CN102186241A - Parallel distributed particle filter based wireless sensor network target tracking method - Google Patents

Abstract

The invention discloses a wireless sensor network target tracking method based on parallel distributed particle filtering, which uses a Gaussian mixture model instead of particles and weights, and calculates local importance sampling, local weights, local weight sums, and local state estimation sums in parallel for each node. variance and sent to the fusion center; the fusion center calculates the overall state estimation based on the transmitted data, and obtains the state estimation value and state estimation variance at the current moment; the state estimation value and state estimation variance value calculated at the previous moment and the Gaussian The particles and weights replaced by the mixture model are transmitted to the cluster head node at this moment to estimate the state of the target position; the present invention only transmits the Gaussian mixture model between the cluster heads, and when the cluster heads are replaced, only the transmission The particles and weights replaced by the Gaussian mixture model can reduce the calculation amount of the algorithm, further effectively reduce the bandwidth, and improve real-time performance and tracking accuracy.

Description

Target Tracking Method for Wireless Sensor Networks Based on Parallel Distributed Particle Filter

technical field

The invention relates to a wireless sensor network target tracking method, which belongs to the field of sensor technology and wireless communication technology.

Background technique

The wireless sensor network is composed of a large number of randomly distributed sensor nodes. It is an ad hoc network that can perceive the information of the detected objects in the coverage area. Because sensor nodes have the characteristics of small size, low price, wireless communication, self-organization, robustness and concealment, they are widely used in the fields of national defense, military and environmental detection. One of the typical applications of wireless sensor networks is target tracking. Through the mutual cooperation between nodes, the targets entering the coverage area are tracked, and the information such as the position of the target is estimated by using multi-node observations and filtering. However, wireless sensor networks also have their own defects: 1) limited energy, the general sensor nodes are powered by batteries, and cannot be recharged, and the transmission of information will consume a lot of energy; 2) the bandwidth is limited, the cooperative work between nodes needs to be transmitted A large amount of data; 3) Node computing energy is limited. At present, most target tracking filters based on wireless sensor networks use angle or signal strength as the measurement value, which will cause too much data transmitted by the sensor, data congestion, and occupy a large amount of bandwidth. Therefore, traditional target tracking is difficult to apply to ordinary wireless sensor network.

Compared with ordinary wireless sensor networks, binary wireless sensor networks only transmit "0" or "1", which can effectively save bandwidth and energy. The existing binary wireless sensor network target tracking uses network structure, sensor detection radius and geometric knowledge for positioning, which is lacking in accuracy; or uses filtering methods to track targets, but most of them are centralized filtering algorithms, which consume too much energy and Real-time can not meet the requirements.

At present, the distributed particle filter wireless sensor network target tracking method can be dynamically clustered according to the current target position, and can be optimized in reducing energy consumption and bandwidth usage, but its shortcomings are: first, the particle filter algorithm in this method It takes a lot of time, and the real-time performance is not good; second, this method is calculated on the cluster head node, which will cause the cluster head node to die prematurely.

Contents of the invention

The purpose of the present invention is to propose a binary wireless sensor network target tracking method based on parallel distributed particle filter in order to overcome the shortcomings of the distributed particle filter wireless sensor network target tracking method in the prior art. The calculation amount of the cluster head node to balance the energy consumption.

本地权重

本地权重和

本地状态估计

和方差

将得到的

传送给融合中心并对本地重采样；最后由融合中心根据传送的数据后计算总体的状态估计，得到当前时刻的状态估计值和状态估计方差P_t；3)在t时刻根据组簇原则组簇，并将上一时刻算出的状态估计值和状态估计方差值以及用高斯混合模型代替的粒子和权重传送给此时刻的簇头节点；4)计算t时刻的状态估计值和状态估计方差，对目标位置进行状态估计；5)t时刻加1，重复步骤3)-4)，直至目标离开覆盖区域。The technical solution of the present invention comprises the following steps: 1) Randomly sow wireless sensor network nodes in the coverage area, select cluster head nodes according to the clustering principle, make the nodes in the cluster observe the target, and send binary data to the cluster head nodes; 2) Form an initial cluster and select cluster head nodes at time t=0, first sample from prior distribution, use Gaussian mixture model to replace particles and weights, divide all particles X _t into s subsets, and divide the subsets into each a node and perform particle updates; then compute local importance samples in parallel for each node

local weight

local weights and

local state estimation

and variance

will get

Send it to the fusion center and resample locally; finally, the fusion center calculates the overall state estimation based on the transmitted data, and obtains the state estimation value at the current moment and state estimation variance P _t ; 3) at time t, group clusters according to the clustering principle, and transmit the state estimation value and state estimation variance value calculated at the previous moment, as well as the particles and weights replaced by the Gaussian mixture model to the Cluster head node; 4) Calculate the state estimation value and state estimation variance at time t, and estimate the state of the target position; 5) Add 1 at time t, repeat steps 3)-4), until the target leaves the coverage area.

The beneficial effects of the present invention are:

1. By using a binary wireless sensor network, only "0" or "1" signals are transmitted between the nodes in the cluster and the cluster head node, which can effectively reduce bandwidth and greatly reduce energy consumption.

2. Using the local parallel distributed particle filter algorithm of the Gaussian mixture model, only the Gaussian mixture model is transmitted between the cluster heads, and the clusters and the cluster heads are continuously updated according to the movement of the target. When the cluster heads are replaced, only the cluster heads need It is enough to transmit the particles and weights replaced by the Gaussian mixture model, and there is no need to transmit a large number of particles between the cluster heads, which further effectively reduces the bandwidth.

3. The parallel particle filter algorithm is adopted, which can adjust the number of particles online according to the filter variance, which reduces the calculation amount of the algorithm, effectively reduces the communication traffic in the target tracking process, greatly improves real-time performance and tracking accuracy, and reduces energy consumption.

Description of drawings

Below in conjunction with accompanying drawing and specific embodiment, the present invention is further elaborated:

Fig. 1 is the flowchart of target tracking method of the present invention;

Fig. 2 is a flowchart of local distributed particle filtering.

Detailed ways

The present invention is a parallel particle filter binary wireless sensor network target tracking method based on Gaussian mixture model, wherein the Gaussian mixture model uses Gaussian probability density function (normal distribution curve) to accurately quantify things, and decomposes one thing into several A model formed based on a Gaussian probability density function. This method uses parallel particle filter to track the position of the target. At the current moment, some sensor nodes are awakened to participate in target tracking according to certain principles, so as to reduce communication traffic and energy consumption. First, let the wireless sensor node be in a dormant state. When the target enters the sensor node, wake up the node, cluster the nodes in the cluster within the specified range, and then select the cluster head node, and re-enter the cluster nodes that do not meet the clustering principle. Sleep state to save energy of sensor nodes. The target tracking node uses the parallel particle filter algorithm, and the cluster head node and the nodes in the cluster cooperate to use parallel particle filter to estimate the target position at the current moment, and continuously update the cluster and the cluster head node according to the movement of the target. When the cluster head node is replaced, Send the information of the last cluster head node to the current cluster head node. Since the sensor nodes transmit binary data to the cluster heads, and the number of particles can be adjusted online, bandwidth and energy can be further saved.

Referring to Fig. 1, the concrete implementation method of the present invention is as follows:

Step 101: At time t=0, initialize the wireless sensor network, randomly sow wireless sensor network nodes in the coverage area, all nodes have uniform specifications, such as communication distance, detection distance, etc., all nodes are in a dormant state, and only keep Simple detection function, that is, it can only detect whether there is a target or not, and close the communication function.

Step 102: The sensor node detects the target and wakes up the nodes within the detection range. Among these nodes, they are clustered according to the principle. The clustering principle is: select the node k with the highest signal reception strength as the cluster head node, and cluster with the cluster head node in a single The nodes within the jump communication range and the cluster head nodes are clustered, the nodes in the cluster are in the wake-up state, and the rest of the nodes continue to enter the dormant state;

Step 103: Nodes in the cluster observe the target and send binary data to the cluster head node.

n＝1，2，L，M，g_n(x_t)是第n个节点接收信号强度的函数，v_n，t是噪声，

μ_v是v_n，t的均值，

是v_n，t的方差，r_n是第n个节点的位置，I_t是目标在时刻t所处的位置，||r_n-I_t||是目标和节点之间的欧几里得距离，Ψ是在目标距离为d₀时的信号能量，α是与传输介质相关的参数。The strength model of the received signal when the nodes in the cluster observe the target is:

n=1, 2, L, M, g _n (x _t ) is the function of the received signal strength of the nth node, v _{n, t} is the noise,

μ _v is the mean of v _n,t ,

is the variance of v _{n, t} , r _n is the position of the nth node, I _t is the position of the target at time t, ||r _n -I _t || is the Euclidean between the target and the node Distance, Ψ is the signal energy when the target distance is d ₀ , and α is a parameter related to the transmission medium.

Process the signal strength y _{n, t} received by the nth node locally, compare the received signal y _{n, t} with the preset threshold γ value, if it is lower than the threshold γ value, no information will be sent ; If it is higher than the threshold γ value, then send the binary information to the cluster head node.

ε_n，t是观测噪声， 为方差，β_n是与传感器种类相关的参数。The cluster head node receives the measurement from the nth node as z _n,t =β _n s _n,t +ε _n,t , where

ε _n,t is the observation noise, is the variance, and β _n is a parameter related to the type of sensor.

i＝1，2，L，N(N为粒子数)，采用常规的方法建立高斯混合模型，以高斯混合模型代替粒子和权重，使用并行粒子滤波算法估计初始状态估计和方差估计。具体如下：Step 104: At time t=0, according to the clustering principle in step 102, form an initial cluster, select a cluster head node, and sample from the prior distribution p(x ₀ ), that is, the initial state

i=1, 2, L, N (N is the number of particles), the Gaussian mixture model is established by the conventional method, the particles and weights are replaced by the Gaussian mixture model, and the initial state estimation and variance estimation are estimated by using the parallel particle filter algorithm. details as follows:

并且：i代表粒子，t代表时刻，

代表t时刻的第i个粒子，z_t为t时刻总量测，p(·)为概率密度函数，因为观测噪声ε_n，t是独立的，所以：

因此可以写成：Firstly, the particles are extracted, and the weight of the particles is

and: i stands for particle, t stands for time,

Represents the i-th particle at time t, z _t is the total measurement at time t, p(·) is the probability density function, because the observation noise ε _{n, t} is independent, so:

therefore can be written as:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>$

$<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

并且

其中Q(·)是为正态分布累积函数。Where z _{n, t} is the measurement of the nth node at time t,

and

where Q(·) is a cumulative function for a normal distribution.

Then, referring to Figure 2, the parallel algorithm is performed as follows:

s＝1，L，S：

with

是指t时刻，第s个节点上，第m_s个粒子，N_s是在s个节点上分到的粒子数。将粒子子集分给每一个节点并进行粒子更新。1) The fusion center divides all particles X _t into some particle subsets

s=1,L,S:

with

It refers to the m _s particle on the s node at time t, and N _s is the number of particles assigned to the s node. Assign a subset of particles to each node and update the particles.

本地权重

本地权重和

归一化本地权重得

本地状态估计

和本地方差将得到的

传送给融合中心；并对本地重采样。2) The fusion center calculates local importance sampling in parallel for each node s=1, L, S

local weight

local weights and

Normalized local weights get

local state estimation

and local difference will get

Send to the fusion center; and resample locally.

同时计算总体的状态估计方差，得到当前时刻的状态估计方差P_t。3) After the fusion center receives the data transmitted by the nodes, it calculates the overall state estimation and obtains the state estimation value at the current moment

At the same time, the overall state estimation variance is calculated to obtain the state estimation variance P _t at the current moment.

本地状态估计

本地方差

当前时刻状态估计值

当前时刻的状态估计方差P_t由下面式子计算：The above local weights and

local state estimation

local difference

Estimated value of current state

The state estimation variance _Pt at the current moment is calculated by the following formula:

为节点s的采样粒子和非归一化权重，那么在节点s给出的本地估计和方差为：

其中

为归一化权重，按以下公式计算：

make

is the sampling particle and non-normalized weight of node s, then the local estimate and variance given at node s are:

in

For the normalized weight, it is calculated according to the following formula:

To combine all non-normalized particles, the global estimate and variance given at the fusion center is:

其中的

为全局的归一化权重，按一下公式计算：

one of them

For the global normalized weight, it is calculated according to the following formula:

According to the several formulas given above, it can be seen that the global state estimation and variance can be calculated according to the local state estimation and variance as follows:

$P_t=\underset{s=1}{\overset{S}{\mathrm{\&Sigma;}}}{w}_t^s[P_t^s+({\hat{X}}_t^s-{\hat{X}}_t)({\hat{X}}_t^s-{\hat{X}}_t)\&prime;],$ 其中，

$P_t=\underset{\mathrm{the\; s}=1}{\overset{S}{\mathrm{\&Sigma;}}}{w}_t^{\mathrm{the\; s}}[P_t^{\mathrm{the\; s}}+({\hat{x}}_t^{\mathrm{the\; s}}-{\hat{x}}_t)({\hat{x}}_t^{\mathrm{the\; s}}-{\hat{x}}_t)\&prime;],$ in,

Step 105: At time t, group clusters according to the clustering principle, and send the state estimation value and variance value calculated by the particle filter at the previous time, as well as the particles and weights replaced by the low-dimensional Gaussian mixture model to the cluster at this time head node;

Step 106: Carry out a parallel particle filter algorithm at time t, the parallel algorithm is like step 104, and performs state estimation on the position of the target.

Step 107: add 1 to the time;

Step 108: Repeat steps 105-107 until the target leaves the coverage area.

Claims (3)

1. A wireless sensor network target tracking method based on parallel distributed particle filtering is characterized by comprising the following steps:

1) randomly broadcasting wireless sensor network nodes in a coverage area, selecting cluster head nodes according to a cluster principle, enabling the cluster nodes to observe targets, and sending binary data to the cluster head nodes;

2) forming an initial cluster at the time when t is 0, selecting cluster head nodes, firstly sampling from the prior distribution, adopting a Gaussian mixture model to replace particles and weights, and enabling all particles X to be_tIs divided into s piecesThe subset is distributed to each node and particle updating is carried out; local importance samples are then computed in parallel for each nodeLocal weight

Local weight sum

Local state estimationSum varianceWill obtainTransmitting to the fusion center and resampling locally; finally, the fusion center calculates the overall state estimation according to the transmitted data to obtain the state estimation value of the current momentAnd the state estimation variance P_t；

3) Grouping clusters at the time t according to a cluster grouping principle, and transmitting the state estimation value and the state estimation variance value calculated at the previous time, and the particles and the weight replaced by the Gaussian mixture model to the cluster head node at the time;

4) calculating a state estimation value and a state estimation variance at the time t, and performing state estimation on a target position;

5) adding 1 at the time t, and repeating the steps 3) -4) until the target leaves the coverage area.

2. The method for tracking the target of the wireless sensor network based on the parallel distributed particle filtering as claimed in claim 1Characterized in that: step 1) the intensity model of the received signal when the node in the cluster observes the target is as follows:n＝1，2，L，M，g_n(x_t) Is a function of the received signal strength of the nth node, v_n，tIs a noise, and the noise is,

μ_vis v_n，tThe average value of (a) of (b),

is v_n，tVariance of r_nIs the position of the nth node, I_tIs the position of the target at time t, | | r_n-I_tI | is the Euclidean distance between the target and the node, Ψ is the distance d at the target₀The signal energy of time, α, is a parameter related to the transmission medium.

3. The method for tracking the target of the wireless sensor network based on the parallel distributed particle filtering as claimed in claim 1, wherein: the global state estimate and variance are calculated from the local state estimate and variance, the calculation formula is:

wherein,

Images