Hyper-Lapse From Multiple Spatially-Overlapping Videos
This publication appears in: IEEE Transactions on Image Processing
Authors: M. Wang, J. Liang, S. Zhang, S. Lu, A. Shamir and S. Hu
Publication Date: Apr. 2018
Hyper-lapse video with high speed-up rate is aneff cient way to overview long videos, such as a human activityin f rst-person view. Existing hyper-lapse video creation methodsproduce a fast-forward video effect using only one videosource. In this paper, we present a novel hyper-lapse videocreation approach based on multiple spatially-overlapping videos.We assume the videos share a common view or location, andf nd transition points where jumps from one video to anothermay occur. We represent the collection of videos using a hyperlapsetransition graph the edges between nodes represent possiblehyper-lapse frame transitions. To create a hyper-lapse video,a shortest path search is performed on this digraph to optimizeframe sampling and assembly simultaneously. Finally, we renderthe hyper-lapse results using video stabilization and appearancesmoothing techniques on the selected frames. Our technique cansynthesize novel virtual hyper-lapse routes, which may not existoriginally. We show various application results on both indoorand outdoor video collections with static scenes, moving objects,and crowds.