CN101950352B - Target detection method capable of removing illumination influence and device thereof - Google Patents

Abstract

The invention provides a target detection method capable of removing illumination influence, including the following steps: a background image is created; the gradients of the current frame image and the background image are calculated and output, wherein the gradient comprises a horizontal gradient and a vertical gradient; the directions and amplitudes of the gradients of the current frame image and the background image are compared, and a foreground outline is extracted and output according to the comparison result; and the extracted foreground outline is filled, so as to obtain a foreground block, and noise is filtered, so as to output the target. The invention also provides a target detection device capable of removing illumination influence. Compared with the prior art, the target detection method and device of the invention effectively solve the problems that the detected target is not accurate and is not reliable due to illumination influence in target detection.

Description

A kind of object detection method and device of removing illumination effect

Technical field

The present invention relates to a kind of object detection method and device, particularly a kind of object detection method and device of removing illumination effect belongs to Flame Image Process, field of video monitoring.

Background technology

Moving object detection is the basis of intelligent video monitoring technology, its testing result directly affect the later stage incident (such as: invasion, article are left over, article are stolen, vehicle reverse driving etc.) the alert rate of mistake and the false alarm rate that detect, therefore obtained paying close attention to widely.Yet when practical application, the situation of illumination variation occurs through regular meeting, this has just influenced the accuracy and the reliability of moving object detection greatly.Therefore, need research to remove the object detection method of illumination effect.

The object detection method of the removal illumination effect of research mainly contains two types at present.Wherein, class methods are based on the method for pixel.In general illumination variation only can be brought the variation of pixel intensity and color does not have too big variation, and these class methods are analyzed with the identification illumination variation pixel value in the HSI space based on this principle.But a lot of situation do not satisfy this hypothesis prerequisite in the true environment the inside, and inside most outdoor scene, are that background or target all do not have colouring information, and this type algorithm is unsatisfactory at the effect of actual environment the inside like this.Another kind of method is based on the method in zone.If all have certain contrast at the illumination variation front and back scene; The variation of illumination can not bring the variation of image texture edge feature so; Method based on the zone is utilized this principle just; If the edge of prospect and background is complementary, then this foreground area is exactly the false foreground area that illumination variation causes.But the hypothesis of " the illumination variation front and back scene all has certain contrast " is false in night, and then this type algorithm lost efficacy.In addition, there is real goal to get in the illumination variation zone simultaneously and also can causes the failure of this type algorithmic match.

Publication number is the method that the one Chinese patent application of CN 101393603A discloses a kind of identification and detection tunnel fire disaster flame.This application provides uses the method for gamma transformation to reject unnecessary illumination.But this method computing complicacy and reliability are not high.

In sum, press at present and propose simply a kind of and remove the object detection method of illumination effect effectively.

Summary of the invention

The objective of the invention is to solve the inaccurate and unreliable problem of the detection target that in target detection, produces owing to illumination effect.In order to achieve the above object, the invention provides a kind of object detection method of removing illumination effect, said object detection method may further comprise the steps:

Step 101: set up background image;

Step 102: calculate and the gradient of output current frame image and the gradient of said background image, said gradient comprises horizontal direction gradient and vertical gradient;

Step 103: the direction and the amplitude of the gradient of more said current frame image and the gradient of said background image, and extract and export prospect profile in view of the above; With

Step 104: the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

Preferably, instep 101, adopt statistical method to set up background image.Said statistical method is to (the x of the pixel in the image of gathering in certain section time t; Y) carrying out statistical study (is exactly to pixel (x; Y) gray-scale value carries out simple number statistics and adds up), select this pixel in this section period (x, stable gray-scale value y) is (with pixel (x; Y) gray-scale value occurrence number is maximum elects stable gray-scale value as) as a setting in the image to should pixel (x; Y) gray-scale value through adding up in this section period t the stable gray-scale value of each pixel in the images acquired, thereby obtains background image.

Preferably, instep 102, adopt gradient operator to calculate horizontal direction gradient, the vertical gradient of said current frame image respectively, and horizontal direction gradient, the vertical gradient of said background image.Said gradient operator is Robert operator or Sobel operator.Robert operator and Sobel operator are the gradient algorithms in the general image treatment, repeat no more here.

Preferably,step 103 may further comprise the steps:

Step 1031: according to the gradient magnitude A2 and the gradient direction θ 2 of each pixel in the horizontal direction gradient of the horizontal direction gradient of the said current frame image ofstep 102 output and vertical gradient and said background image and gradient magnitude A1 that vertical gradient is calculated each pixel in the said current frame image and gradient direction θ 1 and the said background image;

Step 1032: if the pixel in the said current frame image (x, this pixel in gradient magnitude A1 y) and the said background image (x, gradient magnitude A2 y) all>=first threshold T1, then change step 1033 over to; If gradient magnitude A1 and A2 all≤the second threshold value T2, then think this pixel (x y) be noise spot, otherwise calculating | A1-A2|; If | A1-A2|>=the 3rd threshold value T3, think that then (x y) belongs to the foreground point to this pixel; Wherein, first threshold T1 ∈ [8,12], the second threshold value T2 ∈ [3,5], the 3rd threshold value T3 ∈ [4,6];

Step 1033: calculate this pixel (x in the said current frame image; Gradient direction θ 1 y) and this pixel in the said background image (x, the absolute difference of gradient direction θ 2 y) | θ 1-θ 2|, if | θ 1-θ 2| >=the 4th threshold value T4; Think that then (x y) belongs to the foreground point to this pixel; Wherein, the 4th threshold value T4 ∈ [18 °, 22 °]; With

Step 1034: extract the pixel that all belong to the foreground point, thereby obtain prospect profile.

Preferably,step 104 may further comprise the steps:

Step 1041: the prospect profile tostep 103 output is filled to obtain the prospect agglomerate;

Step 1042: calculate the error image of current frame image and background image, adopt thresholding method that this error image is carried out Threshold Segmentation to obtain the variation prospect in the error image; With

Step 1043: said prospect agglomerate and said variation prospect are carried out AND-operation, with the pixel that belongs to said prospect agglomerate and said variation prospect simultaneously as impact point to obtain and export target.

In addition, the present invention also provides a kind of object detecting device of removing illumination effect, and said device comprises: background is set up the unit, is used to adopt statistical method to set up background image; The gradient calculation unit is used to calculate and export the gradient of current frame image and the gradient of background image, and said gradient comprises horizontal direction gradient and vertical gradient; The prospect profile extraction unit is used for the direction and the amplitude of gradient of gradient and the said background image of more said current frame image, and extracts and export prospect profile in view of the above; With the target acquiring unit, be used for the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

Compared with prior art, object detection method of the present invention can detect the target of removing illumination effect exactly with device, has solved the inaccurate and unreliable problem of the detection target that produces owing to illumination effect in the target detection effectively.

Description of drawings

Fig. 1 is the process flow diagram according to the object detection method of removal illumination effect of the present invention;

Fig. 2 is the process flow diagram according to thestep 103 of object detection method of the present invention;

Fig. 3 is the process flow diagram according to thestep 104 of object detection method of the present invention;

Fig. 4 is the frame diagram according to the object detecting device of removal illumination effect of the present invention.

Embodiment

For making your auditor can further understand structure of the present invention, characteristic and other purposes, combine appended preferred embodiment to specify as follows at present, illustrated preferred embodiment only is used to technical scheme of the present invention is described, and non-limiting the present invention.

The object detection method of removal illumination effect provided by the present invention is mainly used in the inaccurate and insecure problem of target detection that produces owing to illumination effect in the monitoring scene that solves.

As shown in Figure 1, Fig. 1 is the process flow diagram according to the object detection method of removal illumination effect of the present invention.Can find out that by Fig. 1 the object detection method of removal illumination effect of the present invention may further comprise the steps:

Step 101: set up background image;

Step 102: calculate and the gradient of output current frame image and the gradient of said background image, said gradient comprises horizontal direction gradient and vertical gradient;

Step 103: the direction and the amplitude of the gradient of more said current frame image and the gradient of said background image, and extract and export prospect profile in view of the above; With

Step 104: the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

Wherein, the background image set up ofstep 101 can be start frame image or specific still image.But in order to ensure the stability and the accuracy of background image, the statistical method of preferred employing is set up background image in step 101.This statistical method is implemented through following steps: to the pixel (x in the image of gathering in certain section time t; Y) carrying out statistical study (is exactly to pixel (x; Y) gray-scale value carries out simple number statistics and adds up), select this pixel in this section period (x, stable gray-scale value y) is (with pixel (x; Y) gray-scale value occurrence number is maximum elects stable gray-scale value as) as a setting in the image to should pixel (x; Y) gray-scale value through adding up in this section period t the stable gray-scale value of each pixel in the images acquired, thereby obtains background image.

Instep 102, can adopt gradient operator to calculate horizontal direction gradient, the vertical gradient of current frame image respectively, and horizontal direction gradient, the vertical gradient of background image.Wherein, this gradient operator is preferably Robert operator or Sobel operator.For example, can adopt 3 * 3Robert operator to calculate horizontal direction gradient, the vertical gradient of current frame image respectively, and horizontal direction gradient, the vertical gradient of calculating background image.

The horizontal direction gradient of 3 * 3Robert operator computed image, vertical gradient utilize 3 * 3Robert operator horizontal direction template, vertical direction template computed image interior pixel to put the horizontal difference of corresponding level, vertical formwork, vertical difference exactly.For example, can select 3 * 3Robert operator horizontal direction template to be:

The vertical direction template is:

Pixel (x, horizontal direction gradient S y) then_H(x, y), vertical gradient S_V(x y) is respectively:

S_H(x，y)＝(f_x+1，y-1+2f_x+1，y+f_x+1，y+1)-(f_x-1，y-1+2f_x-1，y+f_x-1，y+1)

S_V(x，y)＝(f_x-1，y+1+2f_x，y+1+f_x+1，y+1)-(f_x-1，y-1+2f_x，y-1+f_x+1，y-1)

f_{X, y}Remarked pixel point (x, gray-scale value y).)

As shown in Figure 2, Fig. 2 is the process flow diagram according to thestep 103 of object detection method of the present invention.Can find out by Fig. 2, can may further comprise the steps according to thestep 103 of object detection method of the present invention:

Step 1031: according to the gradient magnitude A2 and the gradient direction θ 2 of each pixel in the gradient magnitude A1 of each pixel in the horizontal direction gradient of the horizontal direction gradient of the current frame image ofstep 102 output and vertical gradient and background image and the vertical gradient calculating current frame image and gradient direction θ 1 and the background image.

Image interior pixel point (x, gradient magnitude y), gradient direction computing formula are following:

$A(x,y)=\sqrt{{S_H}^2(x,y)+{S_V}^2(x,y)}$

$\mathrm{\θ}(x,y)=\mathrm{arctg}\frac{S_V(x,y)}{S_H(x,y)}$

Step 1032: if the pixel in the current frame image (this pixel in the x, gradient magnitude A1 y) and background image (x, gradient magnitude A2 y) is all more than or equal to first threshold T1; Then change step 1033 over to, if gradient magnitude A1 and A2, then think this pixel (x all smaller or equal to the second threshold value T2; Y) be noise spot, otherwise calculate | A1-A2|, if | A1-A2| is more than or equal to the 3rd threshold value T3; Think that then (x y) belongs to the foreground point to this pixel.Preferably, first threshold T1 ∈ [8,12], the second threshold value T2 ∈ [3,5], the 3rd threshold value T3 ∈ [4,6].

Step 1033: calculate this pixel (x in the current frame image; Gradient direction θ 1 y) and this pixel in the background image (x, the absolute difference of gradient direction θ 2 y) | θ 1-θ 2|, if | θ 1-θ 2| is more than or equal to the 4th threshold value T4; Think that then (x y) belongs to the foreground point to this pixel.Preferably, the 4th threshold value T4 ∈ [18 °, 22 °].

Step 1034: extract the pixel that all belong to the foreground point, thereby obtain prospect profile.

As shown in Figure 3, Fig. 3 is the process flow diagram according to thestep 104 of object detection method of the present invention.Can find out by Fig. 3, can may further comprise the steps according to thestep 104 of object detection method of the present invention:

Step 1041: the prospect profile tostep 103 output is filled to obtain the prospect agglomerate; The method that profile is filled is a lot; For example can adopt the horizontal direction scanning method, step can be following: the rectangular area with each prospect profile is an object, goes from first of rectangular area to begin scanning; By order from left to right; Scan first point (being leftmost profile) and last point (being the point of right sides), then the pixel between these two point all is made as the foreground point, continues to finish until this line scanning; Begin to scan next line, until last column; The agglomerate that all point after having scanned and foreground point are formed is the prospect agglomerate.

Step 1042: calculate the error image of current frame image and background image, adopt thresholding method that this error image is carried out Threshold Segmentation to obtain the variation prospect in the error image.

Thresholding method is the method for image interior pixel point being cut apart according to threshold value.The choosing method of said threshold value is a lot, and one dimension threshold value, two-dimentional threshold value are arranged.Be example with the simple one dimension fixed threshold of an example below: if the gray-scale value of certain point then is designated as " 1 " to be expressed as the foreground point greater than preset threshold in this error image; Otherwise be designated as " 0 " to be expressed as background dot, obtain the bianry image of prospect thus.

Step 1043: said prospect agglomerate and said variation prospect are carried out AND-operation, with the pixel that belongs to said prospect agglomerate and said variation prospect simultaneously as impact point to obtain and export target.

AND-operation is general a kind of computer operation, if interior certain pixel of image belongs to prospect agglomerate and variation prospect simultaneously particularly, thinks that then this pixel is that impact point then obtains and exports.

As shown in Figure 4, Fig. 4 is the frame diagram according to the object detecting device of removal illumination effect of the present invention.Can find out that by Fig. 4 the object detecting device of removal illumination effect of the present invention comprises:

Background is set up unit 1, is used to set up background image;

Gradient calculation unit 2 is used to calculate and export the gradient of current frame image and background image, and said gradient comprises horizontal direction gradient and vertical gradient;

Prospect profile extraction unit 3 is used for the direction and the amplitude of the gradient of more said current frame image and said background image, and extracts and export prospect profile in view of the above;

Target acquiring unit 4 be used for the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

Compared with prior art, object detection method of the present invention and device have solved the inaccurate and unreliable problem of the detection target that produces owing to illumination effect in the target detection effectively.

What need statement is that foregoing invention content and embodiment are intended to prove the practical application of technical scheme provided by the present invention, should not be construed as the qualification to protection domain of the present invention.Those skilled in the art are in spirit of the present invention and principle, when doing various modifications, being equal to replacement or improvement.Protection scope of the present invention is as the criterion with appended claims.

Claims (5)

1. an object detection method of removing illumination effect is characterized in that, said object detection method may further comprise the steps:

Step 101: adopt statistical method to set up background image;

Step 102: calculate and the gradient of output current frame image and the gradient of said background image, said gradient comprises horizontal direction gradient and vertical gradient;

Step 103 comprises the steps 1031, step 1032, step 1033 and step 1034:

Step 1031: according to the gradient magnitude A2 and the gradient direction θ 2 of each pixel in the horizontal direction gradient of the horizontal direction gradient of the said current frame image of step 102 output and vertical gradient and said background image and gradient magnitude A1 that vertical gradient is calculated each pixel in the said current frame image and gradient direction θ 1 and the said background image;

Step 1032: if the pixel in the said current frame image (x, this pixel in gradient magnitude A1 y) and the said background image (x, gradient magnitude A2 y) all>=first threshold T1, then change step 1033 over to; If gradient magnitude A1 and A2 all≤the second threshold value T2, then think this pixel (x y) be noise spot, otherwise calculating | A1-A2|; If | A1-A2|>=the 3rd threshold value T3, think that then (x y) belongs to the foreground point to this pixel;

Step 1033: calculate this pixel (x in the said current frame image; Gradient direction θ 1 y) and this pixel in the said background image (x, the absolute difference of gradient direction θ 2 y) | θ 1-θ 2|, if | θ 1-θ 2| >=the 4th threshold value T4; Think that then (x y) belongs to the foreground point to this pixel;

Step 1034: extract the pixel that all belong to the foreground point, thereby obtain prospect profile; With

Step 104: the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

2. object detection method according to claim 1; It is characterized in that; In step 102, adopt gradient operator to calculate horizontal direction gradient, the vertical gradient of said current frame image respectively, and horizontal direction gradient, the vertical gradient of said background image.

3. object detection method according to claim 1 is characterized in that, first threshold T1 ∈ [8,12], the second threshold value T2 ∈ [3,5], the 3rd threshold value T3 ∈ [4,6], the 4th threshold value T4 ∈ [18 °, 22 °].

4. object detection method according to claim 1 is characterized in that step 104 may further comprise the steps:

Step 1041: the prospect profile to step 103 output is filled to obtain the prospect agglomerate;

Step 1042: calculate the error image of current frame image and background image, adopt thresholding method that this error image is carried out Threshold Segmentation, to obtain the variation prospect in the error image;

Step 1043: said prospect agglomerate and said variation prospect are carried out AND-operation, with the pixel that belongs to said prospect agglomerate and said variation prospect simultaneously as impact point to obtain and export target.

5. an object detecting device of removing illumination effect is characterized in that, said device comprises:

Background is set up the unit, is used to adopt statistical method to set up background image;

The gradient calculation unit is used to calculate and export the gradient of current frame image and the gradient of background image, and said gradient comprises horizontal direction gradient and vertical gradient;

The prospect profile extraction unit; Be used for, according to the gradient magnitude A2 and the gradient direction θ 2 of each pixel in the horizontal direction gradient of the horizontal direction gradient of the said current frame image of gradient calculation unit output and vertical gradient and said background image and gradient magnitude A1 that vertical gradient is calculated each pixel in the said current frame image and gradient direction θ 1 and the said background image; If the pixel in the said current frame image (this pixel in the x, gradient magnitude A1 y) and said background image (x, gradient magnitude A2 y) all>=first threshold T1; Then calculate this pixel (x in the said current frame image; Gradient direction θ 1 y) and this pixel in the said background image (x, the absolute difference of gradient direction θ 2 y) | θ 1-θ 2|, if | θ 1-θ 2|>=the 4th threshold value T4; Think that then (x y) belongs to the foreground point to this pixel; If gradient magnitude A1 and A2 all≤the second threshold value T2, then think this pixel (x y) be noise spot, otherwise calculating | A1-A2|; If | A1-A2|>=the 3rd threshold value T3, think that then (x y) belongs to the foreground point to this pixel; Extract the pixel that all belong to the foreground point, thereby obtain prospect profile; And

The target acquiring unit be used for the prospect profile that extracts is filled obtaining the prospect agglomerate, and filtering noise is with export target.

Images