Application of K-Nearest Neighbor (KNN) Algorithm for Human Action Recognition
Pengbo Wang, Yongqiang Zhang, Wenting Jiang
Abstract
In this study, the machine learning algorithm, K-Nearest Neighbor (KNN) is introduced for human action recognition. A wearable sensor is employed to collect the acceleration signals, which correspond to six types of human actions, including walking, walking upstairs, walking downstairs, sitting, standing and laying. The signals are normalized and randomly divided into a training set and a test set. A classifier is trained using KNN and the training set. It is aimed to recognize human actions when given acceleration signals. We apply this classifier to the test set and studies the relationship between the value k in KNN and the recognition accuracy. The numerical results indicate that with an appropriate value of k, the recognition accuracy can reach 96.70%, which can meet the engineering requirements. This study not only puts forward an approach for human action recognition but also exhibits the broad prospect of machine learning methods for solving recognition problems.