Recognition human walking and running actions using temporal foot-lift features

Abstract

The recognition of human walking and running actions becomes essential part of many different practical applications such as smart video-surveillance, patient and elderly people monitoring, health care as well as human-robot interaction. However, the requirements of a large spatial information and a large number of frames for each recognition phase are still open challenges. Aiming at reducing the number frames and joint information required, temporal foot-lift features were introduced in this study. The temporal foot-lift features and weighted KNN classifier were used to recognize “Walkin and“Running”actions from four different human action datasets. Half of the datasets were trained and the other half of datasets were experimentally tested for performance evaluation. The experimental results were presented and explained with justifications. An overall recognition accuracy of 88.6% was achieved using 5 frames and it was 90.7% when using 7 frames. The performance of proposed method was compared with the performances of existing methods. Skeleton joint information and temporal foot-lift features are promising features for real-time human moving action recognition.