Auto-Calibration of Multi-view RGB-D Cameras for Virtual Mirrors

In this paper, we propose a new auto-calibration method of multi-view RGB-D cameras for virtual mirrors. In the proposed method, the 3D coordinates of joints in a human body are detected and tracked with multi-view depth cameras. Then, 3D coordinates are transformed from the depth camera coordinate system into the color camera one. Lastly, 3D extrinsic calibration parameters between color cameras are computed from transformed 3D coordinates with considering confidence levels. The effectiveness of the proposed method is verified by applying the virtual mirror system to AR sports contents.