Pose calibration of two cameras with non-overlapped field of view

Abstract

Pose calibration is widely used in the fields of 3D (three-dimensional) shape reconstruction, vision-based robot navigation, augmented reality, and so on. However, if two cameras have the non-overlapped FOV (Field Of View), existing methods cannot calibrate their relative position because the calibration target is invisible to one camera. This paper presents a novel calibration method for two cameras with non-overlapped FOV by using a calibration target and a flat mirror. For the camera invisible to the calibration target, the camera can capture the virtual image via the flat mirror. PNP (Perspective-N-Point) problem is solved via the captured virtual image, and then the relative orientation between the virtual calibration target and the camera is obtained. Based on the reflection theory, moving the mirror to several positions, the orthogonal constraint equations are established. This orthogonal constraint is satisfied by all mirror positions and the reference points on target. Therefore, one can obtain the external orientation between the calibration target and the camera. For the camera visible to the calibration target, the traditional calibration method is used to obtain the external orientation between the target and the camera. After obtaining the relative orientation of the two cameras to the calibration target, an ICP (Iterative Closest Point) algorithm is used to compensate binocular pose matrix. The calibration results are analyzed by using an error model. The experimental results show that the proposed calibration method can obtain the relative orientations between two cameras effectively and accurately.