Low Cost Hand Gesture Learning and Recognition System Based on Hidden Markov Model

Focusing on recognizing some typical gestures based on 3-axis MEMS accelerometer to interact with an application of human-machine interactive game, the hand gesture problem is analyzed firstly, then theory basis of HMM is introduced. In order to obtain training gesture data for HMM, and also to provide a hardware basis for gesture recognition, a low cost data acquisition system hardware is researched and designed. The system can get the acceleration data of user´s gesture and transmit them wirelessly to a personal computer. In the hand gesture recognition approach, k-mean algorithms are applied to cluster and abstract the vector data from sensor. And then the quantized vectors are put into a hidden Markov model to learn and recognize user´s gestures. Finally the gesture recognition library is implemented in C# development environment, and is utilized in a human-machine interactive game application. The results show that the typical gesture emerging in the game can be identified in a high rate, and the user can experience more interest and interaction.