Mid Sweden University

With the intensification of aging population, a growing number of elderly people have to live alone due to domestic and social reasons. Falling becomes one of the most crucial factors in threatening the elderly's lives, which is always difficult to be detected as it is instantaneous and easy to be confused with other motions, such as lying down. In this paper, a new method is proposed for accurate falling detection and activities recognition. It applies hierarchical classifiers to the time series data set including eleven activities of daily living (ADLs), collected by four wearable sensors. The new method combines two machine learning algorithms, performs concrete analysis on the original outcome and then obtains several scarcely-confused groups separately. The experiment indicates that the new method improves the accuracy of classification to a larger extent, reached to more than 90%. Furthermore, the matched algorithm for applying these classifiers, called Hierarchical Classifier Algorithm (HCA), is proposed as well.