云演艺智能虚拟拍摄对于降低拍摄专业门槛和空间人力等成本具有重要意义。本文提出了以机位计算为核心、美学评估为辅助的基于数据驱动的智能拍摄仿真技术，构建虚拟环境下的云演艺智能拍摄系统。首先提取真实视频中的人物关节点信息，然后判断主要演员并构建复曲面坐标系。接着用特征估计网络提取出相机和人物特征，通过相机运动提取门控网络和相机轨迹预测网络计算得到相机的运动参数。然后在Unity3D环境下，控制虚拟相机并生成相机轨迹和相应画面。，再通过美学评价模型对画面进行整体美学评分和主要构图模式判断，选取较优机位。最终，用户可以在小程序端得到较优机位下的运镜轨迹和虚拟环境的仿真画面视频。以中国传媒大学礼堂和威海荣成西霞口剧场的虚拟场景为例展示了智能拍摄系统的效果。

Virtual automatic cinematography is of great importance to lower the professional threshold for filming and reduce space cost and labor cost of shooting. In this paper it was proposed to use data-driven automatic cinematography technology with aesthetic assessment technology to build an automatic cinematography system for cloud performing arts. Firstly, extracted character joint point in the real video, and then the main actorswere judged and the toric space was constructed. Then the feature estimation network was used to extract the camera features and character features, and the motion parameters of the camera were calculated through camera motion extraction gating network and camera trajectory prediction network. After that, these parameters controlled the virtual camera in Unity3D to generate camera trajectory and video. Then through the aesthetic assessment model, the overall aesthetic score of the picture was marked and the main composition pattern was judged. And the system selected the camera position with the best performance. Finally, the user could get the track of camera and the video captured by the chosen camera on Wechat MiniProgram. The virtual environment based on the auditorium of Communication University of China and Xixiakou Theater of Weihai, Rongcheng was taken as an example to demonstrate the effect of the intelligent shooting system.