DepthTiling: A novel way to increase visual SLAM performance in featureless environments

Abstract

The common problem of visual simultaneous localisation and mapping systems is to suffer from featureless environments. It is possible for all environments to have such featureless situations; even though most of the mapped areas contain sufficient textures. This Letter brings a new approach using not only RGB values of the objects but also their positions in the map for feature extraction in order to decrease odometry loss in such situations. DepthTiling recolours RGB image using associated depth data. The experiments give promising results to increase the capability of visual odometry tracking. This study shows that it is possible to increase number of features using related depth data when RGB images are insufficient.