3D街頭立體藝術影像互動式生成系統

Abstract

近幾年來，3D街頭錯視覺藝術深受人們歡迎，我們可以在街上、人行道上、廣場中看到此類藝術作品。同時該藝術也被稱為3D立體粉筆畫，這類畫作本身就是以二維平面形式被畫在街道上，但若在某一特定透視角度下去觀賞，卻能看見畫作以3D立體樣貌來呈現錯視覺之效果。在此，我們提出一個互動式系統，將CG模型場景置入現實照片影像中來模擬3D街頭藝術，使用者無需真的到該場景中去作畫。首先我們透過相機校正來估測該場景中，相機當時拍攝的位置；接著利用現有的NPR技術，來給予我們CG模型圖增添藝術風格化。最後使用Poisson Blending將風格化後之模型圖與原始照片場景圖合併，此方法同時能將場景地面材質細節反映在模型圖上。透過簡單的互動調整，使用者能得到非常生動有趣且相似於真實3D街頭立體藝術之結果圖。

3D Street Art has become more popular in recent years. Many 3D illusion artworks are shown on pavement such as streets, sidewalks, and town squares. It’s often known as 3D chalk art which a 2D artwork drawn on the street itself giving the viewer a 3D optical illusion from a certain perspective. This thesis presents an editing system to synthesis computer graphic models into a realistic photograph without requiring access to the scene to draw 3D street paintings on the ground. First the camera position was recovered using camera calibration algorithm. Second, a NPR stylization technique was applied to create artistic image of 3D models with the same camera position. Finally, the NPR image was composited into source photograph using Poisson approach. All enhanced pavement texture details were also blended on object image. With a small amount of specifications and adjustments, user can create a lively and interesting result which likes those realistic 3D art illusions.