Multiview registration is an important part of the 3D modeling pipeline and it aims to bring all the partial views of a model in to a common co-ordinate system. In case of availability of redundant overlap area among the partial point clouds, motion averaging provides an efficient solution to the multiview registration problem. The averaging of underlying relative motions is performed in the corresponding Lie-algebra elements of the SE(3) transformation matrices. However, in the presence of outliers in the set of relative motions this method is non-robust. We present a graph-based algorithm to filter out the relative motion outliers before performing motion averaging. The relative motions are assigned weights based on their agreement with global motions and other relative motions. The results indicate that our approach can introduce robustness to the motion averaging method of multiview registration by efficiently filtering out the outliers.