Implementation of 3D object reconstruction using a pair of Kinect cameras

In this paper, we propose a process of 3D object reconstruction using a pair of Kinect cameras. After we refine raw depth images from two Kinect cameras using a joint bilateral filter, we find intrinsic and extrinsic parameters by camera calibration. Then, we apply 3D warping to obtain a point cloud model in the 3D space and acquire a smooth surface model of the 3D object. In order to accelerate a processing speed, we employ a CUDA framework for GPU parallel processing. We reconstruct depth data in the integrated 3D space and obtain a 3D object model at 5 fps.