CN104463902A - Stationary target elimination method based on NMI feature - Google Patents

Abstract

The invention belongs to the technical field of photoelectric product application and particularly relates to a stationary target elimination method based on the normalized moment of inertia (NMI) feature. Firstly, foreground areas of a moving object are obtained by utilization of a multi-Gaussian background modeling method and a three-frame cumulative frame differential method and are denoted as Fgb and Fgd, and then Fgb and Fgd are subjected to foreground area marking respectively. The NMI invariant feature of an image is adopted, namely the NMI of the image serves as a feature matching parameter, so that the NMI feature matching accuracy is high, the calculated amount is small, the speed is high, and gray level distortion resistance and good invariance maintaining of geometric distortion are achieved.

Description

A kind of static target removing method based on NMI feature

Technical field

The invention belongs to photovoltaic applied technical field, be specifically related to a kind of static target removing method based on NMI (NormalizedMoment of Inertia) feature.

Background technology

Intelligent Target analysis and video image searching system are that real-time intelligent video monitoring system further expands.This system comprises: man-machine interaction unit, dispatch control unit, intellectual analysis unit and resource database, completes the decoding of video, analysis, concern information extraction and event trace retrieval.System receives user format retrieval and inquisition condition, the relevant massive video data in video storaging equipment is read by network, the video of decoding different coding form, by to the intellectual analysis of video content and concentration extraction, obtain user and pay close attention to target and realize target tracking and target association, matching forms the space historical track paying close attention to target, alleviates the query and search work of clue target in case detecting process, shorten target retrieval required time, improve efficiency of solving a case.

How can from sequence of video images by interested object extraction out, be initial in Intelligent Target analysis and video image searching system be also a most important step.The validity of video frequency motion target detection algorithm directly has influence on the treatment effect of follow-up system, in the application of reality, because in complex scene, moving target exists the possibility becoming stationary state from motion state.The change of this state easily causes the error detection of moving target, thus brings direct impact to the intellectual analysis of video.

Therefore, finding a kind of effective ways eliminating static target, is the vital task improving Design of System Software.

Summary of the invention

The technical problem to be solved in the present invention is to provide a kind of image processing algorithm realizing mobile target in complex background and accurately detect, be mainly used in Intelligent Target analysis and video image searching system, based in the Detection for Moving Target of background modeling, eliminate background residual region when moving target changes stationary state into from motion state, the moving target realized under complex background accurately detects, thus improves the accuracy of video frequency search system intellectual analysis.

In order to realize this purpose, the technical scheme that the present invention takes is:

Based on a static target removing method for NMI feature, comprise the following steps:

1) under capture apparatus stationary state, current video sequence image is gathered;

2) adopt many gauss hybrid models algorithm to detect moving target, obtain moving target foreground picture Fgb;

3) adopt three frame accumulation difference algorithms to detect moving target, obtain moving target foreground picture Fgd;

4) barycenter (x of foreground area in Fgb figure is calculated _o, y _o), if area size is (M, N), f (x, y) is the value at picture position (x, y) place, and prospect is 1, and background is 0:

$\begin{array}{cc}{x}_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}x\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}& y_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}y\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}\end{array};$

5) calculate around barycenter (x in Fgb figure _o, y _o) moment of inertia be designated as

$J_{(x_o,y_o)}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{|(x,y)-(x_o,y_o)|}^{2}f(x,y)=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y);$

6) moment of inertia after normalization is calculated in Fgb figure;

$\mathrm{NMI}=\frac{\sqrt{J_{(x_o,y_o)}}}{m}=\frac{\sqrt{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y)}}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}$

7) adopt identical computing method to divide the foreground area Fgd obtained to be normalized the calculating of moment of inertia to three frame cumulative errors, obtain NMI2;

8) NMI1 and NMI2 alpha value calculated is utilized:

$\mathrm{\α}=\frac{|\mathrm{NMI}1-\mathrm{NMI}2|}{\mathrm{NMI}1}$

9) when α value is less than threshold value Th, then judge the remaining region of this regional background, do not mark.

Further, a kind of static target removing method based on NMI feature as above, step 9) in, threshold value Th is chosen by test according to concrete video scene.

Technical solution of the present invention is based on the static target technology for eliminating of NMI feature, by while background modeling and multiple frame cumulation difference, choose the background residual of NMI feature to static target to differentiate, this technology can eliminate background residual region when target changes stationary state into from motion state, and the moving target realized under complex background accurately detects.By testing the visible light sequential image under street environment, demonstrating this algorithm and there is good effect.

Embodiment

Below technical solution of the present invention is further elaborated.

Know-why of the present invention---when adopting many Gaussian Background modeling technique to detect moving target, due to the impact of background model update method and renewal rate, in moving target foreground segmentation figure, static target can be produced error detection problem as sport foreground (i.e. background residual).First this algorithm utilizes the method for many Gaussian Background modeling method and three frame accumulation frame differences to obtain moving target foreground area, is designated as Fgb and Fgd.Then respectively foreground area mark is carried out to Fgb and Fgd.Due to the motion of target itself and the impact of extraneous factor, the shape of target, size, light and shade can constantly changes.Directly not high based on the matching precision of foreground area template matching method, the extracting method of traditional image invariant features is complicated, and calculated amount is large, and the real-time for massive video searching system has a great impact.This algorithm adopts NMI (the NormalizedMoment of Inertia) invariant feature of image, namely the normalization moment of inertia of image is as characteristic matching parameter, NMI characteristic matching accuracy is high, calculated amount is little, speed is fast, and has anti-tonal distortion and have geometric distortion and maintain the invariance preferably.

Embody rule object of the present invention---Intelligent Target analysis and video image searching system, the algorithm that the present invention proposes adopts C Plus Plus programming realization under VC6.0 platform, in order to realize moving object detection under complex scene and static target is eliminated, comprise the following steps:

1) under capture apparatus stationary state, current video sequence image is gathered;

2) adopt many gauss hybrid models algorithm to detect moving target, obtain moving target foreground picture Fgb;

3) adopt three frame accumulation difference algorithms to detect moving target, obtain moving target foreground picture Fgd;

4) barycenter (x of foreground area in Fgb figure is calculated _o, y _o), if area size is (M, N), f (x, y) is the value at picture position (x, y) place, and prospect is 1, and background is 0:

$\begin{array}{cc}{x}_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}x\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}& y_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}y\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}\end{array};$

5) calculate around barycenter (x _o, y _o) moment of inertia be designated as

$J_{(x_o,y_o)}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{|(x,y)-(x_o,y_o)|}^{2}f(x,y)=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y);$

6) moment of inertia, i.e. NMI1 after calculating normalization;

$\mathrm{NMI}=\frac{\sqrt{J_{(x_o,y_o)}}}{m}=\frac{\sqrt{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y)}}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}$

7) adopt identical computing method to divide the foreground area Fgd obtained to be normalized the calculating of moment of inertia to three frame cumulative errors, obtain NMI2;

8) NMI1 and NMI2 alpha value calculated is utilized:

$\mathrm{\α}=\frac{|\mathrm{NMI}1-\mathrm{NMI}2|}{\mathrm{NMI}1}$

9) when α value is less than threshold value Th (threshold value Th is chosen by test according to concrete video scene).Then judge the remaining region of this regional background, do not mark.

Claims (2)

1., based on a static target removing method for NMI feature, it is characterized in that: comprise the following steps:

1) under capture apparatus stationary state, current video sequence image is gathered;

2) adopt many gauss hybrid models algorithm to detect moving target, obtain moving target foreground picture Fgb;

3) adopt three frame accumulation difference algorithms to detect moving target, obtain moving target foreground picture Fgd;

4) barycenter (x of foreground area in Fgb figure is calculated _o, y _o), if area size is (M, N), f (x, y) is the value at picture position (x, y) place, and prospect is 1, and background is 0:

$\begin{array}{cc}{x}_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}x\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}& y_o=\frac{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}y\·f(x,y)}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}\end{array};$

5) calculate around barycenter (x in Fgb figure _o, y _o) moment of inertia be designated as

$J_{(x_o,y_o)}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{|(x,y)-(x_o,y_o)|}^{2}f(x,y)=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y);$

6) moment of inertia after normalization is calculated in Fgb figure;

$\mathrm{NMI}=\frac{\sqrt{J_{(x_o,y_o)}}}{m}=\frac{\sqrt{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-x_o)}^2+{(y-y_o)}^2)f(x,y)}}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}$

7) adopt identical computing method to divide the foreground area Fgd obtained to be normalized the calculating of moment of inertia to three frame cumulative errors, obtain NMI2;

8) NMI1 and NMI2 alpha value calculated is utilized:

$\mathrm{\α}=\frac{|\mathrm{NMI}1-\mathrm{NMI}2|}{\mathrm{NMI}1}$

9) when α value is less than threshold value Th, then judge the remaining region of this regional background, do not mark.

2. a kind of static target removing method based on NMI feature as claimed in claim 1, is characterized in that: step 9) in, threshold value Th is chosen by test according to concrete video scene.