Gift: A geospatial image and video filtering tool for computer vision applications with geo-tagged mobile videos

摘要

In this paper, we propose a novel geospatial image and video filtering tool (GIFT) to select the most relevant input images and videos for computer vision applications with geotagged mobile videos. GIFT tightly couples mobile media content and their geospatial metadata for fine granularity video manipulation and intelligently indexes FOVs (Field of View) to deal with large volumes of data. To demonstrate the effectiveness of GIFT, we introduce an end-to-end application that utilizes mobile videos to achieve persistent target tracking over large space and time. Our experimental results show promising performance of vision applications with GIFT in terms of lower communication load, improved efficiency, accuracy and scalability.