Research On Target Tracking And Sensor Localization In Wireless Sensor Networks

Wireless Sensor Network(WSN)has great potential for development in the harsh,unattended and resource-limited environments.In recent years,WSN has drawn much attention from the military,industry and academia in various countries.Two of the most popular applications in WSN are sensor localization and target tracking.So far,many algorithms for target tracking have been proposed,which can track the target more accurately.Many scholars also put forward a lot of sensor localization algorithm.However,previous studies mainly took them as two separate tasks.That is to say,they first use the observation information between sensors to locate the sensor,and then combined with these estimated sensor position information and the target information observed by the sensor for target tracking,it may be called"First Localization Then Tracking" FLTT)method.However,the FLTT method is not the optimal solution because the sensor's observation of the target can also be used to optimize the sensors' position estimations.In this thesis,we propose a distributed variational filtering algorithm which can simultaneously locate the sensor and track the target.According to the randomness and the unpredictability of the motion of sensord and target,a serie of hierarchical evolution models are established respectively.The extended Gaussian distribution is used to describe the real-time state of each sensor and target separately.We use the state evolution model as the priori information,based on which the observed information between the sensors and the sensors and targets are fused as the likelihood.Based on the basic idea of Bayesian basic thought,the variational filtering algorithm is used to derive the posteriori probability distribution,thus achieving real-time correction of the sensor and target state estimation.Through matlab experiment concluded as follows,DVF keeping almost the same with particle filter under the condition of execution time,will be significantly improved positioning and tracking accuracy,is mainly due to the DVF will estimate system variables into the system state,the mean and variance of variables set,such a set of variables has a long tail,more able to cope with the reality in nonlinear and non-gaussian situation,such as the target trajectory mutations.In addition,through the experiment of DVF energy consumption is far less than the particle filter,this is because in the previous time in DVF only to the current moment mean and variance,and the particle filter need to pass the simulation selects the number of particle sets with weights.Next,we plan to extend DVF to 3d situation and study the actual application of DVF through field deployment.