Detailed Human Activity Recognition using Wearable Sensor and Smartphones

Use of Human activity recognition is increasing day by day for smart home, eldercare, remote health monitoring and surveillance purpose. Serving these purposes better, needs detailed recognition of activities, viz. sitting on chair or floor, slow or brisk walk, running with load, etc. Very few works aim to distinguish between intense activities (such as walk carrying weight) from its counterpart (walk) which is essential for effective health monitoring of elder adults and patients recovering from surgery. In this work, solution has been proposed for this purpose with the help of wearable and smartphone-embedded sensors. Accordingly, the contribution of this work is to present a framework for detailing in identification for both static and dynamic activities, as well as their intense counterparts by designing an ensemble of classifiers. The ensemble is designed that applies weighted majority voting for classification of test instances. Weights of the base classifiers are determined by feeding their output performance for training dataset in a neural network. We observed that our work achieves above 94% recognition accuracy.