Egomotion estimation for vehicle control

The focus of this thesis is a technique called egomotion estimation, which involves the extraction of motion parameters from a camera based on the nature of the motion field on a frame-by-frame basis. In general, this is a multi-step process that involves estimating the motion field, often referred to as the optical flow, from which the translation direction and rotation are then extracted. The optical flow field is normally generated by tracking a frame's strong features in the subsequent frame of a sequence. Examples of strong features include corners of objects or areas of high contrast within an image. The algorithms described in this thesis have been developed with the hopes of eventually being utilized as the primary sensor on a Draganflyer four-rotor helicopter (also known as a quadrotor) for self-motion estimation. A PD controller was implemented to stabilize the quadrotor, and its effectiveness has undergone initial testing in simulation. -- The algorithms and implementations that follow, in their initial implementations, took over one minute to find a result on an Intel 3.0Ghz Xeon system. They are now running at a rate of about 5Hz, which is certainly a noteable difference. The methods presented are by no means optimal. The author is continuing this research on egomotion estimation as a part of his doctoral studies