可程式操控之魚的實體動畫

Abstract

本論文的目標是以真實世界中所拍攝的影片所製作而成的實體動畫 (realistic animation)。此種技術將動畫中的虛擬的角色，以真實的物體取代，希望能藉此提升動畫的真實性。 在本論文中，我們透過影片模式(video mode)與類神經網路模式(neuralnetwork mode)兩種方法的結合，製作一個以魚為主角，可程式控制的實體動畫，在影片模式的部份，我們發現由不同的取樣間隔(sampling interval)可以產生出不同的彎曲效果，藉由改變取樣間隔可找出最接近目標的影片片段，此種模式保留了原始影像的真實性。而在類神經網路模式部份，則是為了要修正在影片模式中無法完全精確的照著規劃的路徑到達目標，經由實驗結果，利用上述兩種方式的結合可使我們對於魚的真實性與可控制性達到很好的平衡。

The goal of this thesis is producing a realistic animation shot from the real world. This kind of technique replaces the virtual character in animation with real object. Hoping it can enhance the reliability of the animation. In this thesis, we combine two methods with video mode and neural network mode, producing a programming controllable realistic animation using fish as the main character. In the part of video mode, with different sampling interval, we find it can produce different winding effect, so we can find the film fragment most close to the target by changing sampling interval. This mode retains the reliability of the original film. And in the part of neural network mode, we use it to correct the path to the right target. With the result of the experiment, we find it has a good balance in reality and controllability using these two methods above.