Study Of Target Tracking Techniques In Information Fusion System

As a new and developing crossed discipline, study of information fusion aims to how to combine information from multiple sensors and other associated sources such that the resulting estimation and inference is, in terms of precision improvement and uncertainty reduction, better than would be possible if any of these sources were used individually. Target tracking is key technique for low level processing of information fusion and is premise for high level fusion; it has been applied in many military and civil fields widely, and also is theme of the thesis. Maneuvering target tracking, data association and distributed track fusion, three of the most important and general problems in target tracking area, are discussed in this dissertation. The thesis is classified into five chapters and contents are outlined as follows: In chapter 1, an introduction is first given to the general principles and function model of information fusion. And then, fundamental issues about target tracking and architectures of multiple sensors target tracking system are presented. The development and status of information fusion and target tracking are reviewed. Finally, the main achievements and arrangements of the thesis are concluded.In chapter 2, maneuvering target tracking algorithms based on single model estimation techniques are discussed. At first, dynamic model of maneuvering target and tracking algorithms based on maneuver detection and maneuver identification are surveyed. And then, for jerk model of maneuvering target, its performance is proved limited through theoretic deduction, a 'current' statistic jerk model is proposed. The deterministic steady state errors with jerk model are eliminated in the new model, the advantages of 'current' statistic jerk model is also illustrated via simulation.In chapter 3, we focus on maneuvering target tracking algorithms based on multiple-models estimation techniques. Applications of multiple-model smoothing algorithms for maneuvering target tracking are studied via simulation, some important conclusions are obtained. Based on model-set sequential likelihood ratio, an enhanced AGIMM, in which model-set adaptation is implemented by jointly utilizing model posterior probability and predication probability, is proposed, simulation results indicate that improvements of both dynamic and steady state tracking performance are achieved with the enhanced algorithm.In chapter 4, works are concentrated on data association problems. Firstly, an analysis about a JPDA algorithm, which hopes to avoid track coalescence by some decision logic about association, is presented; several problems with its decision logic are pointed out and confirmed. Secondly, a recursive algorithm for tracking maneuvering targets in clutter, based on AECM algorithm, is developed. In this algorithm, model posterior probability and data association probability are approximately computed via HMM filtering respectively. Simulation indicates that the algorithm is valid. Finally, Cramér-Rao low bound of dissimilar multiple sensors filtering system with measurements origin uncertainties are derived, effects of various parameters on steady state CRLB are accessed via simulation, a mistake about these effects in foreign literature is pointed out. In chapter 5, distributed track-to-track fusion algorithms are discussed. Exact estimation covariance matrix for multiple sensors fusion system, in which HF algorithm with complete feedback is adopted and sensor number can be arbitrary, is derived. The relationship between sensor number and steady performance of HF algorithm with partial and complete feedback is evaluated for the first time and results we obtain bring the recent achievement about HF algorithm to completion. Under the same incomplete data rate, performances of BLUE algorithm and HF algorithm with partial feedback are compared for the first time.