Abstract
Human motion synthesis has been an important multidisciplinary research area across computer animation, biomechanics, orthopedics, rehabilitation, bioengineering, ergonomics, etc. The human motion of daily activity is a complex task of movement coordination involving many body parts and joints. Even to generate the simple motion such as raising the arm or getting up from the chair requires complex modeling and computation. Perceptron types of neural networks were used in this study to generalize the movement coordination of a daily human activity, the lifting motion. With only a limited amount of training data, reasonable patterns of the motion could be achieved by the network. The motion patterns were then fitted with Hermite polynomials and initiated in an optimization model to predict the entire motion trajectories. The paper presents the method combining Neural networks and optimization for the generation of human motion. ©2008 IEEE.