Motion Capture And Optimization Of Virtual Robot Based On Kinect

Humanoid robot has been getting widespread attention for its shape and imitative ability to human action.All countries are devoted to the study of humanoid robot in various fields of application.The motion control of humanoid robot is a very complicated work.How to solve this problem effectively is one of the most important research subjects.The experiment is to control the motion of virtual human and robot by somatosensory interaction technology,so that the real robot is able to simulate human motion in real time.This way also realizes the natural human-computer interaction simultaneously.The main contents and innovations of this thesis are as follows:Firstly,the depth data obtained by Kinect is used to generate the depth image sample library of human action.In order to ensure the diversity of samples,we obtain and save the image in real time,and then select the images which contain different people and different action types as samples.The sample library basically contains the commonly used human body action.Next,the feature of image pixels is extracted according to the depth information acquired by Kinect,then the human body parts are classified by the random forest.Random forest classifier is formed by training features' threshold.Then meanshift algorithm is used to obtain the joints of human body parts so as to generate three-dimensional space human skeleton.Features extracted in this thesis can not only meet the performance requirements but also easily distinguish different body parts.The recognition rate of the random forest to the trained samples is 70-95 percent,and the classification time of a single image reaches the millisecond level,which meets the real-time requirement.Finally,this thesis proposes a new method to calculate the motion information of human skeleton,which is called the space vector method,and then the human motion information is transmitted to the robot in real time through the network communication so as to control the robot accomplishing the same action.The space vector method is applicable to the case of limited robot arm structure,if the robot arm structure is more flexible,the Euler angle formula derived in this thesis can be used to calculate joint movements,and the Euler formula is verified by the experiment of virtual human motion.The experimental results indicate that it is feasible to control the motion of robot by using somatosensory interaction technology.Compared with other motion control methods,this approach can better embody the nature of human-computer interaction.