A unified approach to volumetric registration and integration of multiple range images

Registration and integration of measured shape data of real objects are important for 3D modeling for computer graphics and computer-aided design. We propose a new algorithm that solves registration and integration of multiple range images in a unified framework. The closest point of an object from a general 3D point can be determined uniquely. If we have range images of an object and the images are registered correctly, the closest points found on the range images should be matched except for the case where the part is not measured. We register multiple range images by minimizing the error function determined by the variance of the closest points from every point in the 3D volume surrounding the object. We introduce scale factors to control the detail of the shape to be considered. The result contains the information of the closest point on the integrated shape, from which we can determine the distance and surface normal to the object. We experimented this method on synthetic and measured range images.