Track Association Technology In Infrared Multi-target Tracking

With the in-depth application of infrared small target tracking in the fields of national defense,military and civil security,track correlation technology is facing challenges.Missile launches from aerial platforms will cause track splits.Missiles hitting aerial targets will cause track merging.Airborne flying objects will separate after close contact and will cause track crossings.Therefore,it has important practical significance to study the multi-target track correlation technology in the infrared background.At present,the existing track algorithm has many shortcomings.First,there is the problem of dimensional explosion;second,it is easy to correlate errors when the track overlaps;third,most of the algorithms are not suitable for special navigation such as track splitting and track merging.In recent years,random set theory has made continuous progress in the research of multi-target tracking,and the track correlation technology based on random set has also become a research hotspot.In this thesis,by using the random set theory,the multi-prediction model is constructed,and the Gaussian model is penalized,and then the trajectory constraint is introduced.Finally,a track correlation model with low prediction error and high correlation accuracy is obtained.The main research contents are as follows.(1)The track correlation method based on random set is studied,including random set basic model theory,Bayesian filtering architecture,probability hypothesis density filtering implementation.Finally,it focuses on the track correlation strategy based on Gaussian probability hypothesis density filtering.Research on traditional track correlation methods,including motion model construction,threshold judgment,measurement division,similarity measurement methods between prediction targets and measurements,to the formation of the final correlation pair.(2)A multi-model prediction and punishment GM-PHD marked track correlation model is proposed.By establishing multiple prediction and updating Gaussian models,the accuracy of the target prediction value is improved.After that,The non-maximum constraint is applied to the Gaussian weight matrix to enhance the anti-interference ability of the model.Later,the feature compensation strategy is used to compensate and correct the track,which improves the accuracy of the track correlation.Finally,five types of trajectory correlation experiments are designed to explore the correlation accuracy of this model.(3)A joint track association framework of multi-model prediction and punishment GM-PHD and similarity measurement is proposed.By introducing the idea of pre-filtering,the first measurement uses the multi-model prediction and punishment GM-PHD method to filter out the clutter.After that,modules such as confirmation matrix reconstruction,confirmation matrix splitting feasible events,correlation probability calculation,gray correlation analysis and other modules are introduced to complete the track correlation work.(4)Clutter experiments were designed to explore the correlation performance of the proposed model in different density clutter environments.At the same time,the extended target experiment is carried out to explore the correlation of the algorithm in this thesis to non-small targets in the infrared background.