Research On Box-particle Filter Based Multitarget Tracking

Target tracking has draw much attention for the wide application in the real word. In a real tracking scenario, the target of interest more than one. The number of target is also changing with time,because of the appearance and disappearance of the target. Multi-target tracking technology is improved in recent years. Box particle filter is a generalized particle filter newly proposed in latest several years, which has the advantage of less number of particles, low computational complexity and high computational efficiency. The main contents researched on multi-target tracking based on box particle filter in this thesis are as follows.The basic knowledge of box particle filter is introduced. Box particle filter is a generalized particle filter that combined the mathematical tools interval analysis with Monte Carlo algorithm. In this method a box particle is used to instead point particle with the maximum error has already known. The box particle filter is able to deal with non-precision measurement. Compared with the traditional particle filter, the box particle filter has a better performance. It could maintain a good tracking performance with a small number of particle and computation costs. Thus less computation time is needed and the operating efficiency is improved.A novel approach of multi-target tracking called box-particle cardinalized probability hypothesis density(BP-CPHD) is proposed, which is based on the random finite sets and box particle filter. The algorithm maintains the advantages of box particle filter and the CPHD filter. Compared with the traditional SMC-CPHD method, it shows low computational complexity and high operational efficiency. Compared with the box particle probability hypothesis density(BP-PHD) algorithm, it needn't make the assumptions that the number of the target obey Poisson distribution,thus the problem of filter's sensitive to clutter and miss detection is solved. The cardinality distribution of target number is also propagated to get a more accurate estimate of the target number, thereby the tracking performance is improved.An IMM-BP-CPHD algorithm for maneuvering target tracking is proposed, which is based on the proposed BP-CPHD algorithm and the interacting multiple model. The proposed IMM-BP-CPHD algorithm inherits the advantages of the box particle filter, meanwhile, it can cope with maneuvering target tracking problem. Comparing the proposed algorithm with the IMM-SMC-CPHD filter under interval measurements, the advantage of fast running is verified by the simulation results.