System Flow
Source Skeleton: Human

Target Motion Skeleton: Walking Panther

Finding the Correspondence Between Three-Dimensional Skeleton Structural Variations 〔兩相異三維骨架間的結構對應〕

Skeleton correspondence occupies a crucial position in motion retargeting and related animation procedure due to the laborious process of obtaining the appropriate mapping relationship between source and target skeletons. Conventional approaches focused on the objects of similar structure, such as two creatures of same biological species. Hence, we proposed a system to find the correspondence between three-dimensional skeleton structural variations, considered only the structural and topological differences. The proposed system dealt with the strenuous process of mapping skeleton nodes and body-parts in similar and disparate skeleton structural objects by determining the best corresponding skeleton information between two objects with a three-tier mapping algorithm. We evaluated the proposed system through displaying the corresponding node pair of different sets of skeleton structural variations, and performing motion retargeting based on the corresponding result. We also assembled a quantifiable term, Mapped Target Node rate, to describe the mapping result of the node correspondence alongside the animation outcome.

SUMMARY (中文總結):
骨架對應對於動作轉移以及相關製作動畫流程來說都是很重要的一環。 傳統尋找對應的做法多為同屬一種生物的兩骨架來設計,因此,嘗試在兩不同種生物的骨架間尋找對應關係的過程是相當耗時且需要經驗的。 為此,我們提出一個為尋找兩相異三維骨架間結構對應的系統。 本系統注重兩骨架間的結構及拓譜的差異,而不受外觀和骨架所屬物種的影響。 本系統藉由一個三層架構的演算法來找到兩結構相異骨架之間骨架點和身體部位的對應,以減輕手動對應所會遇到的困難。 實驗結果部份,我們除了直接展示骨架點對應的靜態結果外,也利用該對應結果應用於動作轉移實驗中。 我們也提供對應點數比例,有利於以量化數值來輔助評斷骨架對應及動作轉移的結果。

PROJECT MATERIAL: (picture gallery, video, software demo, talk slides, etc.)