CN103152791B

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

Info

Publication number: CN103152791B
Application number: CN201310040487.1A
Authority: CN
Inventors: 谢立; 周圣贤; 宋克兰
Original assignee: Zhejiang University ZJU
Current assignee: Zhejiang University ZJU
Priority date: 2013-01-29
Filing date: 2013-01-29
Publication date: 2015-08-19
Anticipated expiration: 2033-01-29
Also published as: CN103152791A

Abstract

本发明公开了一种基于水下无线传感器网络的目标跟踪方法。该方法首先根据最强信号原则选择簇节点，然后根据单跳距离准则组成簇网络对目标进行观测，如果观测信号强度超过阈值，则发送观测数据给簇头节点。簇头节点接收到粗内节点传送的数据，采用改进重采样的粒子滤波算法对当前时刻的目标位置和方差进行估计。根据目标的运动不断地更新簇头节点，将上一簇头节点状态估计值和方差估计值传送给当前簇头节点，再由当前簇头节点采用改进的重采样粒子滤波算法估计运动目标位置，直到运动目标超出了水下无线传感器网络的跟踪范围；本发明使用改进重采样算法的粒子滤波跟踪方法估计水下目标的位置和方差，提高水下无线传感器网络的目标跟踪性能。The invention discloses a target tracking method based on an underwater wireless sensor network. In this method, cluster nodes are first selected according to the principle of the strongest signal, and then a cluster network is formed to observe the target according to the single-hop distance criterion. If the observed signal strength exceeds the threshold, the observed data is sent to the cluster head node. The cluster head node receives the data transmitted by the coarse inner node, and uses the improved resampling particle filter algorithm to estimate the target position and variance at the current moment. The cluster head node is continuously updated according to the movement of the target, and the state estimation value and variance estimation value of the previous cluster head node are transmitted to the current cluster head node, and then the current cluster head node uses the improved resampling particle filter algorithm to estimate the position of the moving target. Until the moving target exceeds the tracking range of the underwater wireless sensor network; the invention uses the particle filter tracking method of the improved resampling algorithm to estimate the position and variance of the underwater target, and improves the target tracking performance of the underwater wireless sensor network.

Description

A Target Tracking Method Based on Underwater Wireless Sensor Network

技术领域technical field

本发明涉及一种基于水下无线传感器网络的目标跟踪方法。The invention relates to a target tracking method based on an underwater wireless sensor network.

背景技术Background technique

水下无线传感器网络是指在一定的水域中部署大量的传感器节点和自主车协作监测和采集周围环境感兴趣数据的网络，传感器节点能够自组织地建立起网络并进行声通信，经过数据融合技术，指定节点将获取的数据传送到水面或者岸基的控制中心，这样就实现了水下传感器网络与陆地通信网络的融合。水下传感器通常具有低功耗、传输距离短的特点。Underwater wireless sensor network refers to a network in which a large number of sensor nodes and autonomous vehicles are deployed in a certain water area to cooperate in monitoring and collecting data of interest in the surrounding environment. Sensor nodes can self-organize to establish a network and perform acoustic communication. , the designated node transmits the acquired data to the surface or shore-based control center, thus realizing the integration of the underwater sensor network and the land communication network. Underwater sensors usually feature low power consumption and short transmission distances.

水下目标跟踪是水下传感器网络的一个重要应用。水下传感器网络具有节点分布广、数量多，节点间可以相互协作、交换数据，可扩展性强等特点，这有利于扩大目标的跟踪范围、加强目标跟踪的可靠性和实时性。Underwater target tracking is an important application of underwater sensor networks. The underwater sensor network has the characteristics of wide distribution and large number of nodes, and the nodes can cooperate with each other, exchange data, and have strong scalability, which is conducive to expanding the tracking range of targets and enhancing the reliability and real-time performance of target tracking.

水下目标跟踪多为非线性问题，粒子滤波方法在非线性非高斯问题中已经得到了广泛应用，在陆地无线传感器网络中已有粒子滤波用于目标跟踪问题，因此水下目标跟踪问题多采用粒子滤波方法。Underwater target tracking is mostly a nonlinear problem. Particle filter methods have been widely used in nonlinear and non-Gaussian problems. Particle filters have been used for target tracking problems in land wireless sensor networks, so underwater target tracking problems often use particle filter method.

水下传感器网络的目标跟踪方法根据粒子滤波工作方式的不同可以分为集中式粒子滤波跟踪和分布式粒子滤波跟踪。集中式的粒子滤波跟踪方法网络中只有一个中心节点，其余节点将跟踪目标的测量数据传送给中心节点，中心节点负责运用粒子滤波进行数据处理，估计机动目标的运动轨迹。集中式的粒子滤波跟踪方法使得整个网络不够稳定且负载不平衡，分布式的粒子滤波跟踪方法克服了这些缺点，根据粒子滤波存在形式的不同大致上分为四种。第一种方法网络中不同时刻不同数据处理节点运行粒子滤波算法，根据机动目标的预测轨迹，选择离预测位置最近的可行节点作为处理节点。处理节点随着机动目标轨迹的改变而变化，解决了集中式粒子滤波中网络脆弱和负载不平衡的问题。第二种方法为了克服查询过程中可能存在的巨大损耗，采用简单的扩充处理节点的方式，即网络中每个处理节点都同时运行相同的粒子滤波算法，测量值来自整个网络传感器采集的数据。第三种方法假设每个处理节点都有相应数量的传感器与之唯一相连，每个处理节点同时运行不同的粒子滤波算法，更新测量值采用与之相连的传感器所采集的数据。第四种方法考虑到粒子滤波粒子数越多，对机动目标跟踪轨迹的精度越高，而单个处理节点容量有限，所以将粒子集分为若干个子集分别在不同的处理节点中运行粒子滤波。The target tracking method of underwater sensor network can be divided into centralized particle filter tracking and distributed particle filter tracking according to the different working methods of particle filter. The centralized particle filter tracking method has only one central node in the network, and the other nodes transmit the measurement data of the tracking target to the central node. The central node is responsible for data processing using particle filter to estimate the trajectory of the maneuvering target. The centralized particle filter tracking method makes the entire network unstable and the load is unbalanced. The distributed particle filter tracking method overcomes these shortcomings. According to the different forms of particle filter, it can be roughly divided into four types. The first method runs the particle filter algorithm on different data processing nodes in the network at different times, and selects the feasible node closest to the predicted position as the processing node according to the predicted trajectory of the maneuvering target. The processing nodes change as the trajectory of the maneuvering target changes, which solves the problem of network fragility and load imbalance in centralized particle filtering. In order to overcome the huge loss that may exist in the query process, the second method adopts a simple method of expanding processing nodes, that is, each processing node in the network runs the same particle filter algorithm at the same time, and the measured values come from the data collected by the entire network sensors. The third method assumes that each processing node has a corresponding number of sensors uniquely connected to it, each processing node runs different particle filter algorithms at the same time, and updates the measured value using the data collected by the sensors connected to it. The fourth method considers that the more particles in the particle filter, the higher the accuracy of tracking the trajectory of the maneuvering target, and the capacity of a single processing node is limited, so the particle set is divided into several subsets to run the particle filter in different processing nodes.

发明内容Contents of the invention

本发明的目的是克服现有技术的不足，提供一种基于水下无线传感器网络的目标跟踪方法。The purpose of the present invention is to overcome the deficiencies of the prior art and provide a target tracking method based on an underwater wireless sensor network.

基于水下无线传感器网络的目标跟踪方法的步骤如下：The steps of the target tracking method based on the underwater wireless sensor network are as follows:

1）初始化水下无线传感器网络，使所有传感器节点都具有同一规格，并且都处于工作状态；1) Initialize the underwater wireless sensor network so that all sensor nodes have the same specification and are in working condition;

2）选择水下接收信号强度最大的传感器节点作为簇头节点，与簇头节点在单跳通信范围内的传感器节点和簇头节点组簇，其余传感器节点保持在休眠状态；2) Select the sensor node with the highest underwater received signal strength as the cluster head node, cluster with the sensor nodes and cluster head nodes within the single-hop communication range of the cluster head node, and keep the rest of the sensor nodes in a dormant state;

3）簇内传感器节点对目标进行观测，将接收到的信号的强度与预设门限相比较，若高于预设门限则发送数据给簇头节点，反之不发送；3) The sensor nodes in the cluster observe the target, compare the strength of the received signal with the preset threshold, and if it is higher than the preset threshold, send the data to the cluster head node, otherwise, it does not send;

4）设定初始状态估计值和初始方差估计值；4) Set the initial state estimate and initial variance estimate;

5）在k时刻根据步骤2）组簇，并将上一时刻粒子滤波估计的状态估计值和方差估计值打包传送给此k时刻的簇头节点；5) Group clusters according to step 2) at time k, and pack and send the state estimation value and variance estimation value estimated by particle filter at the previous time to the cluster head node at this time k;

6）进行k时刻的改进重采样的粒子滤波，从重要密度函数中采样N个粒子，再更新采样粒子，进行粒子改进重采样，最后输出目标的位置估计值和方差估计值；6) Carry out improved resampling particle filter at time k, sample N particles from the important density function, update the sampled particles, perform particle improved resampling, and finally output the estimated position and variance of the target;

7）k时刻自加1，根据目标的运动不断地更新簇头节点，在簇头节点更换时将上一簇头节点的信息传送给当前簇头节点；7) Increment by 1 at time k, update the cluster head node continuously according to the movement of the target, and transmit the information of the previous cluster head node to the current cluster head node when the cluster head node is replaced;

8）重复步骤5）-步骤7），直至目标脱离水下无线传感器网络的覆盖区域为止。8) Repeat step 5)-step 7) until the target leaves the coverage area of the underwater wireless sensor network.

所述的步骤6）为：从采集粒子i＝1，...,N，并计算重要性权重和归一化重要性权重，其中是建议分布，采集的N个粒子重要性权重为归一化后权重为对粒子进行重采样更新，更新方法为：根据粒子权重的大小来选择保留的粒子，原本粒子穿过概率墙即被保存的方法被舍弃，当粒子权重穿过多个概率墙时，放弃复制粒子的模式，而采用以下策略：The step 6) is: from Collect particles i=1,...,N, and calculate importance weight and normalized importance weight, where is the suggested distribution, and the importance weight of the collected N particles is After normalization, the weight is The particles are resampled and updated. The update method is: select the retained particles according to the weight of the particles. The original method of saving the particles through the probability wall is discarded. When the particle weight passes through multiple probability walls, the copying of the particles is abandoned. mode, and adopt the following strategy:

当复制个数为2n偶数时，产生2(n-1)个新粒子：When the number of copies is 2n even, 2(n-1) new particles are generated:

${{x x - - \frac{22 n no - - 33}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x - - \frac{11}{44 N N} * * Σ Σ}} {{x x}} {{x x}} {{x x + + \frac{11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x + + \frac{22 n no - - 33}{44 N N} * * Σ Σ}},,$

当复制个数为2n+1奇数时，产生2n个新粒子：When the number of copies is 2n+1 odd, 2n new particles are generated:

${{x x - - \frac{22 n no - - 11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x - - \frac{11}{44 N N} * * Σ Σ}} {{x x}} {{x x}} {{x x + + \frac{11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x + + \frac{22 n no - - 11}{44 N N} * * Σ Σ}}$

其中∑为新粒子的分散度变量。where Σ is the dispersion variable of new particles.

本发明使用改进重采样算法的粒子滤波跟踪方法估计水下目标的位置和方差，提高水下无线传感器网络的目标跟踪性能。The invention uses the particle filter tracking method of the improved resampling algorithm to estimate the position and variance of the underwater target, and improves the target tracking performance of the underwater wireless sensor network.

附图说明Description of drawings

图1是水下传感器网络跟踪运动目标示意图。Figure 1 is a schematic diagram of underwater sensor network tracking moving targets.

具体实施方式Detailed ways

基于水下无线传感器网络的目标跟踪方法包括以下步骤：The target tracking method based on underwater wireless sensor network includes the following steps:

2）选择水下接收信号强度最大的传感器节点作为簇头节点，与簇头节点在单跳通信范围内的传感器节点和簇头节点组簇，其余传感器节点保持在休眠状态；2) Select the sensor node with the highest underwater receiving signal strength as the cluster head node, cluster with the sensor node and the cluster head node within the single-hop communication range of the cluster head node, and keep the rest of the sensor nodes in a dormant state;

所述的步骤6）为：从采集粒子i＝1，...,N，并计算重要性权重和归一化重要性权重，其中是建议分布，采集的N个粒子重要性权重为归一化后权重为对粒子进行重采样更新，更新方法为：根据粒子权重的大小来选择保留的粒子，原本粒子穿过概率墙即被保存的方法被舍弃，当粒子权重穿过多个概率墙时，放弃复制粒子的模式，而采用以下策略：当复制个数为2n偶数时，产生2(n-1)个新粒子：The step 6) is: from Collect particles i=1,...,N, and calculate importance weight and normalized importance weight, where is the suggested distribution, and the importance weight of the collected N particles is After normalization, the weight is The particles are resampled and updated. The update method is: select the retained particles according to the weight of the particles. The original method of saving the particles through the probability wall is discarded. When the particle weight passes through multiple probability walls, the copying of the particles is abandoned. mode, and adopt the following strategy: when the number of copies is 2n even, generate 2(n-1) new particles:

实施例Example

步骤101：初始化水下无线传感器网络，在水下环境内随机播撒无线传感器网络节点，所有节点都具有统一的规格，如通信距离、探测距离等，所有节点都处于工作状态，保持探测功能，但是可以关闭通信功能。Step 101: Initialize the underwater wireless sensor network, randomly sow wireless sensor network nodes in the underwater environment, all nodes have uniform specifications, such as communication distance, detection distance, etc., all nodes are in working state, and maintain the detection function, but The communication function can be turned off.

步骤102：传感器节点探测到目标，唤醒在探测范围的节点。这些在探测范围的节点根据原则组簇，即选择信号接收强度最大的节点M作为簇头节点，与簇头节点M在单跳通信范围内的节点和簇头节点组成簇，其余节点继续保持休眠状态。Step 102: The sensor node detects the target, and wakes up the nodes within the detection range. These nodes in the detection range are clustered according to the principle, that is, the node M with the highest signal reception strength is selected as the cluster head node, and the nodes within the single-hop communication range of the cluster head node M and the cluster head node form a cluster, and the remaining nodes continue to sleep state.

步骤103：簇内节点对目标进行观测，对接收到的信号进行本地处理，然后发送数据给簇头节点。节点接收信号的强度模型是：Step 103: The nodes in the cluster observe the target, process the received signal locally, and then send the data to the cluster head node. The strength model of the node's received signal is:

$z z ((k k)) = = \frac{S S ((k k))}{{[[x x ((k k)) - - x x]]}^{22} + + {y the y ((k k)) - - y the y]]}^{22}} + + ϵ ϵ ((k k))$

其中S(k)为目标源级别的声压，x(k)和y(k)是目标在k时刻的二维坐标，x和y是声纳传感器的二维坐标，ε(k)是独立高斯白噪声，所有水下传感器节点的观测都是独立的。where S(k) is the sound pressure of the target source level, x(k) and y(k) are the two-dimensional coordinates of the target at time k, x and y are the two-dimensional coordinates of the sonar sensor, ε(k) is the independent Gaussian white noise, observations of all underwater sensor nodes are independent.

第i个节点接收到的信号强度在本地进行处理，然后根据准则发送数据给簇头节点，该准则是：将k时刻接收到的信号z(k)与预设门限D_threshold相比较，如果值低于门限，则不发送任何信息；如果值高于门限，则发送信息给簇头节点。因此，节点只有当z(k)高于门限D_threshold的时候才向簇头节点传送信息。簇头节点接收到来自第i个节点的量测为：The signal strength received by the i-th node is processed locally, and then the data is sent to the cluster head node according to the criterion. The criterion is: compare the signal z(k) received at time k with the preset threshold D _threshold , if the value If the value is lower than the threshold, no information will be sent; if the value is higher than the threshold, information will be sent to the cluster head node. Therefore, nodes only transmit information to the cluster head node when z(k) is higher than the threshold D _threshold . The cluster head node receives the measurement from the i-th node as:

$z_{i} (k) = \frac{S (k)}{{[x (k) - x_{i}]}^{2} + {[y (k) - y_{i}]}^{2}} * λ_{i} + ϵ_{i} (k),$ 其中 $λ_{i} = \{\begin{matrix} 1 & if & z_{i} (k) > D_{threshold} \\ 0 & if & z_{i} (k) \leq D_{threshold} \end{matrix} .$ $z_{i} (k) = \frac{S (k)}{{[x (k) - x_{i}]}^{2} + {[the y (k) - {the y}_{i}]}^{2}} * λ_{i} + ϵ_{i} (k),$ in $λ_{i} = \{\begin{matrix} 1 & if & z_{i} (k) > {D.}_{threshold} \\ 0 & if & z_{i} (k) \leq {D.}_{threshold} \end{matrix} .$

步骤104：在k=0时刻，根据步骤102中的组簇原则，组成初始簇，并选出簇头节点，设定初始状态估计值和初始方差估计值。Step 104: At time k=0, according to the clustering principle in step 102, an initial cluster is formed, a cluster head node is selected, and an initial state estimate and an initial variance estimate are set.

步骤105：在k时刻，如图1所示，根据步骤102的组簇原则组簇，并将上一时刻粒子滤波算出的状态估计值和方差估计值打包传送给此k时刻的簇头节点。Step 105: At time k, as shown in Figure 1, clusters are formed according to the clustering principle in step 102, and the state estimation value and variance estimation value calculated by the particle filter at the previous time are packaged and sent to the cluster head node at this time k.

步骤106：进行k时刻的改进重采样的粒子滤波算法，对目标的位置进行状态估计。从采集粒子i＝1，...,N，并计算重要性权重和归一化重要性权重。其中是建议分布。采集的N个粒子重要性权重为归一化后权重为 Step 106: Perform an improved resampling particle filter algorithm at time k to estimate the state of the target. from Collect particles i=1,...,N, and calculate importance weights and normalized importance weights. in is the proposed distribution. The importance weight of the collected N particles is After normalization, the weight is

对粒子进行重采样更新，更新方法为：根据粒子权重的大小来选择保留的粒子，原本粒子穿过概率墙即被保存的方法被舍弃。当粒子权重穿过多个概率墙时，放弃复制粒子的模式，而采用以下策略：The particles are resampled and updated. The update method is: select the retained particles according to the size of the particle weight, and the original method of saving the particles after passing through the probability wall is discarded. When particle weights cross multiple probability walls, the pattern of replicating particles is abandoned and the following strategy is used instead:

其中∑为新粒子的分散度变量。改进重采样的粒子滤波算法如下表所示：where Σ is the dispersion variable of new particles. The particle filter algorithm for improved resampling is shown in the table below:

步骤107：k时刻自加1。在下一时刻，当目标移动到另一个位置，按步骤102重新组簇，当选出的簇头节点与上一时刻的簇头节点不相同时，上一簇头节点将信息传送给当前簇头节点，簇头节点之间传送的信息为上一时刻目标的状态估计值和方差估计值。根据目标的运动不断地更新簇和簇头节点，当簇头节点更换时，将上一簇头节点的信息传送给当前簇头。Step 107: Increment by 1 at time k. At the next moment, when the target moves to another location, reorganize the cluster according to step 102. When the selected cluster head node is different from the cluster head node at the previous moment, the previous cluster head node will transmit the information to the current cluster head node , the information transmitted between the cluster head nodes is the estimated state value and variance estimated value of the target at the last moment. According to the movement of the target, the cluster and the cluster head node are constantly updated. When the cluster head node is replaced, the information of the previous cluster head node is transmitted to the current cluster head.

步骤108：重复步骤105-107，直至目标脱离水下无线传感器网络覆盖区域为止。Step 108: Repeat steps 105-107 until the target leaves the coverage area of the underwater wireless sensor network.

Claims

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

1) Initialize the underwater wireless sensor network so that all sensor nodes have the same specification and are in working condition;

2) Select the sensor node with the highest underwater received signal strength as the cluster head node, cluster with the sensor nodes and cluster head nodes within the single-hop communication range of the cluster head node, and keep the rest of the sensor nodes in a dormant state;

3) The sensor nodes in the cluster observe the target and compare the strength of the received signal with the preset threshold. If it is higher than the preset threshold, the data is sent to the cluster head node, otherwise it is not sent;

4) Set the initial state estimate and initial variance estimate;

5) Group clusters according to step 2) at time k, and pack and send the state estimation value and variance estimation value estimated by the particle filter at the previous time to the cluster head node at this time k;

6) Carry out improved resampling particle filter at time k, sample N particles from the important density function, update the sampled particles, perform particle improved resampling, and finally output the estimated position and variance estimated value of the target;

7) Increment by 1 at time k, update the cluster head node continuously according to the movement of the target, and transmit the information of the previous cluster head node to the current cluster head node when the cluster head node is replaced;

8) Repeat step 5)-step 7) until the target leaves the coverage area of the underwater wireless sensor network.

2. a kind of target tracking method based on underwater wireless sensor network according to claim 1, is characterized in that: described step 6) is: from Collect particles i=1,...,N, and calculate importance weight and normalized importance weight, where is the suggested distribution, and the importance weight of the collected N particles is After normalization, the weight is The particles are resampled and updated. The update method is: select the retained particles according to the weight of the particles. The original method of saving the particles through the probability wall is discarded. When the particle weight passes through multiple probability walls, the copying of the particles is abandoned. mode, and adopt the following strategy:

When the number of copies is 2n even, 2(n-1) new particles are generated:

{{x x - - \frac{22 n no - - 33}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x - - \frac{11}{44 N N} * * Σ Σ}} {{x x}} {{x x}} {{x x + + \frac{11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x + + \frac{22 n no - - 33}{44 N N} * * Σ Σ}},,

When the number of copies is 2n+1 odd, 2n new particles are generated:

{{x x - - \frac{22 n no - - 11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x - - \frac{11}{44 N N} * * Σ Σ}} {{x x}} {{x x}} {{x x + + \frac{11}{44 N N} * * Σ Σ}},, . . . . . .,, {{x x + + \frac{22 n no - - 11}{44 N N} * * Σ Σ}}

where Σ is the dispersion variable of new particles.