Experimental airborne NH∞ vision-based simultaneous localization and mapping in unknown environments

Abstract

This article aims to present a robust airborne three-dimensional (3D) visual simultaneous localization and mapping (VSLAM) solution based on a stereovision system embedded in a remotely piloted helicopter. The proposed solution enables an unmanned aerial vehicle to construct a reliable map and localize itself in this map without any user intervention. The main contributions of this article are threefold: first, based on an adaptive scale-invariant feature transform algorithm, a robust approach for feature extraction and matching is proposed. Second, an adaptation of the non-linear H∞ filtering scheme is presented to the VSLAM problem, for which a 3D vision-based observation model is suggested. This filtering scheme avoids issues linked with the classical extended Kalman filtering techniques such as linearization errors, initialization problems, and noise statistics assumptions. Finally, a reliable map management approach, based on the k-nearest landmark concept, is presented to allow efficient loop closing detection and map building. This approach reduces significantly the complexity of the airborne VSLAM algorithm by making it independent of the landmark number. Experimental results using real data sets show the performance and the robustness of the proposed airborne VSLAM solution to build a reliable map and localize the unmanned aerial vehicle within this map.