Home

Resume

Publications

Research

Courses

Contact

River Segmentation and Pathfinding



Overview


In this project, we develop a self-supervised method to segment river in images collected by an octo-rotor, online, to map and follow the river.

Abstract


For a river mapping robot, it is imperative to segment out the river from rest of the scene and decide the direction of travel to traverse the river efficiently. The dynamic appearance of the river makes the task non-trivial. The appearance of a river changes with different lighting conditions, surrounding structures and objects. Reflections can make the river appear very similar to its banks. Hence, the traditional approach of learning a fixed appearance model can fail. We use a dynamic river model while, combining LIDAR and camera data to segment the river online.

Links


Poster (ppt)

Results Video (mp4), Details Video (mp4)

Project Webpage (link)