Human Motion Recognition Using Temporal Foot-Lift Features Extracted from a Small Number of Skeleton Data Frames and Multi Classifiers

Abstract

Human motion recognition becomes an essential part of human–robot collaboration in many different applications such as robot-assisted smart factories, smart warehouse and smart transportation. However, there are still challenges in terms of spatial information and temporal information requirements. Aiming at reducing the number of frames and joint information required, temporal foot-lift features were introduced in this study. The temporal foot-lift features and five different classifiers were applied to recognize “Walking” and “Running” actions from four different human action datasets. Half of the data were trained and the rest were experimentally tested for performance evaluation. The results revealed that the proposed method can give up to 100% accuracy even using a small number of frames. Using KNN classifier and temporal foot-lift features can give the highest performance in recognition. The performance of proposed method was compared with existing methods’ performance. The skeleton joint information and temporal foot-lift features are promising features for real-time human motion action recognition.