一种视频数据驱动的水体表面模型生成方法

针对视频数据噪声及计算误差造成重建流体高度场时域跳变的问题, 提出一种视频数据驱动的流体表面模型生成方法。首先, 在折射的流体表面重建算法基础上, 利用水下场景视频数据生成初始流体表面高度场; 其次, 为了提高模型的时间连贯性, 用数据驱动方式获取浅水波模拟关键参数的最优值; 最后改进了浅水波模拟的计算求解过程, 并将其作为物理约束来修正初始模型。基于真实数据的实验结果表明, 该方法能够有效平滑流体模型高度场的跳变, 使水体表面重建结果更加准确和连贯。

A video-data-driven method to generate water surface models

In order to relieve the jump effects of the height fields caused by video noises and computing errors, this paper proposed a video-data-driven method to generate models of fluid surfaces. First, it used shape from refraction method to reconstruct the initial water surface, and set the initial height fields with the reconstruction results. Then, in order to improve the time coherence of the reconstruction sequences, it chose the optimized parameter of shallow water simulation with a data-driven solution. After that, it improved the computational process of the shallow water simulation and regarded it as the physical constraint to refine the initial models. It tested the proposed method on the real data sets, and the experimental results show the effectiveness of proposed method. The simulated fluid surfaces in videos sequences appear more accurate in spatial domain and much visually satisfied in temporal domain.