Abstract: Automatic registration of partially overlapping range images is the core issue of various 3D laser scanning based reconstruction systems. Inspired by the use of twist to parameterize rigid motion in Hong-Tan workpiece localization algorithm, we present Hong-Tan based ICP (Iterated Closest Point) automatic registration algorithm (HT-ICP) for partially overlapping range images. Using the range images from Stanford 3D Scanning Repository, we compare HT-ICP with the original ICP algorithm and its main variants, and experimental results show that the HT-ICP algorithm can improve both the accuracy and speed of registration. Based on a mesh error measure tool, we measure the error between reconstruction result and benchmark, and experimental results demonstrate that HT-ICP algorithm is sufficient for practical 3D reconstruction applications in industry.