Flexible global calibration of multiple cameras with nonoverlapping fields of view using circular targets

Global calibration of multicamera systems is a difficult problem. The key is to find an appropriate calibration target. This paper proposes a flexible method to construct a global calibration target with circular targets. Any object that is to be measured by the system can be used as a calibration target with the help of a hand-held scanner. The calibration method does not need a special calibration target or multiple shots, which makes it flexible in applications. Circular targets pasted on the calibration target are used as features that are captured by cameras. The particle swarm optimization method is employed to correct the eccentricity error in the projection of circular targets. The eccentricity correction method does not need any prior knowledge except the cameras' intrinsic parameters. A synthetic data experiment was performed to validate the eccentricity correction method, and a physical experiment in a train wheelset inner-side distance measurement system was performed to validate the global calibration method. The measurement accuracy of the system was better than 0.1 mm, and the eccentricity correction method improved the three-dimensional reconstruction accuracy by about 0.1 mm.