基于人体姿态估计的京剧虚拟人物互动系统
Virtual character driving based on pose estimation technology -- a case study of Beijing Opera
投稿时间: 2022/12/20 0:00:00
DOI:
中文关键词: 姿态估计;动作捕捉;单目相机;Unity3D;京剧
英文关键词: pose estimation; motion capture; monocular camera; Unity3D; Beijing Opera
基金项目:
姓名 单位
黄一鸣 中国传媒大学媒体融合与传播国家重点实验室
潘达 中国传媒大学媒体融合与传播国家重点实验室
张宜春 中国艺术科技研究所
点击数:897 下载数:1610
中文摘要:

提出了采用基于计算机视觉的人体动作捕捉技术构建京剧虚拟人物互动系统,首先通过单目相机捕捉京剧表演者的运动画面,经由深度学习姿态估计模型预测24个人体骨骼关键点,再经由滤波处理模块对这些关键点进行平滑操作得到准确的运动信息,最后导入游戏开发引擎Unity3D中,通过将真实人体运动结点和虚拟人物关节点进行绑定,实现驱动虚拟人物仿照真人动作进行京剧表演的舞台效果。实验表明,本互动系统可以实时且准确地估计人体姿态,并能够实现与虚拟人物较好的互动体验感。

英文摘要:

This paper proposes to use the human motion capture technology of computer vision to build a Peking Opera virtual character interaction system. Firstly, the motion pictures of Peking opera performers are captured by monocular camera, and 24 human skeleton key points are predicted by deep learning posture estimation model, Then these key points are smoothed through the filter processing module to obtain accurate motion information. Finally, it is imported into the game development engine unity3d. By binding the real human motion nodes with the virtual character's joint points, the stage effect of driving the virtual Peking opera characters to imitate the real action for Peking opera performance is realized. Experiments show that our proposed interactive system can estimate human posture in real time and accurately, and achieve a better sense of interactive experience with virtual characters.

参考文献: