Dense RGB-D visual odometry using inverse depth

Daniel Gutiérrez-Gómez, Walterio W. Mayol-Cuevas, Josechu J. Guerrero

Abstract: Highlights•A direct method for RGB-D odometry by photometric and geometric error minimisation is presented.•Geometric error is parametrised by inverse depth, fitting better to the depth error model of RGB-D sensors.•The accuracy of the method is increased by introducing a keyframe switching strategy.•Our RGB-D odometry is evaluated in both real and synthetic benchmarking datasets.•We show similar or better accuracy than other state-of-the art approaches in odometry drift and absolute trajectory error.