We present a benchmark dataset for evaluating physical human activity recognition methods from wrist-worn sensors, for the specific setting of basketball training, drills, and games. Basketball activities lend themselves well for measurement by wrist-worn inertial sensors, and systems that are able to detect such sport-relevant activities could be used in applications of game analysis, guided training, and personal physical activity tracking. The dataset was recorded from two teams in separate countries (USA and Germany) with a total of 24 players who wore an inertial sensor on their wrist, during both a repetitive basketball training session and a game. Particular features of this dataset include an inherent variance through cultural differences in game rules and styles as the data was recorded in two countries, as well as different sport skill levels since the participants were heterogeneous in terms of prior basketball experience. We illustrate the dataset’s features in several time-series analyses and report on a baseline classification performance study with two state-of-the-art deep learning architectures.
In the continuous quest to push the boundaries of mobile healthcare and fitness tracking, monitoring respiratory biomarkers emerges as a pivotal frontier. In this paper, we present OptiBreathe, a lightweight on-device earable system designed to decode the respiratory modulations within photoplethysmography (PPG) signals. OptiBreathe computes three clinical respiratory biomarkers towards enabling continuous respiratory health monitoring with wearable devices. In our effort to bridge respiratory research and earable computing, we collected a first-of-its-kind dataset that empowers researchers to explore in-ear PPG alongside gold-standard spirometry-based ground truth in order to measure respiration rate, breathing phases, and tidal volume. OptiBreathe employs multiple algorithms to measure each respiratory parameter, achieving a best mean absolute error (MAE) of 1.96 breaths per minute on respiratory rate. When estimating breathing phases and tidal volume, OptiBreathe attains an MAE of 0.48 seconds on inspiratory time, 0.14 on inhalation-exhalation ratio (inhalation duration divided by exhalation duration), and a best mean absolute percentage error (MAPE) of 17% on tidal volume (averaged across subjects). This work shows that the best performing algorithm depends on individuals' unique physiology, and that future research should investigate the relationship between physiological factors and algorithm performances. As we look forward, we highlight the challenges and nuances in harnessing PPG sensors for respiratory monitoring, inviting researchers to build upon our work.
The COVID-19 pandemic resulted in government restrictions that altered the lifestyle of people worldwide. Studying the impact of these restrictions on exercise behaviors will improve our understanding of the environmental factors that influence individuals’ physical activity (PA). We conducted a retrospective analysis using an stringency index of government pandemic policies developed by Oxford University and digitally-logged PA data from more than 7000 runners collected using a wearable exercise-tracking device to compare the relationship between strictness of lockdowns and exercise habits on a global scale. Additionally, the time-of-day of PA globally, and activity-levels of PA in 14 countries, are compared between the pre-pandemic year of 2019 and the first pandemic year of 2020. We found that during the pandemic year there was a major shift in the time-of-day that runners exercised, with significantly more activity counts logged during standard working hours on workdays (p < 0.001) and fewer activities during the same time frame on weekends (p < 0.001). Of the countries examined, Italy and Spain had among the most strict lockdowns and suffered the largest decreases in activity counts, whereas France experienced a minimal decrease in activity counts despite enacting a strict lockdown with certain allowances. This study suggests that there are several factors affecting PA of dedicated runners, including government policy, workplace policy, and cultural norms.
The COVID-19 pandemic resulted in government restrictions that altered the lifestyle of people worldwide. Studying the impact of these restrictions on exercise behaviors will improve our understanding of environmental factors that influence individuals’ PA. We conducted a retrospective analysis using an index of government pandemic stringency developed by Oxford and a wearable device for runners to compare strictness of lockdowns and exercise habits, using digitally-logged PA data from more than 7,000 runners on a global scale. Additionally, time-of-day of PA globally and levels of PA in 14 countries are compared between the pre-pandemic year of 2019 and the first pandemic year of 2020. We found that during the pandemic the time-of-day that people exercised experienced a major shift, with significantly more activities logged during standard working hours on workdays (p<0.001) and fewer during the same time frame on weekends (p<0.001). Of the countries examined, Italy and Spain had among the most strict lockdowns and suffered the largest decreases in activity counts, whereas France experienced a minimal decrease in activity counts despite enacting a similarly strict lockdown. This study suggests that there are several factors affecting PA, including government policy, workplace policy, and cultural norms.
We pose keystep recognition as a node classification task, and propose a flexible graph-learning framework for fine-grained keystep recognition that is able to effectively leverage long-term dependencies in egocentric videos. Our approach, termed GLEVR, consists of constructing a graph where each video clip of the egocentric video corresponds to a node. The constructed graphs are sparse and computationally efficient, outperforming existing larger models substantially. We further leverage alignment between egocentric and exocentric videos during training for improved inference on egocentric videos, as well as adding automatic captioning as an additional modality. We consider each clip of each exocentric video (if available) or video captions as additional nodes during training. We examine several strategies to define connections across these nodes. We perform extensive experiments on the Ego-Exo4D dataset and show that our proposed flexible graph-based framework notably outperforms existing methods.
We present a benchmark dataset for evaluating physical human activity recognition methods from wrist-worn sensors, for the specific setting of basketball training, drills, and games. Basketball activities lend themselves well for measurement by wrist-worn inertial sensors, and systems that are able to detect such sport-relevant activities could be used in applications toward game analysis, guided training, and personal physical activity tracking. The dataset was recorded for two teams from separate countries (USA and Germany) with a total of 24 players who wore an inertial sensor on their wrist, during both repetitive basketball training sessions and full games. Particular features of this dataset include an inherent variance through cultural differences in game rules and styles as the data was recorded in two countries, as well as different sport skill levels, since the participants were heterogeneous in terms of prior basketball experience. We illustrate the dataset's features in several time-series analyses and report on a baseline classification performance study with two state-of-the-art deep learning architectures.