Gesture recognition for interactive controllers using MEMS motion sensors

In this paper we present our work on real-time human gesture recognition for multimedia interactive controllers through the use of Microelectromechanical Systems (MEMS) 3 axes acceleration sensors. The changes of accelerations in three perpendicular directions due to different gesture motions are detected in real-time by 3-axes MEMS accelerometer embedded in a wireless micro sensing mote, which exports sensor data to a PC via Bluetooth protocol. In the data collection stage, in order to realize real-time recognition, an “auto-cut” algorithm was developed to gather the start and stop motions of an input gesture automatically. After comparing several different data processing methods, we chose Discrete Cosine Transforms (DCT) to reduce the dimension of the input gestures. Subsequently, a series of experiments were performed to analyze the influence of sensor sampling frequency and the number of dominant frequencies for various gestures, and then the best combination was selected for our recognition experiments. Finally, the Hidden Markov Model (HMM) was employed to achieve real-time gesture recognition. We have shown that the gesture recognition accuracy could reach 95.7% when 20 training samples of each gesture and 70 testing samples were used.