Home

Resume

Publications

Research

Courses

Contact

Plane Correspondence Based Odometry



Overview


In this project, we develop a frame to frame odometry for a quadrotor travelling through a ship

Abstract


Flying through long corridors of a ship, offers little or no visual features for state estimation. For computing the state of the robot we rely on 3D data generated by an RGB-D sensor. The corridors have a highly planar structure and offer no feature to estimate the motion along the corridor. We exploit the planar nature of the environment by using frame to frame correspondences for state estimation. We develop a method that naturally reports the directions along which the motion estimate is under-constraint. We also make an argument, given the scope of the environment of an indoor robot flying through corridors, partial state of the robot can be easily calculated. Finally we present some results of plane segmentation and state estimation on data collected during flights.

Links


Project Video (mp4)

Project Presentation (pptx)

Project Report (dvi)