Gesture unit segmentation using support vector machines: segmenting gestures from rest positions

Abstract

Gesture analysis has been widely used for developing new methods of human-computer interaction. The advancement reached in the gesture analysis area is also motivating its application to automate tasks related to discourse analysis, such as the gesture phases segmentation task. In this paper, we present an initiative that aims at segmenting gestures, especially considering the \units" { the larger grain involved in gesture phases segmentation. Thereunto, we have captured the gestures using a Xbox KinectTMdevice, modeled the problem as a classi cation task, and applied Support Vector Machines. Moreover, aiming at taking advantage from the temporal aspects involved in the problem, we have used several types of data pre-processing in order to consider time domain and frequency domain features