摘要
深度图像的研究是当前计算机视觉的研究热点。从图像中获取深度信息有2种方法:1)利用深度感应器,该方法的缺点是成本高;2)基于一个场景的多幅图像或图像序列,通过求取视差,获得深度值,该方法的缺点是需要摄像机参数,专业知识要求较高。针对上述情况,提出了一种简单有效的从单摄像头捕获的人体图像中估计出人体深度信息的方法,利用深度摄像机建立人体的"表观深度"图像对,然后对单摄像头获取的彩色图像进行人体表观特征提取,根据该表观特征检索图像对数据库,并对获得的人体深度进行估计和优化。最后,在厦门大学的深度数据库上,验证了该方法的有效性。
Currently, the research on depth imaging is one of the hotspots concerning computer vision. There are two methods for acquiring depth information from images： 1 ） The utilization of depth sensors, with the disadvantage of this method being its considerable expense. 2） The utilization of multiple images or a sequence of images for the same scene by calculating the optical parallax for getting depth information, with the disadvantages of this method including the requirement of camera parameters and the need for a large amount of professional knowledge. In re- sponse to the circumstances mentioned above, this paper proposes a simple and efficient method that estimates hu- man depth information from images captured by a single camera. The basic ideas of this method include establishing many pairs of human ＇appearance depth＇ images by use of a depth camera, extracting human appearance features from colorful images captured by a monocular camera and then searching the image pairs database according to the appearance features, and estimating and optimizing human depth information obtained from the database of the pairs of images. Finally, simulation experimental results in the Xiamen University depth database established by ourselves were found to validate the effectiveness of the proposed method.
出处
《智能系统学报》
CSCD
北大核心
2014年第2期161-167,共7页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金资助项目(61202143)
福建省自然科学基金资助项目(2013J05100)
厦门市科技重点项目资助项目(3502Z20123017)
湖南省自然科学基金资助项目(12JJ2040)
关键词
深度图像
单摄像头
人体深度估计
基于样本的学习
特征提取
特征匹配
相似样本
深度数据库
depth image
a single camera
human depth estimation
example-based learning method
feature ex- traction
feature matching
similar samples
depth database
