JMIR Publications

ABSTRACT

Background:

Using wearable devices to measure step count is important but step detection algorithms differ across device placements and are typically proprietary rather than open source. Moreover, due to a lack of openly available datasets, most validation studies of these algorithms have been conducted in laboratory settings, frequently with small sample sizes and without a ground-truth reference measure of step count.

Objective:

STEPCAM aims to 1) generate the largest openly available dataset of step count in a free-living environment to date; 2) evaluate step detection algorithms for wrist-, thigh-, and ankle-worn devices against camera-derived ground truth; and 3) examine differences in mean daily step count between different device placements.

Methods:

This will be a cross-sectional study, aiming to recruit 60 participants aged 18 years or older. Participants will wear devices on both wrists, dominant thigh, and dominant ankle for one full 24-hour period. During waking hours, a foot-facing video camera will also be worn to provide a ground-truth measure of step count. Model performance will be assessed using accuracy, recall, specificity, precision, F1 score, and balanced accuracy.

Results:

Recruitment began in May 2025, with twenty-four participants having completed data collection as of September 2026. The study is ongoing and expected to conclude in 2026.

Conclusions:

STEPCAM will collect a dataset compromising more than 700 hours of free-living step count data from healthy adults, which will be made openly available upon study completion. Furthermore, this study will be the first to perform a thorough evaluation of step count algorithms from multiple device placements using a ground-truth reference measure.