Research On Key Technologies Of Multi-target Tracking Under Complex Conditions

In view of the importance of the multi-target tracking technology in the field of information sensing,the researchers have been devoted to the multi-target tracking technology for a few of decades.So far,the tracking methods for cooperative targets have been mature,and those for general non-cooperative targets are still developing,while those for typical opposed non-cooperative military targets still have many problems.Some of the problems are due to the characteristic of targets and circumstances,and the others the sensors themselves.To meet the demand of typical multi-target tracking systems,which works mainly under the condition of sensor observation or in the circumstance of complicated targets,this paper studied and explored the multi-target tracking technology deeply and systematically.The main work of the paper is as follows.Chapter 2 briefly introduced the traditional multi-target tracking methods,the RFS based multi-target tracking approaches,and the multi-target tracking evaluation methods.These introductions make the expounding of the following chapters compact.First,the traditional multi-target tracking methods were introduced: the deduction procedure of the single target Bayes filtering was given;the relationship between the single target Bayes filtering and the Kalman filtering was explained;and how the traditional multi-target tracking methods decompose the multi-target tracking issue into multiple single target tracking issues by the data association technology was introduced.Then the FISST and the multi-target Bayes filtering were specified,and the deduction procedure and the detailed iteration logic of the moment approximation of the multi-target Bayes filtering were introduced.Finally,we given the aim and the basic principle of the multi-target tracking methods,and analyzed the advantages and disadvantages of the different evaluation methods.Chapter 3 was devoted to resolving the track coalescing problem of the joint probabilistic data association(JPDA)algorithm.One method based on bias estimation and removal was proposed,and the typical attribute aided method is investigated.The bias estimation and removal method use only the target state information,and the bias estimation is based on the construction of the so called target to target association hypothesis.The simulation results revealed that this method can deal with track coalescing problem effectively.The research in this paper about the attribute aided tracking methods concerns mainly about the design of the attribute association degree and the threshold.In detail,one degree and threshold determination method based on the famous(Neyman Pearson)NP rule was proposed,which could make the performance of the association methods rather stable and robust.We believe this research is very valuable for the attribute threshold technology.Chapter 4 mainly investigated the spawning targets model integration problem for the cardinalized probability hypothesis density(CPHD)filter.We derived the iteration equations for the general CPHD filter integrating spawning targets.And the derivation is based on the FISST and the famous Faà di bruno's determinant formula.Then,we proposed one feasible iteration solution method for the high order Faà di bruno's determinant,which could make the iteration equations for the general CPHD filter be possibly applied in real project.The simulation results verified the effectiveness of the proposed filter.Chapter 5 proposed one binomial splitting Gaussian mixture unscented Kalman probability hypothesis density(BSGM-UKPHD)algorithm which is suitable for the nonlinear measuring situations.This algorithm computes and evaluates the nonlinearity degree of each predicted probability hypothesis density(PHD)Gaussian element,and once the nonlinearity degree exceeds one predetermined threshold,one binomial splitting strategy is applied and one cluster of Gaussian components with smaller nonlinearity degree could be obtained to approximate this PHD Gaussian element,therefore the update error caused by the nonlinear measuring condition can be constrained and the good performance of the GM-PHD filter is preserved.The simulation results shown that the proposed method can outperform the two typical methods dealing with the nonlinear conditions.Chapter 6 investigated the GM-PHD filter when applied in the asynchronous and heterogeneous measuring situations,and proposed one real update time GM-PHD(RUT-GM-PHD)algorithm.First the main difficulties of the multi-target tracking algorithm under the asynchronous and heterogeneous measuring conditions were analyzed,and found that the key problem is the general measuring model just cannot describe the asynchronous and heterogeneous measuring situations properly and precisely.Then we introduced one time stamp for each PHD Gaussian component and the RUT-GM-PHD algorithm was thus proposed.The rather excellent performance of the proposed method was verified by two simple asynchronous and heterogeneous measuring examples.Finally,we illustrated the several issues should be paid special attention to when arbitrary asynchronous and heterogeneous measuring conditions are applied,and pointed out the corresponding possible resolution methods.