Accurate Image Mosaicing For Bridge Deck Using Graph Optimization with GPS Data

摘要

Bridge deck scanning is a vital task for maintaining the structural health of concrete bridge. A panorama image could provide an overview of the bridge deck as well as details and serves as the reference for other nondestructive evaluation (NDE) method. To obtain a panorama image, image mosaicing technique is often used to assemble multiple local scanned images with sufficient details into a mega one. The conventional idea of creating a panorama with the homography model always suffers from error accumulation. In this paper, we present a novel image mosaicing algorithm for bridge deck surface reconstruction by fusing the GPS data with featurebased image registration in the graph optimization framework. Our proposed approach inherits the drift-less nature from GPS while still maintaining local accuracy of feature-based image registration. Experimental results on real bridges are used to show the accuracy and robustness of our method.