A benchmark for rotation extraction between two images in visual navigation

Abstract

When using camera as an aiding sensor, extraction of accurate pose measurement in the form of translation and rotation is instrumental to the navigation solution. This paper shows the experimental results of accurate angle extraction between a sequence of images taken from a calibrated camera while keeping the scale constant. In this work a complete mathematical model for rotation extraction is discussed comprising of feature extraction and matching, epipolar geometry constraints, structure from motion (SFM), etc. As far as the author’s knowledge, such a benchmarking for rotation extraction has not been previously reported and hence this algorithm can be used as a beginning point for new practitioners in the field of visual navigation. This benchmarking algorithm is also provided as an open-source toolbox for MATLAB available at https://github.com/maazmb/rotation_extraction.git.