Abstract:

The performance of probability hypothesis density (PHD) filter depends heavily on the priori of birth target intensity and the selection of importance sampling (IS) function when the sequential Monte Carlo method is used to implement it. To solve these problems, an improved algorithm is proposed. Firstly, a measurementdriven mechanism is introduced to classify the measurements to get the birth measurements which are used for exploring newborn targets. Secondly, the unscented information filtering is used to incorporate the current measurements information into the IS function, combined with the gate technique to choose the measurements matching with the persistent targets. The results of computer simulation indicate that the proposed algorithm outperforms similar algorithms in its ability to operate in clutter, and can initiate and estimate targets more accurately.