|
|
|||||||||||||||||
|
|
|
|||||||||||||||||
Prelim Announcement: “Commonly Identified Feature-based Stitching System for Stereoscopic Panoramic Video Generation”
Participants: Haoyu Wang
Committee: Dan Schonfeld (Chair/Advisor) Daniel J. Sandin Andrew E. Johnson Hulya Seferoglu Mojtaba Soltanalian Date & Time: July 20th, 2018, 12:00PM - 2:00PM Location: 2068 ERF (Electronic Visualization Laboratory Cyber-Commons) Abstract: In my proposal thesis, we propose a novel stereoscopic panoramic video generation strategies based on sparsely distributed features from the same object, feature, or area. This proposal can improve stitching quality of single-frame stereoscopic panoramas. Additionally, the consistency of these control points for stitching between the consecutive frame in the temporal domain can ensure the stability of the stitched video and avoid most discontinuities. Our proposed strategies begin with the construction of the commonly identified features set. Each set describes the identical features observed and described from two pairs of adjacent camera views. To remove redundant information and repetitive control points in the first frame, we incorporate human visual importance sensitivity to refine these initially detected feature set, producing more reliable and consistent control points in the latter frames of the video sequence. To remove the outliers and to maintain consistency of the selected consensus set, we also extend the conventional random-sample consensus-based homography estimation to stereo mode. Extensive experiments on synthesized and real-captured data indicate that our proposed framework can get better stitching performance than standard feature-based panoramic video generation pipeline in PanoTools. Date: July 20, 2018 |
Related EntriesDirectory:Research: Related Categories |
