Multi-source localization based on approximated kernel density estimator and spatial likelihood function in near-field reverberant environment

In order to cope with the multi-source localization in near-field reverberant environment, approximated kernel density estimator (KDE) algorithm is introduced to provide robust anti-reverberation performance and multi-stage (MS) is used to solve the spectrum aliasing of high frequency on account of wide spacing of microphone array. Then spatial likelihood function (SLF) is built to mix the pairwise KDE or KDEMS function together. Based on the above KDE, MS, SLF, two algorithms SLF-KDE, SLF-KDEMS is proposed. The feasibility of the methods is confirmed by theoretical derivation and computer simulation. The results shows that SLF-KDEMS is a localization algorithm with high robustness and recognition in near-field reverberant environment.