Integration Of Multi-target Tracking Filter Algorithm Based On Radar / Infrared Data

As the data fusion technology developing, to improve the performance of positive sensor and passive sensor in target tracking, radar and infrared sensor are used together. This research focused on the realization of the multi-target tracking filtering algorithm based on radar/infrared data fusion. A summary of the development and status of the algorithm system in target tracking based on radar/infrared sensor can be found in chapter 1. In chapter 2, the optimal Bayesian filter is analyzed and the computational difficulties of multi-target Bayesian filter are discussed. Then the respective collection of targets and measurements are modeled as random finite sets. To solve the computation troubles, a probability hypothesis density (PHD) filter, which propagates the posterior intensity rather than the entire posterior density of multi-target in time, has been introduced.At present, the general close-form solution to the PHD recursion has not been found. In chapter 3 the close-form solution to the multi-target PHD recursion under linear, Gaussian assumptions is discussed. Then a Gaussian mixture probability hypothesis density (GMPHD) filter is introduced and applied to the 3-D coordinate linear Gaussian multi-target tracking problem, the simulate results have been satisfactory.To realize the non-linear Gaussian multi-target PHD recursion, the GMPHD filter is extended to the non-linear case with Extend Kalman Filter (EKF) and Unscented Kalman Filter in Chapter 4. The performance of the two extended GMPHD filters were compared both in accuracy of the target number estimate and the target position estimate.In chapter 5, with the combination of the UKF-extended-GMPHD and the multi-sensor sequential filtering strategy, a sequential GMPHD filter for multi-target tracking based on radar/infrared sensor is proposed. Simulation validates the effectiveness of the filtering algorithm. Comparing to multi-target single radar system, the fusion system can get much more precise results both in target number estimate and the target position estimate.