Research On Data Driven Method For Hand Skeleton Motion Generation

Data driven based method for hand skeleton motion generation is an important research topic in the domain of computer animation and augmented reality.Hand Skeleton motion is also widely used in movies,medical,health,military and so on.Data driven based method can reproduce the motion of a target object with a high fidelity.But data acquisition equipment(DAE),such as data glove and motion capture system are mostly complex and expensive.Taking this factor into consideration,this thesis uses data collected from Kinect and Leap Motion,which are non intrusive and compact data acquisition platforms,to study the data driven method for hand skeleton model motion.The work done are as follows:1?Constructing the skeletal model of the hand.In order to highlight the three-dimensional features of the hand skeleton,this paper employs the OpenGL platform.The three-dimensional spherical and capsule models are constructed by transforming the spherical coordinate system into the Cartesian coordinate system.By referring to the anatomical structure of the hand,I set up the hand skeleton configuration in a reasonable way,hand skeletal motion can be animated by configuring the angles of the hand model.The model has the same degree of freedom as the real hand,and can simulate most of the real hand skeletal motions.2?By employing the data collected from Kinect to explore the estimation of the wrist orientation based on the Euler angle and spherical coordinates.The method based on Euler angle applies Rodrigo rotation formula and 3D graph transformation theorem to get the Euler angle variation between the initial frame and the hand direction vector of the current frame,and then applies the Euler angle change to drive the wrist rotation of the model.The method based on spherical coordinates is used to convert the Cartesian coordinate system into the spherical coordinate system to obtain the angle change between the initial frame and the current hand direction vector.The wrist joint movement is driven by this change.Experiments show that the two methods can reflect the real motion of the hands more accurately when they only consider the influence of the hand direction vector,and the second methods are relatively simple to implement.However,through subsequent analysis,we can see that considering the influence of palm normal vector,the first method is more simple in implementation.3?By employing the data collected from Leap Motion,and discuss an improved method based on key point model.From the original data,the coordinates of the palm joint in the initial frame are estimated,and the method to correct the estimation of the coordinates of the current frame's palm joint coordinates is proposed.According to the structural features of the finger,the fingers are divided into two categories.The attitude of two types of fingers in motion is abstracted into two types of quadrangles and triangles.Using the cosine theorem,these two kinds of models are applied.On the basis of two kinds of mathematical models and prior knowledge,the bending angle of the proximal and distal joints on the finger is estimated.By using the estimated bending angle,the corresponding rotation of the joint is applied to realize the recurrence of the hand attitude.Experiments show that the improved method of key point model has the advantage of quadrangles and triangles.In this thesis,data driven based hand skeleton motion model is studied based on two data sources which are Kinect and Leap Motion.The experimental results show that the Kinect data source can better capture the wrist movement of the real hand,but motions of the fingers often get lost and the data collected contains more noise.Compared with Kinect,Leap Motion data captures more detailed motion of the fingers.Based on the key point model we proposed an improved method that can obtain high fidelity finger movement through LeapMotion data source.