Modeling Living Systems for Computer Vision

Abstract

This paper presents research spanning the fields of computer vision, computer graphics, and artificial life, with implications for AI. Although the modeling of all aspects of living systems is a worthwhile endeavor in its own right, the emphasis here will be on the modeling of animals, including humans, for computer vision. First, I present a new breed of artificial animals in a physics-based virtual marine world, whose muscle-actuated bodies harbor brains with motor, perception, behavior, and learning centers. In these mobile autonomous agents, sensorimotor control for the purposes of perceptuallyguided navigation employs on-board, active computer vision systems that continually analyze the visual world. Second, turning my attention to human animals, I describe new algorithms that can construct artificial human heads with expressive faces. Through the use of range scanners, generic biomechanical facial models may be automatically personalized to individuals. Currently, artificial faces support a model-based approach to facial image analysis. In the future, it should be possible to incorporate brains and some degree of intelligence into them as well.