ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Journal Article Minimization of Parallax Artifacts in Video Stitching for Moving Foregrounds
Cited 7 time in scopus Download 28 time Share share facebook twitter linkedin kakaostory
Authors
MUHAMMAD UMER KAKLI, YONGJU CHO, JEONGIL SEO
Issue Date
2018-10
Citation
IEEE Access, v.6, pp.57763-57777
ISSN
2169-3536
Publisher
IEEE
Language
English
Type
Journal Article
DOI
https://dx.doi.org/10.1109/ACCESS.2018.2871685
Abstract
Parallax handling in the presence of moving foregrounds is a challenging problem in video stitching. The artifacts such as blurring or ghosting become prominent when objects move across multiple views. In this paper, we propose a novel scheme to minimize the parallax artifacts caused by moving objects in a fixed video stitching framework. We first stitch one frame from input videos and generate a lookup table (LUT) for every input video. These LUTs provide the mapping information of input videos to the panorama domain. For every frame of the input videos, we use structural similarity index to identify the LUT points in the overlapping region where parallax exists due to moving object. Therefore, our method does not require to match and track the features across frames to detect the moving object. For every detected LUT point having parallax, a descriptor based on histogram of oriented gradients and local binary patterns is generated and matched such that the parallax is minimized in the overlapping region. This may lead to the undesirable geometric artifacts in the non-overlapping region. To cater this issue, we smoothly propagate the correction in the non-overlapping region closer to the moving object. The proposed scheme is tested on different videos, where the overlapping region is very narrow, ranging from 2.3% to 9%. The experimental results demonstrate that it can significantly reduce the parallax artifacts, which occur due to moving foregrounds.
KSP Keywords
Geometric artifacts, Look Up Table(LUT), Mapping information, Moving Object, Multiple views, Non-overlapping, Structural SIMilarity index, Structure Similarity Index measure(SSIM), Video stitching, histogram of gradient(HOG), local binary pattern(LBP)
This work is distributed under the term of Creative Commons License (CCL)
(CC BY NC ND)
CC BY NC ND