Trinocular Rectification for Various Camera Setups

Convergent stereo images can be rectified such that they correspond to the stereo normal case. The important advantage is that the correspondence analysis is simplified to a linear search along one of the image axes. The proposed method describes a rectification method for three uncalibrated cameras, where the configuration is not generally known. The algorithm automatically determines the best positioning of the cameras and corrects mirroring effects to fit the desired camera setup. Since there is no direct solution for the rectification problem for three cameras, the rectification homographies are linearly determined to within six degrees of freedom from three compatible fundamental matrices. The remaining six parameters are then obtained by enforcing additional constraints.