Asada and Tsuji 1983 : A framework that finds a meaningful representation of rigid and joint motion, given correspondences of points on each object in an image sequence is described. Objects in the input scene show complex motion such that an object rotates around an axis attached to another moving and rotating object. A geometrical model of rotation is introduced to explore relationships between an apparent rotation is introduced to explore relationships between an apparent rotation vector and actual rotations around multiple physical axes. Analysis of the rotational vectors mapped on a Gaussian sphere yields useful information such as a number of the multiple axes and their orientations. The structure of jointly moving objects is found by using these motion cues and is represented in a hierarchical form. In order to find a compact representation of the dynamic scene, the image sequence is partitioned into a small number of subsequences. Properties of translational components are useful for the segmentation; however, their representation is not unique but two degrees of freedom are left. The freedom is used to make the motion representation simple and natural. Each short period in the image sequence is described as a constant velocity or a constant acceleration movement, and then a clustering method is used for the segmentation of the input sequence.