|
||
In recent years, 3D scene construction has become increasingly popular in movies and gaming. However, it is without doubt that the involved effort is significant, and therefore how to simplify such a process has gradually drawn the attentions from many researchers. More specifically, the construction of a 3D scene consists of two parts: the creation of 3D objects, and their deployment. In general, one possible and popular solution is to reuse previous 3D scene construction results. In this regard, there are least two types of approaches. The first type of approaches place more emphasis on the spatial relationships. In particular, by placing a query box in the current scene and comparing its relationships with other objects under the current scene, a desired object in a previous scene can be retrieved if it shares a similar configuration. However, inappropriate representations of previous spatial relationships may lead to ambiguous or superfluous retrieval results. The Second type of approaches focus on the generation of a single object. A method of such could either start from an initial model and gradually evolve into a more complex/specific one by selecting a similar model in the database, or directly synthesize a new model by a combination of more than one model from the database. This paper proposes a framework that not only integrates the two types of approaches just mentioned, but also unifies the previous two different ways for model construction. In addition, the representation of spatial relationships is further refined so that more accurate retrieval results can be returned, together with a more meaningful classification hierarchy to facilitate the involved interaction for model construction.
Keywords:
Shape Retrieval、Data-Driven 3D Modeling、Shape Synthesis
|
||
Fig 1. The System Flow. |
||
Fig 2. Query results for two different spatial relationships. |
||
|
||
[Paper] [Video] |
|
||
現今三維場景越來越被人們廣泛應用於電影、遊戲等地方,但往往重新建置一個三維場景需要耗費大量的時間成本,因此我們想建置一套系統利用前人建置過後的三維場景資訊,來提供建議資訊給使用者,使使用者能方便且快速的建置三維場景。系統主要分兩部分,第一部分:使用者在一三維場景中劃出所要建置的三維模型區域,系統則會利用此區域附近的物件之間空間關係,與資料庫中各場景中物件間空間關係計算其相似度,利用其相似度自動搜尋出可能的出現的物件類別,並將此類別回傳給使用者,讓使用者自行選擇所需放置的類別,當使用者選取該類別後,系統則會自動產生出該類別之基礎模型,讓使用者進行雕塑此模型。第二部分:雕塑該模型的方式為利用形狀的相似性,由原本簡單的形狀,透過不停得選取與目前形狀相似形狀之物件方式,使得該物件快速的達到使用者想要的形狀,在變形的過程中,使用者也可以利用該模型的部位,在資料庫中找尋可替換之部位進行替換,來達到使用者理想中的形狀,最後再由使用者進一步的雕塑此模型,利用上述流程來達到快速建置三維場景的效果。
關鍵字:
形狀檢索、資料驅動之三維建模、形狀合成
|
||
圖一. 流程結果圖 |
||
圖二. 兩不同場景搜尋結果圖 |
||
|
||
[論文] [影片] |