Detection of unseen objects in a 3-D dynamic scene (コンピュータビジョンとイメージメディア)

概要

論文の詳細を見る
Detecting an unseen object in 3-D space is one of the important task on the field of computer vision. Especially, background subtraction is the most popular technique to extract the region of an unseen object from 2-D image taken under the static condition. Though, such condition is rare for practical use; A real scene contains moving object which we previously know its existence. Thus the system must have the ability to distinguish unfamiliar object from moving objects. Although the system can successfully extract the region of an unseen object, we have to decide that the object is intruding into certain region or not. In this paper, we propose a method to extract the region of unseen object from dynamically moving background using iterative projection onto eigenspace by kernel PCA of a background image sequence. Additionally, we realize a calibration-free 3-D intrusion detection system with multiple uncalibrated cameras by integrating the extracted region of unseen object in 2-D captured image.
2008-05-01