Leveraging Activity Indexing for Egocentric Image Retrieval

摘要

Wearable cameras can daily gather large amounts of image data that require powerful image indexing and retrieval techniques in order to find the information of interest. In this work, we address the indexing problem of egocentric data by exploring the relevance of different information sources provided by Convolutional Neural Networks (CNN) combined with image metadata. The proposed method was tested on a public egocentric dataset of 45.000 images and gave encouraging results.