Visual Odometry for autonomous robot navigation through efficient outlier rejection

The ability of autonomous robots to precisely compute their spatial coordinates constitutes an important attribute. In this regard, Visual Odometry (VO) becomes a most appropriate tool, in estimating the full pose of a camera, placed onboard a robot by analyzing a sequence of images. The paper at hand proposes an accurate computationally-efficient VO algorithm relying exclusively on stereo vision. A non-iterative outlier detection technique capable of efficiently discarding outliers of matched features is suggested. The developed technique is combined with an incremental motion estimation approach to estimate the robot's trajectory. The accuracy of the proposed system has been evaluated both on simulated data and using a real robotic platform. Experimental results from rough terrain routes show remarkable accuracy with positioning errors as low as 1.1%.