Elsevier

Journal of Biomechanics

Volume 43, Issue 15, 16 November 2010, Pages 3051-3057
Journal of Biomechanics

Short communication
Evaluation of waist-mounted tri-axial accelerometer based fall-detection algorithms during scripted and continuous unscripted activities

https://doi.org/10.1016/j.jbiomech.2010.07.005Get rights and content

Abstract

It is estimated that by 2050 more than one in five people will be aged 65 or over. In this age group, falls are one of the most serious life-threatening events that can occur. Their automatic detection would help reduce the time of arrival of medical attention, thus reducing the mortality rate and in turn promoting independent living.

This study evaluated a variety of existing and novel fall-detection algorithms for a waist-mounted accelerometer based system. In total, 21 algorithms of varying degrees of complexity were tested against a comprehensive data-set recorded from 10 young healthy volunteers performing 240 falls and 120 activities of daily living (ADL) and 10 elderly healthy volunteers performing 240 scripted ADL and 52.4 waking hours of continuous unscripted normal ADL.

Results show that using an algorithm that employs thresholds in velocity, impact and posture (velocity+impact+posture) achieves 100% specificity and sensitivity with a false-positive rate of less than 1 false-positive (0.6 false-positives) per day of waking hours. This algorithm is the most suitable method of fall-detection, when tested using continuous unscripted activities performed by elderly healthy volunteers, which is the target environment for a fall-detection device.

Introduction

By 2050, the proportion of the world’s population aged 65 or older is set to double to more than 1 in 5, with those aged 80 or older set to almost treble (UN, 2009). Falls and related injuries are not only life threatening to older people (Noury et al., 2008b), but also herald an inability to live independently (Gurley et al., 1996). Conversely, the automatic detection of falls facilitates the provision of early medical attention, reduces the consequences of prolonged lying following a fall (Lord et al., 2001) and promotes an independent lifestyle (Brownsell et al., 2000).

In recent years the number of proposed fall-detection systems developed has increased dramatically (Noury et al., 2008b). The waist is a popular location for fall-detection systems (Noury et al., 2008a), as it provides reliable indications of full-body movement, in addition to its ease of acceptance by allowing attachment to an existing waist band (Mathie et al., 2004).

Recently, Kangas et al. (2009) evaluated a set of fall-detection algorithms (Kangas et al., 2008) on data recorded from 20 middle-aged volunteers (40–65 years old) performing 6 different falls (240 falls in total) and 4 scripted activities of daily living (ADL), these same ADL were also recorded from 21 adults (aged 58–98 years) from a residential care unit (164 ADL in total). They showed that thresholding of impact and posture can achieve 97.5% sensitivity and 100% specificity.

Chao et al. (2009) recorded 7 male subjects (25±1.5 years) performing 8 different fall types and 17 functional ADL (total of 56 falls and 119 ADL). Using a combination algorithm of acceleration cross-product and post-fall posture, a sensitivity of 100% and >98% specificity were obtained.

The aim of this study is to evaluate the performance of 21, novel and existing, fall-detection algorithms of varying complexity on a comprehensive data-set containing simulated falls, normal ADL and including; continuous unscripted ADL performed by both urban and rural based elderly volunteers over extended time periods, which has not been previously performed. It is envisaged, such an evaluation will uncover a more appropriate fall-detection algorithm for the detection of falls in the elderly, using a waist-worn device.

Section snippets

Materials and methods

Waist tri-axial accelerometer readings were recorded (right anterior iliac crest of the pelvis) during simulated falls by young volunteers and ADL performed by young and elderly volunteers, using a custom designed sensor (van de Ven et al., 2008), shown in Fig. 1.

The accelerometer signals were sampled at 200 Hz at 12 bit resolution; each signal was low-pass filtered using an onboard first-order low-pass Butterworth analogue filter at a cut-off frequency of 100 Hz. The University of Limerick

Results

By examining the recorded values for signals RSS, RSSD (defined below), Vve, tFE and tRE (defined in Fig. 2), thresholds that ensure 100% sensitivity were obtained from Fig. 4, Fig. 5 and are displayed in Table 1. Thus all falls are detected correctly.

The minimum peak values for all the falls performed were used to determine the UFT, LFT, UFTD and VT thresholds for the RSS, RSSD and the Vve profiles respectively. The dynamic Root-Sum-of-Squares, RSSD was calculated using the formula by Kangas

Discussion and conclusion

This study evaluated the performance of different combinations of novel and existing fall-detection algorithms of varying complexity on a comprehensive data-set of falls and ADL, including scripted and continuous unscripted activities from urban and rural dwelling elderly volunteers. Previous studies (Kangas et al., 2008, Kangas et al., 2009) found reliable fall-detection can be achieved using simple algorithms with IMPACT+POSTURE (z-axis), which agrees with our findings. However, the most

Conflict of interest

The authors have no commercial stake in the outcomes of this research and no patent has been filed in relation to this work.

Acknowledgments

We are grateful to all volunteers who gave their time to the CAALYX project 〈www.caalyx.eu〉 and the eCAALYX project 〈www.ecaalyx.org〉.

References (17)

There are more references available in the full text version of this article.

Cited by (0)

View full text