Research On Box-Particle Filter Based Multiple Extended Target Tracking

Extended target tracking is a hotspot in the field of target tracking in recent years.It can both obtain the information about the target kinematic state and spatial extent,which not only fits the development trend of sensor technology,but also meets the current military and civilian practical needs.Multi-measure fusion and data association are two of the most difficult problems in multiple extended target tracking.Combining with the high computational efficiency of the box-particle filter and the advantages of random finite sets filter avoiding data association,this thesis focuses on multiple extended target tracking in complex environment.The main work is as follows:Firstly,the box-particle cardinalized probability hypothesis density filter for multiple extended target tracking is proposed to improve the precision of estimating target number.The algorithm replaces the whole multiple measurements within the partition cell by a single regular measure box-particle.At the same time,the proposed filter relaxes the Poisson assumptions of the extended target PHD filter in target numbers,and propagates not only the intensity but also the cardinality distribution.Simulation results show that,dealing with multiple extended target tracking occasions,box-particle CPHD filter can generate more accurate and stable instantaneous estimates of the target number,and admit lower detection probability or more false alarm than the box-particle PHD filter does.Secondly,the concept of the labeled box-particle is proposed and the labeled box-particle cardinalized probability hypothesis density filter for multiple extended target tracking is presented to distinguish the target and give the track.The filter assigns same labels to the box-particles belonging to the same target,and different ones to the particles originating from different target to classify undifferentiated box-particles within the state set.Label management is carried out along with filtering process.Finally,different target is distinguished according to the different label and the target track is achieved by the same label.Simulation results show that,the proposed filter has good performance to differentiate targets and label different tracks.Finally,the ET-BP-CPHD-spawn filter is proposed to track target spawning and target spontaneous birth at the same time.The intensity and cardinality distribution prediction formula of the complete CPHD filter are analyzed and the generations of the spawning and newborn box-particle are given.Moreover,zero-inflated Poisson model is used for approximating spawning and spontaneous birth processes in the ET-BP-ZIP-CPHD filter.Simulation results show that,the two filter both have good performance to track target spawning,while,ET-BP-ZIP-CPHD filtering has higher estimation accuracy and stronger stability.Simulation results also validate the reasonableness and effectiveness of the ZIP model.