基于三维重力加速器人体动作识别与分类

Recognition and classification of human activity by triaxial gravity accelerator

运用单个动作传感器通过机器学习算法——支持向量机(SVM),建立出色的人体日常动作识别模型.通过3个主要步骤对动作数据进行了处理,即小波转换,基于降维和K层交叉验证的主成分分析(PCA)以及自动寻优搜索获得SVM径向基核函数中的最佳参数σ和c,获得识别6种人体动作的最佳分类器.采用SVM(支持向量机)算法获得的动作分类器,在对不同动作识别时,得出的平均准确率达到94.5%.这表明基于人体动作识别的验证方法具有实用价值的,并在不久的将来会有进一步的提升.

Using a posture sensor to validate human daily activity and by machine learning algorithm--support vector machine （SVM） to build an outstanding model. The optimal parameters a and c of radial basis function（RBF） kernel SVM were obtained by searching automatically. Those kinematic data were carried out through three major steps： wavelet transform, principal component analysis （PCA）-based dimensionality reduction and k-fold cross-validation, followed by implementing a best classifier to distinguish 6 different actions. As an activity classifier, the SVM algorithm is used; and we have achieved over 94.5% of mean accuracy in detecting different actions. It is shown that the verification approach based on the recog- nition of human activity detection is valuable and will be further explored in the near future.