Climbing/descending stairs detection using inertial sensors and implementing PCA and a SVM classifier

Abstract

Activity classification have been used in different fields such as energy expenditure measurement or health monitoring. Many combinations of different sensors and machine learning techniques have been proposed in order to do this kind of classification. The aim of this paper is to introduce an activity classification approach for Climbing/Descending stairs detection divided in two phases. In the first phase the signals from accelerometer and gyroscope are filtered, then implementing step detection allows us to extract the relevant features from these signals. The second phase consists of a principal component analysis (PCA) for reducing dimensionality, and a support vector machines (SVM) classifier to identify the motion. Using this methodology, an accuracy of 98.76% is achieved. The data used for classification were taken from an inertial measurement unit carried by three users in their ankles, which was provided by a database from the UCI machine learning repository.