Realtime Edge Based Visual Inertial Odometry for MAV Teleoperation in Indoor Environments

A working solution for control and teleoperation of Micro Aerial Vehicles using a frontal camera and an inertial measurement unit as sole sensors is presented. The system is an extension of an edge based visual odometry algorithm to integrate inertial sensors. A mixed tightly-loosely coupled approach is used, taking advantage of each sensor in this minimalistic setup, while keeping the complexity low. The system runs completely on board a MAV providing a semidense output that is more useful for navigation than the sparse maps generated by most feature based systems. To the best of the author’s knowledge, the system is the first semidense VO method running fully on board a MAV for vision in the loop control. An extensive evaluation of the method is presented using the EuRoC MAV dataset, that is specially targeted for MAV navigation in realistic situations. Some of the practical issues of teleoperation are also addressed, in particular how data is transmitted and presented to the user. Finally, real life experiments are included to illustrate the performance of the complete system and the teleoperation interface.