Abstract

Human activity recognition (HAR) is important for many applications to help healthcare and support systems due to fast increase of senior population worldwide. This paper describes a human activity recognition framework based on feature selection techniques from a waist single accelerometer. The objective is to identify the most important features to recognize static and dynamic human activities based on module acceleration, since a public database. A set of time and frequency features are getting from the module, so to analyze the impact of the features on the performance of the recognition system, a ReliefF algorithm is applied. Finally, a multiclass classification model is implemented thought Support Vector Machine (SVM). Experimental results indicate that the accuracy of the propose model is over of 85%, this percentage is like other works in which use each axes accelerometer. The advantage of this work is the use of the module value that allow identify the activity independently of the sensor position, also it reduces the computer resources.