CN105717505A - Data association method for utilizing sensing network to carry out multi-target tracking - Google Patents

Abstract

The invention provides a data association method for utilizing a sensing network to carry out multi-target tracking. The method comprises the steps of: clustering sensor nodes; collecting measurement information of targets by the nodes; calculating state discrete time equations of the targets; calculating state prediction results of the targets; calculating the posterior probability of a target k in a node I; carrying out target tracking by means of a Bayes framework; calculating the posterior probability distribution of the targets; and repeating the above steps to finish target tracking. In order to carry out accurate tracking on multiple targets, the invention provides the new method to solve the data association problem in wireless sensor network multi-target tracking, the method is based on the Bayes framework, the target position state information is utilized, and the speed state information is also utilized; and compared with the modeling methods, the method provided by the invention is low in calculation amount, high in precision and is capable of accurately carrying out multi-target tracking under a low-cost condition especially in the aspect of cross and maneuvering target tracking.

Description

Sensor Network is utilized to carry out the data correlation method of multiple target tracking

Technical field

The invention belongs to measurement and control area, be specifically related to a kind of data correlation method utilizing Sensor Network to carry out multiple target tracking.

Background technology

In wireless sensor network, the ability following the tracks of target is most important in numerous applications.But due to the limitation in energy, perception, communication, storage and computing capability, they realize conventional target tracking solution and bring challenge during wireless sensor network is applied.Owing to the operation of the collection of any data, process and propagation information causes the increase of resource consumption, wireless sensor network target track algorithm should be able to effectively utilize resource, also to reduce computation complexity.

At present, monotrack obtains good research, but multiple target tracking (MTT) problem still has some to need the key issue solved.Wherein data association is the problem that MTT is the most difficult.The tracking being illustrated in figure 1 under simple scenario two targets, including three phases.First stage, the distance between target is big, and described target can be individually follow the tracks of.Second stage, target is closer to each other, and this eventually results in measurement distortion.This distortion measurement may result in measuring obscuring between target, and data association at this moment can be utilized to process this distortion.Phase III, the distance between target becomes big, and monotrack model is to be suitable for once again.Occurring the mistake in second stage can cause confusion or lost target, it is necessary for then reinitializing in the 3rd stage or again identify.

Now with some research about wireless sensor network MTT, in the framework of IDSQ, such as propose the algorithm of a kind of wireless sensor network MTT.Graphic model is for calculating the association between dbjective state and sensor measurement.Target property information is used for processing the measurement of distortion target.This algorithm achieves Distributed Calculation and localization process by the cooperation between sensor node.But, introduce additional attribute information and add the requirement to hardware complexity, too increase cost, calculating, communication and energy expense.

Summary of the invention

Overhead is less, data association is accurate to it is an object of the invention to provide one such that it is able to the Sensor Network that utilizes that multiple target carries out accurately tracking carries out the data correlation method of multiple target tracking.

This data correlation method utilizing Sensor Network to carry out multiple target tracking provided by the invention, comprises the steps:

S1. the sensor node in monitored area is carried out sub-clustering；

S2., when target enters monitored area, the measurement information of target is gathered by sensor node；

S3. the state discrete time equation of target is calculated；

S4. according to the step S3 result of calculation obtained, following formula is adopted to calculate the status predication result of target:

WhereinFor the target k distributions in the t-1 moment；

And the prediction measuring distribution is calculated as follows:

Wherein,It it is likelihood function；

S5. the result of calculation according to step S4, calculates target k posterior probability in node i:

S6. carry out data association according to step S5, and adopt Bayesian frame to carry out target following；

S7. following formula is adopted to calculate the Posterior probability distribution of target:

Wherein, α is a normalization factor, its objective is byValue be transformed between 0～1；K=1 ..., N is the target k enhancing vector in t, wherein x_kT () is position vector,It is velocity vector,For the target k expectation at the state estimation of t；

S8. repeat step S2～S7, complete the tracking of target.

Sub-clustering described in step S1, for carrying out sub-clustering by node by distance.

Sensor described in step S1 is receiving intensity sensor.

The measurement information of the collection target described in step S2, for adopting the following formula computing node i signal intensity received in t:

Wherein, zⁱT () is the signal intensity that node i receives in t；V_iT () is background noise, N is the quantity of target, a_kT () is the target k signal intensity in t, x_kT () is the target k position vector in t, ξ_iIt is the position vector of sensor node i.

State discrete time equation described in step S3, for adopting following formula to calculate state discrete time equation:

y_k(t)=A y_k(t-1)+ω_k

Wherein,It is state-transition matrix, ω_kTo be average be 0 Gaussian noise sequence；In same sampling period, the change of speed can be expressed asAnd using its selection standard as the value of q, wherein Q is covariance matrix

Employing Bayesian frame described in step S6 carries out target following, for adopting the estimated value of following formula calculated target positions:

Wherein p (x_k| z_1:k) represent according to measuring z_1:k={ z₁,z₂,…z_kEstimation x_kProbability density function；Measure z_k=h_k(x_k,w_k) for state model x_k=f_k(x_k-1,v_k-1) measured value.

The present invention is in order to follow the tracks of accurately multiple target, propose a kind of new method to solve the data association problem during wireless sensor network multi-target is followed the tracks of, the method, based on Bayesian frame, not only uses target location status information, simultaneously also operating speed status information.Compared with the method for some other modeling, method amount of calculation proposed by the invention is little, precision is high, particularly in realizing intersecting machine tracking of maneuvering target, it is possible to exactly multiple target is tracked when low overhead.

Accompanying drawing explanation

Fig. 1 is the example schematic of background technology.

Fig. 2 is the method flow diagram of the present invention.

Fig. 3 is specific embodiments of the invention schematic diagram.

Detailed description of the invention

It is illustrated in figure 2 the method flow diagram of the present invention, this data correlation method utilizing Sensor Network to carry out multiple target tracking provided by the invention, comprises the steps:

S1. for receiving intensity sensor, the sensor node in Sensor Network is carried out sub-clustering by distance；

S2., when target enters the monitored area of Sensor Network, the measurement information of target is gathered by sensor node: the signal intensity that node i receives in t is:

Wherein, v_iT () is background noise, N is the quantity of target, a_kT () is the target k signal intensity in t, x_kT () is the target k position vector in t, ξ_iIt is the position vector of sensor node i.

S3. the state discrete time equation of target is calculated:

y_k(t)=A y_k(t-1)+ω_k

Wherein,It is state-transition matrix, ω_kTo be average be 0 Gaussian noise sequence；In same sampling period, the change of speed can be expressed asAnd using its selection standard as the value of q, wherein Q is covariance matrix

S4. according to the step S3 result of calculation obtained, following formula is adopted to calculate the status predication result of target:

WhereinFor the target k distributions in the t-1 moment；

And the prediction measuring distribution is calculated as follows:

Wherein,It it is likelihood function；

S5. the result of calculation according to step S4, calculates target k posterior probability in node i:

S6. carry out data association according to step S5, and adopt the estimated value of following formula calculated target positions:

Wherein p (x_k| z_1:k) represent according to measuring z_1:k={ z₁,z₂,…z_kEstimation x_kProbability density function；Measure z_k=h_k(x_k,w_k) for state model x_k=f_k(x_k-1,v_k-1) measured value；

S7. following formula is adopted to calculate the Posterior probability distribution of target:

Wherein, α is a normalization factor, its objective is byValue be transformed between 0～1；K=1 ..., N is the target k enhancing vector in t, wherein x_kT () is position vector,It is velocity vector,For the target k expectation at the state estimation of t；

S8. repeat step S2～S7, complete the tracking of target.

Below in conjunction with a specific embodiment and Fig. 2, the method for the present invention is further described:

As in figure 2 it is shown, deploy substantial amounts of inexpensive sensor node on highway.We by distance sub-clustering, have some sensor nodes these nodes in every bunch, bunch head is elected in turn by sensor node (target following node), it would however also be possible to employ Dynamic Cluster head selection algorithm is chosen.Original state each bunch has a small amount of sensor node (target detection node) to be active, and other sensor nodes (target following node) are completely in sleep state；When detecting that target occurs, target detection node will wake bunch other interior sensor nodes up and carry out target following, and now target detection node then enters sleep state.Sensor node, when multiple target is tracked, can collect the measurement information about target, and this measurement information is sent to bunch head carries out data association, and carries out Bayesian Estimation to carry out target following.Meanwhile, the location status information of target is also considered with speed state information to calculate the Posterior probability distribution of target, it will be used to carry out data association in next one estimation.

According to Fig. 2 it can be seen that target A is in the effective scope of detection of node 1,2 and 3, target B is in the effective scope of detection of node 2,3 and 4, and target C is in the effective scope of detection of node 6.The measurement information of these 6 nodes as it can be seen, wherein x represent that dbjective state, z represent the measurement information of node.According to this figure, sensor 2 and 3 is simultaneously by the impact of target A and B；Therefore the measurement of single sensor likely can be subject to the impact of multiple target.On the other hand, single target can be perceived by multiple sensor nodes simultaneously, and such as target A other node 1,2 and 3 simultaneously perceives, and target A is in the state of tWithIt is associated, therefore can obtain target information accurately by the information fusion that multiple sensors measure.

Below multiple target tracking data correlation method being described in detail, flow process is as it is shown on figure 3, specifically comprise the following steps that

S1. initialize: selecting the suitable deployed position of node, member node is deployed in the both sides of road；Assume to have been detected by corresponding target, and the number of target is K, and adopt multiple target tracking algorithm.

S2. target following: utilize inexpensive sensor to carry out information gathering, when multiple targets are close to each other, carries out data association calculating, to carry out multiple target tracking accurately.

This step specifically includes following steps:

1) by bunch member node collection capacity measurement information, and this status information being sent to a bunch head, bunch head selects to be adopt certain mechanism, allows each member node be elected in turn, to realize balancing energy.

Assuming that k is the quantity of signal source, node i is following formula (1) at the signal intensity that t receives:

zⁱ(t)=sⁱ(t)+υ_i(t)

Wherein, v_iT () is background noise, be normally provided as average be 0, variance beGaussian noise, s_iT () is the signal intensity from signal source (monitoring objective) k, be calculated as follows formula (2):

Wherein, a_kT () is the signal source k signal intensity in t, x_kT () is the signal source k position vector in t, ξ_iIt is the position vector of sensor node.

According to formula (1) and (2), node i is at the measurement z of tⁱT () can be calculated as follows formula (3):

2) bunch head receives step 1) after the measurement information that calculates, carry out data association calculating.

Assume t_sBeing a sampling interval, the discrete time equation of state can be expressed as follows formula (6):

y_k(t)=A y_k(t-1)+ω_k

Wherein,It is state-transition matrix, ω_kTo be average be 0 Gaussian noise sequence.In same sampling period, the change of speed can be expressed asAnd using its selection standard as the value of q, wherein Q is covariance matrix

It is understood that the distributions that signal source k is in the t-1 moment isState transition model can be obtained, then status predication according to formula (6)Formula (7) can be calculated as follows:

So, the prediction measuring distribution can be calculated as follows formula (8):

Wherein,It it is likelihood function.

According to formula (7), the measurement of node i includes N number of target, then target k posterior probability in node i can be calculated as follows:

3) calculate a Bayesian Estimation and carry out target following.Target following can be modeled as a dynamic state estimator problem, and the framework based on bayes method can well solve dynamic state estimator problem.Assume that state model is x_k=f_k(x_k-1,v_k-1), wherein x_kIt is dbjective state, v_kIt it is process noise.Meanwhile, measurement model is z_k=h_k(x_k,w_k), wherein w_kIt is measure noise.So bayes method can according to measuring z_1:k={ z₁,z₂,…z_kEstimate x_kProbability density function (PDF): p (x_k| z_1:k).In Sensor Network multiple target tracking is applied, the motion model of target is usually uncertain and unstable, and calculates resource-constrained.So being generally selected the Brownian Model motion model as target.Actually, in the MTT of wireless sensor network applies, what target moved changes more much smaller than the sample rate of sensing node (such as the sample rate of MICAz can more than more than 100 hertz), and therefore constant speed (CV) model can be applied in the sampling interval.

Assume at moment k-1, given probability density function (PDF) p (x_k-1| z_1:k-1).So the PDF of dbjective state is p (x_k| z_1:k-1)=∫ p (x_k| x_k-1)p(x_k-1| z_1:k-1)dx_k-1.In the k moment, when obtaining measuring z_kTime, the estimated value of target location can be calculated according to bayes method, computational methods such as following formula (9):

4) Posterior probability distribution of target is calculated, it will be used to carry out data association in the next one is estimated.

Assuming have m sensor to participate in the target k state estimation in t, the measurement of all the sensors is separate, and the Posterior probability distribution of target k can be calculated as follows:

Wherein, α is a normalization factor,K=1 ..., N is the target k enhancing vector in t, wherein x_kT () is position vector,It it is velocity vector.So, obtain target k after the state estimation of t,Can be obtained by calculating expectation.

S2. judge whether to continue to follow the tracks of, if it is, enter step S1, otherwise terminate.

Claims (6)

1. utilize Sensor Network to carry out a data correlation method for multiple target tracking, comprise the steps:

S1. the sensor in Sensor Network is carried out sub-clustering；

S2., when target enters the monitored area of Sensor Network, the measurement information of target is gathered by sensor node；

S3. the state discrete time equation of target is calculated；

S4. according to the step S3 result of calculation obtained, following formula is adopted to calculate the status predication result of target:

$\hat{p}\left(y_k\right(t\left)\right)={\∫}_{\mathrm{\ψ}(t-1)}p\left(y_k\right(t\left)\right|y_k\left(t-1\right))\·\tilde{p}\left(y_k\right(t-1\left)\right){\mathrm{dy}}_k(t-1)$

WhereinFor the target k distributions in the t-1 moment；

And the prediction measuring distribution is calculated as follows:

$\hat{p}\left(z_k^i\right(t\left)\right)={\∫}_{\mathrm{\ψ}\left(t\right)}p\left(z_k^i\right(t\left)\right|y_k\left(t\right))\·\hat{p}\left(y_k\right(t\left)\right){\mathrm{dy}}_k\left(t\right)$

Wherein,It it is likelihood function；

S5. the result of calculation according to step S4, calculates target k posterior probability in node i:

$\tilde{p}\left(z_k^i\right(t\left)\right)={\∫}_{R^{N-1}}p\left(z^i\right|z_1^i,...,z_N^i)\[\underset{k=1}{\overset{N}{\Π}}\hat{p}\left(z_k^i\right(t\left)\right)\]{\mathrm{dz}}_1^i...{\mathrm{dz}}_{k-1}^i{\mathrm{dz}}_{k+1}^i...{\mathrm{dz}}_N^i$

S6. carry out data association according to step S5, and adopt Bayesian frame to carry out target following；

S7. following formula is adopted to calculate the Posterior probability distribution of target:

$\tilde{p}\left(y_k\right(t\left)\right)=\mathrm{\α}\·\hat{p}\left(y_k\right(t\left)\right)\·\∫\[\underset{i=1}{\overset{m}{\Π}}\tilde{p}\left(z_k^i\right(t\left)\right)\·p\left(z_k^i\right(t\left)\right|y_k\left(t\right))\]{\mathrm{dz}}_k^1...{\mathrm{dz}}_k^m$

Wherein, α is a normalization factor, its objective is byValue be transformed between 0～1；It is the target k enhancing vector in t, wherein x_kT () is position vector,It is velocity vector,For the target k expectation at the state estimation of t；

S8. repeat step S2～S7, complete the tracking of target.

2. the data correlation method utilizing Sensor Network to carry out multiple target tracking according to claim 1, it is characterised in that the sub-clustering described in step S1, for carrying out sub-clustering by node by distance.

3. the data correlation method utilizing Sensor Network to carry out multiple target tracking according to claim 1, it is characterised in that the sensor described in step S1 is receiving intensity sensor.

4. carry out the data correlation method of multiple target tracking according to the Sensor Network that utilizes one of claims 1 to 3 Suo Shu, it is characterised in that the measurement information of the collection target described in step S2, for adopting the following formula computing node i signal intensity received in t:

$z^i\left(t\right)=\underset{k=1}{\overset{N}{\Σ}}\frac{a_k\left(t\right)}{\left|\right|x_k\left(t\right)-{\mathrm{\ξ}}_i\left|\right|}+{\mathrm{\υ}}_i$

Wherein, zⁱT () is the signal intensity that node i receives in t；V_iT () is background noise, N is the quantity of target, a_kT () is the target k signal intensity in t, x_kT () is the target k position vector in t, ξ_iIt is the position vector of sensor node i.

5. carry out the data correlation method of multiple target tracking according to the Sensor Network that utilizes one of claims 1 to 3 Suo Shu, it is characterised in that the state discrete time equation described in step, for adopting following formula to calculate state discrete time equation:

y_k(t)=A y_k(t-1)+ω_k

Wherein, $A=\left[\begin{array}{cc}1& t_s\\ 0& 1\end{array}\right]$ It is state-transition matrix, ω_kTo be average be 0 Gaussian noise sequence；In same sampling period, the change of speed can be expressed as $\sqrt{{\left(Q\right)}_{22}}=\sqrt{{\mathrm{qt}}_s},$ And using its selection standard as the value of q, wherein Q is covariance matrix $Q=\left[\begin{array}{cc}{t_s}^3/3& {t_s}^2/2\\ {t_s}^2/2& t_s\end{array}\right]q.$

6. carry out the data correlation method of multiple target tracking according to the Sensor Network that utilizes one of claims 1 to 3 Suo Shu, it is characterised in that the employing Bayesian Estimation described in step S6 carries out target following, for adopting the estimated value of following formula calculated target positions:

$p\left(x_k\right|z_{1:k})=\frac{p\left(z_k\right|x_k)p\left(x_k\right|z_{1:k-1})}{p\left(z_k\right|z_{1:k-1})}$

Wherein p (x_k|z_1:k) represent according to measuring z_1:k={ z₁,z₂,…z_kEstimation x_kProbability density function；Measure z_k=h_k(x_k,w_k) for state model x_k=f_k(x_k-1,v_k-1) measured value.