Eye on the Prize : Enhancing Realism during Interaction towards Non-Player-Characters with Natural Eye Movements

Abstract: In the wake of motion capture and visual animation one aspect seems to be lacking. A realistic representation of the dynamic and unpredictable human visual perception. A human examines her surroundings in a unpredictable, saliency based and top-down task oriented manner. In the field of computer science, interaction design and the industry of game development, great leaps have been taken when it comes to capture motion of bodies. Motion capture helps developers and film makers to portray realistic humans in virtual environments. Where motion has come far, eyes and perception has not. As of yet a virtual representation of human visual perception, has not been mimicked as close as body motion. This thesis will examine perceived realism in virtual agents, with a focus on eye motion. In this study a virtual agent has been given eye movements of human beings and been compared to an agent based on current virtual agents in games. This is the first step towards synthesizing more than just human motion in virtual agents. It will provide future research with the data and tools needed to produce an algorithm based on the gathered data. Prior context research includes a study of current games. Two participant experiments have been be conducted, both has recorded eye positional data for analysis. The first experiment helps build the second as it compares virtual agents using a Likert scale for a subjective rating of realism. The results offers some very interesting data, indeed data that lie at the core of the study as well as data for further studies. While statistical analyses of Likert scales might be considered ambiguous this study has done so and reached a conclusion. A virtual agent enhanced with eye-motion based on human eye movements does portray a more realistic human like behaviour.