Boissy Patrick, Choquette Stéphane, Hamel Mathieu, Noury Norbert
Research Centre on Aging, Sherbrooke Geriatric University Institute, Department of Kinesiology, FEPS, Université de Sherbrooke, Sherbrooke, Quebec, Canada.
Telemed J E Health. 2007 Dec;13(6):683-93. doi: 10.1089/tmj.2007.0007.
More than one third of community-dwelling older adults and up to 60% of nursing home residents fall each year, with 10-15% of fallers sustaining a serious injury. Reliable automated fall detection can increase confidence in people with fear of falling, promote active safe living for older adults, and reduce complications from falls. The performance of a 2-stage fall detection algorithm using impact magnitudes and changes in trunk angles derived from user-based motion sensors was evaluated under laboratory conditions. Ten healthy participants were instrumented on the front and side of the trunk with 3D accelerometers. Participants simulated 9 fall conditions and 6 common activities of daily living. Fall conditions were simulated on a protective mattress. The experimental data set comprised 750 events (45 fall events and 30 nonfall events per participant) that were classified by the fall detection algorithm as either a fall or a nonfall using inputs from 3D accelerometers. Significant differences for impacts recorded, trunk angle changes (p<0.01), and detection performances (p<0.05) were found between fall and nonfall conditions. The proposed algorithm detected fall events during simulated fall conditions with a success rate of 93% and a false-positive rate of 29% during nonfall conditions. Despite a slightly superior identification performance for the accelerometer located on the front of the trunk, no significant differences were found between the two motion sensor locations. Automated detection of fall events based on user-based motion sensing and fuzzy logic shows promising results. Additional rules and optimization of the algorithm will be needed to decrease the false-positive rate.
每年有超过三分之一的社区老年人以及高达60%的疗养院居民跌倒，其中10%至15%的跌倒者会受到重伤。可靠的自动跌倒检测可以增强那些有跌倒恐惧者的信心，促进老年人积极安全地生活，并减少跌倒带来的并发症。在实验室条件下，对一种使用基于用户的运动传感器得出的冲击强度和躯干角度变化的两阶段跌倒检测算法的性能进行了评估。10名健康参与者在躯干的前部和侧面安装了三维加速度计。参与者模拟了9种跌倒情况和6种日常常见活动。跌倒情况在保护垫上模拟。实验数据集包含750个事件（每位参与者45次跌倒事件和30次非跌倒事件），这些事件通过跌倒检测算法利用三维加速度计的输入被分类为跌倒或非跌倒。在跌倒和非跌倒情况之间，记录的冲击、躯干角度变化（p<0.01）和检测性能（p<0.05）存在显著差异。所提出的算法在模拟跌倒情况下检测跌倒事件的成功率为93%，在非跌倒情况下的误报率为29%。尽管位于躯干前部的加速度计的识别性能略优，但在两个运动传感器位置之间未发现显著差异。基于用户的运动传感和模糊逻辑的跌倒事件自动检测显示出有希望的结果。需要额外的规则和算法优化来降低误报率。
