A gesture recognition algorithm for vision-based unicursal gesture interfaces

Abstract

Network communication services such as Skype or gaming applications using interactive TVs with cameras and IR-sensors have begun to appear. Vision-based recognition systems using hand gestures to control applications such as TV menus are gaining popularity in the market. It is crucial that the VGI is simple, easily used and allows users to comfortably command TVs. For simple and easy to use VGIs, we focus on vision-based unicursal gesture interfaces (VUGI) which recognizes both the shape of a unicursal figure drawn by a user and orientation indicated by the start position of the unicursal figure. In this paper, we propose Support Vector Machines (SVM) using Distance to Border (DtB) features to increase the number of recognized unicursal gestures. We evaluated the performance of the proposed algorithm in an experiment and the mean recognition rate of eight unicursal figures was 94.6% and the maximum recognition processing time was 9.56ms.