Human fallen pose detection by using feature selection and a generative model

In this paper we are interesting in knowing which features provide useful information for detecting a fall and how the set of selected characteristics impact the accuracy of detection. For this purpose two sets of features are used. The first one describes the shape of the detected person, and the second one, the change of the shape over the time. All of features are extracted from a cloud of points of a detected person by the Kinect device. To determinate a fallen pose, a generative model is used. Two experiments are carried out to analyze the effect of using two different subset of features, one of them selected by a Genetic Algorithm and the second by Principal Component Analysis (PCA). The obtained results suggest that the success of detection of fall depends on the selected features, and the genetic algorithm is a good technique to select them, when compared with PCA.