Department of Electronics and Radio Engineering, Kyung Hee University, Yongin 446-701, Korea.
Sensors (Basel). 2012 Sep 27;12(10):13185-211. doi: 10.3390/s121013185.
In this paper, we propose a non-parametric clustering method to recognize the number of human motions using features which are obtained from a single microelectromechanical system (MEMS) accelerometer. Since the number of human motions under consideration is not known a priori and because of the unsupervised nature of the proposed technique, there is no need to collect training data for the human motions. The infinite Gaussian mixture model (IGMM) and collapsed Gibbs sampler are adopted to cluster the human motions using extracted features. From the experimental results, we show that the unanticipated human motions are detected and recognized with significant accuracy, as compared with the parametric Fuzzy C-Mean (FCM) technique, the unsupervised K-means algorithm, and the non-parametric mean-shift method.
在本文中，我们提出了一种非参数聚类方法，使用从单个微机电系统（MEMS）加速度计获得的特征来识别人体运动的数量。由于所考虑的人体运动的数量是未知的，并且由于所提出的技术是无监督的，因此不需要为人体运动收集训练数据。采用无限高斯混合模型（IGMM）和崩溃吉布斯抽样器来使用提取的特征对人体运动进行聚类。从实验结果可以看出，与参数模糊 C-均值（FCM）技术、无监督 K-均值算法和非参数均值移动方法相比，该方法可以以较高的准确性检测和识别意外的人体运动。
