Reliable filter-based step counting algorithm for low-intensity aided walking

Parastoo Alinia

doi:10.7273/000004001

Regular physical activity (PA) is a modifiable risk factor in maintenance and improvement of physical fitness and cardiovascular health. Yet, more than half of the U.S. adults do not accumulate enough physical activity. One simple, yet effective, approach to monitor daily physical activity is to keep track of the number of steps taken daily. Various wearable step counters, also called activity trackers, have been developed for this purpose. As these activity trackers (e.g., Fitbits) are being utilized in clinical trials, the research community remains uncertain about reliability of the trackers, particularly in studies that involve walking aids and low-intensity activities (metabolic equivalent of task < 3). While these trackers have been tested for reliability during walking and running activities, there has been limited research on validating these trackers during low-intensity activities and walking with assistive tools. In this master thesis, we first conduct a pilot study which aims to investigate the performance of three Fitbit devices (i.e., Zip, One, and Flex) at different wearing positions (i.e., pants pocket, chest, and wrist) during walking at three different speeds including 2.5 km/h, 5 km/h, and 8 km/h performed by healthy adults during treadmill walking, walking with a shopping cart, walking with a walker, and eating. The results show that off-the-shelf activity trackers produce unreliable measurements during slow walking and walking with assistive devices (i.e., aided walking). To address this issue, we propose a reconfigurable filter-based step counting algorithm and investigate how filter parameters learned in one setting, called source, can be transfered to a new setting, called target. Our algorithm learns optimal cut-off frequency to filter the target data using knowledge from a filter-bank. The filter-bank, which includes a set of features and frequency labels, is constructed during an off-line training of the source data. We evaluate the proposed step counting algorithm on a dataset of 15 subjects performing 5 walking activities. The proposed algorithm, ParaLabel, achieves an average accuracy of 86.4%, 86.3%, and 84.5% on the data from sensor on the chest, pocket and wrist, respectively.

Reliable filter-based step counting algorithm for low-intensity aided walking

Files and links (1)

Abstract

Metrics

Details