Home security surveillance based on acoustic scenes analysis

Recent years have witnessed a growing interest in the automatic detecting abnormal and dangerous events through the analysis of audio streams. Indeed, there are kinds of events (such as scream, gaspand cry, et al.) that can be effectively detected by using audio sensors for audio scene analysis. In this paper, a novel method based on a cycle supervised learning (CSL) is proposed for acoustic scene analysis in home security monitoring. Because the abnormal audio signal is an important clue for the safety awareness, in the process of the CSL we take the input signal cyclically into the trained model and the best distinguishing features are selected in each surveillance learning. The system adapts to deal with the specific issues of security surveillance by making a joint analysis of the background scene and abnormal event. To test the proposed method in complex, realistic scenarios, we have built an available dataset of home background scene with abnormal acoustic events, which allows us to evaluate the robustness of the method with respect to five kinds of abnormal audio events occurring frequently in the older members or sick people. The experimental results have confirmed its applicability under real world conditions.