论文快速发表网

社科类论文科技类论文医学类论文管理类论文教育类论文农林类论文新闻类论文建筑类论文文艺类论文法学类论文

EI Compendex Source List（2022年1月） EI Compendex Source List（2020年1月） EI Compendex Source List（2019年5月） EI Compendex Source List（2018年9月） EI Compendex Source List（2018年5月） EI Compendex Source List（2018年1月）中国科学引文数据库来源期刊列 CSSCI(2017-2018)及扩展期刊目录 2017年4月7日EI检索目录（最新） 2017年3月EI检索目录最新公布北大中文核心期刊目录 SCI期刊（含影响因子）

本站服务项目浅谈青年教师如何开展教科研大学青年教师如何处理教学和科

论文范文

Fusing Depth and Silhouette for Scanning Transparent Object with RGB-D Sensor 时间:2017-09-07 22:32 来源:未知作者:admin 点击: 次 Abstract:3D reconstruction based on structured light or laser scan has been widely used in industrial measurement, robot navigation, and virtual reality. However, most modern range sensors fail to scan transparent objects and some other special materials, of which the surface cannot reflect back the accurate depth because of the absorption and refraction of light. In this paper, we fuse the depth and silhouette information from an RGB-D sensor (Kinect v1) to recover the lost surface of transparent objects. Our system is divided into two parts. First, we utilize the zero and wrong depth led by transparent materials from multiple views to search for the 3D region which contains the transparent object. Then, based on shape from silhouette technology, we recover the 3D model by visual hull within these noisy regions. Joint Grabcut segmentation is operated on multiple color images to extract the silhouette. The initial constraint for Grabcut is automatically determined. Experiments validate that our approach can improve the 3D model of transparent object in real-world scene. Our system is time-saving, robust, and without any interactive operation throughout the process.
1. Introduction
3D reconstruction based on structured light, including fringe pattern, infrared speckle, TOF, and laser scan, is widely used in industrial measurement, robot navigation, and virtual reality for its accurate measurement. In spite of the good performance in specific settings, it is troublesome for structured light to scan transparent objects. The transparent object which belongs to nonspecular surface can not reflect correct depth due to the properties of light absorption, reflection, and refraction. Therefore, some 3D acquisition systems have been specially developed for transparent object [1–3].
On the other hand, the popularity of consumer-grade RGB-D sensor, such as Kinect, makes it easier to combine depth and RGB information to improve a 3D scanning system. It occurs to us that we can recover the transparent surface by combining a passive reconstruction method as transparent objects appear in a stabler shape on color images. Since the transparent object is commonly with less texture, shape from silhouette (SFS) is considered more suitable to address the transparent issue. In addition, the flaw of SFS that fails to shape the concave objects can be remedied by structured light.
Some researchers have tried to fuse the depth and silhouette information for 3D scan. Yemez and Wetherilt [4] present a 3D scan system which fuses laser scan and SFS to fill holes of the surface. Narayan et al. [5] fuse the visual hull and depth images on the 2D image domain. And their approach can obtain high-quality model for simple, concave, and transparent objects with interactive segmentation. However, both of them only achieve good results in the lab environment but are not applied for natural scene with complex background.
Lysenkov et al. [6] propose a practical method for dealing with transparent objects in real world. Our idea is similar to theirs. We also try to look for approximate region of transparent object and some other nonspecular objects cued by noise from depth sensor before we use Grabcut [7] (classical image matting method) to extract their silhouettes on color images.
The main contributions of this paper are
(i)a complete system tackling the problem of volumetric 3D reconstruction of transparent objects based on multiple RGB-D images with known poses,
(ii)a novel pipeline that localizes transparent object before recovering the model by SFS,
(iii)a robust transparent object localization algorithm cued by both zero depth (ZD) and wrong depth (WD),
(iv)our system which is able to cope with real-world data and does not need any interactive operations.