Research On Multi-target Tracking Methods In Radio Tomographic Network

Radio tomographic im age(RTI) is an emerging device-free lo calization technology based on wireless tom ography network. The technology uses the change of received signal s trength(RSS) caused by obstruction of the objects to locate and track the objects within the area surrounded by wireless sensor network. In general, RTI uniformly divides the area into voxels, and solves the linear m odel relating the attenuation of RSS to voxels of attenuati on field, to reconstruct im age and achieve target localization and tracking.The method of multi-target tracking(MTT) based on RTI needs to assign cluster observations obtained during clustering to di fferent targets to update the positions of the targets through Kalman filter. However, the blob corresponding to a target may be divided into several clusters in the process of clustering. T he phenomenon is called over-clustering, i.e., there will be s everal cluster observations originated from the same target. Global nearest neighbor data association(GNNDA) which only chooses one cluster to obtain the updated position of the target will result in inf ormation loss and low tracking performance when over-clustering occurs. In this paper we e mploy joint probabilistic data association(JPDA) to MMT based on RTI. JPDA calcula tes the probabilities of all the c lusters originated from the sam e target to obtain th e updated position of the targ et. The experimental results show that the trackin g performance based on JPDA is better than those based on GNNDA, which indicates that the method based on JPDA is robust to over-clustering.Despite that the MTT m ethod based on RTI is simple, it is based on the linear model between the attenuation of RSS a nd the attenuation of voxels, and m oreover the tracking performance is not high. W e employ the MTT method based on Particle Filtering(PF) which address es non-linear observation models, for example, diffraction model, exponential model to track the target. Moreover, the MTT m ethod based on R TI suffers from latency, resulting in the exis tence of a la g between theestimatedtarget number and the tr ue one for time-varying MTT. Additionally, the tracking accuracy of the traditional method should be improved. The PF-based time-varying MTT method utilizes particles with variable dimensions to estimate the target number and track the targets. The e xperimental results show that the proposed method solves the latency problem and improves the tracking accuracy.