Abstract: In this paper, we focus on the problem of fast and accurate featureless registration of outdoor large scale 3D pointclouds which possess great differences in the aspects of both resolution and view of point. There are two main methods generally used to solve this problem: feature based algorithm and point based one. However, feature based method can only be used in very special environments with clear geometric structure, while traditional point based method can only obtain a relative coarse estimation and is sensitive to initial alignment. Thus, in this paper, a registration algorithm, called Octree Based Multiresolution Heuristic ICP, is proposed. Without relying on the good initial registration and marked features, hybrid-ICP combines different ICP algorithms, and improve the alignments using finer levels of representation. In our outdoor riverside environments experiments, our method outperform the classical point based registration algorithm with an accuracy of 7 times better than classical Generalized-ICP and a speedup 1.6 times.