Constrained-covisibility marginalization for efficient on-board stereo SLAM

PIRE, TAIHÚ; NITSCHE, MATIAS; FISCHER, THOMAS; CASTRO, GASTON; DE CRISTÓFORIS, PABLO

Paris

Conferencia; European Conference on Mobile Robotics (ECMR); 2017

ENSTA ParisTech, Université Paris-Saclay, France

When targeting embedded applications such ason-board visual localization for small Unmanned Air Vehicles(UAV), available hardware generally becomes a limiting factor.For this reason, the usual strategy is to rely on pure motion in-tegration and/or restricting the size of the map, i.e. performingvisual odometry. Moreover, if monocular vision is employed,due to the additional computational cost of stereo vision, thisrequires dealing with the problem of unknown scale.In this work we discuss how the cost of the tracking task canbe reduced without limiting the size of the global map. To do so,the notion of covisibility is strongly used which allows choosinga fixed and optimal set of points to be tracked. Moreover, thiswork delves into the concept of parallel tracking and mappingand presents some finer parallelization opportunities.Finally, we show how these strategies improve the com-putational times of a stereo visual SLAM framework calledS-PTAM running on-board an embedded computer, close tocamera frame-rates and with negligible precision loss.