12 September 2017 An improved silhouette for human pose estimation
Author Affiliations +
Abstract
We propose a novel method for analyzing images that exploits the natural lines of a human poses to find areas where self-occlusion could be present. Errors caused by self-occlusion cause several modern human pose estimation methods to mis-identify body parts, which reduces the performance of most action recognition algorithms. Our method is motivated by the observation that, in several cases, occlusion can be reasoned using only boundary lines of limbs. An intelligent edge detection algorithm based on the above principle could be used to augment the silhouette with information useful for pose estimation algorithms and push forward progress on occlusion handling for human action recognition. The algorithm described is applicable to computer vision scenarios involving 2D images and (appropriated flattened) 3D images.
© (2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Anthony H. Hawes, Anthony H. Hawes, Khan M. Iftekharuddin, Khan M. Iftekharuddin, } "An improved silhouette for human pose estimation", Proc. SPIE 10395, Optics and Photonics for Information Processing XI, 1039512 (12 September 2017); doi: 10.1117/12.2274449; https://doi.org/10.1117/12.2274449
PROCEEDINGS
9 PAGES


SHARE
RELATED CONTENT


Back to Top