Improving Energy Expenditure Estimation through Activity Classification and Walking Speed Estimation Using a Smartwatch.

Energy expenditure (EE) estimation is an important factor in tracking personal activity and preventing chronic diseases, such as obesity and diabetes. The challenge is to provide accurate EE estimations in free-living environment through portable and unobtrusive devices. In this paper, we present an experimental study to estimate energy expenditure during sitting, standing and treadmill walking using a smartwatch. We introduce a novel methodology, which aims to improve the EE estimation by first separating sedentary (sitting and standing) and non-sedentary (walking) activities, followed by estimating the walking speeds and then calculating the energy expenditure using advanced machine learning based regression models. Ten young adults participated in the experimental trials. Our results showed that combining activity type and walking speed information with the acceleration counts substantially improved the accuracy of regression models for estimating EE. On average, the activity-based models provided 7% better EE estimation than the traditional acceleration-based models.

Combined Regression and Classification Models for Accurate Estimation of Walking Speed Using a Wrist-worn IMU.

Walking speed is an important quantity not only in fitness applications but also for Iifestyle and health monitoring purposes. With the recent advances in MEMS technology, miniature body-worn sensors have been used for ambulatory walking speed estimation using regression models. However, studies show that these models are more prone to errors in slow walking regime compared to normal and fast walking regimes. To address this issue, our study proposes a combined classification and regression walking speed estimation model. An experimental evaluation was performed on 10 healthy subjects during treadmill walking trials using a smartwatch. The experimental results show that including the classification model can improve the accuracy of walking speed estimation in the slow speed regime by about 22%. The results show that the proposed combined model has error of less than around 13% for various walking speed regimes.

Use of Average Vertical Velocity and Difference in Altitude for Improving Automatic Fall Detection from Trunk Based Inertial and Barometric Pressure Measurements.

Despite the extensive research that has been carried out on automatic fall detection using wearable sensors, falls in the elderly cannot be detected effectively yet. Although recent fall detection algorithms that evaluate the descent, impact and post impact phases of falls, often using vertical velocity, vertical acceleration and trunk angle respectively, tend to be more accurate than the algorithms that do not consider them, they still lack the desired accuracy required to be used among frail older adults. This study aims to improve the accuracy of fall detection algorithms by incorporating average vertical velocity and difference in altitude as additional parameters to the vertical velocity, vertical acceleration and trunk angle parameters. We tested the proposed algorithms on data recorded from a comprehensive set of falling experiments with 12 young participants. Participants wore waist-mounted accelerometer, gyroscope and barometric pressure sensors and simulated the most common types of falls observed in older adults, along with near-falls and activities of daily living (ADLs). Our results showed that, while the base algorithm with the three parameters provided 91.8% specificity, the addition of difference in altitude and average vertical velocity improved the specificity to 98.0% and 99.6%, respectively.

A Gaussian process regression model for walking speed estimation using a head-worn IMU.

Miniature inertial sensors mainly worn on waist, ankle and wrist have been widely used to measure walking speed of the individuals for lifestyle and/or health monitoring. Recent emergence of head-worn inertial sensors in the form of a smart eyewear (e.g. Recon Jet) or a smart ear-worn device (e.g. Sensixa e-AR) provides an opportunity to use these sensors for estimation of walking speed in real-world environment. This work studies the feasibility of using a head-worn inertial sensor for estimation of walking speed. A combination of time-domain and frequency-domain features of tri-axial acceleration norm signal were used in a Gaussian process regression model to estimate walking speed. An experimental evaluation was performed on 15 healthy subjects during free walking trials in an indoor environment. The results show that the proposed method can provide accuracies of better than around 10% for various walking speed regimes. Additionally, further evaluation of the model for long (15-minutes) outdoor walking trials reveals high correlation of the estimated walking speed values to the ones obtained from fusion of GPS with inertial sensors.

Regression Model-Based Walking Speed Estimation Using Wrist-Worn Inertial Sensor.

Walking speed is widely used to study human health status. Wearable inertial measurement units (IMU) are promising tools for the ambulatory measurement of walking speed. Among wearable inertial sensors, the ones worn on the wrist, such as a watch or band, have relatively higher potential to be easily incorporated into daily lifestyle. Using the arm swing motion in walking, this paper proposes a regression model-based method for longitudinal walking speed estimation using a wrist-worn IMU. A novel kinematic variable is proposed, which finds the wrist acceleration in the principal axis (i.e. the direction of the arm swing). This variable (called pca-acc) is obtained by applying sensor fusion on IMU data to find the orientation followed by the use of principal component analysis. An experimental evaluation was performed on 15 healthy young subjects during free walking trials. The experimental results show that the use of the proposed pca-acc variable can significantly improve the walking speed estimation accuracy when compared to the use of raw acceleration information (p<0.01). When Gaussian process regression is used, the resulting walking speed estimation accuracy and precision is about 5.9% and 4.7%, respectively.

Experimental evaluation of regression model-based walking speed estimation using lower body-mounted IMU.

This study provides a concurrent comparison of regression model-based walking speed estimation accuracy using lower body mounted inertial sensors. The comparison is based on different sets of variables, features, mounting locations and regression methods. An experimental evaluation was performed on 15 healthy subjects during free walking trials. Our results show better accuracy of Gaussian process regression compared to least square regression using Lasso. Among the variables, external acceleration tends to provide improved accuracy. By using both time-domain and frequency-domain features, waist and ankle-mounted sensors result in similar accuracies: 4.5% for the waist and 4.9% for the ankle. When using only frequency-domain features, estimation accuracy based on a waist-mounted sensor suffers more compared to the one from ankle.

A magnetometer-free indoor human localization based on loosely coupled IMU/UWB fusion.

The magnetic distortions in indoor environment affects the accuracy of yaw angle estimation using magnetometer. Thus, the accuracy of indoor localization based on inertial-magnetic sensors will be affected as well. To address this issue, this paper proposes a magnetometer-free solution for indoor human localization and yaw angle estimation. The proposed algorithm fuses a wearable inertial sensor consisting of MEMS-based accelerometer and gyroscope with a portable ultra-wideband (UWB) localization system in a cascaded two-step filter consisting of a tilt Kalman filter and a localization Kalman filter. By benchmarking against an optical motion capture system, the experimental results show that the proposed algorithm can accurately track position and velocity as well as the yaw angle without using magnetometer.

Fitness activity classification by using multiclass support vector machines on head-worn sensors.

Fitness activity classification on wearable devices can provide activity-specific information and generate more accurate performance metrics. Recently, optical head-mounted displays (OHMD) like Google Glass, Sony SmartEyeglass and Recon Jet have emerged. This paper presents a novel method to classify fitness activities using head-worn accelerometer, barometric pressure sensor and GPS, with comparisons to other common mounting locations on the body. Using multiclass SVM on head-worn sensors, we obtained an average F-score of 96.66% for classifying standing, walking, running, ascending/descending stairs and cycling. The best sensor location combinations were found to be on the ankle plus another upper body location. Using three or more sensors did not show a notable improvement over the best two-sensor combinations.

Adaptive Kalman filter for indoor localization using Bluetooth Low Energy and inertial measurement unit.

This paper proposes a novel indoor localization method using the Bluetooth Low Energy (BLE) and an inertial measurement unit (IMU). The multipath and non-line-of-sight errors from low-power wireless localization systems commonly result in outliers, affecting the positioning accuracy. We address this problem by adaptively weighting the estimates from the IMU and BLE in our proposed cascaded Kalman filter (KF). The positioning accuracy is further improved with the Rauch-Tung-Striebel smoother. The performance of the proposed algorithm is compared against that of the standard KF experimentally. The results show that the proposed algorithm can maintain high accuracy for position tracking the sensor in the presence of the outliers.