Arm gesture recognition using continuous DP for user-defined gestures

Abstract

The purpose of our study is to construct a system where a user can register his/her own arm gestures as templates and entered gestures are recognized precisely on a real-time basis. In order to realize them, we propose the following method: (1) 3D positions of both user's wrists are captured by using a Kinect sensor. (2) Sequences of motion vectors of both wrists are detected from the time series 3D position data. (3) The entered gesture is recognized based on the similarity between the entered gesture and each template. The similarity can be calculated by applying continuous DP matching to the sequences of motion vectors of them. The templates are gestures which are recorded by the user before the recognition process. In our experiments, good results were obtained. © Springer-Verlag Berlin Heidelberg 2013.