Research Of Motion Design Based On Kinect Of Virtual Robot

In recent years, the robot began to walk into people’s daily life, and it is changing people’s life. But the human-computer interaction between robot and human is not developed correspondingly. The current human-robot interaction way is still the traditional approaches, such as mouse, keyboard, remote controller, single and not flexible, it is difficult to meet the growing needs of the robot, human-computer interaction in a certain extent restricts the development of robots.As a digital form of real robot, virtual robot has the same movement way as the real robot, it has unique advantages in the robot simulation. In this thesis, we analyze the design and control of real robot’s movement through the research of virtual robot’s human-computer interaction. Which relates to the problem of capturing the motion data, the common motion capture methods includes mechanical way, optical way, acoustic way, electromagnetic way and so on, however these methods are costly, high environmental requirements, so they are not convenient. Therefore, this thesis proposes a new method to extract the motion data which uses the depth image. In this thesis, we use the kinect device to obtain the depth image, which contains the three-dimensional information, and is not affected by the environmental factors such as illumination. The main work and content of this thesis is mainly focused on the following aspects:Firstly, we processes the depth image, we use the background subtraction method to extract the foreground depth image of the moving body, as the background is fixed. For the depth image of the human body, we use the median filter for noise reduction and edge smoothing, laying the foundation for the subsequent processing.Secondly, this thesis puts forward a new motion data acquisition method based on the depth image, which is calculating the movement of each joint according to the skeleton extraction of the human movement depth image with 3D information and positioning joints through the skeleton, and the method is proven to be effective.Again, in view of the skeleton extraction, this thesis combined two methods, the thinning method of the skeleton extraction and distance method of skeleton extraction, and got the advantages of the two. In the occlusion problem encountered in the skeleton extraction, this thesis proposes using the depth information to extract the occluded parts and combine with the Hough Transformation and the initial human body calibration results to repair the movement skeleton. For the burr problem, this thesis proposes a fast and effective method to solve the problem according to the method of connecting length and slope. And the experimental analysis and comparison of the above methods are carried out.Then, for the joint point positioning, we give a solution based on initial body calibration, which solves the sinking of shoulder joints. For the position of the joint point, this thesis introduces the calculation method and process of the motion data of the joint, and makes the interpolation of the motion between the two frames to smoothing movement. And we propose a simple gesture matching based on joint rotation angle.Finally, we designed and implemented the virtual robot motion control system based on kinect. In this thesis, we use the Unity3 D game development engine to use the obtained data of motion to achieve the system, and did a comprehensive test.