Recursive Context Reasoning for Human Detection and Part Identification - Robotics Institute Carnegie Mellon University

Recursive Context Reasoning for Human Detection and Part Identification

Liang Zhao and Chuck Thorpe
Workshop Paper, CVPR '00 Workshop on Human Modeling, Analysis, and Synthesis, June, 2000

Abstract

Human detection and body parts identification are important and challenging prob lems in computer vision. High performance human detection depends on reliable contour extraction, but contour extraction is an under constrained problem without the knowledge about the objects to be detected. This paper proposes a recursive context reason ing (RCR) approach to solving the above dilemma. A TRSfootnote{TRS is the abbreviat ion of translation, rotation, and scaling}-invariant probabilistic model is designed to encode the shapes of t he body parts and the context information --- the size and spatial relationships between body parts. A Bayesian framework is developed to perform human detection and part identification under partial occlusion. A contour reconstruction procedure is introduced to integrate the human model an d the identified body parts to predict the shapes and locations of the parts missed by the contour detector; the refined contours are used to reevaluate the likelihood ratio. Therefore, contour extraction, part identification, and human detection are improved iteratively. The experimental results of the RCR approach to human detection and body parts identification in cluttered scenes are very encouraging.

BibTeX

@workshop{Zhao-2000-8052,
author = {Liang Zhao and Chuck Thorpe},
title = {Recursive Context Reasoning for Human Detection and Part Identification},
booktitle = {Proceedings of CVPR '00 Workshop on Human Modeling, Analysis, and Synthesis},
year = {2000},
month = {June},
keywords = {human detection, body part identification, human shape modeling},
}