Robust direct visual inertial odometry via entropy-based relative pose estimation

Visual solution methods, like monocular visual odometry and monoSLAM, have attracted increasingly interests in robotics area. However, due to the large computational burden around volume sequential images processing, it is still hard to make numerous visual-based algorithms applying in highly agile platforms like Micro Aerial Vehicle (MAV) in real-time circumstance. In this paper, we present a method, which combines the direct image information from monocular camera and the measurements from inertial sensor in an Extend Kalman Filter (EKF) framework to perform an effective odometry solution. In contrast to other odometry methods, our solution gets rid of traditional feature extraction and expression, using the mutual information between images to perform the tracking. This entropy based tracking method enhances the robustness to illumination variation. The result of our method has been tested on real data.