摘要
针对虚拟拍摄子弹时间系统存在不连续感、无法扣除背景、系统不稳定和设备成本昂贵等问题,提出一种基于人工智能的子弹时间虚拟拍摄方法。方法采用廉价的网络摄像头阵列,构建拍摄系统,采用自定义的标定和数据传输方案,采用嵌入式设备作为数据中转。对于输出结果的后期处理,采用预训练的神经网络辅助完成后期画面的拼接处理和优化。借助这一方法,研制出一种简单易行的子弹时间虚拟拍摄系统。实验表明,上述系统解决了子弹时间视频的不连续感和扣除背景问题,保证了系统稳定性和输出结果的可用性,并且,大幅度降低了构建虚拟拍摄子弹时间系统的成本。
Aiming at the problems of discontinuity sense,inability to deduct background,system instability and ex-pensive equipment cost of bullet time virtual system,we propose an artificial intelligence-based bullet time system.Our system employs an inexpensive webcam array to construct the shooting system,adopts a customized calibration and data transmission scheme,and employs embedded devices as data relays.For the post-processing of the output,a pre-trained neural network is used to assist in completing frame stitching and optimization work.Experiments show that the system solves the problem of discontinuity sense and deduction of background in bullet time video,ensures the stability of the system and the usability of the output,and,drastically reduces the cost of building such systems.
作者
顾乃林
王锐
GU Nai-lin;WANG Rui(School of Economics and Management,Suqian University,Suqian Jiangsu 223800,China;Beijing ManaVR Company Limited,Beijing 100123,China)
出处
《计算机仿真》
2024年第11期481-485,共5页
Computer Simulation
关键词
子弹时间
摄像头阵列
背景扣除
人工智能
Bullet time
Camera array
Background deduction
Artificial intelligence
