This paper discusses technique used for construction of highresolution image mosaic from a video sequence and the synchronously logged camera attitude information. It allows one to infer geometric characteristics of the imaged terrain and hence improve mosaic quality and reduce computational burden. The technique is demonstrated using numerical modeling and is applied to video data collected on Rainsford Island, Mass. Calculation of the transformation relating consecutive image frames is an essential operation affecting reliability of the whole mosaicing process. Improvements to the algorithm are suggested, which significantly decrease the possibility of convergence to an inappropriate solution.
Yuri Rzhanov, Lloyd Huff, G. Randy Cutter