Combining compressive sensing with particle filter for tracking moving wideband sound sources

Recent work has demonstrated the superiority of compressive sensing (CS) for the sound source localization. Using the microphone array, the direction-of-arrival (DOA) estimation is obtained by using the wideband extensions of CS technique. For tracking moving wideband sound sources, a method combined CS with particle filter (PF) algorithm is proposed. Each source is approximated by a set of particles with different weights assigned. Based on the DOA estimation, the kernel density estimator is as the likelihood function to estimate the probability of particle position. The uniform linear array and random linear array are employed in the study. For the uniform linear array, the spatial aliasing is the main offset for DOA estimation using CS method. The result shows that the spatial aliasing is disappeared and the tracking trajectory is consistent with the true source trajectory. For the random linear array, the offset at endfire is the main offset for DOA estimation using CS method. Results show that the DOA tracking for moving wideband sound sources using the proposed method can obtain a good performance with a maximum offset no more than 3 degree and an average error no more than 1 degree.