Wisdom of the Crowd in Egocentric Video Curation

Yedid Hoshen, Gil Ben-Artzi, Shmuel Peleg; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2014, pp. 573-579

Videos recorded by wearable egocentric cameras can suffer from quality degradations that cannot always be fixed by current methods. When several wearable video cameras are viewing the same scene, each having highly variable quality, it is possible to combine them into a single high-quality video. Current techniques select for each point in time the highest quality video stream, but the highest quality video may not be relevant. E.g. the best quality video can come from a person that happen to look sideways from the main attraction. We propose the curation of a single video stream from multiple egocentric videos by requiring that the selected video will also view the most interesting region in the scene. Importance of a region is determined by the "wisdom of the crowd", i.e. the number of cameras looking at a region. The resulting video is more interesting and of higher quality than any individual video streams can possibly obtain. Several examples are presented demonstrating the effectiveness of this technique.