摘要
提出和实现了由非精确摄像机运动下的图象序列自动建立三维环境全景模型的方法。首先通过运动滤波和运动分解获得运动稳定的图象序列，然后采用无特征提取的时空纹理方向精确估计、深度边界确定和遮挡恢复算法，建立全局自然景物的真实感三维环境模型。提出了深度分层布景表示，可用于机器人全局定位的自然路标提取和真实环境虚拟再现的图象合成。推广和结合了外极面图象和全景图象的方法，放宽了对运动的要求，即平移或／和平稳运动的条件，从而可使该方法适用于更广泛的场合。
A novel approach is presented to automatically construct the 3 D natural scene model from video sequences captured by a vibrated camera with only approximately known motion. First, stabilized image sequence is obtained by using a motion filtering technique so that smooth panoramic view image and epipolar plane images can be generated. Then direct methods are proposed for accurate estimation of spatio temporal texture orientation, localization of depth boundary and recovery of occluded regions. The two stage approach not only combines the panoramic image method with epipolar plane image analysis, but also generalizes them to handle image sequence vibrations due to the uncontrollable fluctuation of the camera. Finally a 3 D panoramic scene model called depth layered setting is constructed, which can be used in landmark selection of robot navigation and image synthesis of virtualized reality.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1998年第S1期87-91,共5页
Journal of Tsinghua University(Science and Technology)
基金
国家"八六三"高技术项目
国防预先研究课题
关键词
虚拟化现实
图象稳定
外极面图象
全景图象
分层表示
virtualized reality
image stabilization
epipolar plane image
panoramic view image
layered representation
