Multiple-feature based image retrieval techniques for large-scale visual localization

Multiple-feature based image retrieval techniques for large-scale visual localization

Funded by: Zhejiang Provincial Natural Science Foundation of China, Grant No. LY19F030022

Period: 2019-01/2022-12

About the project:

Localization using image retrieval is a new genre of visual localization techniques. As an effective complement to classical localization methods, it can be used for autonomous driving, augmented reality, culture and education, etc. The basic idea is to compare a query image with a geo-referenced image database, and perform location recognition or camera pose estimation using retrieved geo-references. Existing methods in this domain typically use single features, and offer limited application scales and accuracy due to large amounts of data. This project is dedicated to realizing large-scale accurate place recognition with multiple-feature based image retrieval. The new method can extend visual localization to larger applications scales, and possesses certain generality to build the foundation of retrieval and linkage of massive image content in other domains.