Research On Object Tracking Algorithm Based On Particle Filter

The location-aware function in indoor space is of great significance to people.It is a key basic requirement for location-based products and services.It has important uses in the areas of elderly care,fire rescue,construction management,and intelligent transportation.High-precision indoor tracking and navigation is one of the main research hotspots at present.In the work of this thesis,a wireless sensor network(WSN)based on Zigbee technology was used,which had the advantages of low cost,small power consumption,close range,and large network scale.In this thesis,the indoor positioning problem was modeled as a nonlinear Bayesian filtering problem to solve the problems of noise and positioning accuracy in wireless indoor target tracking.Based on the improved particle filters(PF),the indoor target tracking problem using Received Signal Strength Indication(RSSI)ranging was studied.The advantage of particle filter is that it can be applied to any system,and there are no prerequisites for linear or Gaussian distributions,and the accuracy is independent of the model and distribution.Because of its strong adaptability and more uniform accuracy,it is an ideal solution in the actual state estimation problem.The accuracy of the algorithm is related to the selected importance function and the number of particles.Therefore,for the problem of particle depletion and instability of RSSI measurement data,this thesis mainly improved the particle distribution optimization and preprocessed RSSI data to improve the accuracy of target tracking.First,the particle distribution was optimized to improve the estimation performance of the filtering algorithm.After the Gaussian particle filter(GPF)obtained the posterior estimated particle distribution,the gravitational search algorithm(GSA)was introduced,which caused the particles to move to a high likelihood region,which could reduce the loss of effective particles.Moreover,the proposed algorithm did not require the resampling stage in standard PF,which also guaranteed the diversity of particles.Numerical simulation experiments showed that the improved GSA-GPF suppressed the divergence phenomenon that occured in standard PF.When a small number of particles were extracted,the tracking error was reduced by about 64.1% compared to PF,and it provided better accuracy than the particle swarm algorithm optimized GPF.Secondly,a cascade filter was proposed by preprocessing the measurement data.Unscented Kalman Filter(UKF)could reduce the impact of environmental noise,reduced the fluctuation of RSSI due to noise,and then inputed the filtered RSSI value into PF.The designed and implemented UKF-GPF cascade algorithm had improved positioning accuracy by about 33.72% compared to using only GPF in a 2D environment.In addition,a Zigbee-based WSN was built,and target tracking tests were performed in the established low-cost and low-power indoor positioning system to verify the effectiveness of the improved algorithm.