A Simple Indoor Fall Control System for the Elderly Based on the Analysis of Object Bounding Box Parameters

Abstract

In this paper a human fall detection problem using video recordings features such as bounding box height to width ratio and speed of the bounding box movement is considered. We examine two datasets, the first of which, in addition to video recordings, also contains accelerometer readings. Support vector machine, decision tree and random forest were used as classification models. These models were built on secondary features generated with the use of the tsfresh library. The experimental results showed that the analized datasets are almost linearly separable according to some newly generated features. Thus, the use of the tsfresh library allows to apply simpler models and at the same time to provide a higher classification accuracy up to 1.0 in comparison with more complex LSTM models.