A Kinect Joint Data Processing Algorithm And Its Application In Robot Motion-sensing Control

With the development of robot technology and human-computer interaction technology,motion-sensing control technology with its natural and convenient control experience is playing more and more important role in the process of the development of robot for civil use.Meanwhile,with the development of computer vision technology,robot motion-sensing control technology based on computer vision can reduce the complexity of motion-sensing robot control,so visual sensor has become the major equipment to capture the information for motion-sensing robot control.Kinect is a kind of low-cost visual motion-sensing sensor,it uses the technology of TOF(time of flight)to capture motion-sensing information.Kinect can not only overcome the influence of environment illumination changing and background complexity in traditional visual sensor,but also can provide plentiful 3D information for robot motion-sensing control.So this thesis intends to use Kinect in the system of robot motion-sensing control.To realize this system,this thesis need to focus on solving the following problems.Firstly,because of occlusion problems and the stability of hardware,Kinect maybe generate some inaccurate motion-sensing data.These error data can lead to inaccurate body control.Secondly,the coordinate of Kinect is different from the coordinate of the robot,so we need a series of transformation algorithms to make use of motion-sensing information on robot motion-sensing control.This research is launched based on these two questions.First of all,a Kinect joint optimization algorithm is proposed in this thesis.The algorithm firstly captures human joint position data.Then,by methods of intra-frame constraint and inter-frame constraint,the algorithm calculates the reliability of joint data.Next,this algorithm uses a joint optimization method based on improved Holt filter to deal with joint data.This algorithm deals with inaccurate data according to the prediction formula of Holt filter and human physiological constraints and deals with unoptimized joint data according to the smoothing formula of Holt filter and joints reliability.Finally,this algorithm generates the optimized joint data.The experimental results show that the proposed joint optimization algorithm is obviously effective.Secondly,a coordinate transformation algorithm based on solid geometry method andDenavit-Hartenberg(D-H)matrix method is proposed in this thesis.Solid geometry method transforms human joint position data to human motion angle data.Then D-H matrix method is used to build robot kinematic model and generate robot motion coordinate transformation matrix.Finally,this algorithm combines human motion angle data with robot motion coordinate transformation matrix to generate robot motion data founded on robot coordinate.Last but not least,a Nao robot motion-sensing control platform is realized by the proposed algorithm.The virtual robot platform based on Choregraphe software realizes the motion-sensing control of virtual robot by Kinect.It is a transformation of the theoretical research to practical application.