CN103777187A - Weak target track-before-detect method based on traversal random Hough conversion - Google Patents

Abstract

The invention discloses a weak target track-before-detect method based on traversal random Hough conversion. In the method, clutter suppression first threshold processing is firstly performed on all frames of radar echo data and different frames of data exceeding a first threshold are stored in a separate manner; then all the radar echo data exceeding the first threshold are superimposed in a data space; during Hough conversion on all the points in the data space, the invention proposes a traversal pairing idea, which is a method for performing one-to-one pairing on all the different points in the data space so as to obtain linear parameters; then traversal Hough conversion is performed on all the points in the data space so as to obtain the linear parameters; then voting is performed on the to-be-solved linear parameters; and at last a linear parameter with the maximum vote value is extracted so that track backtracking is realized. The weak target track-before-detect method based on the traversal random Hough conversion is capable of realizing weak-target detection and tracking under a complex background environment and greater in detection performance and tracking precision at the same time.

Description

Tracking before weak target detection based on traversal random Hough transformation

Technical field

The invention belongs to radar data process field, relate to a kind of based on traversal random Hough transformation weak target detection before tracking.

Background technology

The timely detection of low signal-to-noise ratio target and accurately Continuous Tracking problem are one of radar-probing system gordian techniquies that need to solve.After traditional detection, follow the tracks of, judge by thresholding is set whether single frames target exists, follow the tracks of again detecting after target.This method has good detection tracking performance in the time that target signal to noise ratio is higher, but in the time that target signal to noise ratio is lower, because target is submerged in noise signal, utilize this kind of method to detect and follow the tracks of the loss that will cause echo signal target, be unfavorable for that the target detection under low signal-to-noise ratio environment is followed the tracks of.Before detecting, tracking TBD is that radar weak target detects a kind of effective ways (the Track Before Detect following the tracks of, TBD), it does not do threshold processing or adopts less thresholding original metric data, thereby improve target signal to noise ratio and realize the raising of the detectability of surveillance radar to target by the mode of multiframe energy accumulation, in the time obtaining testing result, announce tracking results simultaneously.Its essence is and exchange energy for the time, improve the signal to noise ratio (S/N ratio) of target.TBD method based on Hough conversion does not need target prior imformation, the backward energy that comes from same target can be carried out to non-coherent accumulation, strengthens target signal to noise ratio, and then detects weak target signal.Due to standard Hough convert required computing time and memory space larger, thereby researcher is below this technical improvement, has proposed random Hough transformation, revise the technology such as Hough conversion and Generalized Hough Transform.Tradition random Hough transformation detects targetpath from the cumulative echo data midplane of multiframe stack, target data is not chosen to the method for any constraint in addition, can cause detection probability lower, the situation that false track is on the high side, straight line parameter in the table of dynamic link simultaneously just can not change after selecting, the point falling in its certain error being voted, find optimum solution for traversal dynamic link table, will there is larger deviation with true flight path in the flight path recovering like this.For these problems, based on traversal random Hough transformation to all cross threshold data travel through pairing ask for straight line parameter, then from every frame radar return data, choose to straight line parameter have contribution method realize recalling of flight path.This method has higher detection probability, and can also avoid occurring the situation of false dismissal.

Summary of the invention

The present invention is directed to the deficiencies in the prior art, provide a kind of based on traversal random Hough transformation weak target detection before tracking.It is not high that the method has solved the detection probability existing in tracking before the existing target detection based on random Hough transformation, easily occurs the problem of false-alarm.Can further improve tracking accuracy simultaneously.

The present invention includes following steps:

(1) Hough parameter space is carried out to quantification treatment.

(2) the radar return data that exceed the first thresholding different k frame in each emulation are stored in respectively in the matrix that K is different, the storage matrix that wherein k frame exceedes the first threshold data is expressed as z _k.

(3) stacked data of all frames being crossed to the first thresholding is added in datum plane W, and all different pieces of information points are traveled through and ask for straight line parameter (ρ, θ), and puts into dynamic link Table A.

(4) the straight line parameter in dynamic link table is contrasted to ballot one by one, if both meet certain error scope, the ballot value of selected parameter value in dynamic link Table A is added to 1.

(5) ask in A ballot value maximum point and extract last straight line parameter value (ρ, θ).

(6), according to straight line parameter (ρ, θ), cross in the first thresholding echo data and extract required parameter is had to contribution point at each single frames.

(7) realize the estimation to dbjective state according to required point.

(8) flight path is recalled, and shows tracking results, prepares to accept next group multiframe data.

Gordian technique of the present invention is to travel through the different pieces of information in paired data space, asks for straight line parameter.Dynamic link table is relatively ballot mutually, chooses optimal straight line parameter; Single frames extracts data and recalls targetpath.The present invention has two advantages compared with traditional random Hough transformation: (1), detection probability are higher, because its traversal has been matched all data in datum plane, thereby having utilized is likely the information of target position data, having avoided in traditional random Hough transformation is not the problem that parameter influence detection probability is asked in 2 pairings of target trajectory; (2), single frames is recalled flight path.In tradition random Hough transformation, during to the recalling of targetpath, extract the method that targetpath is had to contribution point in datum plane, when thresholding being set when too high, target actual position information is not stored, thereby causes the situation of false dismissal.Select single frames radar return data to carry out target while recalling, can store Multi simulation running metric data, and then the phenomenon of avoiding target location to be dropped.

Accompanying drawing explanation

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

Embodiment

Below in conjunction with Fig. 1, step of the present invention is made the following instructions:

Step 1. parameter space quantizes.

Hough conversion is that θ asks for to straight line parameter ρ, first will carry out quantification treatment to Hough parameter space so utilize the TBD method converting based on Hough to carry out realize target detection tracking.

θ _n＝(n-1/2)Δθn＝1,2…,N _θ （1）

ρ _n＝(n-1/2)Δρn＝1,2…,Nρ

Wherein Δ θ=π/N _θ, N _θfor the hop count of cutting apart of parameter θ; Δ ρ=L/N _ρ, L=sqrt (n _x ²+ n _y ²), n _xand n _yfor the corresponding dimension in radar data space, N ρ is that parameter ρ is cut apart hop count.

Step 2. echo data does threshold processing and the radar return data that single frames exceedes the first thresholding is stored in to different matrixes.

Radar return data are carried out to Hough conversion, obtain target component.In the time that realistic objective is processed, in order to reduce computing time and to reduce false alarm rate, so introduce the first thresholding η.

The radar return data that each frame exceeded to the first thresholding are stored in respectively in different z matrixes.

z _k＝(x _i,k,y _i,k)1≤k≤K1≤i≤l _k （2）

Z _krepresent all the first threshold data positional information set, (x of exceeding of k frame _i,k, y _i,k) be the positional information of i data of k frame data, l _kbe that k frame data exceed the first threshold data number.

Step 3. is all the first threshold data data spaces that are added to that exceed

All stacked datas are added to datum plane W.If different frame data all exist target in same position in additive process, choose in different frame target amplitude maximal value as the range value of this position, and its correspondence position information is stored in matrix U.

U _i＝(x _i,y _i)1≤i≤L （3）

In formula, L is the number of data space mid point.

Step 4. traversal is asked for straight line parameter, and putting into one by one dynamic link table

Then according to formula (4) and (5), the Hough that travels through a little in U is converted, store successively the straight line parameter of at every turn asking for into dynamic link Table A.

$\left\{\begin{array}{cc}\mathrm{\ρ}=x_i\cos \left(\mathrm{\θ}\right)+y_i\sin \left(\mathrm{\θ}\right)& 1\≤i\≤L-1\\ \mathrm{\ρ}=x_j\cos \left(\mathrm{\θ}\right)+y_j\sin \left(\mathrm{\θ}\right)& i\≤j\≤L\end{array}\right.---\left(4\right)$

$\left\{\begin{array}{c}\mathrm{\θ}={\mathrm{tg}}^{-1}\left(\frac{x_j-x_i}{y_j-y_i}\right)\\ \mathrm{\ρ}=\left(\begin{array}{c}{x}_i\cos \left(\mathrm{\θ}\right)+y_i\sin \left(\mathrm{\θ}\right)+\\ x_j\cos \left(\mathrm{\θ}\right)+y_i\sin \left(\mathrm{\θ}\right)\end{array}\right)/2\end{array}\right.---\left(5\right)$

Step 5. is voted to the straight line parameter in dynamic link table, extracts ballot value maximum point.

Traversal pairing finishes rear data in Table A to be voted one by one according to formula (6).If meet certain error α and β between two straight line parameters in Table A, its corresponding ballot value is added to 1.

|ρ _i-ρ _j|≤α,|θ _i-θ _j|≤β1≤i＜j≤N （6）

A(ρ _i,θ _i)＝A(ρ _i,θ _i)+1

In formula, i and j are respectively i and j parameter value in table, and N is number of parameters in Table A.

Extract maximum ballot value parameter (ρ, θ) in Table A, if there be M the parameter that ballot is maximum in Table A, according to following formula, it be weighted and on average try to achieve optimum (ρ, θ).

$(\mathrm{\ρ},\mathrm{\θ})=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}({\mathrm{\ρ}}_i,{\mathrm{\θ}}_i)/M---\left(7\right)$

Step 6. is extracted has contribution data to straight line parameter, estimating target state.

Utilize required straight line parameter extract in each frame the contributive point of straight line parameter and be stored in matrix S by formula (8):

|ρ-x _i,kcos(θ)-y _i,ksin(θ)|≤mΔρ1≤i≤l _k （8）

S _k＝(x _i,k,y _j,k)1≤i≤l _k,1≤k≤K

In formula, m is positive integer.According to the data in S, dbjective state is estimated, and estimated value is stored in B.

Step 7., according to dbjective state matrix B export target flight path, prepares to accept next group data.

Claims (1)

1. tracking before the weak target detection based on traversal random Hough transformation, is characterized in that the method comprises the following steps:

Step 1, Hough parameter space is carried out to quantification treatment;

Step 2, the radar return data that exceed the first thresholding different k frame in each emulation are stored in respectively in the matrix that K is different, the storage matrix that wherein k frame exceedes the first threshold data is expressed as zk;

Step 3, the stacked data that all frames are crossed to the first thresholding are added in datum plane W, and all different pieces of information points are traveled through and ask for straight line parameter (ρ, θ), and put into dynamic link Table A;

Step 4, the straight line parameter in dynamic link table is contrasted to ballot one by one, if both meet certain error scope, the ballot value of selected parameter value in dynamic link Table A is added to 1;

Step 5, ask in dynamic link Table A ballot value maximum point and extract last straight line parameter value (ρ, θ);

Step 6, according to straight line parameter (ρ, θ), crossing in the first thresholding echo data to extract at each single frames has contribution point to required parameter;

Step 7, realize the estimation to dbjective state according to required point;

Step 8, flight path are recalled, and show tracking results, prepare to accept next group multiframe data.