CAREER: Learning and Using Models of Geo-Temporal Appearance

点击翻译按钮获取中文摘要

Billions of geotagged and time-stamped images are publicly available via the Internet, providing a rich record of the appearance of people, places, and things across the globe. These images are a largely untapped resource that could be used to improve our understanding of the world and how it changes over time. This project develops automated methods of extracting useful information from this imagery and fusing it into high-resolution global models that capture geo-temporal trends. Once the trends have been captured, these models are used to improve performance on computer vision tasks and make geotagged imagery a usable and navigable resource for education and research in other disciplines. The project includes an education and outreach component that brings real-world problems to computer science (CS) students, mentors students across the educational spectrum, and makes the research accessible to the public.This project develops computer vision technologies to capture spatial and temporal appearance trends and is organized into four main research thrusts: (1) investigating novel methods for extracting information from Internet imagery using weakly supervised learning, (2) developing techniques that integrate ground-level imagery with aerial and satellite data to model the expected image appearance anywhere in the world at any time, (3) evaluating methods for using such models to improve the performance of computer vision algorithms, and (4) automatically creating visual representations that make it possible for novice users to explore the learned geo-temporal trends via the Internet.Project webpage: http://geotemporal.csr.uky.edu