A simulation environment for visual-inertial sensor fusion

The complementary nature of visual- and inertial-based trackers allows for robust hybrid tracking systems capable of handling fast motion and occlusions. We present a tool to investigate the fusion of these two types of modalities, in the form of a simulation environment. The environment generates the ground-truth data of a nine-axis inertial measurement unit (IMU) as well as the individual camera frames from an array of cameras. Using the simulation environment, we demonstrate how visual trackers and inertial-based trackers can be benchmarked individually, and then show how the environment can be used to study visual-inertial sensor fusion.