LAUSR

Abstract Camera calibration plays a fundamental role for 3D computer vision since it is the first step to recover reliable metric information from 2D images. The calibration of a stereo-vision system is a two-step process: firstly, the calibration of the individual cameras must be carried out, then the two individual calibrations are combined to retrieve the relative placement between the two cameras, and to refine intrinsic and extrinsic parameters. The most commonly adopted calibration methodology uses multiple images of a physical checkerboard pattern. However, the process is time-consuming since the operator must move the calibration target into different positions, typically from 15 to 20. Moreover, the calibration of different optical setups requires the use of calibration boards, which differ for size and number of target points depending on the desired working volume. This paper proposes an innovative approach to the calibration, which is based on the use of a conventional computer screen to actively display the calibration checkerboard. The potential non-planarity of the screen is compensated by an iterative approach, which also estimate the actual screen shape during the calibration process. The use of an active display greatly enhances the flexibility of the stereo-camera calibration process since the same device can be used to calibrate different optical setups by simply varying number and size of the displayed squared patterns.