Sound Source Localization And Tracking Based On Microphone Array In High Reverberation

In the past few decades,sound source localization(SSL)and sound source tracking(SST)technologies based on microphone arrays have attracted increased attentions.The purpose of SSL and SST includes determining the localization of the sound source and tracking its movements,as well as detecting and locating multiple sound source targets.In the SST system,useless sound is regarded as noise,and the interference caused by multipath propagations is regarded as reverberation.The anti-noise and anti-reverberation abilities of SSL and SST technology has become the most important research direction.In this thesis,two aspects of SSL and SST of a single target sound source and multiple target sound sources in the sound source localization and tracking system are studied separately.For the localization and tracking task of a single sound source,the positioning and tracking performance of the sound source is usually affected by the signal-to-noise ratio and reverberation.Due to the presence of reverberation,the localization result may deviate significantly from the true position of the sound source,resulting in a decrease in tracking performance.In order to reduce the influence of reverberation,the multiple results of the localization estimation are preserved.This thesis proposes an improved multiple hypothesis tracking(MHT)algorithm to determine the most likely trajectory and eliminate false estimates.In order to improve the tracking performance,the interacting multiple model(IMM)is integrated into the sound source tracking to deal with the complex motions of the sound sources.Finally,the experimental results of simulation and actual data show that the proposed method demonstrates an good positioning and tracking performance,especially in the case of high reverberations.In the case of localization and tracking tasks for multiple sound sources.This thesis introduces the power factor corresponding to the localization estimate and converts it into weights corresponding to the position to assist the multi-target tracking task.At the same time,in order to eliminate the false peaks caused by reverberation,combined with the search space characteristics of the positioning technology,proposes a density clustering algorithm is used for clutter suppression to eliminate isolated false peaks.Fusing the clustered position information into the multi-target tracking algorithm,and propose a new adaptive targets birth model based on the probability hypothesis density(PHD)tracking algorithm is developed.In the update iteration process of the tracking algorithm,the weights corresponding to the estimated positions update the movement status of multiple targets.From simulation experiments and measured data,the proposed tracking algorithm shows a strong tracking ability when tracking multiple target sound sources,and is robust to reverberation and noise.