ROBUST SOURCE COUNTING AND ACOUSTIC DOA ESTIMATION USING DENSITY-BASED CLUSTERING

Direction-of-Arrival (DOA) estimation for multiple simultaneously active acoustic sources without knowledge of the number of sources and the noise level remains a challenging task. A method of source counting for DOA estimation using density-based clustering is proposed. Multiple Density-based Spatial Clustering of Applications with Noise (DBSCAN) with varying noise sensitivity is applied in an evolutionary procedure to obtain weighted centroids. An autonomous DBSCAN is finally run on the weighted centroids to extract the final DOA estimates. The results using generated and estimated DOAs show that the proposed technique significantly outperforms the conventional histogram peak picking as well as the original DBSCAN and variations of Kmeans with <= 4 degrees DOA estimation accuracy and improves the source counting.