Feature Point Selection Scheme of Stereo Visual Odometry for Planetary Exploration Rover

Planetary exploration rovers are required to estimate their position in GPS-denied environments accurately. Visual Odometry is one of the solutions in such environments. By tracking feature points in the image, it is possible to estimate the positionwith high accuracy, even in extreme environments. If all the feature points in the image are used for estimation, however, the computational cost increases. Especially in stereo camera methods, the calculation time required for stereo matching is highly extended. Therefore, this paper proposes a method for selecting feature points in the image before stereo matching. The accuracy will bediminished if the number of feature points is reduced. To address this issue, the feature points are selected separately for rotation and translation. As a result of verification using an actual rover, it is confirmed that the proposed method can reduce the computational cost by up to $33 \%$ compared to the conventional method.