Automatic recognition of sound events can be valuable for efficient situation analysis of audio scenes. In this article we address the problem of detecting human activities in natural environments based solely on the acoustic modality. The primary goal is the continuous acoustic surveillance of a particular natural scene for illegal human activities (trespassing, hunting, etc.) in order to promptly alert an authorized officer for taking the appropriate measures. We constructed a novel system that is mainly characterized by its hierarchical structure as well as by its acoustic parameters. Each sound class is represented by a hidden Markov model created using descriptors from the time, frequency, and wavelet domains. The system has the ability to automatically adapt to acoustic conditions of different scenes via the feedback loop that serves unsupervised model refinement. We conducted extensive experiments for assessing the performance of the system with respect to its recognition and detection capabilities. To this end we employed confusion matrices and Detection Error Tradeoff curves while we report that high performance was achieved for both detection and recognition.

Acoustic detection of human activities in natural environments / S. Ntalampiras, I. Potamitis, N. Fakotakis. - In: AES. - ISSN 1549-4950. - 60:9(2012), pp. 686-695.

Acoustic detection of human activities in natural environments

S. Ntalampiras;
2012

Abstract

Automatic recognition of sound events can be valuable for efficient situation analysis of audio scenes. In this article we address the problem of detecting human activities in natural environments based solely on the acoustic modality. The primary goal is the continuous acoustic surveillance of a particular natural scene for illegal human activities (trespassing, hunting, etc.) in order to promptly alert an authorized officer for taking the appropriate measures. We constructed a novel system that is mainly characterized by its hierarchical structure as well as by its acoustic parameters. Each sound class is represented by a hidden Markov model created using descriptors from the time, frequency, and wavelet domains. The system has the ability to automatically adapt to acoustic conditions of different scenes via the feedback loop that serves unsupervised model refinement. We conducted extensive experiments for assessing the performance of the system with respect to its recognition and detection capabilities. To this end we employed confusion matrices and Detection Error Tradeoff curves while we report that high performance was achieved for both detection and recognition.
Engineering (all); Music
Settore INF/01 - Informatica
AES
http://www.aes.org/e-lib/browse.cfm?elib=16373
Article (author)
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

Caricamento pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/2434/615182
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 25
  • ???jsp.display-item.citation.isi??? 19
social impact