The Research Of Robot Arm Driver System Based On SCM And Kinect

The human-computer interaction and robot arm control system base on human action recognition technology is widely used in medical rehabilitation?virtual reality and industrial?civil domain.The recognition accuracy and precision based on 2D color image is not high due to the environment noises,such as light?complex background?shadow.With the appearance of Kinect,the pattern recognition based on the depth image becomes the mainstream.Due to the depth image can directly represent the 3D features of things and avoid the negative effects from the texture?background and light.It solve the problem of pattern recognition based on 2D image.In this paper,with the demand of real-time robot arm control and based on the Kinect depth image human action recognition technology.Firstly,we obtained the data set through the Kinect hardware,and then we use the color image data and depth image data to calibrate the Kinect color sensor and depth sensor and map the skeleton data to color data.Using the maps,we abstract the features from the skeleton data set.Afterwards,two algorithm based on the bone joints features.The 'drive'has two meanings,one is perform reaction to the recognized human body semantic action command,the other is mapping the joints of the human body to the joints of robot arm,the robot arm can trace the position of the bone joints.For the former,firstly we analysis the human action characters through obtain the data resource and extract the angle features from the bone vector,we also train the classifier with the help of SVM algorithm.Use this classifier to recognize the human body action,In addition,we extract the relative position of human body bone joints,and rapidly obtain the recognition result of single action with the discriminant of threshold.For the latter,we directly map the joints of the arm to the robot arm joints,and we can control the robot arm to imitate the human arm motion by calculating the angle of joints.Finally,with the algorithms,we designed the WPF program and the Arduino SCM program on the windows platform based on the open source SCM and self-designed servo controlled robot arm.We achieve the goal to control the robot arm by using the Kinect to obtain the skeleton data set real-timely.At the end of this thesis,we verified each of those algorithms.Compared with the HMM model and the human body recognition based on 2D image.The algorithms in this paper is better than others in real-time recommendation.And the Kinectcan recognize the action in a speed of 20-25frames per second and in an accuracy of 95%.The algorithms complexity are low.