Estimating camera rotation from vanishing points in underwater video

Abstract

The knowledge of camera pose aids in extracting information from images such as the subject's size and position. In this research, we developed an algorithm that estimates the camera rotation angles relative to a static scene without the use of geometric features or any calibration object. For a camera with translation relative to a static scene, the features' motion generates vanishing points in the image that can be used for rotation estimation using perspective transformation. Numerical simulation and experimental validation was performed to verify the accuracy of the proposed technique and test the algorithm's robustness to noise.