Research Of Multi-targets Tracking Algorithm Based On Random Set Theory

With the continuous development of computer science and information technology, targettracking technology has penetrated into many applications; especially the multi-target trackingtechnology has been widely applied to video surveillance, traffic control, and human-computerinteraction and other fields. With the further development of the theory of random sets, targettracking technology based on random set has become a new area of research for scholars athome and abroad. Compared to traditional methods of data association, it does not need tosolve the problem of the correlation between the data. But it can be in the form of thecollection, so it can avoid the multi-dimensional integral calculation, which can reduce theamount of time. As the random set theoretical, scholar Mahler proposed based on random set ofprobability density filter (PHDF) algorithm, the algorithm has laid the foundation for the futureof multi-target tracking. Multi-target tracking with the random set theory is also the focus ofthis study. The following is the research content of this article:1) First of all, the target tracking technology is based on the sequence of the targetdetection and target tracking, which can be divided into detection before tracking, detectionwith tracking and tracking before detection, how to to used in specific environmental. There isa brief analysis about three types of technology in the paper. Then the classic filter targettracking algorithms can be introduced, such as Bayes filtering, Kalman filtering and particlefiltering.2) There is focusing on the basic definition and development of a random set theory. Inaddition, definition about set of integral and set derivative concept can be introduced. With thedevelopment of the theory, the concept provides a theoretical basis for target tracking andinformation fusion. It turns out that PHDF algorithm based random set theory could arise,According to multi-target Bayesian filtering, PHDF algorithm associated with Bayes filteringand it makes some analysis. Now, the above theory based on random set of multi-targettracking technology has laid a theoretical foundation.3) Because there some limit for PHD filtering algorithms, so there are someimprovement method based PHDF, such as Particle PHDF (called P-PHDF) and Gaussianmixture PHDF (called GM-PHDF), which divided into the implementation of the particle filterapplications in nonlinear non-Gaussian PHDF (called P-PHDF). Because the newborn targets’position is uncertain and strong noise affect the clustering algorithm, the improvement ofP-PHDF algorithm is solutions to track multi-targets with changing the target presentation. Ithas better tracking performance than the P-PHDF. All the same time, GM-PHDF only used toapply under specific linear Gaussian environment, so there some limitation in the normal environment, such as non-Linear and no-Gaussian. So it can be allowed to have a widerapplication of UK with the GM-PHDF algorithm, called UK-GM-PHDF. The new methodcould have better tracking results. And last, the tracking system will be better real-time andhigher accuracy.4) When it used PHDF algorithm, which would exaggerate the impact of estimates withmissing target. Because of them, Australia scholar Vo proposed a new algorithm which isMulti-Bernoulli multi-target tracking based on the random set. It deals with the multi-targettracking through simplify multi-state space to single-target state space, and the posteriorprobability density obtained for the single-target collection. Multi-Bernoulli random setsdirectly fitting a set of collections and obtained the posterior probability of the multi-targetstate. Mixed operator of dynamic model has the advantage to get the optimal matching, soparticle filter tracking with dynamic model improved Multi-Bernoulli multi-target trackingalgorithm. Finaly, the improved algorithm will get some effective performance with themulti-target detecting and tracking in low SNR sequence of video images.