Research Of Multi-target Tracking And Sensor Fusion And Control Based On Random Finite Set

Multi-sensor multi-target tracking problem is one of the hot and difficult point of academe and engineering domain and has received tremendous attention for half a century.In 1994,Mahler firstly proposed a new random finite sets theory and claimed that it can be used in multi-target tracking application.Then random finite sets theory has been developed very rapidly in recent years.Based on random finite sets theory,especially labeled random finite sets theory,the paper addresses the problem of multi-target tracking and multi-sensor fusion and control.The main content comprises:1.The multi-target Bayes filter is the core of random finite sets and has been expounded,the labeled random finite set has been introduced.Also,the approximation techniques of multi-target Bayes filter,including three set filters and two labeled set filters,have been analyzed comparatively.2.For tracking highly dynamic targets,a jump Markov system multiple model solution has been presented using the generalized labeled multi-Bernoulli filter.The proposed multiple model generalized labeled multi-Bernoulli filter which is a new labeled set filter can solve multiple maneuvering targets tracking problem and achieve better performance.The simulation results demonstrate that in multi-target tracking applications,the proposed method can outperform the competing techniques in terms of tracking accuracy and robustness.3.A new method for distributed multi-target tracking with multiple sensors is presented.The proposed method is particularly designed for generalized labeled multi-Bernoulli model and able to allow sequential fusion.The proposed method can handle the label space mismatching phenomenon and achieve better fusion performance.The simulation results show that the proposed fusion method is advantageous in more challenging applications where targets move in close proximity with frequent deaths.4.Based on labeled random finite sets,the problem of multi-sensor control for multi-target tracking in the sensor network systems has been focused on.We propose two novel multi-sensor control approaches in the framework of generalized Covariance Intersection.The first joint decision making method is optimal and can achieve overall good performance,while the second independent decision making method is suboptimal as a fast realization with smaller amount of computations.Simulation in challenging situation shows that both two proposed approaches can make sensors take right actions and performances are better than random control method.