Virtual Posture Recognition Research Based On Kinect Sensor

In recent years, Human-Computer Interaction(HCI) applications in virtual environment have been developing under the technology and market demands, which involve computer vision, sensor technology, artificial intelligence, pattern recognition and so on. Meanwhile, it has been widely used in the fields of motion sensing game, virtual classes, online shopping and military training. Many studies based on virtual human posture recognition help interaction systems more natural, so it is more significant to make the researches of virtual human posture recognition in virtual environment. But there are still some problems about the interaction systems, like high costs of motion capture devices, undesirable recognition rate and so on. It follows that the study based on virtual human posture recognition of Kinect is very meaningful.In this thesis, combined with synthesize motion-sensor technology, virtual reality,behavioral science and other comprehensive knowledge, the research is about human motion posture recognition and interaction method,with the goal of establishing a virtual environment interaction. And on the top of the domain of virtual room, a natural interaction method based on somatosensory technologies in virtual environment is formed.The main works are summarized as follows:1. the research presents the architecture of virtual environment interaction based on virtual human posture recognition. Then the thesis proposes framework of virtual human posture recognition in virtual environment based on Kinect. The analysis is conducted on operating principle of Kinect, virtual character model and the design of VR scenes and other key technologies.2. The key technologies of depth information acquisition and human skeleton tracking were studied, then the method drive the avatar real time using the preprocessed data based on motion retargeting in inverse kinematics. On top of this, the real-time multiple avatar interaction is presented. At last, we model agent behaviors to achieve the interaction in sync between agents and users.3. The thesis focuses on recognition method of virtual posture. First, creating the human structure vectors by human skeleton joints, then selecting the combination of angles and module ratios between vectors to establish the body posture description vector, Finally, Multi-class Support Vector Machine based on Gaussian kernel function and the improved posture recognition based joint angle are used in our experiments. The experimental result shows that the proposed body posture recognition based Support Vector Machine has higher recognition rate?4. A somatosensory interaction system in Virtual Home based on Kinect posture recognition is designed for the demand of natural human-computer interaction applications based on Virtual Home. This system defines the Virtual Home interactive posture. In addition, the accuracy and effectiveness can be achieved by testing and verifying the proposed method in a typical interactive scenario.