CN103985127A - Detecting method and device for weak and small target with intensive fixed star background - Google Patents

Abstract

The invention belongs to the field of image processing and discloses a detecting method for a weak and small target with an intensive fixed star background. The detecting method aims at achieving detection of a weak and small target under the conditions of large view field of the fixed star background and high detecting capacity. First, fixed star interference is weakened through inter-frame registration difference, and then an image is cut through a simple and efficient method of a self-adaption mean value and a K time variance; then a binary statistic filter is adopted to conduct target clustering and remove isolated noise points; characteristics of the position, size, gray level and external rectangle of the target are obtained through segment marking and the like; a static target is extracted through a method of frame-to-frame position correlation (a fixed star edge and double-start edges which are not completely removed through difference are removed at the same time); track detection is conducted through a nearest neighbor method based on logics to finally achieve detection of the weak and small target. After a series of preprocessing, candidate targets sent to the track correlation are few, the number of the candidate targets is controlled in a range of a few to a dozen, and therefore track correlation efficiency and target detecting efficiency are high.

Description

A kind of detection method of small target of intensive star background and device

Technical field

The invention belongs to image processing field, relate to a kind of object detection method, particularly a kind of detection method of small target that is applicable to intensive star background.

Background technology

Available radar and optoelectronic device are followed the tracks of in the detection of extraterrestrial target, and the target track that detects tracking when need is higher, and when area is very little, detections of radar is followed the tracks of and just seemed some difficulty.To this type of target, ground photo-electric telescope just demonstrates its huge advantage, want the fractionlet of high orbit in surveying, telescope must have enough detectivities, enhancing along with telescope detectivity, in range of telescope, a lot of faint fixed stars are also detected, and existing target in visual field, has again a large amount of fixed stars like this.Target how from intensive star background, fast and effeciently to be detected, and target is extracted is the technological difficulties that image is processed.When ground photo-electric telescope visual field is large, detectivity is when high, in image, fixed star is intensive, the general method that adopts difference to remove fixed star, but difference method can not be removed fixed star completely and disturb, therefore the succeeding target that participates in track association is more, and calculated amount and computing velocity are all difficult to meet the demand of real-time system, may cause very high false-alarm in addition, for these problems, the present invention has provided a set of simple solution efficiently.

Summary of the invention

The technical matters that the present invention solves: the present invention proposes a kind of detection method of small target that is applicable to intensive star background, the large visual field, the Weak target under high detectivity condition that have solved star background detect.Reduce the false alarm rate of target detection, improved correctness and the detection efficiency of target detection, improved targetpath association and starting velocity simultaneously.

The present invention is by the following technical solutions to achieve these goals:

The invention provides a kind of detection method of small target that is applicable to intensive star background, it is characterized in that step is as follows:

Step (1), difference: receiving sequence image, carries out interframe registration difference;

Step (2), image are cut apart: adopt m+k σ thresholding computing method to Image Segmentation Using, m is full figure average, and σ is the mean square deviation of full figure, and k is empirical value, and k span is 3.0-7.0;

Step (3), target cluster: the bianry image after step (2) image is cut apart is by pixels statistics N * N neighborhood internal object number of pixels, and when it is greater than C, what this pixel represented is target, is retained, otherwise weeds out this pixel; Be expressed as:

$B^{\′}(x,y)=\left\{\begin{array}{ccc}1& \mathrm{if}& \underset{(x,y)\∈(N\×N)}{\mathrm{\Σ}}B(x,y)>C\\ 0& \mathrm{else}& \end{array}\right.---\left(1\right)$

Wherein, B (x, y) represents the image after binaryzation, B'(x, y) be the bianry image after target cluster, N * N represents the neighborhood of current pixel, the pixel that in the neighborhood that C is current pixel, value is 1 with;

Step (4), target label: use the method for line segment mark, obtain the position coordinates of size, gray scale and each pixel of target of target;

Step (5), static target detect: static target is defined as and on image, shows as the target that interframe change in location is very little

Step (6), double star are removed: remove the fixed star edge that is detected as static target in step (5);

Step (7), flight path detect: use the track detecting method of logic-based arest neighbors to detect target trajectory, confirm real target.

In technique scheme, calculate the registration amount of interframe fixed star, fixed star interframe registration amount and the telescopical latitude of ground, point to orientation, just, pixel resolution is relevant with working frame frequency, the registration amount computing method of the middle registration difference of described step (1) are as follows:

A) calculate fixed star angular altitude speed

$\frac{\mathrm{dE}}{\mathrm{dt}}=\cos \mathrm{\φ}\sin A---\left(1\right)$

Wherein, E is photo-electric telescope angular altitude, and A is photo-electric telescope position angle, and φ is hour angle, is the latitude at photo-electric telescope place;

B) calculate fixed star Azimuth, Speed, Altitude

$\frac{\mathrm{dA}}{\mathrm{dt}}=\sin \mathrm{\φ}-\frac{\cos \mathrm{\φ}\·\cos A}{\tan z}---\left(2\right)$

Wherein, z is zenith angle, the same formula of other meaning of parameters (1);

C) by formula (1) formula (2) and the interframe time interval, calculate synchronous target at interframe movement amount e and a, interframe movement amount a and e are rounded to the pixel matching amount that just obtains horizontal and vertical again divided by pixel level and vertical resolution respectively, are designated as: obtain registration amount and adjacent two frames are carried out to difference after registration later, remove fixed star.

In technique scheme, in step (5), static target detects specific as follows:

Difference, cut apart, the position of target is T after cluster and two continuous frames mark _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, present frame sequence number is i, if meet:

abs(x _i-x _i-1)≤nTh and abs(y _i-y _i-1)≤nTh (3)

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be potential inspection target, by T _{i-1, i}(x _i, y _i) retain and give subsequent step to process, wherein nTh is predefined threshold value.

In technique scheme, in step (5), in static target detection, the interval of nTh is [2,10], and the value maximum of nTh is no more than the radius of target to be detected.

In technique scheme, in step (6), the removal method of double star is as follows:

After two continuous frames mark, the position of target is T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, and present frame sequence number is i, and registration amount is if meet formula (4):

$\mathrm{abs}(x_i-x_{i-1}-\stackrel{\‾}{x})\≤n{\mathrm{Th}}_x,\mathrm{and},\mathrm{abs}(y_i-y_{i-1}-\stackrel{\‾}{y})\≤n{\mathrm{Th}}_y---\left(4\right)$

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be the residual edge of double star after difference, by T _{i-1, i}(x _i, y _i) from result, delete, do not give flight path and detect, wherein nTh _xand nTh _yinterval be [2,10].

The present invention also provides a kind of Weak target pick-up unit that is applicable to intensive star background, it is characterized in that comprising:

Differential attachment: receiving sequence image, carries out interframe registration difference;

Image segmenting device: adopt m+k σ thresholding computing method to Image Segmentation Using, m is full figure average, and σ is the mean square deviation of full figure, and k is empirical value, and span is 3.0-7.0;

Target clustering apparatus: by pixels statistics N * N neighborhood internal object number of pixels, when it is greater than C, what this pixel represented is target, is retained, otherwise weeds out this pixel to the image after step image segmenting device binaryzation; Be expressed as:

$B^{\′}(x,y)=\left\{\begin{array}{ccc}1& \mathrm{if}& \underset{(x,y)\∈(N\×N)}{\mathrm{\Σ}}B(x,y)>C\\ 0& \mathrm{else}& \end{array}\right.---\left(2\right)$

Wherein, B (x, y) represents the image after binaryzation, B'(x, y) be the bianry image after target cluster, N * N represents the neighborhood of current pixel, the pixel that in the neighborhood that C is current pixel, value is 1 with;

Target label device: use the method for line segment mark, obtain the position coordinates of size, gray scale and each pixel of target of target;

Static target pick-up unit: static target is defined as and shows as the target that interframe change in location is very little on image

Double star removal device: remove the fixed star edge that is detected as static target in step static target pick-up unit;

Flight path pick-up unit: use the track detecting method of logic-based arest neighbors to detect target trajectory, confirm real target.

In technique scheme, calculate the registration amount of interframe fixed star, fixed star interframe registration amount and the telescopical latitude of ground, point to orientation, just, pixel resolution is relevant with working frame frequency, the registration amount computing method of the middle registration difference of described step (1) are as follows:

A) calculate fixed star angular altitude speed

$\frac{\mathrm{dE}}{\mathrm{dt}}=\cos \mathrm{\φ}\sin A---\left(5\right)$

Wherein, E is photo-electric telescope angular altitude, and A is photo-electric telescope position angle, and φ is hour angle, is the latitude at photo-electric telescope place;

B) calculate fixed star Azimuth, Speed, Altitude

$\frac{\mathrm{dA}}{\mathrm{dt}}=\sin \mathrm{\φ}-\frac{\cos \mathrm{\φ}\·\cos A}{\tan z}---\left(6\right)$

Wherein, z is zenith angle, the same formula of other meaning of parameters (1);

C) by formula (1) formula (2) and the interframe time interval, calculate synchronous target at interframe movement amount e and a, interframe movement amount a and e are rounded to the pixel matching amount that just obtains horizontal and vertical again divided by pixel level and vertical resolution respectively, are designated as: obtain registration amount and adjacent two frames are carried out to difference after registration later, remove fixed star.

In technique scheme, in static target pick-up unit, static target detects specific as follows:

Difference, cut apart, the position of target is T after cluster and two continuous frames mark _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, present frame sequence number is i, if meet:

abs(x _i-x _i-1)≤nTh and abs(y _i-y _i-1)≤nTh (7)

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be potential inspection target, by T _{i-1, i}(x _i, y _i) retain and give subsequent step to process, wherein nTh is predefined threshold value.

In technique scheme, in static target pick-up unit, in static target detection, the interval of nTh is [2,10], and the value maximum of nTh is no more than the radius of target to be detected.

In technique scheme, in double star removal device, the method for the removal of double star is as follows:

After two continuous frames mark, the position of target is T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, and present frame sequence number is i, and registration amount is if meet formula (4):

$\mathrm{abs}(x_i-x_{i-1}-\stackrel{\‾}{x})\≤n{\mathrm{Th}}_x,\mathrm{and},\mathrm{abs}(y_i-y_{i-1}-\stackrel{\‾}{y})\≤n{\mathrm{Th}}_y---\left(8\right)$

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be the residual edge of double star after difference, by T _{i-1, i}(x _i, y _i) from result, delete, do not give flight path and detect, wherein nTh _xand nTh _yinterval be [2,10].

Principle of the present invention: the Weak target that this method is applicable to intensive star background detects, in image, fixed star is intensive and target is very weak, therefore principle of the present invention is first to adopt the method for fixed star registration difference to remove a large amount of background fixed stars, then fails to remove clean fixed star edge while removing difference; Finally utilize the track initiation method based on arest neighbors principle, detect Weak target.The present invention compares tool with conventional target detection method and has the following advantages:

One, adopt the pre-service of interframe registration difference, obvious and computing velocity is fast to motion fixed star inhibition;

Two,, based on average+k times of variance dividing method, the desirable computing velocity of segmentation effect is fast;

Three, detect to adopt adjacent two frame difference results to carry out position relevant for static target, can effectively remove fixed star edge and method simple, counting yield is high;

Four, to adopt in step (3) remaining candidate target and the candidate target of former frame mark to carry out position relevant for double star elimination method, can reject the false-alarm that double star edge causes, and reduces follow-up track association number of targets, reduction false-alarm and operand;

Five, adopt the track detecting method of logic-based arest neighbors, the method is a kind of effective object detection method, it utilizes the velocity characteristic of target travel to be detected, when carrying out track initiation, limit the initial velocity range of target, reduce initial flight path number, accelerate arithmetic speed and target detection track initiation speed.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is that static target of the present invention detects schematic diagram

Fig. 3 is that double star of the present invention is rejected schematic diagram;

Fig. 4 is the track detecting method schematic diagram of logic-based arest neighbors of the present invention;

Fig. 5 is original image in embodiment;

Fig. 6 is for to carry out the filtered result of registration difference, binaryzation and two-value to Fig. 5;

Fig. 7 is by the result (result behind removal fixed star edge) of interframe position coherent detection static target to Fig. 6;

Fig. 8 is for removing the schematic diagram of double star to Fig. 7;

Fig. 9 final detection result design sketch.

Embodiment

The Weak target that the present invention is specially adapted to intensive star background detects.This method flow process as shown in Figure 1, comprising:

(1) interframe registration difference: remove background fixed star;

(2) image is cut apart: reject extended background, outstanding target;

(3) target cluster: remove spuious noise spot, filling cavity and cluster target;

(4) target label: obtain the information such as target location, size, gray scale and boundary rectangle;

(5) static target detects (fixed star edge is removed, and removes the fixed star edge staying after difference): reduce the number of targets that follow-up flight path is processed, avoid to greatest extent detecting false target (reduction false-alarm);

(6) flight path of logic-based arest neighbors detects, and by pretreated target being carried out to multi-frame processing (track initiation), confirms final target.

Concrete, the present invention includes following steps:

1. interframe registration difference: when telescope detects the speed motion of target by band (being generally theoretical guiding or fixed point to the position of synchronous target), adopting interframe registration difference to remove that fixed star disturbs is optimum solution.Interframe registration difference can be removed most of fixed star interference, and in outstanding image, interframe shows as static target.

2. Target Segmentation: Target Segmentation algorithm of the present invention adopts simple self-adaptation m+k σ segmentation threshold computing method, the background mean value of m presentation video, the variance that σ is image, wherein k is the empirical value obtaining by great many of experiments, is generally between 3-7.

3. target cluster: adopt two-value statistical filtering to carry out target cluster.To the image after step 2 binaryzation by pixels statistics N * N neighborhood internal object number of pixels, when it is greater than certain value C, what this pixel represented is target, is retained, otherwise weed out this pixel, two-value statistical filtering can be eliminated spuious noise spot, filling cavity and cluster target; Two-value statistical filtering is expressed as:

$B^{\′}(x,y)=\left\{\begin{array}{ccc}1& \mathrm{if}& \underset{(x,y)\∈(N\×N)}{\mathrm{\Σ}}B(x,y)>C\\ 0& \mathrm{else}& \end{array}\right.---\left(1\right)$

Wherein, B (x, y) represents the image after binaryzation, B'(x, y), be the image after cluster.N * N represents the neighborhood of current pixel, the present embodiment is chosen 3 * 3, C be value is 1 in neighborhood pixel with, the present embodiment value is 3.

4. target label: target label is the bianry image after step 3 target cluster to be adopted to line segment mark (labeling method is referring to multi-Target Image tracker > > mono-literary composition of < < real-time mark, Zhang Guilin, Cao Weiheng, Li Qiang etc., HUST's journal, 1994,22 (5): 36-41.), by mark, can calculate the features such as position, size, gray scale and boundary rectangle of target.

5. static target detects: the interframe static target relevant based on interframe position detects (removal of fixed star edge).Suppose difference, cut apart, the position of target is T after cluster and two continuous frames target label _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, present frame sequence number is i, if meet:

abs(x _i-x _i-1)≤nTh and abs(y _i-y _i-1)≤nTh

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be potential inspection target, T _{i-1, i}(x _i, y _i) retain and give subsequent step to process, wherein nTh is predefined threshold value;

6. the rejecting at double star edge: suppose that the position of target is T after two continuous frames mark _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, and present frame sequence number is i, and registration amount is if meet:

$\mathrm{abs}(x_i-x_{i-1}-\stackrel{\&OverBar;}{x})\&le;n{\mathrm{Th}}_x,\mathrm{and},\mathrm{abs}(y_i-y_{i-1}-\stackrel{\&OverBar;}{y})\&le;n{\mathrm{Th}}_y$

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be the residual edge of double star after difference, by T _{i-1, i}(x _i, y _i) from result, delete, do not give flight path and detect, wherein nTh _xand nTh _yinterval be [2,10].

The track detecting method of logic-based arest neighbors: the needs of considering real-time, the target detecting when need is the Weak target of non-dense set, non-maneuverability, and when background clutter is weak, (the method is referring to real-time detection method > > mono-literary composition of the large visual field of < < deep space Weak target for the arest neighbors correlating method of employing logic-based, week enters, Wu Qinzhang. state's optical technology, 2006,32 (1): 134-137).The arest neighbors correlating method of logic-based as shown in Figure 5, is in given data window, according to the data acquisition obtaining, by least square method, data point is carried out to matching, then the speed of target is estimated.If the speed of estimating is (maximum according to middle high rail target, minimum movement speed and set) in the span of appointment, generate a temporary transient flight path, then the position of the 3rd frame target is predicted, and centered by predicted position, determine an associated region, any some mark that drops on associated region is by a temporary transient flight path of expansion, continue estimating speed value estimated acceleration value, and then according to the estimated value of speed and acceleration, the position of next frame is predicted and is set up corresponding associated region, any some mark that drops on associated region will generate a new flight path.Then, the flight path of all generations is carried out to matching with straight line or quafric curve, the point on flight path and the error of matched curve within the specific limits, are confirmed this flight path; If do not satisfied condition, delete this flight path.When there is the Targets Dots of not being correlated with, need this mark to carry out flight path search.Next frame point mark is relevant to already present flight path, and the rule of handle complex situations is as follows:

Article (1) one, track and several point are simultaneously relevant, the new some mark that to get the nearest point of point predicted with track be track.

(2), when several tracks are relevant with same some mark, this point belongs to trajectory predictions and puts nearest track.

(3) when several tracks and several somes marks are simultaneously relevant, if a track is only relevant to a unique point, this track is preferential, and it is relevant to remaining point that other tracks are only considered.

(4) if the Bo Mennei appearance point mark not of extrapolation, extrapolation is a bit as new some mark, and it is relevant at next frame, to proceed flight path.

Above-mentioned Data Association is only based target track, and obviously, when the eigentransformations such as the size of the target in the target trajectory detecting, gray scale, variance are larger, probably this is not the true flight path of an objective.So in carrying out the process of track association, the feature of combining target is carried out the associated probability that will greatly improve the efficiency of track association and reduce generation false track.

The flight path of combining target feature is relevant, and cardinal rule is exactly when carrying out track association, and the changing features such as size, gray scale and variance of only having target in allowed limits time, are just correlated with, and flight path is correlated with and is still adopted the arest neighbors correlation principle of logic-based.As for choosing of clarification of objective, the scope of changing features, or the variation evaluation criteria of target population feature, due to the diversity that target signature changes, the difference of kinetic characteristic need to be measured and adjust in actual engineering application.

Found through experiments, the feature of size, gray scale and the variance of general Weak target is generally more stable at short notice, and can be used as feature, to carry out flight path relevant.And the thresholding of changing features generally can be decided to be 1/3～1/5.

The present invention has chosen the image sequence of a medium star density, and it is carried out to target detection, comprises 1 of target in described image sequence.Fig. 5 is the frame original image in sequence image, and Fig. 6, for Fig. 5 is carried out to the filtered result of difference, binaryzation and two-value, wherein calculates the k=3.0 of binarization segmentation thresholding, and two value filterings are got 3 * 3 neighborhoods and calculated, two-value filter threshold c=3; Fig. 7 carries out the result that the relevant static target in interframe position detects (removed most difference after residual fixed star edge) to image, and the thresholding that wherein position is relevant is (3,3); Fig. 8 is that the image thresholding that wherein position is relevant that Fig. 7 is removed behind double star fixed star edge is again (3,3); Fig. 9 carries out the net result of multiframe flight path detection by arest neighbors method.

Claims (10)

1. be applicable to a detection method of small target for intensive star background, it is characterized in that step is as follows:

Step (1), difference: receiving sequence image, carries out interframe registration difference;

Step (2), image are cut apart: adopt m+k σ thresholding computing method to Image Segmentation Using, m is full figure average, and σ is the mean square deviation of full figure, and k is empirical value, and k span is 3.0-7.0;

Step (3), target cluster: the bianry image after step (2) image is cut apart is by pixels statistics N * N neighborhood internal object number of pixels, and when it is greater than C, what this pixel represented is target, is retained, otherwise weeds out this pixel; Be expressed as:

$B^{\&prime;}(x,y)=\left\{\begin{array}{ccc}1& \mathrm{if}& \underset{(x,y)\&Element;(N\&times;N)}{\mathrm{\&Sigma;}}B(x,y)>C\\ 0& \mathrm{else}& \end{array}\right.$

Wherein, B (x, y) represents the image after binaryzation, B'(x, y) be the bianry image after target cluster, N * N represents the neighborhood of current pixel, the pixel that in the neighborhood that C is current pixel, value is 1 with;

Step (4), target label: use the method for line segment mark, obtain the position coordinates of size, gray scale and each pixel of target of target;

Step (5), static target detect: static target is defined as and on image, shows as the target that interframe change in location is very little;

Step (6), double star are removed: remove the fixed star edge that is detected as static target in step (5);

Step (7), flight path detect: use the track detecting method of logic-based arest neighbors to detect target trajectory, confirm real target.

2. a kind of detection method of small target that is applicable to intensive star background according to claim 1, it is characterized in that: the registration amount of calculating interframe fixed star, fixed star interframe registration amount and the telescopical latitude of ground, point to orientation, just, pixel resolution is relevant with working frame frequency, the registration amount computing method of the middle registration difference of described step (1) are as follows:

A) calculate fixed star angular altitude speed

Wherein, E is photo-electric telescope angular altitude, and A is photo-electric telescope position angle, and φ is hour angle, is the latitude at photo-electric telescope place;

B) calculate fixed star Azimuth, Speed, Altitude

$\frac{\mathrm{dA}}{\mathrm{dt}}=\sin \mathrm{\&phi;}-\frac{\cos \mathrm{\&phi;}\&CenterDot;\cos A}{\tan z}---\left(2\right)$

Wherein, z is zenith angle, the same formula of other meaning of parameters (1);

C) by formula (1) formula (2) and the interframe time interval, calculate synchronous target at interframe movement amount e and a, interframe movement amount a and e are rounded to the pixel matching amount that just obtains horizontal and vertical again divided by pixel level and vertical resolution respectively, are designated as: obtain registration amount and adjacent two frames are carried out to difference after registration later, remove fixed star.

3. a kind of detection method of small target that is applicable to intensive star background according to claim 1, is characterized in that: in step (5), static target detects specific as follows:

Difference, cut apart, the position of target is T after cluster and two continuous frames target label _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, present frame sequence number is i, if meet:

abs(x _i-x _i-1)≤nTh and abs(y _i-y _i-1)≤nTh

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be potential inspection target, by T _{i-1, i}(x _i, y _i) retain and give subsequent step to process, wherein nTh is predefined threshold value.

4. a kind of detection method of small target that is applicable to intensive star background according to claim 3, it is characterized in that: in step (5), in static target detection, the interval of nTh is [2,10], the value maximum of nTh is no more than the radius of target to be detected.

5. a kind of detection method of small target that is applicable to intensive star background according to claim 1, is characterized in that: in step (6), the removal method of double star is as follows:

After two continuous frames target label, the position of target is T _{iz2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, and present frame sequence number is i, and registration amount is if meet formula:

$\mathrm{abs}(x_i-x_{i-1}-\stackrel{\&OverBar;}{x})\&le;n{\mathrm{Th}}_x,\mathrm{and},\mathrm{abs}(y_i-y_{i-1}-\stackrel{\&OverBar;}{y})\&le;n{\mathrm{Th}}_y$

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be the residual edge of double star after difference, by T _{i-1, i}(x _i, y _i) from result, delete, do not give flight path and detect, wherein nTh _xand nTh _yinterval be [2,10].

6. a Weak target pick-up unit that is applicable to intensive star background, is characterized in that comprising:

Differential attachment: receiving sequence image, carries out interframe registration difference;

Image segmenting device: adopt m+k σ thresholding computing method to Image Segmentation Using, m is full figure average, and σ is the mean square deviation of full figure, and k is empirical value, and span is 3.0-7.0;

Target clustering apparatus: by pixels statistics N * N neighborhood internal object number of pixels, when it is greater than C, what this pixel represented is target, is retained, otherwise weeds out this pixel to the image after step image segmenting device binaryzation; Be expressed as:

$B^{\&prime;}(x,y)=\left\{\begin{array}{ccc}1& \mathrm{if}& \underset{(x,y)\&Element;(N\&times;N)}{\mathrm{\&Sigma;}}B(x,y)>C\\ 0& \mathrm{else}& \end{array}\right.$

Wherein, B (x, y) represents the image after binaryzation, B'(x, y) be the bianry image after target cluster, N * N represents the neighborhood of current pixel, the pixel that in the neighborhood that C is current pixel, value is 1 with;

Target label device: use the method for line segment mark, obtain the position coordinates of size, gray scale and each pixel of target of target;

Static target pick-up unit: static target is defined as and shows as the target that interframe change in location is very little on image

Double star removal device: remove the fixed star edge that is detected as static target in step static target pick-up unit;

Flight path pick-up unit: use the track detecting method of logic-based arest neighbors to detect target trajectory, confirm real target.

7. a kind of Weak target pick-up unit that is applicable to intensive star background according to claim 6, it is characterized in that: the registration amount of calculating interframe fixed star, fixed star interframe registration amount and the telescopical latitude of ground, point to orientation, just, pixel resolution is relevant with working frame frequency, the registration amount computing method of the middle registration difference of described step (1) are as follows:

A) calculate fixed star angular altitude speed

$\frac{\mathrm{dE}}{\mathrm{dt}}=\cos \mathrm{\&phi;}\sin A---\left(3\right)$

Wherein, E is photo-electric telescope angular altitude, and A is photo-electric telescope position angle, and φ is hour angle, is the latitude at photo-electric telescope place;

B) calculate fixed star Azimuth, Speed, Altitude

$\frac{\mathrm{dA}}{\mathrm{dt}}=\sin \mathrm{\&phi;}-\frac{\cos \mathrm{\&phi;}\&CenterDot;\cos A}{\tan z}---\left(4\right)$

Wherein, z is zenith angle, the same formula of other meaning of parameters (1);

C) by formula (1) formula (2) and the interframe time interval, calculate synchronous target at interframe movement amount e and a, interframe movement amount a and e are rounded to the pixel matching amount that just obtains horizontal and vertical again divided by pixel level and vertical resolution respectively, are designated as: obtain registration amount and adjacent two frames are carried out to difference after registration later, remove fixed star.

8. a kind of Weak target pick-up unit that is applicable to intensive star background according to claim 6, is characterized in that: in static target pick-up unit, static target detects specific as follows:

Difference, cut apart, the position of target is T after cluster and two continuous frames mark _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, present frame sequence number is i, if meet:

abs(x _i-x _i-1)≤nTh and abs(y _i-y _i-1)≤nTh

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be potential inspection target, by T _{i-1, i}(x _i, y _i) retain and give subsequent step to process, wherein nTh is predefined threshold value.

9. a kind of Weak target pick-up unit that is applicable to intensive star background according to claim 8, it is characterized in that: in static target pick-up unit, in static target detection, the interval of nTh is [2,10], the value maximum of nTh is no more than the radius of target to be detected.

10. a kind of Weak target pick-up unit that is applicable to intensive star background according to claim 6, is characterized in that: in double star removal device, the method for the removal of double star is as follows:

After two continuous frames mark, the position of target is T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i), i-2 wherein, i-1, i represents number of frames, and present frame sequence number is i, and registration amount is if meet formula:

$\mathrm{abs}(x_i-x_{i-1}-\stackrel{\&OverBar;}{x})\&le;n{\mathrm{Th}}_x,\mathrm{and},\mathrm{abs}(y_i-y_{i-1}-\stackrel{\&OverBar;}{y})\&le;n{\mathrm{Th}}_y$

T _{i-2, i-1}(x _i-1, y _i-1) and T _{i-1, i}(x _i, y _i) be the residual edge of double star after difference, by T _{i-1, i}(x _i, y _i) from result, delete, do not give flight path and detect, wherein nTh _xand nTh _yinterval be [2,10].