Ground plane based visual odometry for RGBD-Cameras using orthogonal projection

This work presents a method for visual odometry that allows robust 3 degrees of freedom trajectory estimation for wheeled robots using a downward facing RGBD-camera. Assuming that the robot moves on a ground plane while the environment itself can have arbitrary geometry allows to estimate the frame to frame motion from orthographic projections of the RGBD-data. Instead of directly aligning these projections, the reference frame is split into blocks, which are individually registered using Efficient Second Order Minimization, and thus create several estimates of the current motion. These estimates are combined using an outlier rejection scheme to create a robust estimate of the actual motion even under challenging conditions. The results of this method are compared to the results of other state-of-the-art methods to show its accuracy and robustness.