Research On Clothing Animation Method With Realistic Visual Perception

Clothing animation is an important multi-disciplinary technology such as computer vision,computer graphics,virtual reality and so on.It aims to simulate the clothing movement with realistic visual effect by computer,and is widely used in large-scale three-dimensional games,CG film production,virtual fitting and other fields.Because in the clothing animation simulation,each frame needs to calculate a large number of point-line-surface dynamic equations contained in the three-dimensional clothing model,and deal with a large number of collision detection tasks.Therefore,although high-resolution clothing animation can show more detailed clothing folds,but the calculation cost is high.Based on this,this paper studies an adaptive clothing animation method,focusing on the research of clothing visual salience model and spatiotemporal smooth constraint model,and combining it with the underlying dynamics driven clothing animation to realize the adaptive multi precision clothing animation with realistic visual perception.The main achievements of this paper are as follows:(1)In this paper,a visual saliency prediction model of clothing animation based on visual perception is studied.The visual attention mechanism reflects that the human eye mainly focuses on a few regions of interest and ignores most regions of non interest.This paper combines the eye selective attention rule with the clothing animation technology,and studies the visual saliency prediction model in the clothing animation scene,so as to improve the efficiency of animation simulation while preserving the fidelity of the model as much as possible.Firstly,eye tracking technology is used to construct visual saliency map of video frames,and spatial saliency features are generated based on full convolution neural network;Secondly,the motion field information between adjacent video frames is extracted by FlowNet as temporal saliency features of video frames;finally,spatial-temporal saliency features are fused as final saliency map;(2)An adaptive dynamic mesh reconstruction technique based on spatio-temporal smoothing is studied.When using saliency constraint to directly control the mesh resolution,it may lead to the instantaneous and rapid change of different mesh resolution,and then lead to the rigid effect of clothing deformation.In addition,due to the movement of the camera viewpoint,the mesh visual state changes from a low-resolution invisible state to a high-resolution visible state,which will reduce the continuity of the simulation process.In order to solve this problem,this paper makes a spatial smoothing of the mesh resolution based on the saliency factor,which uniformly transits the mesh resolution from the high salient region to the low salient region in the visible field of view,and based on the camera viewpoint movement,the time smoothing processing based on saliency factor is carried out,which makes the mesh resolution steadily and continuously refined from the current coarse mesh state over time,and gradually becomes a high-resolution mesh.(3)In this paper,a method of generating adaptive clothing animation with realistic visual perception is studied.Firstly,the clothing saliency prediction model is used to predict the saliency of the animation frame,and the saliency information of the vertices of the clothing model is obtained according to the camera model.Then the spatio-temporal variation and dynamics-driven deformation are combined to carry out multi-resolution mesh dynamic reconstruction.The high-resolution mesh is used in the high saliency region of the clothing model,and the coarse-resolution mesh is used in the low saliency area and the invisible area of the back.According to the saliency information,camera viewpoint information and the underlying drive of clothing animation,a realistic and stable clothing animation based on visual saliency is constructed.