Human Pose Tracking based on Cascaded Pose Regression
Download as PDF
DOI: 10.23977/iset.2019.026
Corresponding Author
Dengwei lv
ABSTRACT
Human pose tracking is the first-step for videos in social and scientific applications. In this paper, we propose a method for human pose tracking based on Deep Neural Networks (DNNs) using Cascaded Pose Regression (CPR) framework and contextual information. We first introduce a cascade of DNN-based regressors to obtain precision human pose. Moreover, a context-based pose tracking strategy is proposed to improve the tracking rate. We analyze the performance of the proposed method with detailed evaluation metrics and challenging dataset, and obtain comparable or better performance to the state-of-the-art.
KEYWORDS
Deep learning, pose estimation, CPR, contextual information, track