Author ORCID Identifier
0000-0001-5862-5239
Date of Award
5-1-2023
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science
First Advisor
Ying Zhu
Second Advisor
Rajshekar Sunderraman
Third Advisor
Greg Smith
Fourth Advisor
Zhisheng Yan
Abstract
Game developers increasingly consider the degree to which character animation emulates facial expressions found in cinema. Employing animators and actors to produce cinematic facial animation by mixing motion capture and hand-crafted animation is labor intensive and therefore expensive. Emotion corpora and neural network controllers have shown promise toward developing autonomous animation that does not rely on motion capture. Previous research and practice in disciplines of Computer Science, Psychology and the Performing Arts have provided frameworks on which to build a workflow toward creating an emotion AI system that can animate the facial mesh of a 3d non-player character deploying a combination of related theories and methods. However, past investigations and their resulting production methods largely ignore the emotion generation systems that have evolved in the performing arts for more than a century. We find very little research that embraces the intellectual process of trained actors as complex collaborators from which to understand and model the training of a neural network for character animation. This investigation demonstrates a workflow design that integrates knowledge from the performing arts and the affective branches of the social and biological sciences. Our workflow begins at the stage of developing and annotating a fictional scenario with actors, to producing a video emotion corpus, to designing training and validating a neural network, to analyzing the emotion data annotation of the corpus and neural network, and finally to determining resemblant behavior of its autonomous animation control of a 3d character facial mesh. The resulting workflow includes a method for the development of a neural network architecture whose initial efficacy as a facial emotion expression simulator has been tested and validated as substantially resemblant to the character behavior developed by a human actor.
DOI
https://doi.org/10.57709/35151307
Recommended Citation
Schiffer, Sheldon, "An Actor-Centric Approach to Facial Animation Control by Neural Networks For Non-Player Characters in Video Games." Dissertation, Georgia State University, 2023.
doi: https://doi.org/10.57709/35151307
File Upload Confirmation
1