Variation of pose and gesture recognition accuracy using two kinect versions
Alina Delia Călin · 2016
The aim of this paper is to analyse the variation of the gesture recognition accuracy of several classifiers, based on input provided by two different sensors: Kinect for XBox 360 (Kinect 1) and its improved, newer version, Kinect for XBox One (Kinect 2). The accurate and meaningful recognition of body postures and gestures is a problem that depends on multiple factors, including the machine learning methods used, the characteristics of the database of poses and gestures (sample size, number of samples from each class, data noise, input sensor data) and the selected combination of gestures or poses we are interested to distinguish from each other (their duration/size and joints implied, complexity or similarity). The original contribution of this article consists of creating a database of meaningful poses using Kinect 2 and comparing the efficiency of several classifiers trained and tested on this database with previously obtained results for Kinect 1, considering sensor type, gesture complexity, as well as data size and the characteristics of the poses.