期刊
DIGITAL SIGNAL PROCESSING
卷 106, 期 -, 页码 -出版社
ACADEMIC PRESS INC ELSEVIER SCIENCE
DOI: 10.1016/j.dsp.2020.102853
关键词
Acoustic source tracking; Measurement association; Robust least squares; Microphone selection; Particle filtering
资金
- Basic Scientific Research project [JCKY2017207B042]
- National Natural Science Foundation of China [61673313, 61673317]
- Science and Technology on Underwater Test and Control Laboratory [6142407190508]
In this work, we address the problem of tracking multiple acoustic sources by using an array of microphones in reverberant and noisy environments. Generally, the time difference of arrival (TDOA) measurements are obtained by the peak extraction of the Generalized Cross Correlation (GCC) function, and then used to update source states in the framework of target tracking. In adverse environments, due to the possible false alarms, the missed detections and the poor observability resulting from the unfavorable microphone-source geometry, some microphone pairs may fail to provide efficient measurements. To improve the quality of the measurements and the performance of tracking, two adaptive methods are developed to associate TDOAs across microphone pairs to obtain the position estimates which serve as the pseudo-measurements for the following filtering step. One method is based on the robust least squares, which can associate the TDOAs by assigning TDOAs with different weights adaptively and produce better position estimates. Another method is derived from the idea of microphone selection, which aims to select a subset of microphones rather than using all the microphones. Besides, we develop the multiple targets tracking method and utilize the filter states to predict the source-nearby microphone pairs to avoid the brute-force searching. Experimental results demonstrate that, by introducing the adaptive fusion of TDOAs, performance improvements can be made in both the state and the number estimation. (C) 2020 Elsevier Inc. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据