Selecting Feature Representation for Online Summarisation of Egocentric Videos

Abstract:

Visualising the content of a video through a keyframe summary has been a long-standing quest in computer vision. Using real egocentric videos, this paper explores the suitability of seven feature representations of the video frames for the purpose of online summarisation. Computational speed is an essential requirement in this set-up. We found that simple feature spaces such as RGB moments and CENTRIST are a good compromise between speed and representativeness in comparison with semantically richer but computationally more cumbersome spaces obtained through convolutional neural networks.

@inproceedings{ Paria18, 
author = { Yousefi, Paria and Kuncheva, Ludmila I. and Matthews, Clare E. }, 
title = { Selecting Feature Representation for Online Summarisation of Egocentric Videos }, 
booktitle = {Computer Graphics \& Visual Computing (CGVC) Conference 2018 - Poster}, 
year = 2018, 
address = {Swansea, UK}, 
month = sep, 
organization = {The Eurographics Association} }