VVS: Fast Similarity Measuring of FoV-Tagged Videos

Abstract

In this article, we propose a novel method that can measure the similarity of FoV-tagged videos in two dimensions. Recently many researchers have focused on measuring the similarity of FoVtagged videos. The similarity measurement of FoV-tagged videos plays an important role in various societal applications, including urban road networks, traffic, and geographic information systems. Our preliminary work introduced the Largest Common View Subsequences (LCVS) algorithm for computing the similarity of FoV-tagged videos. However, LCVS requires a high computational cost for calculating common viewable regions between two FoV-tagged videos. To handle this limitation, we propose the largest View Vector Subsequence (VVS) algorithm for reducing the computational cost of FoV-tagged videos. VVS uses the movement distances and the viewable direction distances to support the simplified vector-based similarity measurement. We demonstrate the superiority of our approach by comparing it with the Longest Common Subsequences (LCSS) and our prior work (LCVS). Our experimental results show that VVS outperforms the prior work in terms of the computational cost and enhances the versatility and stability of the similarity measurement.INDEX TERMS FoV-tagged video, similarity measurement, recommendation system.

Discovering Homogeneous Groups from Geo-Tagged Videos

“…In a previous study, the authors collected eye-tracking data from 1003 images of 15 viewers, recorded each observer's look tracking path and look position, and used that as ground truth data for training a saliency model using machine learning. The term geo-tagged video refers to a video obtained using a camera device with an embedded GPS sensor to capture the video's shooting time, location, and direction [4][5][6]. GeoUGV [4] is an extensive geo-tagged video dataset.…”