Mobile phone-based internet of things human action recognition for E-health

Human action recognition plays an important role in E-health, such as risk assessment, disease treatment, rehabilitation and so on. We proposes a mobile phone-based internet of things method for human action recognition. In the work, data are collected from a smart phone worn on the waist and transmitted to the application server on the internet. The application server program cuts these data into segments of 128 samples with 50% overlap. And each segment is embedded into a 6-dimensional pseudo phase space, then a geometric template matching algorithm is applied to classify them into different actions. Last, Bayesian principle and voting rule are combined to confuse the results of the k-nearest neighbor classifiers. Experimental results on UCI HAR datasets show that this method can obtain a significant improvement in accuracy compared with the traditional SVM methods.