Aircraft Attitude Estimation Using Panoramic Images

Abstract

This thesis investigates the problem of reliably estimating attitude from panoramic imagery in cluttered environments. Accurate attitude is an essential input to the stabilisation systems of autonomous aerial vehicles. A new camera system which combines a CCD camera, UltraViolet (UV) filters and a panoramic mirror-lens is designed. Drawing on biological inspiration from the Ocelli organ possessed by certain insects, UV filtered images are used to enhance the contrast between the sky and ground and mitigate the effect of the sun. A novel method for real–time horizon-based attitude estimation using panoramic image that is capable of estimating an aircraft pitch and roll at a low altitude in the presence of sun, clouds and occluding features such as tree, building, is developed. Also, a new method for panoramic sky/ground thresholding, consisting of a horizon– and a sun–tracking system which works effectively even when the horizon line is difficult to detect by normal thresholding methods due to flares and other effects from the presence of the sun in the image, is proposed. An algorithm for estimating the attitude from three–dimensional mapping of the horizon projected onto a 3D plane is developed. The use of optic flow to determine pitch and roll rates is investigated using the panoramic image and image interpolation algorithm (I2A). Two methods which employ sensor fusion techniques, Extended Kalman Filter (EKF) and Artificial Neural Networks (ANNs), are used to fuse unfiltered measurements from inertial sensors and the vision system. The EKF estimates gyroscope biases and also the attitude. The ANN fuses the optic flow and horizon–based attitude to provide smooth attitude estimations. The results obtained from different parts of the research are tested and validated through simulations and real flight tests.