Posture and Activity Recognition and Energy Expenditure Estimation in a Wearable Platform

Abstract

The use of wearable sensors coupled with the processing power of mobile phones may be an attractive way to provide real-time feedback about physical activity and energy expenditure (EE). Here we describe the use of a shoe-based wearable sensor system (SmartShoe) with a mobile phone for real-time recognition of various postures/physical activities and the resulting EE. To deal with processing power and memory limitations of the phone, we compare use of Support Vector Machines (SVM), Multinomial Logistic Discrimination (MLD), and Multi-Layer Perceptrons (MLP) for posture and activity classification followed by activity-branched EE estimation. The algorithms were validated using data from 15 subjects who performed up to 15 different activities of daily living during a four-hour stay in a room calorimeter. MLD and MLP demonstrated activity classification accuracy virtually identical to SVM (~95%), while reducing the running time and the memory requirements by a factor of >103. Comparison of perminute EE estimation using activity-branched models resulted in accurate EE prediction (RMSE=0.78 kcal/min for SVM and MLD activity classification, 0.77 kcal/min for MLP, vs. RMSE of 0.75 kcal/min for manual annotation). These results suggest that low-power computational algorithms can be successfully used for real-time physical activity monitoring and EE prediction on a wearable platform.

Energy Expenditure in People with Diabetes Mellitus: A Review

“…Kwapisz et al (108) found similar result (91.7%) with a smartphone in detecting six activities of daily life. Several studies reported great accuracy in estimating AEE compared with IC (RMSE = 0.69–1.25 METs and MSE = 0.25 METs) and a lower error as compared to another method (MSE = 2.05 METs for the Actiheart) (109–111). Inertial sensors consisting of a three dimensions accelerometer and gyroscope and a magnetometer can be used to identify the type of activity performed like previous accelerometers (112).…”

A Comparative Review of Footwear-Based Wearable Systems

“…Figure 3a shows a picture of SmartStep insole [80] and the associated Android application for daily activity monitoring. PC post processing for activity classification using neural network [81], decision trees [74], and support vector machines (SVM) [81] PC post processing for activity classification utilizing multinomial logistic discrimination (MLD) [80] PC post processing for activity classification using decision trees [78] and linear discriminant analysis [78] Four different decision trees to classify activity from four sets of sensors (left shoe, right shoe, accelerometer, and gyroscope. Majority voting to decide the activity [79] Activities [55,82] and healthy subjects [73,81] Initial validation study on five healthy subjects [80] Validated on five healthy subjects and one subject with amputee NA Accuracy~99% 96% 98.4%~99%…”

“…In terms of data processing, as reported in [81], support vector machines (SVM), being computationally expensive, are not suitable for implementing in portable electronic devices, such as smartphones, for real-time activity classification purposes. The study in [81] also reports that activity prediction models based on multinomial logistic discrimination (MLD) is computationally less expensive in terms of required memory space and execution time, and performs equally well in terms of accuracy, as compared to SVM.…”

“…The study in [81] also reports that activity prediction models based on multinomial logistic discrimination (MLD) is computationally less expensive in terms of required memory space and execution time, and performs equally well in terms of accuracy, as compared to SVM. Binary decision trees [74,78,79], being a light-weight classifier, can possibly enable the implementation of predictive models on the sensor itself.…”

“…The SmartShoe platforms have been extensively validated in EE (in controlled laboratory environments) [81,86]. SmartShoes were compared against other accelerometry-based wearable devices and have proven to be equally or more accurate [87].…”