Near-real-time implementation of an adaptive anomaly detection system for monitoring elderly people living alone

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K12757
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 90150:Medical assistive technology-related
• Research Institution: Akita Prefectural University
• Principal Investigator: Abe Koji 秋田県立大学, システム科学技術学部, 助教 (50315652)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: 異常検知システム / 音響特徴量抽出 / 自動セグメンテーション / ニューラルネットワーク / 見守りシステム

Outline of Final Research Achievements

The target application of this research is monitoring the daily lives of elderly people living alone. Therefore, we conducted research on a system that enables appropriate anomaly detection while minimizing privacy violations. First, we proposed a system that appropriately divides sound signals using acoustic features as clues so that they contain one acoustic event. For sound segments using this division system, a previous anomaly detection system based on GMM and three anomaly detection systems that apply neural networks were constructed. The performance of these proposed systems was evaluated using test sounds in which an abnormal signal was embedded in a sound signal recorded in a real space. The experimental results showed that automatic segmentation contributed to improving anomaly detection performance. In addition, a system that automated the extraction of acoustic features (based on DAGMM) showed the highest performance.

Free Research Field

音響信号処理，音響工学，情報工学

Academic Significance and Societal Importance of the Research Achievements

本研究成果は，音信号を対象とした異常検知の性能向上について一定の寄与をしたと言える．各種物理センサを用いて独居老人を対象とした見守りシステムを構築する上で，プライバシーの侵害は大きな問題を引き起こす可能性がある．音信号を用いたシステムはカメラをはじめとする視覚的なモニタリングよりもプライバシー侵害が少なく，かつ死角が発生しづらいという利点がある．そのため複数のセンサを用いた総合的な見守りシステムにおいて一番初めのトリガ的役割を果たすことができる．このような運用の場合，誤検知を恐れるよりも異常検知漏れが問題となるが，本提案システムでは，設定により再現率を1に出来ており検知漏れを完全に回避できた．