Second year report.
Full text not available from this repository.
The performance of visible and infrared cameras which can be used for guidance is improving rapidly. These sensor improvements and the increase in computing power now allow image based guidance techniques, such as structure from motion, to be more readily utilised in applications such as intelligent cruise control and convoying. This project seeks to extract vehicle specific cues from image sequences, e.g. red tail-lights, and utilise 2D image plane tracking to obtain estimates for the (relative) 3D vehicle motion, e.g. bearing and time-to-contact. These image based estimates should provide complementary information to other conventional sensors, e.g. millimetric radar, for data fusion. This document is intended to introduce the problems of tracking vehicles in (monochrome filtered) image sequences. A brief review of generic segmentation techniques for extracting vehicle specific cues is presented, together with details of Fourier spectral filtering required to enhance a far infrared image sequence. Details of the specification and acquisition of filtered monochrome image motorway sequences are presented, together with a brief comment on the relationship between filtered monochrome and colour images.
Actions (login required)