Computational Transient Imaging: Instrumentation, Image Formation Models, Inverse Methods, and Applications

瞬态成像是一个相对年轻的研究方向，主要研究在宏观尺度上对非平稳光分布的时间分辨捕获，或称“飞行中的光视频”。瞬态图像有望使计算机视觉中的许多问题得到更深入的理解和优雅的解决方案，就像脉冲响应在科学和工程中被证明是一种多功能和有价值的工具一样。到目前为止，获取瞬态图像需要非常昂贵和高度敏感的光学设备（超短脉冲激光器和条纹相机），因此保留给少数高度专业化的实验室。拟议的项目是基于我们自己开发的光电相关测量的瞬态图像的重建技术，这些测量可以很容易地从飞行时间测距相机中发现的锁定图像传感器中获得。我们的工作的目标是更好地了解这种替代硬件平台的瞬态成像，并找到以下核心问题的答案：如何可以瞬态图像测量在更高的分辨率，在更短的时间和降低硬件成本？现有的模型和数值方法如何更好地利用瞬态信号固有的数据稀疏性，以获得更可靠的重建？如何反转多路径图像形成模型来解决长期存在的计算机视觉问题？哪些新的应用机会出现天文成像变得广泛可用？通过这个项目，我们将以跨学科的方式解决所有这些问题，并将计算作为关键推动因素。

The relatively young research direction of transient imaging deals with the time-resolved capture of non-stationary light distribution on macroscopic scales, or "videos of light in flight". Transient images are expected to enable a deeper understanding of, and elegant solutions to, many problems in computer vision, just as impulse responses have proven a versatile and valuable tool in the sciences and engineering. To date, the acquisition of transient images required very expensive and highly sensitive optical devices (ultrashort pulse lasers and streak cameras) and thus was reserved to a few highly specialized laboratories. The proposed project is based on our own development of a reconstruction technique of transient images from opto-electronical correlation measurements that are readily available from lock-in image sensors found in time-of-flight range cameras. The goal of our work is to better understand this alternative hardware platform for transient imaging, and to find answers to the following central questions:How can transient images be measured at a higher resolution, in shorter time and at reduced hardware cost? How can existing models and numerical methods make better use of the data sparsity inherent to transient signals, in order to obtain more reliable reconstructions? How can multi-path image formation models be inverted to solve long-standing computer vision problems? Which new application opportunities emerge astransient imaging becomes widely available?With this project, we will address all of these questions in an interdisciplinary way, with computation as the key enabler.