不特定条件でも頑健な音情報のみを用いた転倒識別に関する研究

抄録

Recently, we have faced on the problem such as increasing elderly people who have occurred accidents including lonely dead and falling and so on at their houses. To deal with this problem, many researchers and companies are proposing and releasing systems monitored by sensors for keeping them safety. However, conventional systems using sensors including cameras and microphones give little privacy. Thus we proposed the falling discrimination using sound information for watching elderly people. Proposed system expects the falling discriminations using acoustic features extracted from sound. For discrimination accurately, we focused on Spectrum gravity between subjects/things and floorings. The feature characterized and modeled as Spectrum ratio of delta power as dynamic parameter. Finally, experiments of two-class discriminations using feature parameters achieved to 96.83 % with dataset of unspecific conditions.