Human Activity Recognition using Inertial Data

Human activity recognition is an emerging research domain due to its applications in medicine, sport, surveillance and human-robot interaction. Usually, the classification is based either on data obtained from sensors placed on the human's body or on image or video sequences analysis. In this paper is proposed a procedure based on the k-nearest neighbors, J48 and Random forests classifiers which use data acquired from the accelerometer of a wearable device. The features are extracted using the overlapping time window technique. The recognized activities are: walking, standing and going up-down stairs. The obtained results are better than those obtained in other approaches.