Human pose estimation from silhouettes: a consistent approach using distance level sets
Year: 2002
Type: conference-paper
Abstract: We present a novel similarity measure (likelihood) for estimating\nthree-dimensional human pose from image silhouettes in model-based vision\napplications. One of the challenges in\nsuch approaches is the construction of a model-to-image likelihood that\ntruly reflects the good configurations of the problem.\nThis is hard, commonly due to the violation of consistency principle resulting in the\nintroduction of spurious, unrelated peaks/minima that make the search\nfor model localization difficult. We introduce an entirely continuous\nformulation which enforces model estimation consistency by means of an\nattraction/explanation silhouette-based term pair. We subsequently show how the\nproposed method provides significant consolidation and improved attraction zone\naround the desired likelihood configurations and elimination of some of the spurious ones.\nFinally, we present a skeleton-based smoothing method for the image\nsilhouettes that stabilizes and accelerates the search process.
Source: TU/e Research Portal
Authors: Cristian Sminchisescu, Alexandru Telea
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2145399221