Calibration of 3D lidars: A fully automatic and robust method for calibrating multiple 3D lidars using only point cloud data

Abstract

In order to use lidars for perception in autonomous vehicles, they must be properly calibrated. Commonly used techniques for automatic calibration, such as iterative closest point, often requires an accurate guess of the calibration parameters, which is challenging to obtain. Additionally, these techniques are often dependent on feature-extraction or designated calibration environments, which are highly application-specific. We propose an alternative calibration algorithm for an arbitrary number of lidars based on particle swarm optimization. Using our method, accurate calibration parameters can be produced from extremely rough initial guesses, without the aforementioned application-specific limitations. When tested on synthetic data, the algorithm is shown to be superior to conventional methods. Additionally, a method for allowing calibration during vehicle movement is explored and proposed.