CN101344967B - Detection method for small mobile objective in astronomical image - Google Patents

Abstract

The invention discloses a detecting method of a dim moving target in an astronomical image. The method comprises the steps that: first, a feature integrated model of a selective visual centralized mechanism is adopted to acquire a two-dimension trajectory projection image so as to obtain an integral image of focus of attention, namely, a projection image containing a target trajectory; then a join-point problem model in a curve is adopted for carrying out the detection of the target trajectory, and the target trajectory is obtained finally by backward pass; later, the target trajectory detected in the two-dimension projection image is mapped back to every frame in an initial image sequence before the projection and a square weighted centroid extraction method is adopted to carry out the centroid localization of the target. Since the detection of a trajectory is carried out by the utilization of the trajectory continuity of the target, and the detection of the dim moving target is carried out by the method of the combination of the selective visual centralized mechanism and the curve discovery of the join-point problem model, the detection rate of the method is improved to 85 percent to 89 percent from less than 50 percent of the prior art when signal-to-noise ratio is less than 2.

Description

The detection method of small and weak moving-target in the astronomic graph picture

Technical field

The present invention relates to the detection method of small and weak moving-target in a kind of small and weak moving target detection method, particularly astronomic graph picture.

Background technology

The small and weak moving-target of low signal-to-noise ratio detects problem and has directly determined the operating distance of astronomical sight system and detected performance that head it off has very important significance for improving system performance effectively.The detection method of existing Weak target mainly contains: based on the detection method of track projection and the method for searching for based on three-dimensional track.Document " real time detection algorithm of weak targets in astronomical images, photoelectric project, 2005,32 (12): 1-4 " discloses a kind of algorithm that detects the Weak target track on track two-dimensional projection basis.The method adopts background inhibition and Threshold Segmentation to obtain containing the binary map of moving-target, adopts spatial domain differential filtering and two steps of target point set matching method to detect small and weak moving-target afterwards.Contain the moving-target binary map stage obtaining, the method utilizes first three two field picture of sequence image to carry out background estimating, and adopts the automatic threshold dividing method to carry out gray level image to cut apart, and therefore, has limitation for the small and weak degree of target; At spatial domain differential filtering and target point set matching stage, because the precision of coupling has determined that directly the Weak target verification and measurement ratio is low, and the noise immunity of differential filtering is relatively poor.Less than 2 o'clock, verification and measurement ratio was lower than 50% at target signal to noise ratio.

Summary of the invention

The prior art verification and measurement ratio is low in order to overcome, the deficiency of noise immunity difference, the invention provides the detection method of small and weak moving-target in a kind of astronomic graph picture, employing finds that based on the selective visual centralized mechanism with based on the curve of tie point problem model the method that combines carries out the detection of small and weak moving-target, can improve verification and measurement ratio, and noise immunity is strong.

The technical solution adopted for the present invention to solve the technical problems: the detection method of small and weak moving-target in a kind of astronomic graph picture is characterized in that comprising the steps:

(a) adopt the feature integration model in the selective visual centralized mechanism to carry out obtaining of two-dimentional track perspective view, global scene figure is carried out level and vertical even division, obtain mutually disjoint zonule, selecting the gray feature that wherein has vision significance gathers as focus, reject the characteristic element that does not wherein partly satisfy restrictive condition and constitute final focus set, described restrictive condition is:

${\hat{G}}_{\mathrm{ij}}=\left\{\begin{array}{cc}{t}_{\mathrm{ij}}=g_{\mathrm{ij}},& g_{\mathrm{ij}}\&GreaterEqual;3\mathrm{\&sigma;}\\ t_{\mathrm{ij}}=0,& g_{\mathrm{ij}}<3\mathrm{\&sigma;}\end{array}\right.---\left(2\right)$

$T=\left\{{\hat{G}}_{\mathrm{ij}}\right|i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m;j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n\}---\left(3\right)$

In the formula, t _IjExpression potential target point gray-scale value, Expression potential target point position, T represents the set of potential target point, σ is the image variance;

At last the focus of every two field picture is gathered according to rule:

$F(x,y)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}[f_i(x,y)-f_{i+1}(x,y)]---\left(4\right)$

Carry out feature and integrate, in the formula, F (x, the image after y) expression is integrated, f _i(x, y) the i frame focus-of-attention set in the expression sequence image;

Constitute the two-dimentional track perspective view that contains target trajectory;

(b) adopt the tie point problem model of curve in finding to carry out the detection of target trajectory, regard each data point stage of detection problem as, each stage has only a state, with the condition of continuity of curve

$M\left(p_i\right)=1+\underset{p_i\&Element;\mathrm{Cone}\left(p_j\right)}{\max }\left(M\left(p_j\right)\right)---\left(5\right)$

As decision condition, and the maximum number of employing curve process point is as target function, travel through the point of process in order, calculate the desired value in each stage, the point of desired value maximum is last point of aim curve, uses the backstepping method at last, can finally obtain having a few on the aim curve, promptly obtain target trajectory;

(c), and finally carry out the barycenter location of target by square weighting barycenter extraction method with on each frame in the original sequence before the target trajectory mapping backing up shadow that detects in the two-dimension projection in the step (a).

The invention has the beneficial effects as follows: owing to utilize the track continuity of target to carry out the detection of track, the method that the curve discovery of employing selective visual centralized mechanism and tie point problem model combines is carried out the detection of small and weak moving-target, method based on the selective visual centralized mechanism has reduced pending quantity of information and ground unrest effectively, curve discover method based on the tie point problem model has stronger noise immunity, therefore, to the movement velocity of moving-target without limits, less than 2 o'clock, verification and measurement ratio 50% brought up to 85～89% by being lower than of prior art in signal to noise ratio (S/N ratio).

Below in conjunction with drawings and Examples the present invention is elaborated.

Description of drawings

Accompanying drawing is the detection method process flow diagram of small and weak moving-target in the astronomic graph picture of the present invention.

Embodiment

With reference to accompanying drawing.At first carry out two-dimentional track projection.

The original image of 1024 * 1024 sizes of input is carried out the even division of level and vertical direction, obtain 4 * 4 impartial zonule set of size, they mutually disjoint separately.For the zonule of even division, the pixel in the zone under fixed star and the target has significant gray scale optimal characteristics, and promptly in each zonule, the focus-of-attention of target area should be one of one or several of gray-scale value maximum.Therefore, according to formula (1), ask for the maximal value in the zonule, the gamma characteristic that is about in each zonule is arranged, and obtains an order statistic, and the maximal value of getting this statistic promptly obtains the focus-of-attention of zonule.Wherein, G _IjExpression piece R _IjThe picture element of middle gray-scale value maximum, g _IjExpression G _IjGray-scale value, (x, y) presentation video is at (x, the gray-scale value of y) locating for I.

$g_{\mathrm{ij}}=\underset{(x,y)\&Element;R_{\mathrm{ij}}}{\max }\left(I(x,y)\right)---\left(1\right)$

After the focus-of-attention of having asked for each zonule, the focus-of-attention of each zonule is merged, constitute the focus-of-attention set of original image, wherein not only comprise the Weak target that will detect to some extent, also comprise the decoy that some fixed star backgrounds and much noise are produced.Therefore, for the rejecting of the focus-of-attention element of loseing interest in, can significantly reduce the valueless quantity of information of focus-of-attention set.According to formula (2), (3), each element and σ=25 in the initial focus-of-attention set are compared, reject the focus-of-attention of loseing interest in, can obtain final focus-of-attention set.Wherein, t _IjExpression potential target point gray-scale value,

Expression potential target point position, T represents the set of potential target point, σ is the image variance.

${\hat{G}}_{\mathrm{ij}}=\left\{\begin{array}{cc}{t}_{\mathrm{ij}}=g_{\mathrm{ij}},& g_{\mathrm{ij}}\&GreaterEqual;3\mathrm{\&sigma;}\\ t_{\mathrm{ij}}=0,& g_{\mathrm{ij}}<3\mathrm{\&sigma;}\end{array}\right.---\left(2\right)$

$T=\left\{{\hat{G}}_{\mathrm{ij}}\right|i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m;j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n\}---\left(3\right)$

According to the FIT model theory of Treismna, will be integrated into the vision object with serial mode at these visual signatures of attention stage.We have obtained the focus-of-attention set of a two field picture processing stage of preceding attention, owing to institute studies a question is at sequence image, therefore, can obtain corresponding focus-of-attention set to each two field picture, in the focus-of-attention set of different frame, focus-of-attention for the fixed star background of extracting, because the unchangeability (unanimity of focus-of-attention position) of background, by formula (4), these focus-of-attentions are integrated again, comprised target trajectory information in interior new focus-of-attention set obtaining.Wherein, F (x, the image after y) expression is integrated, f _i(x, y) the i frame focus-of-attention set in the expression sequence image.

$F(x,y)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}[f_i(x,y)-f_{i+1}(x,y)]---\left(4\right)$

In second step, carry out target trajectory and detect.

Regard each data point stage of detection problem as, each stage has only a state.As decision condition, and adopt the curve negotiating current generation to pass through a little maximum number the condition of continuity of curve as target function.Travel through all stages in order, calculate the desired value in each stage, the point of desired value maximum is last point of aim curve, uses the backstepping method at last, can finally obtain having a few on the aim curve, promptly obtains target trajectory.

At first, note Lip _cFor β equals the class of a curve of c, L _N(Lip _c) be the maximum number of points of curve negotiating, Cone (p _i) expression point p _iThe set of the point on the right, and point and p in the satisfied set _iLine and horizontal angle be no more than arctan (c).

Data point is from left to right sorted, each some p _iBe divided into a stage, there is a state in each stage.Decision-making is selected for the next one of some points, here we will with a p _iSatisfy the point set of certain serial relation and make a strategic decision, be designated as Cone (p as it _i), target function is M (p _i), promptly represent stage p _iThe maximum number of points of curve negotiating during point.Can calculate by following strategy.

$M\left(p_i\right)=1+\underset{p_i\&Element;\mathrm{Cone}\left(p_j\right)}{\max }\left(M\left(p_j\right)\right)---\left(5\right)$

Calculate the desired value of each state according to following formula, maximum in all M values at last is exactly the maximum number of points of curve negotiating, that is:

$L_N=\left({\mathrm{Lip}}_c\right)=\underset{i}{\max }M\left(p_i\right)---\left(6\right)$

Note phase points p _iPrevious phase points, and guarantee $\underset{i}{\max }M\left(p_i\right)>\mathrm{\&mu;},$ μ is a specified rate, and decision detects the curve obtain and whether satisfies the track requirement of counting, and at last by the backtracking method, seeks point on all curves from the some backstepping of M maximum, has promptly finally realized the detection of target trajectory.

In the 3rd step, carry out target barycenter location.

On each frame in the target trajectory that detects in the two-dimension projection original sequence before according to corresponding coordinate position mapping backing up shadow, and locate by the barycenter that formula (7) finally carries out target.

$x_0=\frac{\underset{x=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{n}{\mathrm{\&Sigma;}}}{F}^2(x,y)x}{\underset{x=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{n}{\mathrm{\&Sigma;}}}{F}^2(x,y)}$ $y_0=\frac{\underset{x=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{n}{\mathrm{\&Sigma;}}}{F}^2(x,y)y}{\underset{x=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{n}{\mathrm{\&Sigma;}}}{F}^2(x,y)}---\left(7\right)$

Wherein, x, y are the horizontal ordinate of pixel, and (x y) is that ((m n) is the size of window, x for x, the gray-scale value of y) locating to F ₀, y ₀Be the centre coordinate of trying to achieve.

Utilize the inventive method to carry out the detection of small and weak moving-target, less than 2 o'clock, verification and measurement ratio was 85～89% in signal to noise ratio (S/N ratio).

Claims (1)

1. the detection method of small and weak moving-target in the astronomic graph picture is characterized in that comprising the steps:

(a) adopt the feature integration model in the selective visual centralized mechanism to carry out obtaining of two-dimentional track perspective view, global scene figure is carried out level and vertical even division, obtain mutually disjoint zonule, the gray feature that selects the point that wherein has vision significance is gathered as focus, reject the characteristic element that does not wherein partly satisfy restrictive condition and constitute final focus set, described restrictive condition is:

${\hat{G}}_{\mathrm{ij}}=\left\{\begin{array}{cc}{t}_{\mathrm{ij}}=g_{\mathrm{ij}},& g_{\mathrm{ij}}\&GreaterEqual;3\mathrm{\&sigma;}\\ t_{\mathrm{ij}}=0,& g_{\mathrm{ij}}<3\mathrm{\&sigma;}\end{array}\right.---\left(3\right)$

$T=\left\{{\hat{G}}_{\mathrm{ij}}\right|i=\mathrm{1,2},...,m;j=\mathrm{1,2},...,n\}---\left(3\right)$

In the formula, t _IjExpression potential target point gray-scale value,

Expression potential target point position, g _IjExpression G _IjGray-scale value; T represents the set of potential target point, and σ is the image variance;

At last the focus of every two field picture is gathered according to rule:

$F(x,y)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}[f_i(x,y)-f_{i+1}(x,y)]---\left(4\right)$

Carry out feature and integrate, in the formula, F (x, the image after y) expression is integrated, f _i(x, y) the i frame focus-of-attention set in the expression sequence image;

Constitute the two-dimentional track perspective view that contains target trajectory;

(b) adopt the tie point problem model of curve in finding to carry out the detection of target trajectory, regard each data point stage of detection problem as, each stage has only a state, with the condition of continuity of curve

$M\left(p_i\right)=1+\underset{p_i\&Element;\mathrm{Cone}\left(p_j\right)}{\max }\left(M\left(p_j\right)\right)---\left(5\right)$

As decision condition, and the maximum number of employing curve process point is as target function, travel through the point of process in order, calculate the desired value in each stage, the point of desired value maximum is last point of aim curve, uses the backstepping method at last, can finally obtain having a few on the aim curve, promptly obtain target trajectory;

(c), and finally carry out the barycenter location of target by square weighting barycenter extraction method with on each frame in the original sequence before the target trajectory mapping backing up shadow that detects in the two-dimension projection in the step (a).

Images