Research On Data-driven Simulation Techniques For Chinese-shadow-play

In recent years,people have been done a lot of research and exploration in the protection of shadow play.In order to alter the present situation of the traditional culture of shadow play,we have done research and exploration in the digital protection of shadow play.In view of the lack of simulation of the overall comprehensive scene of shadow play in the current simulation form,this paper carries out digital modeling and simulation from the comprehensive scene of shadow play.In addition,to solve the problem of interactive fast simulation of shadow animation,a data driven animation generation method is proposed,which extracts motion control elements of video data and generates shadow play animation.First of all,we simulate the motion control of shadow play animation from the point of view of data driven.It mainly includes two parts: digital modeling of the shadow play scene and the extraction of the motion control data.Then the motion control data are quickly applied to the shadow play model to generate shadow play animation.In the digital modeling of the comprehensive scene of the shadow play,we adopt the second generation modeling method.The model,curtain,lighting and fixed the scene of the shadow play are modeled on the whole,and the OGRE engine software is used to render and display.The control data are organized by a general motion data format file.We implement the transformation of the pose trajectory data sequence to the BVH motion data format file.And the motion controls of the model in the rendering engine by the BVH motion data.Secondly,we have done research and Exploration on the extraction of control data.In combination with the characteristics of the shadow play,we have realized the extraction of human posture data from RGB video.Then,the pose data in the video is organized into a motion sequence to control the motion of the shadow play model,and the shadow play model is generated to control the animation.Based on the framework of deep convolution neural network for 2D pose data,this paper proposes a structured learning method for spatio-temporal data in video.In the deep learning network framework,the whole structured learning process is integrated into the network frame structure,while obtaining the more precise data of the single video frame,it also achieves the purpose of maintaining the consistency of the temporal and spatial in the video.After obtaining 2D pose data,this paper completes the extraction of 3D pose data.Combined with the characteristic of the limited motion scene in the shadow play,the 2.5D pose data are constrained.During the constraint process of the 2.5D pose data,the motion tracking compensation and correction of the constraint data is complemented,which makes the motion control data obtained more stable and accurate.