving object and its shadow region can be separated by making the difference, obtain shadow region.
6. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 5, its feature exists In：The step 5 is：
Contours extract is carried out to differential chart, is filtered, by a series of morphological operations, it is determined that accurately shadow region S1。
7. the self adaptive elimination method of moving vehicle shade in highway monitoring video according to claim 6, its feature exists In：The step 6 is：
Doing difference due to gradient can be so that region of a part of moving object itself be eliminated, therefore by using YUV color spaces, is thrown Shadow feature is further detected with the mode that Gradient Features are combined to vehicle shadow region, is comprised the following steps that：
The first step：Shadow Detection is carried out to foreground moving object using YUV color spaces and draws mask M1, YUV color spaces are one The method for planting color coding, wherein Y represents the gray value of lightness and image, and U and V represent aberration, and U and V are to constitute colour Two components, YUV color spaces are defined as：
Y=0.299R+0.587G+0.114B
U=-0.1687R-0.3313G+0.5B=128
V=0.5R-0.4187G-0.0813B+128；
Second step：Mask M is determined according to the RGB triple channel pixels that the RGB triple channels pixel of moving object is less than correspondence background2：
Wherein i is correspondence image frame, and t is location of pixels, μ for pair Answer background model；
3rd step：Because view field does not have light source direct projection, only air scattering, therefore moving object view field is relative to colour Shadow region has bluenessization change, and the rate of change of channel B is less than R, and G passage rates of change draw the shadow region of final optimization pass S2：
Wherein i is correspondence image frame, and t is location of pixels, and μ is correspondence background model, then by by S1And S2Region is carried out and fortune Calculate and determine final shadow region, and contrast original image frame and carry out shadow removal.
CN201710390395.4A 2017-05-27 2017-05-27 The self adaptive elimination method of moving vehicle shade in highway monitoring video Pending CN107220949A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710390395.4A CN107220949A (en) 2017-05-27 2017-05-27 The self adaptive elimination method of moving vehicle shade in highway monitoring video

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710390395.4A CN107220949A (en) 2017-05-27 2017-05-27 The self adaptive elimination method of moving vehicle shade in highway monitoring video

Publications (1)

Publication Number Publication Date
CN107220949A true CN107220949A (en) 2017-09-29

Family

ID=59946897

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710390395.4A Pending CN107220949A (en) 2017-05-27 2017-05-27 The self adaptive elimination method of moving vehicle shade in highway monitoring video

Country Status (1)

Country Link
CN (1) CN107220949A (en)

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108682019A (en) * 2018-04-25 2018-10-19 高雪亚 Height-adjustable is hurdled system
CN109064411A (en) * 2018-06-13 2018-12-21 长安大学 A kind of pavement image based on illumination compensation removes shadow method
CN109544605A (en) * 2018-05-23 2019-03-29 安徽大学 Mobile shadow detection method based on time-space relationship modeling
WO2019076326A1 (en) * 2017-10-20 2019-04-25 苏州科达科技股份有限公司 Shadow detection method and system for surveillance video image, and shadow removing method
CN109754440A (en) * 2018-12-24 2019-05-14 西北工业大学 A kind of shadow region detection method based on full convolutional network and average drifting
CN110164153A (en) * 2019-05-30 2019-08-23 哈尔滨理工大学 A kind of adaptive timing method of traffic signals
CN110213460A (en) * 2019-06-24 2019-09-06 华南师范大学 A kind of prospect scattering removing method based on inverse locking phase spatial modulation
CN110232717A (en) * 2019-06-10 2019-09-13 北京壹氢科技有限公司 A kind of target identity recognition methods suitable for multipair multi-targets recognition
CN110288538A (en) * 2019-05-23 2019-09-27 南京理工大学 A kind of the moving target shadow Detection and removing method of multiple features fusion
CN110443762A (en) * 2019-07-24 2019-11-12 南京末梢信息技术有限公司 A kind of road surface cast shadow suppressing method
CN111444359A (en) * 2020-04-15 2020-07-24 西安超嗨网络科技有限公司 Method for identifying putting-in behavior by image difference
CN111768450A (en) * 2020-06-10 2020-10-13 北京的卢深视科技有限公司 Automatic detection method and device for line deviation of structured light camera based on speckle pattern
CN112015033A (en) * 2019-05-28 2020-12-01 夏普株式会社 Shadow correction signal generating apparatus
CN112116000A (en) * 2020-09-16 2020-12-22 深圳印像数据科技有限公司 Image identification method for clothing type
CN112365519A (en) * 2020-11-25 2021-02-12 广州杰赛科技股份有限公司 Foreground image extraction method
CN113269694A (en) * 2021-06-17 2021-08-17 山东能之源核电科技有限公司 Image shadow elimination method based on content perception information
CN113297990A (en) * 2021-05-28 2021-08-24 西安理工大学 Human foot moving object detection method based on Gaussian mask light stream
CN114943833A (en) * 2022-03-28 2022-08-26 台州学院 Bubble identification image processing method for bubble flow in gas-liquid reactor
CN115272142A (en) * 2022-09-30 2022-11-01 南通市通州区华凯机械有限公司 Scene image preprocessing method of immersive driving simulator
CN115482474A (en) * 2022-08-24 2022-12-16 湖南科技大学 Bridge deck vehicle load identification method and system based on high-altitude aerial image
CN116452459A (en) * 2023-04-25 2023-07-18 北京优酷科技有限公司 Shadow mask generation method, shadow removal method and device

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1588431A (en) * 2004-07-02 2005-03-02 清华大学 Character extracting method from complecate background color image based on run-length adjacent map
CN101686338A (en) * 2008-09-26 2010-03-31 索尼株式会社 System and method for partitioning foreground and background in video
KR20110121261A (en) * 2010-04-30 2011-11-07 (주)임베디드비전 Method for removing a moving cast shadow in gray level video data
US20110273620A1 (en) * 2008-12-24 2011-11-10 Rafael Advanced Defense Systems Ltd. Removal of shadows from images in a video signal
CN102298781A (en) * 2011-08-16 2011-12-28 长沙中意电子科技有限公司 Motion shadow detection method based on color and gradient characteristics
CN102332157A (en) * 2011-06-15 2012-01-25 夏东 Method for eliminating shadow
CN103440475A (en) * 2013-08-14 2013-12-11 北京博思廷科技有限公司 Automatic teller machine user face visibility judging system and method
CN103679704A (en) * 2013-11-22 2014-03-26 中国人民解放军第二炮兵工程大学 Video motion shadow detecting method based on lighting compensation
CN104899881A (en) * 2015-05-28 2015-09-09 湖南大学 Shadow detection method for moving vehicle in video image
KR101717613B1 (en) * 2016-12-27 2017-03-17 주식회사한맥아이피에스 The moving vehicle detection system using an object tracking algorithm based on edge information, and method thereof

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1588431A (en) * 2004-07-02 2005-03-02 清华大学 Character extracting method from complecate background color image based on run-length adjacent map
CN101686338A (en) * 2008-09-26 2010-03-31 索尼株式会社 System and method for partitioning foreground and background in video
US20110273620A1 (en) * 2008-12-24 2011-11-10 Rafael Advanced Defense Systems Ltd. Removal of shadows from images in a video signal
KR20110121261A (en) * 2010-04-30 2011-11-07 (주)임베디드비전 Method for removing a moving cast shadow in gray level video data
CN102332157A (en) * 2011-06-15 2012-01-25 夏东 Method for eliminating shadow
CN102298781A (en) * 2011-08-16 2011-12-28 长沙中意电子科技有限公司 Motion shadow detection method based on color and gradient characteristics
CN103440475A (en) * 2013-08-14 2013-12-11 北京博思廷科技有限公司 Automatic teller machine user face visibility judging system and method
CN103679704A (en) * 2013-11-22 2014-03-26 中国人民解放军第二炮兵工程大学 Video motion shadow detecting method based on lighting compensation
CN104899881A (en) * 2015-05-28 2015-09-09 湖南大学 Shadow detection method for moving vehicle in video image
KR101717613B1 (en) * 2016-12-27 2017-03-17 주식회사한맥아이피에스 The moving vehicle detection system using an object tracking algorithm based on edge information, and method thereof

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
IVAN HUERTA等: "Detection and removal of chromatic moving shadows in surveillance scenarios", 《2009 IEEE 12TH INTERNATIONAL CONFERENCE ON COMPUTER VISION》 *
S. NADIMI等: "Physical models for moving shadow and object detection in video", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 》 *
张峰: "复杂背景下基于分层码本模型的运动目标检测与阴影消除", 《中国优秀硕士学位论文全文数据库 信息科技辑（月刊）》 *
罗铁镇: "基于混合高斯模型的运动检测及阴影消除算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑（月刊）》 *
高新波等编著: "《视觉信息质量评价方法》", 30 September 2011, 西安：西安电子科技大学出版社 *

Cited By (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2583198B (en) * 2017-10-20 2022-04-06 Suzhou Keda Technology Co Ltd Shadow detection method and system for monitoring video images, and shadow removal method
GB2583198A (en) * 2017-10-20 2020-10-21 Suzhou Keda Technology Co Ltd Shadow detection method and system for surveillance video image, and shadow removing method
WO2019076326A1 (en) * 2017-10-20 2019-04-25 苏州科达科技股份有限公司 Shadow detection method and system for surveillance video image, and shadow removing method
CN108682019B (en) * 2018-04-25 2019-03-22 六安荣耀创新智能科技有限公司 Height-adjustable is hurdled system
CN108682019A (en) * 2018-04-25 2018-10-19 高雪亚 Height-adjustable is hurdled system
CN109544605B (en) * 2018-05-23 2022-01-21 安徽大学 Moving shadow detection method based on space-time relation modeling
CN109544605A (en) * 2018-05-23 2019-03-29 安徽大学 Mobile shadow detection method based on time-space relationship modeling
CN109064411A (en) * 2018-06-13 2018-12-21 长安大学 A kind of pavement image based on illumination compensation removes shadow method
CN109064411B (en) * 2018-06-13 2021-08-17 长安大学 Illumination compensation-based road surface image shadow removing method
CN109754440A (en) * 2018-12-24 2019-05-14 西北工业大学 A kind of shadow region detection method based on full convolutional network and average drifting
CN110288538A (en) * 2019-05-23 2019-09-27 南京理工大学 A kind of the moving target shadow Detection and removing method of multiple features fusion
CN112015033B (en) * 2019-05-28 2022-04-05 夏普株式会社 Shadow correction signal generating apparatus
CN112015033A (en) * 2019-05-28 2020-12-01 夏普株式会社 Shadow correction signal generating apparatus
CN110164153A (en) * 2019-05-30 2019-08-23 哈尔滨理工大学 A kind of adaptive timing method of traffic signals
CN110232717A (en) * 2019-06-10 2019-09-13 北京壹氢科技有限公司 A kind of target identity recognition methods suitable for multipair multi-targets recognition
CN110213460A (en) * 2019-06-24 2019-09-06 华南师范大学 A kind of prospect scattering removing method based on inverse locking phase spatial modulation
CN110443762A (en) * 2019-07-24 2019-11-12 南京末梢信息技术有限公司 A kind of road surface cast shadow suppressing method
CN110443762B (en) * 2019-07-24 2023-09-22 南京末梢信息技术有限公司 Pavement shadow suppression method
CN111444359A (en) * 2020-04-15 2020-07-24 西安超嗨网络科技有限公司 Method for identifying putting-in behavior by image difference
CN111444359B (en) * 2020-04-15 2023-09-15 西安超嗨网络科技有限公司 Method for identifying placement behavior related to image difference
CN111768450B (en) * 2020-06-10 2023-10-13 合肥的卢深视科技有限公司 Automatic detection method and device for structured light camera row deviation based on speckle pattern
CN111768450A (en) * 2020-06-10 2020-10-13 北京的卢深视科技有限公司 Automatic detection method and device for line deviation of structured light camera based on speckle pattern
CN112116000A (en) * 2020-09-16 2020-12-22 深圳印像数据科技有限公司 Image identification method for clothing type
CN112365519A (en) * 2020-11-25 2021-02-12 广州杰赛科技股份有限公司 Foreground image extraction method
CN113297990A (en) * 2021-05-28 2021-08-24 西安理工大学 Human foot moving object detection method based on Gaussian mask light stream
CN113269694A (en) * 2021-06-17 2021-08-17 山东能之源核电科技有限公司 Image shadow elimination method based on content perception information
CN114943833A (en) * 2022-03-28 2022-08-26 台州学院 Bubble identification image processing method for bubble flow in gas-liquid reactor
CN115482474A (en) * 2022-08-24 2022-12-16 湖南科技大学 Bridge deck vehicle load identification method and system based on high-altitude aerial image
CN115482474B (en) * 2022-08-24 2023-07-04 湖南科技大学 Bridge deck vehicle load identification method and system based on aerial image
CN115272142A (en) * 2022-09-30 2022-11-01 南通市通州区华凯机械有限公司 Scene image preprocessing method of immersive driving simulator
CN115272142B (en) * 2022-09-30 2023-04-18 南通市通州区华凯机械有限公司 Scene image preprocessing method of immersive driving simulator
CN116452459A (en) * 2023-04-25 2023-07-18 北京优酷科技有限公司 Shadow mask generation method, shadow removal method and device
CN116452459B (en) * 2023-04-25 2024-05-24 北京优酷科技有限公司 Shadow mask generation method, shadow removal method and device

Similar Documents

Publication Publication Date Title
CN107220949A (en) The self adaptive elimination method of moving vehicle shade in highway monitoring video
CN109636905B (en) Environment semantic mapping method based on deep convolutional neural network
EP3614308B1 (en) Joint deep learning for land cover and land use classification
CN108573276B (en) Change detection method based on high-resolution remote sensing image
CN111738314B (en) Deep learning method of multi-modal image visibility detection model based on shallow fusion
CN103971386B (en) A kind of foreground detection method under dynamic background scene
CN108537239B (en) Method for detecting image saliency target
CN104966085B (en) A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features
CN103093203B (en) A kind of human body recognition methods again and human body identify system again
CN111539273A (en) Traffic video background modeling method and system
CN106709568A (en) RGB-D image object detection and semantic segmentation method based on deep convolution network
CN103914699A (en) Automatic lip gloss image enhancement method based on color space
CN109685045B (en) Moving target video tracking method and system
CN109215053B (en) Method for detecting moving vehicle with pause state in aerial video shot by unmanned aerial vehicle
CN110738676A (en) GrabCT automatic segmentation algorithm combined with RGBD data
CN103955945B (en) Self-adaption color image segmentation method based on binocular parallax and movable outline
CN109685732A (en) A kind of depth image high-precision restorative procedure captured based on boundary
CN108985169A (en) Across the door operation detection method in shop based on deep learning target detection and dynamic background modeling
CN106204594A (en) A kind of direction detection method of dispersivity moving object based on video image
CN104182968B (en) The fuzzy moving-target dividing method of many array optical detection systems of wide baseline
CN104751466B (en) A kind of changing object tracking and its system based on conspicuousness
CN109858487A (en) Weakly supervised semantic segmentation method based on watershed algorithm and image category label
CN114627052A (en) Infrared image air leakage and liquid leakage detection method and system based on deep learning
CN106991686A (en) A kind of level set contour tracing method based on super-pixel optical flow field
CN106529432A (en) Hand area segmentation method deeply integrating significance detection and prior knowledge

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20170929

RJ01 Rejection of invention patent application after publication