Application of neural networks for stereo-camera calibration

Abstract

The position of a world point can be measured by the use of calibrated stereo cameras. Although simple linear methods for the calibration assuming an ideal projection model are available, their solutions are usually not accurate since most off-the-shelf lenses used in machine vision applications sustain considerable amount of nonlinear distortion. Recent research efforts on the problem have been thus concentrated on the modeling of lens distortion and its correction techniques. However, the types of lens distortion are various and the equations derived are more complicated if more precise model is employed for higher accuracy. In this paper methods for calibrating stereo vision systems with neural networks are described. Different approaches are tested under various conditions and their results are compared.