CN103152791B - A kind of method for tracking target based on underwater wireless sensor network - Google Patents

Abstract

The invention discloses a kind of method for tracking target based on underwater wireless sensor network.First the method selects cluster node according to peak signal principle, then observes target according to single-hop distance criterion group clustered network, if observation signal intensity exceedes threshold value, then sends observation data to leader cluster node.Leader cluster node receives the data that thick interior nodes transmits, and adopts the particle filter algorithm improving resampling to estimate the target location of current time and variance.Motion according to target constantly upgrades leader cluster node, send a upper leader cluster node state estimation and estimate of variance to current cluster head node, the resampling particle filter algorithm improved is adopted to estimate moving target position, until moving target is beyond the following range of underwater wireless sensor network by current cluster head node again; The present invention uses the particle filter tracking method improving resampling methods to estimate position and the variance of submarine target, improves the performance of target tracking of underwater wireless sensor network.

Description

A kind of method for tracking target based on underwater wireless sensor network

Technical field

The present invention relates to a kind of method for tracking target based on underwater wireless sensor network.

Background technology

Underwater wireless sensor network refers to the network disposing a large amount of sensor nodes and Autonomous Vehicles cooperation monitoring and collection surrounding environment data of interest in certain waters, sensor node can be set up network to self-organizing and carry out sound communication, through Data fusion technique, the data of acquisition are sent to the control centre of the water surface or bank base by specified node, so just achieve the fusion of underwater sensor network and terrestrial communications network.Underwater sensor has the short feature of low-power consumption, transmission range usually.

Underwater target tracking is an important application of underwater sensor network.Underwater sensor network has that Node distribution is wide, quantity is many, can mutually cooperate, swap data between node, and the features such as extensibility is strong, this is conducive to expanding the following range of target, the reliability strengthening target following and real-time.

Underwater target tracking mostly is nonlinear problem, particle filter method is widely applied in nonlinear and non-Gaussian problem, in the wireless sensor network of land, existing particle filter is used for Target Tracking Problem, therefore underwater target tracking problem many employings particle filter method.

The method for tracking target of underwater sensor network can be divided into centralized particle filter tracking and distributed particle filter to follow the tracks of according to the difference of particle filter working method.Only have a Centroid in centralized particle filter tracking method network, all the other nodes send the measurement data of tracking target to Centroid, and Centroid is responsible for using particle filter to carry out data processing, estimates the movement locus of maneuvering target.Centralized particle filter tracking method makes whole network stable not and laod unbalance, and distributed particle filter tracking method overcomes these shortcomings, and the difference according to particle filter existence form is divided into four kinds haply.In first method network, different pieces of information processing node does not run particle filter algorithm in the same time, according to the prediction locus of maneuvering target, selects from the nearest feasible node of predicted position as processing node.Processing node changes along with the change of maneuvering target track, solves the problem of networks vulnerable and laod unbalance in centralized particle filter.Second method is in order to overcome in query script the huge loss that may exist, adopt the simple mode expanding processing node, namely in network, each processing node runs identical particle filter algorithm simultaneously, the data that measured value gathers from whole network sensor.The third method supposes that each processing node has the transducer of respective numbers to be uniquely connected with it, and each processing node runs different particle filter algorithms simultaneously, upgrades the data that measured value adopts the transducer be attached thereto to gather.4th kind of method considers that particle filter population is more, higher to the precision of maneuvering target tracking track, and single processing node finite capacity, so be divided into several subsets to run particle filter respectively in different processing nodes particle collection.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, a kind of method for tracking target based on underwater wireless sensor network is provided.

Step based on the method for tracking target of underwater wireless sensor network is as follows:

1) initialization underwater wireless sensor network, makes all the sensors node all have same specification, and all in running order;

2) select received signal strength is maximum sensor node as leader cluster node under water, with the sensor node of leader cluster node within the scope of single-hop communication and leader cluster node group bunch, all the other sensor nodes remain on resting state;

3) a bunch inner sensor node is observed target, by the intensity of signal that receives compared with pre-determined threshold, if higher than pre-determined threshold, sends data to leader cluster node, otherwise does not send;

4) Initial state estimation value and initial variance estimated value is set;

5) in the k moment according to step 2) group bunch, and the state estimation estimated by a upper moment particle filter and estimate of variance packing send the leader cluster node in this k moment to;

6) carry out the particle filter of the improvement resampling in k moment, N number of particle of sampling from importance density function, then upgrade sampling particle, carry out particle and improve resampling, finally export location estimation value and the estimate of variance of target;

7) the k moment is from adding 1, and the motion according to target constantly upgrades leader cluster node, sends the information of a upper leader cluster node to current cluster head node when leader cluster node is changed;

8) step 5)-step 7) is repeated, depart from the overlay area of underwater wireless sensor network until target till.

Described step 6) is: from gather particle i=1 ..., N, and calculate weights of importance and normalization weights of importance, wherein be suggestion distribution, N number of particle weights of importance of collection is after normalization, weight is resampling renewal is carried out to particle, update method is: select the particle retained according to the size of particle weights, and the method that namely particle is saved through probability wall is originally rejected, when particle weights is through multiple probability wall, abandon the pattern copying particle, and adopt following strategy:

When copying number and being 2n even number, produce 2 (n-1) individual new particle:

$\{x-\frac{2n-3}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\}\left\{x\right\}\left\{x\right\}\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-3}{4N}*\mathrm{\Σ}\},$

When copying number and being 2n+1 odd number, produce 2n new particle:

$\{x-\frac{2n-1}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\left\}\right\{x\left\}\right\{x\left\}\right\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-1}{4N}*\mathrm{\Σ}\}$

Wherein ∑ is the decentralization variable of new particle.

The present invention uses the particle filter tracking method improving resampling methods to estimate position and the variance of submarine target, improves the performance of target tracking of underwater wireless sensor network.

Accompanying drawing explanation

Fig. 1 is underwater sensor network pursuit movement object delineation.

Embodiment

Method for tracking target based on underwater wireless sensor network comprises the following steps:

1) initialization underwater wireless sensor network, makes all the sensors node all have same specification, and all in running order;

2) select received signal strength is maximum sensor node as leader cluster node under water, with the sensor node of leader cluster node within the scope of single-hop communication and leader cluster node group bunch, all the other sensor nodes remain on resting state;

3) a bunch inner sensor node is observed target, by the intensity of signal that receives compared with pre-determined threshold, if higher than pre-determined threshold, sends data to leader cluster node, otherwise does not send;

4) Initial state estimation value and initial variance estimated value is set;

5) in the k moment according to step 2) group bunch, and the state estimation estimated by a upper moment particle filter and estimate of variance packing send the leader cluster node in this k moment to;

6) carry out the particle filter of the improvement resampling in k moment, N number of particle of sampling from importance density function, then upgrade sampling particle, carry out particle and improve resampling, finally export location estimation value and the estimate of variance of target;

7) the k moment is from adding 1, and the motion according to target constantly upgrades leader cluster node, sends the information of a upper leader cluster node to current cluster head node when leader cluster node is changed;

8) step 5)-step 7) is repeated, depart from the overlay area of underwater wireless sensor network until target till.

Described step 6) is: from gather particle i=1 ..., N, and calculate weights of importance and normalization weights of importance, wherein be suggestion distribution, N number of particle weights of importance of collection is after normalization, weight is resampling renewal is carried out to particle, update method is: select the particle retained according to the size of particle weights, originally the method that namely particle is saved through probability wall is rejected, when particle weights is through multiple probability wall, abandon the pattern copying particle, and adopt following strategy: when copying number and being 2n even number, produce 2 (n-1) individual new particle:

$\{x-\frac{2n-3}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\}\left\{x\right\}\left\{x\right\}\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-3}{4N}*\mathrm{\Σ}\},$

When copying number and being 2n+1 odd number, produce 2n new particle:

$\{x-\frac{2n-1}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\left\}\right\{x\left\}\right\{x\left\}\right\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-1}{4N}*\mathrm{\Σ}\}$

Wherein ∑ is the decentralization variable of new particle.

Embodiment

Step 101: initialization underwater wireless sensor network, sows wireless sensor network node in environment under water at random, and all nodes all have unified specification, as communication distance, detection range etc., all nodes are all in running order, keep detecting function, but can communication close function.

Step 102: sensor node detects target, wakes the node in investigative range up.These nodes in investigative range are organized bunch on principle, and the node M namely selecting signal receiving strength maximum is as leader cluster node, and form bunch with the node of leader cluster node M within the scope of single-hop communication and leader cluster node, all the other nodes continue maintenance resting state.

Step 103: a bunch interior nodes is observed target, carries out processing locality to the received signal, then sends data to leader cluster node.The strength model of node Received signal strength is:

$z\left(k\right)=\frac{S\left(k\right)}{{[x\left(k\right)-x]}^2+{y\left(k\right)-y]}^2}+\mathrm{\ϵ}\left(k\right)$

Wherein S (k) is the acoustic pressure of target source rank, x (k) and y (k) is the two-dimensional coordinate of target in the k moment, x and y is the two-dimensional coordinate of sonar sensor, ε (k) is independent white Gaussian noise, and the observation of all underwater sensor nodes is all independently.

The signal strength signal intensity that i-th node receives processes in this locality, and then send data to leader cluster node according to criterion, this criterion is: signal z (k) arrived by k reception and pre-determined threshold D _thresholdcompare, if value is lower than thresholding, then do not send any information; If value is higher than thresholding, then the information that sends is to leader cluster node.Therefore, node only has when z (k) is higher than thresholding D _thresholdtime just transmit information to leader cluster node.The measurement that leader cluster node receives from i-th node is:

$z_i\left(k\right)=\frac{S\left(k\right)}{{[x\left(k\right)-x_i]}^2+{[y\left(k\right)-y_i]}^2}*{\mathrm{\λ}}_i+{\mathrm{\ϵ}}_i\left(k\right),$ Wherein ${\mathrm{\λ}}_i=\left\{\begin{array}{ccc}1& \mathrm{if}& z_i\left(k\right)>D_{\mathrm{threshold}}\\ 0& \mathrm{if}& z_i\left(k\right)\≤D_{\mathrm{threshold}}\end{array}\right..$

Step 104: in the k=0 moment, according to bunch principle of the group in step 102, composition initial cluster, and select leader cluster node, setting Initial state estimation value and initial variance estimated value.

Step 105: in the k moment, as shown in Figure 1, according to the group bunch principle group bunch of step 102, and the state estimation calculated by a upper moment particle filter and estimate of variance packing send the leader cluster node in this k moment to.

Step 106: the particle filter algorithm carrying out the improvement resampling in k moment, carries out state estimation to the position of target.From gather particle i=1 ..., N, and calculate weights of importance and normalization weights of importance.Wherein it is suggestion distribution.The N number of particle weights of importance gathered is after normalization, weight is

Carry out resampling renewal to particle, update method is: select the particle retained according to the size of particle weights, and the method that namely particle is saved through probability wall is originally rejected.When particle weights is through multiple probability wall, abandon the pattern copying particle, and adopt following strategy:

When copying number and being 2n even number, produce 2 (n-1) individual new particle:

$\{x-\frac{2n-3}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\}\left\{x\right\}\left\{x\right\}\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-3}{4N}*\mathrm{\Σ}\},$

When copying number and being 2n+1 odd number, produce 2n new particle:

$\{x-\frac{2n-1}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\left\}\right\{x\left\}\right\{x\left\}\right\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-1}{4N}*\mathrm{\Σ}\}$

Wherein ∑ is the decentralization variable of new particle.The particle filter algorithm improving resampling is as shown in the table:

The step 107:k moment is from adding 1.At subsequent time, when target moves to another position, again to organize bunch by step 102, when the leader cluster node selected is not identical with the leader cluster node in a upper moment, a upper leader cluster node sends information to current cluster head node, and the information transmitted between leader cluster node is state estimation and the estimate of variance of a upper moment target.Motion according to target constantly upgrades cluster and clusterhead node, when leader cluster node is changed, sends the information of a upper leader cluster node to current cluster head.

Step 108: repeat step 105-107, till target departs from underwater wireless sensor network overlay area.

Claims (2)

1., based on a method for tracking target for underwater wireless sensor network, it is characterized in that its step is as follows:

1) initialization underwater wireless sensor network, makes all the sensors node all have same specification, and all in running order;

2) select received signal strength is maximum sensor node as leader cluster node under water, with the sensor node of leader cluster node within the scope of single-hop communication and leader cluster node group bunch, all the other sensor nodes remain on resting state;

3) a bunch inner sensor node is observed target, by the intensity of signal that receives compared with pre-determined threshold, if higher than pre-determined threshold, sends data to leader cluster node, otherwise does not send;

4) Initial state estimation value and initial variance estimated value is set;

5) in the k moment according to step 2) group bunch, and the state estimation estimated by a upper moment particle filter and estimate of variance packing send the leader cluster node in this k moment to;

6) carry out the particle filter of the improvement resampling in k moment, N number of particle of sampling from importance density function, then upgrade sampling particle, carry out particle and improve resampling, finally export location estimation value and the estimate of variance of target;

7) the k moment is from adding 1, and the motion according to target constantly upgrades leader cluster node, sends the information of a upper leader cluster node to current cluster head node when leader cluster node is changed;

8) step 5)-step 7) is repeated, depart from the overlay area of underwater wireless sensor network until target till.

2. a kind of method for tracking target based on underwater wireless sensor network according to claim 1, is characterized in that: described step 6) is: from gather particle i=1 ..., N, and calculate weights of importance and normalization weights of importance, wherein be suggestion distribution, N number of particle weights of importance of collection is after normalization, weight is resampling renewal is carried out to particle, update method is: select the particle retained according to the size of particle weights, and the method that namely particle is saved through probability wall is originally rejected, when particle weights is through multiple probability wall, abandon the pattern copying particle, and adopt following strategy:

When copying number and being 2n even number, produce 2 (n-1) individual new particle:

$\{x-\frac{2n-3}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\}\left\{x\right\}\left\{x\right\}\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-3}{4N}*\mathrm{\Σ}\},$

When copying number and being 2n+1 odd number, produce 2n new particle:

$\{x-\frac{2n-1}{4N}*\mathrm{\Σ}\},...,\{x-\frac{1}{4N}*\mathrm{\Σ}\left\}\right\{x\left\}\right\{x\left\}\right\{x+\frac{1}{4N}*\mathrm{\Σ}\},...,\{x+\frac{2n-1}{4N}*\mathrm{\Σ}\}$

Wherein ∑ is the decentralization variable of new particle.