@inproceedings{56ac8d9d59424f2b87c9014edef66e92,
title = "Multiple camera based human motion estimation",
abstract = "We propose a human motion detection method using multipleviewpoint images. We employ a simple elliptic model and a small number of reliable image features detected in multiple-viewpoint images to estimate the pose (position and normal axis) of a human body, where feature extraction is employed based on distance transformation. The COG (center of gravity) position and its distance value are extracted in the process. These features are robust against changes in human shapes caused by hand/leg bending and produce stable pose estimation results. After a pose estimation, a {"}best view{"} is selected based on the estimation result and further processing is performed including body-side detection and gesture recognition (in a 2D image of the selected view). This viewpoint selection approach can overcome the problem of self-occlusions. We confirmed the stability of the system through experiments.",
author = "Akira Utsumi and Hiroki Mori and Jun Ohya and Masahiko Yachida",
year = "1997",
month = jan,
day = "1",
doi = "10.1007/3-540-63931-4_274",
language = "English",
isbn = "3540639314",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "655--662",
editor = "Roland Chin and Ting-Chuen Pong",
booktitle = "Computer Vision - ACCV 1998 - 3rd Asian Conference on Computer Vision, Proceedings",
note = "3rd Asian Conference on Computer Vision, ACCV 1998 ; Conference date: 08-01-1998 Through 10-01-1998",
}