Motion Level-of-Detail: A Simplification Method on Crowd Scene

Recent technological improvement in character animation has increased the number of characters that can appear in a virtual scene. Besides, skeletal and mesh structures are expected to be more complex in the future. Therefore, simulating massive characters' joints in a real-time crowd environment without any preprocessing is unaffordable. We propose a preprocessing method called 'motion level-of-detail' to overcome this limitation. Our 'motion level-of-detail' framework not only minimizes the simulation cost of the joints, but also maintains the similarity between the original and the simplified motion. 'Joint posture clustering (JPC)', which is the skeletal simplification method of our framework, reduces skeletal node by the clusters of similar postures. A cluster is a set of continuous frames, where each frame has similar posture. Because our approach depends on motion trajectory, simplified result preserves the quality of the motion. We also applied a geometric simplification on deformable character mesh, to increase performance. Our approach was particularly useful for the complex skeletal motions that have a monotonous trajectory.