Study Of Multi-Target Tracking Algorithm Based On Random Sets

As a typical signal processing problem, target tracking have been paid much attention to by the people, and has broad application prospect. As more and more complicated application, single target tracking is hard to meet the demand, consequently, multiple target tracking has gradually become research point of domestic and overseas scholars. Traditional multiple target tracking algorithms are based on data association, their tracking performance is influenced by data association to a large extent. Differently, random sets based multi target tracking algorithm does not need data association, and effectively overcomes the drawback of traditional methods, it has become a research hotspot recently as well as the research emphases of this paper.Firstly, basic theory of tracking filtering and several classic filtering algorithms are introduced in this paper, and simulation experiments are performed to compare and analyse their performance. Then, basic multi target tracking theory is introduced, and traditional multi target tracking algorithms based on data association are introduced on basis of the theory. As the emphases of this paper, principle of random sets based multi target tracking algorithms and its classical algorithms are deeply studied. At the same time, validation and analysis are performed through simulation experiments. At last, several improved algorithms are proposed in this paper aiming at the lack of traditional probability hypothesis density filtering algorithms. First, Aiming at the drawback of the particle probability hypothesis density filter rely excessively on the distribution of measurement noise when updating particles, a new multi target tracking algorithm under unknown probabilistic distribution of measurement noise is proposed. Second, Aiming at the drawback of the gaussian particle probability hypothesis density ?lter needs repeated particle approximation and resampling in predict step and update step, a modi?ed gaussian particle probability hypothesis density ?ltering algorithm is proposed. Thirdly, Aiming at the drawback of the gaussian mixture probability hypothesis density tracker can not track multiple maneuvering targets, a gaussian mixture probability hypothesis density tracker algorithm with the ability of tracking multiple maneuvering targets is proposed. Simulations show that proposed algorithms possess better tracking performance in comparison with original ones.