Learnt-based Inverse Kinematics For Human Motion Synthesis

Human motion synthesis technology plays an important role in video games, films production, motion simulation and virtual reality filed, it's a one research focus of computer animation field. In recent years, because of the rapid development of motion capture hardware and software, machine learning techniques are widely used in three-dimensional human animation generation, and made a lot of research results.Existing work on human animation synthesis can be roughly split into two approaches, those that combine segment of motion capture data, and those that perform inverse kinematics. In this paper, a learnt inverse kinematics based on minimal set of motion capture data was present, which can perform inverse kinematics on special articulated object. The learnt inverse kinematics model was designed in inverse kinematics question mind, which is a double-layer motion model with the first layer is end position and joint is the second layer. Firstly, we will describe the hierarchical model of the human skeleton and introduced the common human motion synthesis technology and machine learning techniques. Then train the learnt inverse kinematics model by clustering motion capture data. When the motion model has been established, we will introduce how to synthesis new human motion by the learnt inverse kinematics in the end of the paper.Massive simulation examples are carried out on model establishment and motion synthesis and experimental results can show the effectiveness of the method. The method proposed in this paper inherits the flexibility on motion control of inverse kinematics and the newly synthesized motion maintains the style of the sample motion. The main feature of the work is the simplicity of implementation.