Creating a Database for Objective Comparison of Gesture Recognition Systems

In the past decade, accelerometer sensors became small enough to embed in mobile devices that people are using every day. Thus, arm movements became attractive to control mobile games and game consoles. However, there is no widely-used gesture-vocabulary for controlling devices and gesture recognition is not standardized. Furthermore, there is no widely-accepted benchmark for measuring performance of the developed gesture recognition methods, and it is impossible to decide objectively, which gesture recognition system is better than another. We present how to effectively build a benchmark database for gesture recognition systems, and we developed a data collection application which shortens the measurement process. We collect data for 23 gesturewords by Nintendo Wii Remote and HTC Touch Diamond devices. We recruited 30 people to participate in our data collection task and they produced the largest gesture recognition database to date.