Free context smartphone based application for motor activity levels recognition

Abstract

Despite being considered as simple everyday objects, smartphones have the most innovative sensors and electronics technology built in. These features make them powerful, nonintrusive tools for monitoring the user's physical and cognitive performance. This study aims at exploiting smartphone-based physical activity identification, implementing a classification algorithm that makes use of data extracted from in-built smartphone's accelerometer and gyroscope. Data were gathered from three subjects carrying a standard smartphone equipped with a devoted application able to acquire data from the smartphone' sensors and send them to a remote server. We implemented a specific software that uses K-Nearest Neighbours (KNN) and Support Vector Machines (SVM) classifiers to recognize the type of activity performed with 1.5 seconds granularity. We evaluated the performances of the two classifiers in the cases of 3 (low/medium/high intensity activity) and 4 (rest/walk/stairs/run) activity levels classification. The 3 levels classification showed accuracy and F-1 scores always >90% for both classifiers, whereas the 4 levels classification was not effective in distinguishing between walk and climbing stairs. A reliable classification among low, medium, and high intensity activity demonstrates to be a meaningful achievement for overall monitoring of physical activity level, giving a precise and fairly accurate estimation of type and duration of the activity.

Pattern recognition techniques for the identification of Activities of Daily Living using mobile device accelerometer

“…The SVM and KNN methods was implemented by the authors of [66], using only the mean and standard deviation as accelerometer features for the recognition of resting, walking, going up stairs, going down stairs, and running, reporting an accuracy higher than 90%.…”

Identification of activities of daily living through data fusion on motion and magnetic sensors embedded on mobile devices

“…According to [26], the Gaussian mixture model (GMM) and the time series shapelets, applied to the accelerometer and gyroscope data, allow the recognition of sitting, standing, walking, and running activities with mean and standard deviation as features, reporting an accuracy of 88.64%. The authors of [27] also used the mean and standard deviation as features for the application of KNN and SVM methods, in order to recognize walking, resting, running, going downstairs, and going upstairs with a reported accuracy higher than 90%.…”