Multiplicative Extended Kalman Filter based on Visual Data for Attitude Estimation

In this paper a method for rigid body attitude estimation based on a new feature-tracking algorithm and a Multiplicative Extended Kalman Filter (MEKF) is proposed and developed. The new feature-tracking algorithm allows us to detect and to track the image features (points and lines). This algorithm is able to cancel the drift due to a prediction error accumulation when a long sequence of images is used. Indeed, in this algorithm we have added a correction of the prediction given by optical flow features, using criterion based on Euclidean distance only in the predicted search area. Thus, the extracted points and lines are used as visual data measurements, which enable us to estimate attitude when they are fused with gyros measurements using MEKF. Simulation results has been presented to show the effectiveness of the applied method.