基于混合高斯模型的非固定握持姿势手势识别

Gesture recognition with unfixed holding position based on Gaussian mixture model

针对手势识别研究中普遍要求用户以严格固定方式握持数据采集设备,致使用户体验差的问题,使用混合高斯模型(Gaussian mixture model,GMM)对非固定握持姿势的手势识别算法进行改进,以提高手势人机交互时的舒适性.首先通过GMM从加速度传感器数据中提取用户握持姿势数据,然后借助握持信号实现手势命令数据提取与坐标转换,使识别系统能够自适应不同的握持姿势.为使GMM可以同时满足手势识别应用中对稳定性和适应速度的要求,优化了GMM的学习机制,包括增加备则模态和改善优先级计算.实验结果表明,所述系统在滚转角和俯仰角+60°^-60°、偏摆角+20°^-20°范围内,握持姿势对手势识别正确率没有明显影响,实现了非固定握持姿势的手势识别,起到了提高用户体验的作用.

In most gesture recognition researches,participants are asked to hold the data collecting device in fixed position strictly,which causes poor user experience.To solve this problem,an en-hancement algorithm based on the Gaussian mixture model （GMM）is presented.It can achieve ges-ture recognition with unfixed holding position and improve the interaction comfortability.First,the holding position information is abstracted from raw acceleration data by the GMM.Then the coordi-nate transformation is conducted and the gesture operating information is separated with the holding position information.To meet the requirements for stability and recognition speed,the parameter up-dating strategy of the GMM is improved by adding backup component and optimizing the priority consideration.The experimental results show that when the roll angle and the pitching angle are be-tween -60°to ＋60°,the yaw angle is between -20°to ＋20°,the holding position has no signifi-cant impact on recognition accuracy.So the gesture recognition algorithm is improved without fixed holding position,thus achieve better user experience.