AeVIO: Asynchronous Event based Visual Inertial Odometry

Event cameras have several advantages such as motion blur-free data output, high dynamic range, and better low light sensitivity. Visual-Intertial Odometry (VIO) solutions can benefit with the use of these sensors instead of traditional frame based cameras. However, their sparse and asynchronous data pose a challenge for traditional computer vision algorithms. To address these shortcomings, asynchronous (data-driven) approaches are needed for event-camera-based VIO solutions. This paper introduces an end-to-end data-driven event camera based Visual-Inertial Odometry (AeVIO) algorithm that performs the state update based on camera velocity. The scheme performs event feature detection and tracking asynchronously on the event stream and fuses feature measurements with IMU data using a structureless version of the Extended Kalman Filter to update state estimates. The algorithm’s performance is evaluated on various datasets, including simulated and real environments, and demonstrates better performance in the tested scenarios when compared with image-based MSCKF-mono.