An approach for registration of sparse feature sets detected in stereo images taken from two different views is proposed.
The approach is based on the RANSAC paradigm which ensures robustness to outliers. The strategy proposed in this work is to
apply a multi-step procedure in the hypothesis generation stage of RANSAC using approximate information about the relative
camera pose between the two views obtained e.g. by odometry. The data is taken from the input data set one by one and used
to pose a geometric constraint for selection of the next data. This strategy reduces the number of false hypothesis in the
RANSAC procedure. The presented technique is evaluated using both synthetic data and real data obtained by a stereo camera
system mounted on a mobile robot.
For more details see our paper "Stereo Image Registration Based on RANSAC with Selective Hypothesis Generation" (