Investigation of an emotional virtual human modelling method

Abstract

In order to simulate virtual humans more realistically and enable them life-like behaviours, several exploration research on emotion calculation, synthetic perception, and decision making process have been discussed. A series of sub-modules have been designed and simulation results have been presented with discussion.

A visual based synthetic perception system has been proposed in this thesis, which allows virtual humans to detect the surrounding virtual environment through a collision-based synthetic vision system. It enables autonomous virtual humans to change their emotion states according to stimuli in real time. The synthetic perception system also allows virtual humans to remember limited information within their own First-in-first-out short-term virtual memory.

The new emotion generation method includes a novel hierarchical emotion structure and a group of emotion calculation equations, which enables virtual humans to perform emotionally in real-time according to their internal and external factors. Emotion calculation equations used in this research were derived from psychologic emotion measurements. Virtual humans can utilise the information in virtual memory and emotion calculation equations to generate their own numerical emotion states within the hierarchical emotion structure. Those emotion states are important internal references for virtual humans to adopt appropriate behaviours and also key cues for their decision making.

The work introduces a dynamic emotional motion database structure for virtual human modelling. When developing realistic virtual human behaviours, lots of subjects were motion-captured whilst performing emotional motions with or without intent. The captured motions were endowed to virtual characters and implemented in different virtual scenarios to help evoke and verify design ideas, possible consequences of simulation (such as fire evacuation).

This work also introduced simple heuristics theory into decision making process in order to make the virtual human’s decision making more like real human. Emotion values are proposed as a group of the key cues for decision making under the simple heuristic structures. A data interface which connects the emotion calculation and the decision making structure together has also been designed for the simulation system.