A Sliding Window Based Approach With Majority Voting for Online Human Action Recognition using Spatial Temporal Graph Convolutional Neural Networks
Mejdi Dallel, Vincent Havard, Yohan Dupuis, David Baudry · 2022
Nowadays, Human Action Recognition (HAR) has become an important issue since it is widely used in video surveillance, human-robot collaboration in industry, etc. Developing such accurate and efficient algorithms remains a difficult task because of the high variability of the human shapes, postures, as well as the complexity of their movements but more importantly when using continuous/untrimmed data streams. Since HAR from Segmented/trimmed sequences has been intensively studied and developed in the recent years, Online HAR in the other hand remains a challenging task and is less developed. In this paper, we propose a Sliding Window and Majority Voting skeleton-based approach for Online HAR using Spatial Temporal Graph Convolutional Neural Networks (STGCN-SWMV). Our method is evaluated on two Online skeleton-based datasets named OAD and UOW. In comparison with existing methods, the obtained results exceed state-of-the-art algorithms and show the efficiency of the proposed approach.