With the development of mobile devices, gesture recognition based on MEMS accelerometer draws researchers much attention. Due to the restriction of recognition accuracy and efficiency, present methods are still challenging to be applied as a UI. So a simple but effective gesture recognition approach was proposed here. Firstly the ten gestures were classified into four categories according to their linguistic and operation similarity in gesture definition phase. In the recognition phase,the captured gesture was pre-classified by a three-stage classifier with the kinematic features extracted from gesture acceleration. Then the gesture was recognized according to its acceleration changing patterns. Meanwhile,with strict feature threshold restrictions to gestures,the unconscious movements were eliminated significantly. Experiment among 15 volunteers achieved an average accuracy of 95.2% and a recognition time within 0.01 s,which validates the feasibility of the proposed method in terms of accuracy and efficiency.%随着智能手机等移动电子设备的发展,基于MEMS加速度传感器的手势识别成为移动设备人机交互的研究热点.由于准确率及实时性的限制,目前的手势识别方法仍不足以推向实用.针对这一问题,提出了一种简单有效的手势识别方法:在手势定义阶段根据语义及操作的相似性将10个手势分为4个类别,通过提取反映各类手势运动学特征的加速度特征量,利用决策树分类器对手势进行预分类,然后根据各类手势的加速度变化规律识别具体的手势；同时通过严格的特征量阈值,有效地去除了无意识的误动作.该方法在15位实验者中获得了95.2％的平均准确率,识别时间小于0.01 s,对基于MEMS加速度传感器的手势识别研究具有一定参考价值.
展开▼
