Title: Real-time online action detection and segmentation using improved efficient linear search

Authors: Shiye Wang; Zhezhou Yu; Xiangchun Yu

Addresses: School of Computer Science and Technology, Jilin University, Changchun, China ' School of Computer Science and Technology, Jilin University, Changchun, China ' School of Computer Science and Technology, Jilin University, Changchun, China

Abstract: More and more attention has been paid to linear-time online action detection and video segmentation, due to wide application in the fields of human-computer interaction, games and surveillance. In this paper we propose a new descriptor which can be adopted for action recognition, online action detection and segmentation. In addition, we propose the improved efficient linear search (improved ELS) whose scheme is modified to solve the problem of the existence of many action classes' maximum subarray sums exceeding their thresholds. Then we evaluated our approach on MSRC-12 and MSR-Action3D datasets. The results show that our descriptor achieves the state-of-the-art results on action recognition and the performance of the improved ELS is much higher than that of the ELS.

Keywords: linear-time; skeleton data; action recognition; action detection and segmentation; moving pose descriptor; improved efficient linear search; improved ELS.

DOI: 10.1504/IJCSM.2019.10019869

International Journal of Computing Science and Mathematics, 2019 Vol.10 No.2, pp.129 - 139

Received: 13 Jun 2017
Accepted: 25 Jun 2017

Published online: 02 Apr 2019 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article