Real-time, image-based motion estimation for the Dervish landmine-clearance vehicle

• Main Author: Haworth, Christopher D.
• Published: University of Edinburgh 2002
• Subjects: 526.9
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.652245

This thesis proposes and demonstrates that a low-cost commercial DSP platform can be used to perform real-time motion estimation for the Dervish landmine-clearance vehicle based on digital imaging of the ground. The Dervish landmine-clearance vehicle provides specific <i>application</i> requirements that are first examined in detail. The motion estimation <i>algorithm</i> requirements are then examined and, based on the tracking accuracy and theoretical computational performance, correlation-based feature tracking is selected. The individual parameters, including block size, motion model and interpolation technique, are carefully examined and suitable values selected. Finally, the <i>hardware</i> requirements are examined, a DSP platform chose and a trial implementation reported. A difficult problem within correlation-based motion estimation is how to update the reference block. This thesis describes novel methods for reference block updating. The reference block update strategies are examined in detail for parameter selection, tracking accuracy and computational performance. Particular attention is given to the tracking accuracy under increasing levels of image noise. Using the <i>application, algorithm</i> and <i>hardware</i> requirements and decisions, a system is proposed that meets the main requirements of the Dervish landmine-clearance vehicle. The issues of robustness within long-term tracking and a novel dissimilarity measure based on the reference block update strategies are also presented. Parts of the research presented in this thesis have also been reported in conference and journal papers, which are listed at the end of the document.