Multistep Explicit Stereo Camera Calibration Approach to Improve Euclidean Accuracy of Large-Scale 3D Reconstruction

Abstract

The spatial accuracy of point clouds generated from a mobile stereo camera set is very sensitive to the intrinsic and extrinsic camera parameters (i.e., camera calibration) used in stereo image-based three-dimensional (3D) reconstruction methods. The existing camera calibration algorithms induce a significant amount of error owing to poor estimation accuracy in camera parameters when they are used for large-scale scenes such as mapping civil infrastructure. This leads to higher uncertainties in the location of 3D points, and may result in the failure of the whole reconstruction process. This paper proposes a novel procedure to address this problem. It hypothesizes that a set of multiple calibration parameters created by videotaping a moving calibration pattern along a specific path can increase overall calibration accuracy and ultimately enhance the Euclidean accuracy of the generated point cloud. This is achieved by using conventional camera calibration algorithms to perform separate estimations for certain predefined distance values. The result, which includes multiple sets of camera parameters, is then used in the structure from motion process. The proposed method has been tested on infrastructure scenes and experimental analyses indicate 25% improvement in the spatial distance accuracy of 3D points.