The Target Tracking Algorithm Research Based On Particle Filtering In Wireless Sensor Network

Recent advances in the communications, embedded system, micro-electro-mechanical systems (MEMS),and sensor technology have enabled the appearance of wireless sensors, which have the ability of apperception, computation, and communication. Compared with traditional sensor, wireless sensors are small in size,low-cost, and low power. Target tracking, which aims at confirming the position, number and movement direction of target, is now popularly used in military and civil aspects, such as monitor and control, navigation and so on. With the development of wireless sensor, target tracking based on wireless sensor network has been more important.In the problems of modeling target tracking based on wireless sensor networks, this paper lists some network model of target tracking algorithms, such as the dual detection, information-driven, convey tree. This paper chooses convey tree algorithm as the target tracking algorithm for the network model, which by comparing the complexity and the energy consumption. Then we chose prediction mechanism and the localization of the reconstruction. The localization strategy can reduce the cost of reconstruction. The costs reduce more pronouncedly when the node density is more or remodeling need more frequently. However, it must consume more energy in the data transmission of localization strategy reconstruction when the node density is low. Therefore, it should be based on different characteristics of the sensor networks to choose a different method.This paper bring particle filter into the problem of data fuse between sensor nodes. Particle filter is a new filtering algorithm. It is another form of Recursive Bayesian filters. The Bayes filters idea is using the random samples to describe the Probability distribution. These random samples were named as "particles". Then at the basis of measurement, particle filter describe the probability distribution of the actual implementation by adjusting the weight of the particle size and the location of the sample and using the random sample mean the estimated value of as a system. In theory, the particle filter can be used for estimating any problem of nonlinear/non-Gaussian stochastic systems. Particle filter overcome the disadvantage of the extended Kalman filter effectively. The paper discussed some variants of filtering such as EKF, Grid-Based Method, EKF and Approximate Grid-Based Method. Aiming at the nonlinear/non-Gaussian filter problems, the paper give generic ideas of particle filter, based on the analysis of standard algorithm of sampling-importance-resampling filter, the problems of particle filter are discussed and some improvement methods are illustrated. And pseudo-code of every variants of particle filter has been given. The paper compared several variants of particle filter such as SIR,ASIR,RPF and discussed the advantages and disadvantages of them.Firstly, the prediction samples of the general particle filter do not take into account the latest measurements of the system state. That has a larger deviation between the samples of the general particle with the samples of the true posterior probability, which will affect the estimated accuracy. Secondly, the general particle filter will have a particle degeneracy phenomenon. The phenomenon will make a little number of particles have a heavy weight and the majority of the particles have a light weight when the algorithm after several iterations. That makes more calculation of waste in the calculation of the particle which does not work on update. So use a resampling algorithm. The resampling algorithm reduces the number of particles of light weights and reduces the complexity of algorithm. However, the introduction of resampling algorithm brings sample impoverishment. This paper will research from these two areas.Aim for the dynamic analysis and concluded need two models at least: one is the system model which used to describe the time evolution of the system state model. The other is measurement model which is about the status of the noise system. We assume that these models are gave the form of the probability. The probability express of state space and the receipt information of the new measurement value of the information necessary to update. The characteristics of which coincide with the Bayesian approach, so they can together with each other ideally. That provides a framework of dynamic state estimation problem for a common solution.Target tracking estimation includes two basic stages: prediction and update. System model predict the distribution of the posterior probability density function, which in a measurement interval of time. Update amendments the distribution density function by using the latest measurements.The basic idea of particle filter is to use the concept of sequence of the importance sampling approximation and discrete the probability density functions using the approximation of the random sample. In the particle filter, the probability density function describe by a series of discrete samples with particle weight. Particle filter is looking for the random sample of the group of spread in the state space and use them to approximate the probability density function. Particle filter use the sample mean to replace integral operator to obtain minimum variance estimation of the status process. The sample is referred to as "particle".This paper is to examine the particle filter algorithm. First of all, in improving the density function, we select the important density function based on a likelihood function. It is a good use of the latest observations. It is better to improve the accuracy of particle filter. Secondly, the particle filter has the question of sample impoverishment in the process of resampling. I chose the regular particle filter. The distinction between the new particle filter and the sampling importance resampling (SIR) is: SIR resample from the approximate of the discrete distribution, the new algorithm resample from the similar of the continuous distribution.From the compare of the new algorithm and extended Kalman filter, particle filter, we can see that the effect of the new algorithm in tracking is better than the original filter. Although the tracking accuracy had been improved, the particle filter as a new algorithm have yet to be further improved whether in the particle filter on the convergence of its own or in the tracking algorithm accuracy and complexity.It is worthwhile to do the further research on target tracking base on wireless sensor network because of its enormous potential.