Single-shot extrinsic calibration of a generically configured RGB-D camera rig from scene constraints

摘要: With the increasing use of commodity RGB-D cameras for computer vision, robotics, mixed and augmented reality other areas, it is significant practical interest to calibrate relative pose between a depth (D) camera an RGB in these types setups. In this paper, we propose new single-shot, correspondence-free method extrinsically generically configured rig. We formulate extrinsic calibration problem as one geometric 2D-3D registration which exploits scene constraints achieve single-shot calibration. Our first reconstructs sparse point clouds from single-view 2D image. These are then registered with dense camera. Finally, directly optimize warping quality by evaluating 3D clouds. does not require correspondences across multiple color images or different modalities more flexible than existing methods. The can be very simple demonstrate that containing three sheets paper sufficient obtain reliable lower error