Extrinsic Calibration of Multiple Depth Cameras for 3D Face Reconstruction

This paper presents a reliable and robust approach for 3D face reconstruction using low-cost depth cameras. The solution is designed to withstand both camera noise and motion artifacts, making it potentially suitable for dealing with living subjects. Our method utilizes Iterative Closest Point (ICP) registration with a calibration pattern, allowing for scalable acquisition with multiple devices. Experiments are conducted adopting two low-cost depth cameras, and producing the 3D reconstruction of a dummy head to favour a metrological evaluation. The findings indicate that the suggested approach outperforms the alternative method of directly applying ICP to the facial point cloud. Additionally, the outcomes demonstrate that the low-cost solution deviates from the high-quality professional equipment by an average of 0.5 mm, showing the notable accuracy of the proposed method.