Benchmark for Evaluating Initialization of Visual-Inertial Odometry

In this paper, we present a novel benchmark for the evaluation of the initialization of visual-inertial odometry (VIO). First, we collect a specific set of sequences in three challenging scenarios from existing public datasets. Second, we design benchmarking metrics for VIO initialization algorithms, with which the rapidity and accuracy of state estimation are evaluated comprehensively. We also evaluate state-of-the-art VIO initialization approaches on our benchmark. To facilitate related research, we publicly release the full benchmark described in this paper.