3D Multiple Human Pose Estimation from Multi-View Images
Sara Ershadi-Nasab, Erfan Noury, Shohreh Kasaei, Esmaeil Sanaei
3D multiple human pose estimation is a challenging task because of the large variations in scale and pose of humans, fast motions, multiple persons, and arbitrary number of visible body parts due to occlusion or truncation.
Using multi-view images, some of these ambiguities can be resolved. This is due to the fact that more evidence of body parts is available from multiple views.
In this work, a novel method for 3D multiple human pose estimation using evidences from multi-view images is proposed.
The proposed method utilizes a fully connected pairwise conditional random field that has two types of pairwise terms.
The first pairwise term encodes the spatial dependencies between human body joints based on an articulated human configuration.
The second pairwise term is based on the output of a 2D deep part detector.
An approximate inference is then performed using the loopy belief propagation algorithm.
The proposed method is evaluated on the KTH Football II, Campus, and Shelf datasets.
Experimental results indicate that the proposed method achieves substantial improvements in terms of the probability of correct pose metric in comparison with existing state-of-the-art methods.
Results on the Campus Dataset
Results on the Shelf Dataset
Copyright 2017 IPL - All Rights Reserved
Developed by: Mohammad Reza Bagheri