Accuracy Evaluation of Stereo Vision Aided Inertial Navigation for Indoor Environments

Abstract

Accurate knowledge of position and orientation is a prerequisite for many applications regarding unmanned navigation, mapping, or environmental modelling. GPS-aided inertial navigation is the preferred solution for outdoor applications. Nevertheless a similar solution for navigation tasks in difficult environments with erroneous or no GPS-data is needed. Therefore a stereo vision aided inertial navigation system is presented which is capable of providing real-time local navigation for indoor applications. A method is described to reconstruct the ego motion of a stereo camera system aided by inertial data. This, in turn, is used to constrain the inertial sensor drift. The optical information is derived from natural landmarks, extracted and tracked over consequent stereo image pairs. Using inertial data for feature tracking effectively reduces computational costs and at the same time increases the reliability due to constrained search areas. Mismatched features, e.g. at repetitive structures typical for indoor environments are avoided. An Integrated Positioning System (IPS) was deployed and tested on an indoor navigation task. IPS was evaluated for accuracy, robustness, and repeatability in a common office environment. In combination with a dense disparity map, derived from the navigation cameras, a high density point cloud is generated to show the capability of the navigation algorithm.