[image 02624] 講演会のご案内（10/13 13:30-15:00, Osaka U., Prof. Shmuel Peleg and Prof. Michael Werman)

[Yasuyuki Matsushita]

Image-MLの皆様

大阪大学の松下です。お世話になっております。

下記の通り、大阪大学吹田キャンパスにてShmuel Peleg先生とMichael Werman先生をお迎えする講演会を開催いたします。
登録等必要ございませんので、ふるってご参加いただけましたら幸いです。

—

Time: 13:30-15:00 October 13, 2017
Place: A110, Graduate School of Information Science and Technology, Osaka University 
（大阪大学吹田キャンパス　情報科学研究科A棟 A110）

Speaker 1: Professor Shmuel Peleg, The Hebrew University of Jerusalem 

Title: Video without Photographers

Abstract:
In ordinary video a photographer selects the area of interest, and presses the record button at the time of interest. But most existing video is taken without a photographer. Surveillance cameras, car-mounted cameras, and wearable cameras are turned on and record continuous videos without a photographer aiming the camera or pressing the record button. The resulting video is long and unstructured, and mostly boring. Most such video may therefore never be watched. In this talk I will describe some approaches to make such video accessible to human viewers. Topics will include:
(1) Stationary cameras: 
Video Synopsis - Summarization of video from static cameras by showing simultaneously objects that appeared in different times. The has been made into a product by Briefcam Ltd.
(2) Wearable Cameras:
* Egosampling - Fast forward of egocentric video without the shake.
* Biometrics - 4 seconds of video from a wearable camera are enough to recognize the person (camera shake is unique!).
* Temporal Segmentation: Optical flow allows the determination of the activity of the wearer


Speaker 2: Professor Michael Werman, The Hebrew University of Jerusalem

Title: Epipolar Geometry from Epipolar Lines

Abstract:
The fundamental matrix is the basic building block of multiple view geometry and its computation is the first step in many vision tasks. It is usually computed from pairs of corresponding points. 
The fundamental matrix can also be computed from three matching epipolar lines. Given three such epipolar line correspondences, the one dimensional homography between the lines can be recovered. The 3 degrees of freedom for the 1-D homography together with the 4 degrees of freedom of the epipoles yield the required 7 degrees of freedom needed to compute the fundamental matrix. 
We show various scenarios where the corresponding line method is feasible and superior to using point correspondences.

--------
Yasuyuki Matsushita
Graduate School of Information Science and Technology
Osaka University
yasumat ＠ ist.osaka-u.ac.jp