Monocular Visual-Inertial Odometry (VIO) has become ubiquitous for navigation of autonomous Micro Air Vehicles (MAVs). Yet, state-of-the-art VIO is still very failure-prone, which can have dramatic consequences. To prevent this, VIO must be able to re-initialize in mid-air, either during a free fall or on a constant velocity trajectory after attitude control has been re-established. However, for both of these trajectories, the visual scale cannot be observed with VIO batch initializers because of the absence of acceleration change. We propose to use a small and lightweight laser-range finder (LRF) and a scene facet model to initialize vision-based navigation at the right scale under any motion condition and over any scene structure. This new range constraint is integrated into a visual-inertial bundle-adjustment initializer. We evaluate our approach in simulation, including robustness to various parameters, and demonstrate on real data how this approach can address midair state estimation failure in real-time.