Monocular Camera Calibration for Autonomous Driving — a comparative study

Autonomous driving is currently a widely researched topic worldwide. With a large research effort being taken by industrial research units in the automotive sector, it is no longer exclusive to academic research labs. Essential to this ongoing effort towards level-5 vehicle autonomy, are the sensors used for tracking and detection, mainly lasers, radars and cameras. Most of the cameras for automotive application systems use wide-angle or fish-eye lens, which present high distortion levels. Cameras need to be calibrated for correct perception, particularly for capturing geometry features, or for distance-based calculations. This paper describes a case-study concerning monocular camera calibration for a small scale autonomous driving vehicle vision system. It describes the fundamentals on camera calibration and implementation, with results given for different lenses and distortion models. The aim of the paper is not only to provide a detailed and comprehensive review on the application of these calibration methods, but to serve also as a reference document for other researchers and developers starting to use monocular vision in their robotic applications.