This paper addresses the problem of flexible and accurate global calibration for multiple non-overlapping vision sensors in a confined workspace. Instead of using an auxiliary calibration pattern, the proposed method uses one laser tracker and only its accessory target sphere to obtain all the 3D calibration points and then accomplish the initial estimation of pose between the vision sensors. Then, the 3D calibration points and the extrinsic parameters between vision sensors are further optimized via the bundle adjustment algorithm based on the distance and reprojection constraints. Experiments were conducted to validate the performance and the experimental results demonstrate that the distance error can be decreased from 3.5 mm to 0.8 mm after introducing the distance and reprojection constraints.
Keywords: bundle adjustment; calibration; nonoverlapping vision sensors.