Application of adaptive GM-PHD filters to multi-target tracking

Yan CANG,Di CHEN,Xiaojun BI

DOI: https://doi.org/10.11990/jheu.201409014

2015-01-01

Abstract:The problem is that when initializing prior probability in a probability hypothesis density( PHD) filter,the position of newborn targets is indefinite,and the targeting computations must search the whole monitoring domain,noticably decreasing efficiency. This paper expands on the original PHD filter,to adaptively distinguish newborn targets from surviving targets in the prediction and updating steps. It then measures again,updating newborn target strength after each scan,thus avoiding the problem of prior probability parameter initialization. The optimal sub pattern assignment( OSPA) function was used to benchmark the performance of the algorithm,plus both simulated and tested data were used for validation. Using such measurements greatly enhances discrimination between newborn targets and surviving targets. The results show that estimation accuracy for the number of targets has increased to 97%,while the OSPA distance has decreased 50% than the original GM-PHD algorithm.