Markerless Augmented Reality for Robots in Unprepared Environments

Abstract

Augmented Reality (AR) can assist humans in understanding complex robot information, and improve Human and Robot Interaction (HRI). However, many restrictions are imposed by the underlying technology used and thus have limited current AR systems to operate in controlled or modified robot environments. This hinders the wide spread use of AR for different robot applications. This paper presents a markerless AR system that combines recent tracking and detection techniques for AR visualisation of robot task relevant information. We employ natural feature tracking techniques to compute the camera pose for accurate registration of virtual objects. Automatic relocalisation of the camera pose is achieved using a planar object detection algorithm which recovers from tracking failures. Experiments using a camera mounted on a mobile ground robot demonstrated accurate tracking and successful recovery of planar features in an unprepared indoor environment. 1