Accelerated search of human activity registered by non-invasive sensors

In this paper we present a new approach for accelerated search of human activity registered by noninvasive sensors. A standard data representation in three combined directions x, y and z for the feature vectors are used as input. A k-NN (k-Nearest Neighbors) classifier is trained with the full 3D descriptors. The aim is to recognize one of the following actions - running, walking, going upstairs and down-stairs, sitting, and standing up in a less execution time at the test stage. There a search for particular action could be implemented over a record of prolonged time, e.g. on a daily basis, for an observed person. Following the initial normalization procedures the test vectors are split into groups by components x, xy, yz and xyz which prove to be representative enough for sequential lookup of some of the actions of interest. The approach is considered perspective for application in medical assistive systems for various types of patients where medical personnel could not be involved in permanent superintendence.