This paper presents a registration approach totackle the problem of rigid alignment of local data acquiredfrom leapfrog locations along elongated structures such aspipelines or other types of pipes. As cylindrical geometrydoes not have enough constraints for alignment, classicalregistration algorithms fail to converge since a sliding effectcan occur between the views to be aligned. We propose todecompose the transformation matrix into its rotational andtranslational components and optimize at first the rotationand then the translation parameters. Experimental resultsand comparison with other methods show the validity of theproposed framework.