Prioritized Cascade Search and Fast One-Many RANSAC for On-Device Scalable Image-Based Localization

PROJECT TITLE :

On-Device Scalable Image-Based Localization via Prioritized Cascade Search and Fast One-Many RANSAC

ABSTRACT:

We describe a complete on-device solution for large-scale image-based urban localisation. Compact image retrieval and two-dimensional to three-dimensional correspondence search can be used to estimate the location of a target inside large urban areas. A network connection is not required to use our design. The scalability benefit of image retrieval and the precision of 3D model-based localization are combined in a system design that we believe will alleviate the resource limitations of mobile devices. For efficient computing of 2D-3D correspondences, we present a new hashing-based cascade search based on cascading. A new one-many RANSAC for precise pose estimation is also proposed. One-many RANSAC is a new approach to urban localization that addresses the issue of repeating building structures (such as windows and balconies). On a wide range of benchmark datasets, our 2D-3D correspondence search delivers state-of-the-art localization accuracy. A big dataset of Google Street View images shows that a standard mobile device may be used for large-scale localisation.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here