A 5-Parameter Bundle Adjustment Method for Image Mosaic
Xianyong Fang, Bin Luo, Haifeng Zhao, Yiwen Zhang · 2009
A 5-parameter bundle adjustment method is proposed in this paper for global mosaic of an image sequence. By decomposing the rotation matrix into a 3-parameter rotation axis and a rotation angle, to each image, there are 5 parameters that need estimate for rotation axis, rotation angle and focal length. The proposed method minimizes the distance between feature matches in 3D space so that these 5 parameters are refined gradually for final stitching of all images together. Experimental results suggest that this method is an useful extension to existing research with the advantages of real 3D camera motion, accurate 5-parameter rotation decomposition and reliable 3D feature definition.