A Globally Consistent Solution to Multiview Image Registration
Multiview image registration is to compute the globally consistent transformations of a sequence of images. Due to various uncertainties, multiview image registration is considered as stochastic estimation problem. In order to improve global consistency of registration, the states of all viewpoints are estimated in a common state vector and covariance matrix. The system state consists of the position and orientation of viewpoints. System augmentation model is based on the coarsely pairwise matching. System observation model is constructed with feature correspondence. The position and orientation of viewpoints are augmented and estimated recursively with augmented Kalman filter.