Life Gallery: event detection in a personal media collection

摘要

Usage of camera-equipped mobile devices raises the need for automatically organizing large personal media collections. Event detection algorithms have been proposed in the literature to tackle this problem but a shortcoming of these studies resides in the lack of comparison of the different solutions. The goal of this work is then to provide a set of tools for the assessment of an event detection method. Our first contribution is our prototype Life Gallery, which aims to retrieve metadata of a personal media collection in order to facilitate the building of a benchmark. Our second contribution is an event detection algorithm based on successive refinements of an initial temporal partition. Our solution uses a simple algorithm that regroups consecutive similar media or split groups with dissimilar media. The choice to provide a basic solution is to compare in a future work our results with more complex solutions, in order to really assess the benefit of these latter. Finally, our third contribution is a benchmark of 6 different users' collections obtained thanks to our Life Gallery application.