The Recognition of Human Daily Actions with Wearable Motion Sensor System

This paper develops a method for recognition of human daily actions by using wearable motion sensor system. It can recognize 13 daily actions with the data in WARD1.0 efficiently. We just extract 11 features including the means and variances of vertical acceleration data of five sensors and the mean of horizontal angular speeds of the waist sensor. Then we randomly select 80i¾?% of the samples as the training set, and the remaining samples as the test set. By removing the abnormal samples based on the confidence interval of the distance among the same type samples and using the SVM as the classifier, we present a new method for recognition of the human daily actions. Moreover, we optimize the parameters of SVM with K-CV K-fold Cross Validation method. The results of the experiments show that the proposed method can efficiently identify the 13 kinds of daily actions. The rate average recognition can approach to 98.5i¾?%.