运动传感驱动的3D直观手势交互

3D Intuitive Gesture Interaction via Motion Sensing

为了使手势交互方式较少受到场地和光线的限制,提出利用加速度传感器作为输入设备进行手势识别的方法.对每种手势只要求用户做一次示范表演,通过添加噪声等手段来提高训练数据生成的自动化程度;将训练数据经过预处理和特征提取之后用于训练机器学习模型(隐马尔科夫模型和支持向量机).在包含70种手势的测试集上进行实验,平均识别率超过90%;并开发了幻灯片手势控制和手势拨号2个基于手势的人机交互原型系统,结果表明文中方法能够显著地提升用户在人机交互中的体验.

Motion sensing techniques are less limited in space and lighting from the point of view of human computer interaction. We employ accelerometers to study on how to effectively build natural and intuitive gesture-based user interfaces with machine learning methods. The gestures of the users are respectively recognized by hidden Markov model and support vector machine, in which the training data are semi-automatically generated by noise-adding techniques. It greatly reduces the workload of training and accordingly facilitates the development of gesture-based interaction. Experiments on a dataset of 70 gestures show that the average recognition rate is more than 90%. Two prototype interfaces, gesture-based phone-dialer and slide-presentation controller, are developed to verify the usability of our gesture-based interaction.