CN102833882B - Multi-target data fusion method and system based on hydroacoustic sensor network - Google Patents

Abstract

The invention provides a multi-target data fusion method and system based on a hydroacoustic sensor network. The multi-target data fusion method comprises the following steps: (101) arranging a plurality of collecting nodes of the hydroacoustic sensor network in an observation area, arranging N collecting nodes in a water area, obtaining the azimuth information of a plurality of target elements, and transmitting the obtained azimuth information of the target elements to a main node, the N collecting nodes adopts line array hydrophones or vector hydrophones as detection parts, and the detection parts can process signals in real time, and the azimuth information of the target elements is observed by each collecting node in the observation area; and (102) fusing the data of the azimuth information of the target elements by the main node: associating data by the main node according to the azimuth information of the target elements and the information data of the targets stored in the main node by adopting the data associating strategy, and tracking and positioning the multiple targets to finish the track initiating, maintaining and deleting of the multiple targets.

Description

A kind of Multi-target Data fusion method based on water sound sensor network and system

Technical field

Native system relates to signal processing field.Be related specifically to a kind of Multi-target Data fusion method based on water sound sensor network and the system that utilize water sound sensor network to carry out Multi-target Data fusion.

Background technology

Water sound sensor network is exactly lay several underwater sound sensor nodes in certain region under water, underwater sensor node both can obtain underwater information, also the function of subsurface communication each other and networking can be completed, also there is certain signal processing function, the underwater information of acquisition is transferred to specific node (host node) by water sound sensor network by each sensor node, then host node includes the information of all the sensors node on the bank general networks by wireless or wired mode, and sends to the network under water of observer.

Research about water sound sensor network is just at the early-stage, although be faced with many challenges, but still has unrivaled advantage.Water sound sensor network can be widely used in military and civilian field, can be used to environmental monitoring, resource exploration, seabed exploration, disaster prevention, the navigation of supplementary navigation device, anti-to dive, anti-terrorism early warning under water etc.For these application, target localization and tracking technique are the bases of above-mentioned application.For target navigation, anti-location of diving, early warning under water, seabottom geology Reconnaissance Survey etc. all needs to carry out accurate locating and tracking process to target, so be the of paramount importance technology of water sound sensor network based on the target locating of water sound sensor network, for some application scenario, the location of submarine target and tracking are the application purposes of underwater sensor network.

Underwater sensor network is different from the wireless sensor network of land, because the propagation velocity of acoustical signal in water is very slow, therefore its message transmission rate is very little, it is impossible that the raw information of all nodes is transferred to host node by sensor network, therefore each node is needed self to have certain signal processing function, the information that can obtain its transducer carries out certain preliminary treatment, then the pre-processed results of node is transferred to host node by water sound sensor network, host node carries out data fusion by the result of being come by each node-node transmission to be positioned target and follows the tracks of.

When adopting active mode to carry out target localization, be easy to expose self while reaching target localization, disguised poor, so employing passive mode location is the focus of this research field.But passive detection node is generally difficult to the range information detecting target under water, this adds increased the difficulty of target localization and tracking.For Passive Positioning field under water, the azimuth information that single-sensor detects is much more accurate than range information, therefore azimuth information transducer is utilized to be a feasible suggestion as the probe portion of water sound sensor network interior joint, for underwater detectoscope, the transducer of energy directional bearing information has various hydrophone array sensor, also can be vector hydrophone, the azimuthal measurement of underwater sound source can be realized by various Array Signal Processing algorithm.

Because single orientation detection node cannot carry out target localization and tracking, therefore need multiple node to carry out target cooperative location and tracking simultaneously, and the multinode characteristic of water sound sensor network is for which providing application platform, therefore utilize water sound sensor network to carry out target cooperative location and tracking has certain feasibility, but prior art is about being applied in by wireless sensor network in field that underwater multi-target follows the tracks of.The present invention has filled up forward-lookingly and has utilized water sound sensor network to carry out underwater multi-target location and follow the tracks of the blank of this technical field, and the data fusion for water sound sensor network provides a kind of new thinking.

When utilizing Pure orientation information to carry out multi-sensor multi-target tracking, the very important point solves data correlation problem, namely judge each azimuthal measurement value is from which target, when target sets up flight path, need to carry out measured value-track data association, association algorithm has nearest neighbor algorithm, joint probability density association algorithm etc.; If targetpath is not set up, then need to carry out measured value-measured value data correlation, association algorithm has minimum distance method, maximum likelihood method etc.; The present invention considers from system perspective, first find target by measured value-measured value data correlation, set up initial flight path, then maintain flight path by measured value-measured value data correlation, whole data fusion process is carried out automatically, does not need artificial participation.

Summary of the invention

The object of the invention is, provides a kind of Multi-target Data fusion method based on water sound sensor network and system for overcoming prior art in the blank realizing underwater multi-target real-time tracking process field.

The present invention aims to provide a kind of Multi-target Data fusion method that may be used for Real-Time Monitoring, and the method can carry out real time data processing, automatically can set up flight path, maintains flight path, delete flight path, for water sound sensor network data fusion provides a kind of new approaches.

The invention provides a kind of multiple target real time data fusion method based on water sound sensor network, this data fusion method is used for carrying out multiobject track initiation in real time, flight path maintains and flight path is deleted, and described method comprises following steps:

Step 101, the acquisition node step of some wireless sensor networks is laid in observation area, N (N >=3) individual acquisition node is laid in waters, described N number of acquisition node adopts linear array hydrophone or vector hydrophone as probe portion, this probe portion also carries out real time signal processing, obtain the described target bearing information of each acquisition node in described observation area, and the azimuth information obtained is transferred to host node by net under water;

Step 102, described host node carries out the step of data fusion to described azimuth information, and this host node adopts data correlation strategy to carry out track and localization to multiple target, completes multiobject track initiation, flight path maintains and flight path is deleted;

Wherein, described multiple target comprises: certainty goal set and tentative goal set; Wherein, described certainty target is the target of continued presence in described observation area in the section sometime that detects of described acquisition node; Described tentative target is that described acquisition node detects but the uncertain target that whether can occur continuously within a certain continuous print time period, if the number of times that can associate described azimuth information continuously when certain target in described tentative goal set reaches a certain setting threshold C ₁time secondary, this tentative object element is just converted to a target in certainty goal set.

In technique scheme, described data correlation strategy is: the azimuth information data of the up-to-date transmission of described acquisition node and the target left on described host node in qualitative objective set really, target in tentative goal set are carried out to measured value-track data and associated, show the azimuth information data of described acquisition node come from which object element in described certainty goal set or described tentative goal set; And between the not associated orientation values to any target, carry out measured value-measured value association, whether have new tentative target occur, if having, then utilize the method for target localization to carry out target states initialization if detecting.

Describe based on above, preferred described employing data correlation strategy carries out track and localization to multiple target and comprises following steps further:

Step 201, described azimuth information is associated with the target left on described host node in qualitative objective set really, then the remaining azimuth information data be not associated with associated with the target in the tentative goal set of leaving on described host node, each orientation values for determining in described azimuth information comes from which target in described certainty goal set and described tentative goal set;

Step 202, flight path upgrades and the step of prediction, if a target in described certainty goal set or described tentative goal set is by plural azimuth information data correlation, then upgrades the flight path information of this target; If the data amount check of the azimuth information that object element can be associated with is less than 2, then using the updated value of the status predication value of this object element as current time;

Step 203, the step that certainty targetpath initial sum flight path is deleted, the state of each object element is checked according to the threshold value of setting, carry out the diversification in role of target, even have certain target in described certainty goal set not exceeded a certain setting threshold C by the number of times that described azimuth information associates continuously ₂, then this certainty target is deleted; If there is certain target in described tentative goal set to be exceeded another setting threshold C by the number of times that node azimuth information associates continuously ₁, then by target that this tentative target transition is in certainty goal set;

Step 204, in described orientation values, carries out measured value-measured value data correlation between the not associated orientation values to any target, checks whether that new tentative target occurs.If find that there is new tentative target to occur, then carry out initialization to this tentative target, positional information initialization adopts least square method to position, and speed is initialized as 0;

Wherein, the data association algorithm in technique scheme described in step 201 adopts classical nearest neighbor algorithm; State updating algorithm described in step 202 adopts EKF; Data association algorithm described in step 204 adopts minimum distance method, and target location algorithm described in step 204 adopts least square method.

The present invention also provides a kind of multiple target real time data emerging system based on water sound sensor network, and this system is based on water sound sensor network, and for carrying out multiobject track initiation in real time, flight path maintains and flight path is deleted; Described system comprises:

The information collection node of water sound sensor network: for gathering the data in a certain region, process, obtains several azimuth informations;

Host node, for the azimuth information that acquisition node of receiving information sends, and adopts data correlation to carry out data fusion, obtains multiobject track initiation in real time, flight path maintains and flight path is deleted;

Wherein, described multiple target comprises: the target that certainty goal set comprises and the target that tentative goal set comprises; Described certainty target is the target of continued presence in described observation area in the section sometime that detects of described acquisition node, described tentative target is that described acquisition node detects but the uncertain target that whether can occur continuously within a certain continuous print time period, if the number of times that can associate described azimuth information continuously when certain object element in described tentative goal set reaches a certain setting threshold C ₁time secondary, this tentative target is just converted to a target in certainty goal set.

In technique scheme, described acquisition node comprises further:

Adopt the probe unit of linear array hydrophone or vector hydrophone, for the primary data information (pdi) of acquisition node overlay area; Signal processing unit, processing for the real-time primary signal to gathering, obtaining the azimuth information that raw information comprises; Transmitting element, crosses net be under water transferred to host node for obtaining azimuth information.

Described host node comprises further:

Communication unit, for accepting the azimuth information that all acquisition nodes send;

Memory cell, the azimuth information that the data message detecting all targets in the data message of all targets in qualitative objective set really, tentative goal set for depositing all acquisition nodes and all node-node transmission received are come;

Certainty object element associative cell, deposits qualitative objective really for the nearest azimuth information that received by communication unit and described memory cell and carries out data correlation;

Tentative object element associative cell, carries out data correlation for the tentative target nearest azimuth information do not received by the described communication unit of certainty target association and described memory cell deposited;

Data associating unit between orientation values, carries out data correlation between the nearest azimuth information do not received by the described receiving element in certainty target and tentative target association;

Target tracking unit, when described certainty object set and tentative target tightening have target association to measured value, utilizes the measured value be associated with to carry out Trajectory Prediction and renewal to this target, thus completes the maintenance of flight path.

Target localization unit, if carrying out in the measured value between azimuth information-measured value association process, has been found to new tentative target, is then needing to utilize the method for target localization to carry out initialization to the state of tentative target.

Threshold value is arranged and state conversion unit, for arranging the threshold value C deleting this certainty object element ₂with another the threshold value C tentative target being converted to certainty target ₁; If meet threshold value C ₁, then this tentative target is converted to certainty target.

Wherein, the data association algorithm of described certainty object element associative cell adopts classical nearest neighbor algorithm, described in the unit of the described azimuth information data correlation with depositing, data association algorithm adopts minimum distance method, described method for tracking target adopts EKF method, and described target localization adopts least square method to carry out.

Important technical advantage of the present invention has:

(1) location and track algorithm amount of calculation simply, are suitable for real-time process completely.

(2) two kinds of association algorithms are all classical algorithms most in use, are convenient to understand and Be very effective.

(3) track initiation, flight path maintenance and flight path are deleted and can automatically be carried out, and do not need artificial participation.

(4) for the data fusion of water sound sensor network provides a kind of method being suitable for process in real time newly.

Data fusion method of the present invention carries out data fusion at characteristic layer, and automatically can carry out multiobject track initiation, flight path maintains and flight path is deleted, can overcome false-alarm to a certain degree and false dismissal, amount of calculation is simple, and Be very effective, is suitable for real-time process.

Accompanying drawing explanation

Fig. 1 is the structural representation that the present invention adopts the data fusion model of prior art;

Fig. 2 is the data fusion flow chart that the present invention adopts;

Fig. 3 is measured value-track data association flow chart that the present invention adopts;

Fig. 4 is measured value-measured value data correlation flow chart that the present invention adopts;

Fig. 5 is the water sound sensor network illustraton of model that the present invention adopts;

Fig. 6 is the acquisition node schematic diagram that the present invention adopts;

Fig. 7 is the host node schematic diagram that the present invention adopts.

Embodiment

Below in conjunction with block diagram, to the present invention,---the multiple target real time data fusion method based on water sound sensor network---is described in detail.

Data correlation is being divided into two parts by the present invention, that is: the algorithm of the association of measured value-track data, measured value-measured value data correlation wherein measured value-track association has nearest neighbor algorithm, JPDA method etc., and nearest neighbor algorithm amount of calculation is simple, is easy to implement; JPDA algorithm is considered from the angle of probability, calculate various may the joint probability density of associative combination, the then state of weighted calculation target, along with the increase of target number and azimuthal measurement value number, the amount of calculation of the method be the growth of multiple shot array formula.In the association process of measured value-measured value, classic algorithm has minimum distance method, maximum likelihood method etc., maximum likelihood method needs exhaustive various associative combination, calculate the probability of often kind of combination, therefore amount of calculation is large, and although minimum distance method interrelating effect is slightly poor, its amount of calculation is simple, is applicable to process in real time.

The application model of water sound sensor network of the present invention can represent with the block diagram shown in Fig. 1, the initial data collected is carried out the azimuth information that preliminary treatment obtains some targets by each sensor node, azimuth information is transferred to host node and carries out data fusion, data fusion comprises the initial of flight path, the maintenance of flight path, the deletion of flight path, then host node the result after data fusion is sent to observer or on the bank base station carry out decision-making or target in real time aobvious not.

The present invention is by the following technical solutions: in waters, lay N (N >=3) individual node, node adopts linear array hydrophone or vector hydrophone as probe portion, probe portion can carry out real time signal processing, obtain the azimuth information of target, the azimuth information obtained is transferred to host node by net under water by each node, these azimuth informations are carried out data fusion by host node, draw the target information in observation area.The azimuth number that each node dealt in each moment is unknown, occurs that the probability of false-alarm is P in azimuth information _f>=0, the detection probability P of target _d≤ 1, if the node orientation processing time be spaced apart T _s.In data fusion, target is divided into two kinds: certainty target and tentative target, certainty target is the target of continued presence in this waters that sensor network detects, tentative target is that sensor network detects but the uncertain target that whether can occur continuously, when tentative target the number of times of correlation measurement value continuously can reach C ₁time secondary, tentative target just can be converted to certainty target.

The key step of the technical program is as follows, and the described measured value wherein in following description is the azimuth information data that node measurement obtains:

(1) measured value-track data association; Carrying out data correlation by measured value with observing the target obtained, first being associated with certainty target by measured value, then being associated with tentative target by the remaining measured value be not associated with, data association algorithm adopts classical nearest neighbor algorithm.

(2) flight path upgrades and prediction (flight path maintenance); If a target can be associated by plural measured value, then flight path renewal is carried out to target, adopt the state of EKF to target to upgrade;

If the measured value that target can be associated with is less than 2, then using the updated value of the status predication value of target as target current time.

(3) dbjective state (deletion of certainty targetpath initial sum flight path) is checked; According to the threshold value set, (one is tentative target transition is the continuous degree of incidence C of target that certainty target needs ₁, one is the continuous not associated number of times C of target of the deleted needs of certainty target ₂) check the state of each target, carrying out the diversification in role of target, if there is certainty target number of times not associated continuously to exceed threshold value, then delete target, if there is the continuous degree of incidence of tentative target to exceed threshold value, is then certainty target by tentative target transition.

(4) measured value-measured value association; Do not carried out measured value-measured value by the measured value of target association associate remaining, data association algorithm adopts minimum distance method.

(5) if the target that had measured value to be associated with in step (4), then produce a new tentative target, the position least square positioning mode of target calculates, and speed is initialized as 0.

Fig. 2 shows the data fusion process of whole water sound sensor network, and as can be seen from the figure, the step that this method comprises has measured value-track data association, monotrack, target following state-detection, measured value-measured value prediction, single goal location and track initiation.

Step 1): when there being measured value to arrive, first needing to carry out measured value-track data association, measured value is associated with the target observed.If current time has M (M>=2) individual node to have data to arrive, the population of measured values of node j is L _j(k), measured value is expressed as { θ _jl(k), l=1 ..., L _j(k) }, then the data of carrying out each node during data correlation need to carry out separately, and the present invention adopts classical nearest neighbor algorithm to associate.

When carrying out target following, need to predict (introduce in detail and see step 2), if the status predication value of i-th target current time is X the state value of target _i(k) '=[x _i(k) ', y _i(k) ', vx _i(k) ', vy _i(k) '] ^t, i=1 ..., N.Can calculate target i thus relative to the bearing prediction value of node j is:

${\mathrm{\θ}}_{\mathrm{ij}}{\left(k\right)}^{\′}={\tan }^{-1}\left(\frac{y_i{\left(k\right)}^{\′}-y_{\mathrm{sj}}\left(k\right)}{x_i{\left(k\right)}^{\′}-x_{\mathrm{sj}}\left(k\right)}\right)$

Wherein (x _sj(k), y _sj(k)) represent that node j is in the position of moment k, be a known quantity.According to the principle of nearest neighbor algorithm, each target should be selected in its tracking gate and the measured value minimum with its predicted value error associates.

Fig. 3 provides the basic flow sheet of arest neighbors association algorithm.If have multiple target association to a measured value, just create so-called conflict, then can solve this problem by competition mechanism.Competition mechanism is exactly the angle from measured value, selects the target minimum with its error to associate, does not have selected target again to associate other measured values again.

It should be noted that in this step, the mode that we adopt certainty target priority to select, namely first measured value and certainty target are carried out data correlation, then not associated measured value is associated with tentative target.Method about measured value-track association also has a lot, such as JPDA etc., but its amount of calculation is explosive growth along with the number of angle value and the number of target, is unfavorable for real-time operation, and the present invention introduces JPDA method no longer in detail.

Step 2): at the end of the data correlation of measured value and target, just can carry out state updating to target (comprising certainty target and tentative target).For certainty target, it should be noted that if this target association is to plural measured value (do not need all nodes to be associated with target, can overcome part false dismissal like this), then utilize the measured value be associated with to carry out dbjective state renewal; If this target association to measured value be less than two, then the measured value be associated with can not be utilized to carry out state updating, and for this situation, the present invention utilizes the predicted value of current time as the measured value of current time to carry out dbjective state renewal.In state updating process, the present invention adopts expanded Kalman filtration algorithm, and this algorithm steps is as follows.

In 2 dimensional region, kth moment target location is [x (k), y (k)], target velocity is [vx (k), vy (k)], the state vector X (k) of target=[x (k), y (k), vx (k), vy (k)] ^t, target movement model adopts the simplest uniform rectilinear motion model (CV model), if its acceleration obeys zero-mean gaussian distribution, then the equation of motion of target is:

X(k+1)＝AX(k)+Γw (1)

$A=\left[\begin{array}{cccc}1& 0& \mathrm{Ts}& 0\\ 0& 1& 0& \mathrm{Ts}\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right],$ $\mathrm{\Γ}=\left[\begin{array}{cc}{\mathrm{Ts}}^2/2& 0\\ 0& {\mathrm{Ts}}^2/2\\ \mathrm{Ts}& 0\\ 0& \mathrm{Ts}\end{array}\right]$

Wherein Ts is the sampling interval, w=[w ₁, w ₂] Gaussian distributed, if its covariance matrix is Q.

Measured value be target association to node azimuth that it is estimated, i.e. Z (k)=[φ ₁(k) ..., φ _n(k)], N is the angle number that target association arrives.Then measure equation to be written as:

$\left\{\begin{array}{c}{\mathrm{\φ}}_1\left(k\right)=\arctan \left(\frac{y\left(k\right)-y_1\left(k\right)}{x\left(k\right)-x_1\left(k\right)}\right)+v_1\\ .\\ .\\ .\\ {\mathrm{\φ}}_N\left(k\right)=\arctan \left(\frac{y\left(k\right)-y_N\left(k\right)}{x\left(k\right)-x_N\left(k\right)}\right)+v_N\end{array}\right.---\left(2\right)$

(2) formula is abbreviated as φ _i(k)=h _i(X (k))+v _i(i=1 ..., N), observation noise v=[v ₁..., v _n] Gaussian distributed, its covariance matrix is R, and (1) and (2) two formulas constitute state equation and the observational equation of target respectively.

Follow the tracks of in theory, when linear Gaussian noise, Kalman filtering is optimal algorithm, but in above-mentioned trace model, observational equation is nonlinear, therefore can not directly utilize Kalman filtering to carry out target following.For nonlinear situation, being normally linear problem by nonlinear problem approximate transform, then utilize Kalman filtering to follow the tracks of, the general principle of EKF that Here it is, and Taylor expansion being for which providing good instrument.Obtain after (2) formula is carried out Taylor expansion:

Z(k)≈HX(k)+v (3)

$H=\left[\begin{array}{cccc}\frac{{\∂h}_1}{\∂x\left(k\right)}& \frac{{\∂h}_1}{\∂y\left(k\right)}& \frac{{\∂h}_1}{\∂\mathrm{vx}\left(k\right)}& \frac{{\∂h}_1}{\∂\mathrm{vy}\left(k\right)}\\ ...& ...& ...& ...\\ \frac{{\∂h}_N}{\∂x\left(k\right)}& \frac{{\∂h}_N}{\∂y\left(k\right)}& \frac{{\∂h}_N}{\∂\mathrm{vx}\left(k\right)}& \frac{{\∂h}_N}{\∂\mathrm{vy}\left(k\right)}\end{array}\right]$

Then classical Kalman filtering just can be utilized to carry out target following.EKF process is as follows:

(1) forecasting process

$X^{\′}\left(k\right)=A\hat{X}(k-1)$ (calculating state variable forward)

P ' (k)=AP (k-1) A ^t+ Γ Q Γ ^t(forward reckon error covariance)

(2) renewal process

K (k)=P ' (k) H ^t(HP ' (k) H ^t+ R) ^-1(calculating kalman gain)

$\hat{X}\left(k\right)=X^{\′}\left(k\right)+K\left(k\right)(Z\left(k\right)-{\mathrm{HX}}^{\′}\left(k\right))$ (by observational variable more new estimation)

P (k)=(I-K (k) H) P ' (k) (renewal error covariance)

Step 3): dbjective state detects.For certainty target, if having certainty target continuously the not associated number of times to measured value exceeded threshold value C ₂secondary, then this target will be deleted; For tentative target, if the associated continuously number of times of tentative target has exceeded C ₁secondary, then include this tentative target in certainty target ranks, if the continuous associated number of times of tentative target is less than C ₁, and current time is not associated with, then this tentative target deleted.In this step, C ₁and C ₂choose and need dependence experience to select.

Step 4): measured value-measured value data correlation, step 2 carry out complete after, if also have measured value not to be associated with any target, then need to carry out measured value-measured value data correlation, this step is the key producing fresh target, and at initial time, because certainty target and tentative target number are all 0, do not need to carry out step above, directly carry out the data correlation of this step.

Data association algorithm in this step adopts minimum distance method, and algorithm flow chart as shown in Figure 4, in the middle of this association process, when needing to find target more than two nodes, could be calculated this target and be present in region.In this step, the localization method used in step 5 is needed.

Flow chart as shown in Figure 4, the key step of minimum distance method is as follows:

A) first produce all possible associative combination, calculate the angular error of each combination.

B) combination of angular error not within threshold value is deleted.

C) optimum combination is selected.

Superincumbent step a) in, angular error calculates down:

If a certain combination selects any one angle value from each node, if from i-th (i=1 ..., M) and select measured value in individual node (1≤l _i≤ L _i), utilize the location algorithm in step 5 can calculate the least square positioning result of this associative combination the angle value of target relative to node can be finally inversed by by this positioning result (i=1,2,3), [x _i, y _i] be node i position, definition angular error is:

$\mathrm{Err}\left(\mathrm{\θ}\right)=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}{({\mathrm{\θ}}_{l_i}-{\widehat{\mathrm{\θ}}}_{l_i})}^2$

Obviously, the angular error of faulty combination will be very large.

Superincumbent step b) in, threshold value, by manually empirically arranging, recommends to be set to (N is node number), namely allows the error average out to 15 degree of each node.This step can overcome the impact that false-alarm is brought.

Superincumbent step c) in, select the principle of optimum combination as follows:

The combination of correct association is selected to have following principle according to error:

[1] measured value can not appear in two correct associative combination.

[2] angle error value of correct associative combination should be minimum value as far as possible, and error should within a threshold range, and threshold value is rule of thumb manually arranged.

[3], in correct associative combination, target number should be many as much as possible.

It is maximum likelihood method that measured value-measured value is associated in theoretic optimum correlating method, but maximum likelihood method needs exhaustive all combinations, then the associating likelihood probability of each combination is calculated, select a kind of combination of maximum probability as association results, amount of calculation is very large, be not suitable for real-time process, and the minimum distance method that the present invention adopts calculates simple, successful.

Step 5): tentative object initialization.After step 4, if create new target association combination, then need to carry out target localization to carry out initialization to dbjective state according to the measured value of association.The present invention adopts amount of calculation simple and the least square method of Be very effective positions.This step main processes of calculation is as follows:

Suppose that total N number of node has been associated with target, use vector (i=1,2 ... N) coordinate of i-th node in two dimensional surface network is represented.Vector represent the associated coordinate of target in two dimensional surface.Azimuth between target and i-th node is set to φ _i, it meets following relational expression:

${\tan \mathrm{\φ}}_i=\frac{y_T-y_i}{x_T-y_i},(i=\mathrm{1,2},\·\·\·N)---\left(4\right)$

When nodes N >=3, equation group (1) is an overdetermined equation, separates overdetermined equation and generally adopts least square method.(4) formula is rewritten as follows:

-x _Ttgφ _i+y _T＝-x _itgφ _i+y _ii＝1，2…N (5)

Order:

$A=\left(\begin{array}{cc}-\mathrm{tg}{\widehat{\mathrm{\φ}}}_1& 1\\ -\mathrm{tg}{\widehat{\mathrm{\φ}}}_2& 1\\ .& .\\ .& .\\ .& .\\ -\mathrm{tg}{\widehat{\mathrm{\φ}}}_N& 1\end{array}\right)$

Then can obtain:

$A\·{\stackrel{\→}{x}}_T=b---\left(6\right)$

Linearly model can be found out after conversion, can be in the hope of by Linear least square estimation algorithm closed solutions be:

${\widehat{\stackrel{\→}{x}}}_{\mathrm{PLLS}}={\left(A^{T}A\right)}^{-1}{A}^T\stackrel{\→}{b}---\left(7\right)$

After target localization, whole system just can obtain a new tentative target.The position of this tentative target is exactly the positioning result of least square method, and speed is initialized as 0, and it is initial that this just can complete tentative targetpath.

So far, the present invention is complete about the multiple target real time data fusion method statement of water sound sensor network.

The present invention also provides a kind of multiple target real time data emerging system based on water sound sensor network, and this system is based on water sound sensor network, and for carrying out multiobject track initiation in real time, flight path maintains and flight path is deleted; Described system comprises as shown in Figure 5, in observation area, lay several sensor nodes, each node gathers environmental data separately, and carry out real time signal processing, then result is transferred to host node, host node utilizes the data receiving arrival to carry out data fusion, obtain the target information in observation area, be then transferred to user by wired or wireless mode.

The information collection node block diagram of water sound sensor network as shown in Figure 6, for gathering the environmental information in a certain region, carrying out real time signal processing, obtaining several azimuth informations; The transducer that wherein signal gathering unit adopts generally adopts linear array hydrophone or vector hydrophone, for the environmental information in acquisition node overlay area; Collecting unit by gather real-time data transmission to signal processing unit, signal processing unit processes the primary signal gathered in real time, obtain the azimuth information that raw information comprises, then result is transferred to communication unit, then result is sent to host node by communication unit.

As shown in Figure 7, wherein communication unit, for receiving the azimuth information that all acquisition nodes send for described host node block diagram; After communication unit collects the azimuth information of all nodes of single moment point, node azimuth information is transferred to data fusion center by communication unit, after the azimuth information that data fusion center process transmits, obtain the target information in observation area, then target information is transferred to user by wired or wireless mode.Wherein data fusion unit is specifically divided into again following unit:

Memory cell, the azimuth information that the data message detecting all targets in the data message of all targets in qualitative objective set really, tentative goal set for depositing all acquisition nodes and all node-node transmission received are come;

Certainty object element associative cell, deposits qualitative objective really for the nearest azimuth information that received by communication unit and described memory cell and carries out data correlation;

Tentative object element associative cell, carries out data correlation for the tentative target nearest azimuth information do not received by the described communication unit of certainty target association and described memory cell deposited;

Data associating unit between orientation values, carries out data correlation between the nearest azimuth information do not received by the described receiving element in certainty target and tentative target association;

Target tracking unit, when described certainty object set and tentative target tightening have target association to measured value, utilizes the measured value be associated with to carry out Trajectory Prediction and renewal to this target, thus completes the maintenance of flight path.

Target localization unit, if carrying out in the measured value between azimuth information-measured value association process, has been found to new tentative target, is then needing to utilize the method for target localization to carry out initialization to the state of tentative target.

Threshold value is arranged and state conversion unit, for arranging the threshold value C deleting this certainty object element ₂with another the threshold value C tentative target being converted to certainty target ₁; If meet threshold value C ₁, then this tentative target is converted to certainty target.

Wherein, in described integrated unit the concrete implementation step of unit and method before this invention section describe in introduce in detail.

So far, the specific embodiment of the present invention statement is complete.

It should be noted that, embodiment of the present invention of above introduction and and unrestricted.It will be understood by those of skill in the art that any amendment to technical solution of the present invention or the equivalent alternative spirit and scope not departing from technical solution of the present invention, it all should be encompassed in right of the present invention.

Claims (11)

1., based on a Multi-target Data fusion method for water sound sensor network, this data fusion method is used for carrying out multiobject track initiation in real time, flight path maintains and flight path is deleted, and described method comprises following steps:

Step 101, the acquisition node step of some water sound sensor networks is laid in observation area, N number of acquisition node is laid in waters, described N number of acquisition node adopts linear array hydrophone or vector hydrophone as probe portion, this probe portion also carries out real time signal processing, obtain the azimuth information of some object elements that each acquisition node described observes in described observation area, and the azimuth information obtained is transferred to host node by net under water;

Step 102, described host node carries out the step of data fusion to described azimuth information, the information data of the target that first described azimuth information and this host node have stored by this host node adopts data correlation strategy to carry out data correlation, then follow the tracks of multiple target and locate, to complete multiobject track initiation, flight path maintains and flight path is deleted;

Wherein, described target comprises: certainty goal set and tentative goal set; Wherein, described certainty goal set is the set of the target of continued presence in described observation area in the section sometime that detects of described acquisition node; Described tentative goal set is the set of that described acquisition node detects but the uncertain target that whether can occur continuously within a certain continuous print time period, if the number of times that can associate described azimuth information continuously when certain target in described tentative goal set reaches a certain setting threshold C ₁time secondary, this tentative target is just converted to a target in certainty goal set; Described N>=3;

Described data correlation strategy is: associate with the flight path of target the azimuth information value that the orientation values of the up-to-date transmission of described acquisition node is measured one by one with each object element in each object element left on described host node in qualitative objective set really, tentative goal set successively, and then determine each azimuth information value of described each acquisition node comes from which target in described certainty goal set or described tentative goal set; If after above-mentioned steps, measured value is also had not associate any target, then remaining azimuthal measurement value is carried out azimuthal measurement value therein to associate with azimuthal measurement Value Data, judge whether that new tentative target occurs further, fail to produce any tentative target if the azimuthal measurement value between remaining azimuth information associates with azimuthal measurement Value Data, then abandon this remaining azimuthal measurement value; If create new tentative target, then least square location algorithm is utilized to carry out the track initiation of tentative target.

2. the Multi-target Data fusion method based on water sound sensor network according to claim 1, is characterized in that, described employing data correlation strategy carries out track and localization to multiple target and comprises following steps further:

Step 201, described azimuth information is carried out measured value with the target left on described host node in qualitative objective set really associate with track data, then the remaining azimuth information data be not associated with are carried out measured value with the target in the tentative goal set of leaving on described host node to associate with track data, each orientation values for determining in described azimuth information comes from which target in described certainty goal set or described tentative goal set;

Step 202, the step that flight path upgrades and predicts, if a target in described certainty goal set or described tentative goal set by plural azimuth information data correlation, then upgrades by the flight path information of described up-to-date azimuth information to this target; If the data amount check of the azimuth information that object element can be associated with is less than 2, then using the updated value of the status predication value of this object element as the flight path information of current time;

Step 203, the step that certainty targetpath initial sum flight path is deleted, the state of each object element is checked according to the threshold value of setting, carry out the diversification in role of object element, even have certain object element in described certainty goal set not exceeded a certain setting threshold C by the number of times that described azimuth information associates continuously ₂, then this certainty object element is deleted; If there is the number of times of the continuous collected node azimuth information association of certain object element in described tentative goal set to exceed another setting threshold C ₁, then this tentative object element is changed into an object element in certainty goal set;

Step 204, carries out measured value and measured value data correlation by between the not associated azimuth information to any object element, checks whether that new tentative target occurs among association process; If find that there is new tentative object element to occur, then initialization is carried out to this tentative object element,

Wherein, described initialization comprises: adopt the positional information initialization that least square method positions, speed is set to 0.

3. the Multi-target Data fusion method based on water sound sensor network according to claim 2, is characterized in that, the data association algorithm of step 201 adopts classical nearest neighbor algorithm.

4. the Multi-target Data fusion method based on water sound sensor network according to claim 2, is characterized in that, the state updating algorithm of step 202 adopts EKF.

5. the Multi-target Data fusion method based on water sound sensor network according to claim 2, is characterized in that, the data association algorithm of step 204 adopts minimum distance method.

6., based on a Multi-target Data emerging system for water sound sensor network, this system is based on water sound sensor network, and for carrying out multiobject track initiation in real time, flight path maintains and flight path is deleted; Described system comprises:

The information collection node of water sound sensor network: for gathering the data in a certain region, process, obtains several azimuth informations about target;

Host node, for receiving the azimuth information that described information collection node sends, carries out data fusion, carries out multiobject track initiation in real time, flight path maintains and flight path is deleted;

Wherein, described multiple target comprises: the target that certainty goal set comprises and the target that tentative goal set comprises; Described certainty goal set is the set of the object element of continued presence in observation area in the section sometime that detects of described acquisition node, described tentative goal set is the set of that described acquisition node detects but the uncertain object element that whether can occur continuously within a certain continuous print time period, if the number of times that can associate described azimuth information continuously when certain object element in described tentative goal set reaches a certain setting threshold C ₁time secondary, this tentative object element is just converted to an object element in certainty goal set;

Described host node comprises further:

Communication unit, for receiving the azimuth information that all acquisition nodes send, and sends to memory cell by this information;

Memory cell, the azimuth information that the data message detecting all object elements in the data message of all object elements in qualitative objective set really, tentative goal set for depositing all acquisition nodes and all node-node transmission received are come;

Certainty object element associative cell, the data message of each certainty object element that the described memory cell for the up-to-date azimuth information that received by communication unit and extraction is deposited carries out data correlation one by one;

Tentative object element associative cell, for carrying out data correlation one by one by all elements in the tentative goal set of not deposited by the nearest azimuth information in the association of the data message of object elements all in certainty goal set and described memory cell;

Data associating unit between orientation values, for not carried out measured value-measured value data correlation between the azimuth information on any target association, to judge whether that new tentative target occurs;

Target tracking unit, when described up-to-date azimuth information associates with any one object element, utilizes the up-to-date azimuth information that is associated with to carry out Trajectory Prediction and renewal to the object element in this association, thus completes and maintain this object element flight path;

Target localization unit, if create new tentative target in the middle of the azimuthal measurement value between azimuth information and azimuthal measurement Value Data association process, then needs to utilize the state of the method for target localization to this new tentative object element to carry out initialization;

Threshold value is arranged and state conversion unit, for arranging the threshold value C deleting this certainty object element ₂with another the threshold value C tentative object element being converted to certainty object element ₁; If meet threshold value C ₁, then this tentative object element is converted to certainty object element.

7. the Multi-target Data emerging system based on water sound sensor network according to claim 6, it is characterized in that, described acquisition node comprises further:

Adopt the probe unit of linear array hydrophone or vector hydrophone, for the primary data information (pdi) of acquisition node overlay area;

Signal processing unit, processing for the real-time primary signal to gathering, obtaining the azimuth information of target; With

Communication unit, crosses net be under water transferred to host node for obtaining azimuth information.

8. the Multi-target Data emerging system based on water sound sensor network according to claim 6, is characterized in that, the data association algorithm of described certainty target data association unit and tentative target association unit adopts classical nearest neighbor algorithm.

9. the Multi-target Data emerging system based on water sound sensor network according to claim 6, is characterized in that, the azimuth information between described azimuth information and azimuth information are carried out data correlation and adopted minimum distance method.

10. the Multi-target Data emerging system based on water sound sensor network according to claim 6, is characterized in that, described target following adopts EKF method.

The 11. Multi-target Data emerging systems based on water sound sensor network according to claim 6, is characterized in that, described target localization adopts least square method.