The Researching On Multiple Sensors Data Fusion For Multi-target Tracking

Target tracking is derived from war times, and there have been lots of substantialaccomplishments so far. Usually there are many targets in the monitor region, and thenumber of the targets is time varying. Therefore, multi-target tracking has alwaysbecoming the hot and difficult point. While multi-target tracking needs many sensors todetect targets in multi-level aspects, and the redundancy and complementary data frommany sensors needs to use data fusion in order to associating and filtering, and thenmakes the tracks of the targets. In addition, because of the sensors character and thesmall or low altitude targets, the multi-target tracking becomes more difficult.Based on the concept of mixed form and multiple steps form, a new tandem typemixed form data fusion structure is proposed in this article firstly. This structure hasboth the distributed advantage and the centralized advantage. Based on the new formstructure and the distributed form structure, two realizable structures of the stateestimation method of the probability hypothesis density filtering are proposed.Then, according to the data fusion structure before the data fusion center, the dataassociation studied intensively is composed of the distributed track association and thecentralized data association. The data association based on fuzzy reasoning is comparedwith the joint probabilistic data association by simulating, and the results demonstratethat the former is better.Finally, based on Bayes filter frame, the Kalman filtering and the particle filteringare studied intensively. The probability hypothesis density filter based on the randomfinite set is focused. According to the nonlinear and non-gaussian system, a modifiedmarginal particle filtering-probability hypothesis density method is proposed, and thetarget loss problem of multi-target tracking is solved by using this method. According tothe condition of low detection probability of sensors, a probability hypothesis densitysmoother method is proposed, and the target loss rate and the wrong target tracking rateunder the condition of low detection probability of sensors can be reduced by using this method.