Research On Particle Filter In Multi-sensor Measurement For Maneuvering Target Tracking

Target tracking, using the prior information of a target to estimate the posterior information of its state,has been widely used in many fields including automatic control, navigation, precise guidance system,artificial intelligence, information fusion and fault detection. For the simple state estimation in which onlythe linear model and Gaussian noise be considered, the traditional methods could obtain effective tracking.However, it fails to produce reasonable result for the complex state estimation, which involves nonlinear,non-Gaussian, multi-model, multi-scale and high dimension. the particle filter has been proposed toconquer this problem. Based on the technology of the multi-sensor data fusion, particle filtering methodswas studied to improve the property of the target state estimation, the main contribution of this paper are asfollows:First, for the reason that the accuracy of the interacting multiple model with particle filter is restrictedin tracking maneuvering target, a new interacting multiple model which adopts the multi-sensor sequentialparticle filter algorithm is proposed. In the novel algorithm, the interacting multiple models mechanism isused to confirm the motion pattern of target. Based on utilization of the redundancy and complementarymeasurement information from the single sensor and multi-sensor, a sequential re-sampling method is usedto optimize the distribution of particles and integrated into the interacting multiple model with particle filter.Simulation results show that the new algorithm can approximate the state of the maneuvering target welland is superior to the interacting multiple models with particle filter in the precision.Second, according to multi-sensor data fusion method and in order to overcome the low efficiency ofthe interactive multiple model particle filter in tracking maneuvering target, the multi-rate interactivemultiple model particle filter algorithm (MRIMMPF) is put forward. In the basis of the measurementinformation which the multi-sensor data fusion technology is provided, using the interactive model particlefilter as a framework, adopts the model of the multi-rate to rational use the multi-sensor measurementinformation, and reduce the computational complexity of the system. Simulation shows that MRIMMPFalgorithm effectively uses the measurement information of the multi-sensor. And further more, enhance theefficiency of the tracking system. Third, the interactive multiple model particle filter are promoted to track the reentry target which hasthe features of the strong nonlinear, multiple model and high dimension. In the unknown ballistic priorinformation, builds corresponding the ballistic coefficient model-set, and adopts interactive multiple modelto deal with the uncertainty of the system. The result of the simulation shows that this algorithm is betterthan the unscented Kalman method of the interactive multiple model in the aspects of the convergencespeed and the estimation precision in the case of tracking the unknown ballistic coefficient of the reentrytarget.