ECHO

The automated analysis of visual data is a key enabler for industrial and consumer technologies and of immense economic and social importance. Its main challenge is in the inherent ambiguity of images due to the very mechanism of image capture: light reaching a pixel on different paths or at different times is mixed irreversibly. Consequently, even after decades of extensive research, problems like deblurring or descattering, geometry/material estimation or motion tracking are still largely unsolved and will remain so in the foreseeable future.

Transient imaging (TI) tackles this problem by recording ultrafast optical echoes that unmix light contributions by the total pathlength. So far, TI used to require high-end measurement setups. By introducing computational TI (CTI), we paved the way for a lightweight capture of transient data using consumer hardware. We showed the potential of CTI in scenarios like robust range measurement, descattering and imaging of objects outside the line of sight – tasks that had been considered difficult to impossible so far. The ECHO project is rooted in computer graphics and computational imaging. In it, we will overcome the practical limitations that are hampering a large-scale deployment of TI: the time required for data capture and to reconstruct the desired information, both in the order of seconds to minutes, a lack of dedicated image priors and of quality guarantees for the reconstruction, the limited accuracy and performance of forward models and the lack of ground-truth data and benchmark methods. Over the course of ECHO, we will pioneer advanced capture setups and strategies, formation models, priors and numerical methods, for the first time enabling real-time reconstruction and analysis of transient light transport in complex and dynamic scenes. The methodology developed in this far-reaching project will turn TI from a research technology into a family of practical tools that will immediately benefit many applications.

ECHO is funded by ERC Starting Grant 802192 (funding period 12/2018 – 11/2023). The ECHO project is funded by the European Research Council.

Project Team

Prof. Dr.-Ing. Matthias B. Hullin


Principal Investigator
E-mail: hullin@…
Phone: +49 228 73-54169

Clara Callenberg


PhD Researcher
E-mail: callenbe@…
Phone: +49 228 73-60636

M.Sc. Javier Grau Chopite


PhD Researcher
E-mail:
Phone: +49 228 73-60642

M.Sc. Weizhen Huang


PhD Researcher
E-mail: whuang@…
Phone: +49 228 73-60740

M.Sc. Markus Plack


PhD Researcher
E-mail: mplack@…
Phone: +49 228 73-4546

Patrick Hähn


Student Helper

Former members / graduates

Dr. rer. nat. Jonathan Klein


PhD Researcher
E-mail: kleinj@…

Dr. rer. nat. Sebastian Thiel (né Werner)


PhD Researcher. Now at Max Planck Institute for Radio Astronomy.
E-mail: werners@…

Dr. rer. nat. Julian Iseringhausen


PhD Researcher / Postdoc (now at Google Research)
E-mail:

Candice Ottersbach

Bachelor’s Student

Publications

Super-Resolution Time-Resolved Imaging Using Computational Sensor Fusion

Clara Callenberg, Ashley Lyons, Dennis den Brok, Areeba Fatima, Alejandro Turpin, Vytautas Zickus, Laura M. Machesky, Jamie A. Whitelaw, Daniele Faccio, Matthias B. Hullin
Scientific Reports (Nature Publishing Group) 11, 1689 (2021), https://doi.org/10.1038/s41598-021-81159-x, 2021. Super-Resolution Time-Resolved Imaging Using Computational Sensor Fusion

Low-Cost SPAD Sensing for Non-Line-Of-Sight Tracking, Material Classification and Depth Imaging

Clara Callenberg, Zheng Shi, Felix Heide, Matthias B. Hullin
ACM Transactions on Graphics 40 (4), Article 61 (Proc. SIGGRAPH 2021), 2021. Low-Cost SPAD Sensing for Non-Line-Of-Sight Tracking, Material Classification and Depth Imaging

Non-Line-of-Sight Reconstruction using Efficient Transient Rendering

Julian Iseringhausen, Matthias B. Hullin
ACM Transactions on Graphics 39 (1), 2020. Non-Line-of-Sight Reconstruction using Efficient Transient Rendering

In this paper, we present an efficient renderer for three-bounce indirect transient light transport, and use it to reconstruct objects around corners to unprecedented accuracy.

Computational Parquetry: Fabricated Style Transfer with Wood Pixels

Julian Iseringhausen, Michael Weinmann, Weizhen Huang, Matthias B. Hullin
ACM Transactions on Graphics 39 (2), 2020. Computational Parquetry: Fabricated Style Transfer with Wood Pixels

A new computational woodworking technique enabled by analysis of features found in natural materials.

Chemomechanical Simulation of Soap Film Flow on Spherical Bubbles

Weizhen Huang, Julian Iseringhausen, Tom Kneiphof, Ziyin Qu, Chenfanfu Jiang, Matthias B. Hullin
ACM Transactions on Graphics 39 (4) (Proc. SIGGRAPH), 2020. Chemomechanical Simulation of Soap Film Flow on Spherical Bubbles

A framework for simulating the intricate flow within spherical soap films.

Deep Non-Line-of-Sight Reconstruction

Javier Grau Chopite, Matthias B. Hullin, Michael Wand, Julian Iseringhausen
Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020. Deep Non-Line-of-Sight Reconstruction

The first deep-learning framework for reconstructing object shapes around a corner.

Trigonometric moments for editable structured light range finding

Sebastian Werner, Julian Iseringhausen, Clara Callenberg, Matthias B. Hullin
Proc. Vision, Modeling and Visualization, Rostock, Germany , 2019. Trigonometric moments for editable structured light range finding

We enhance existing structured light phase shifting methods by using trigonometric moments.

A Quantitative Platform for Non-Line-of-Sight Imaging Problems

Jonathan Klein, Martin Laurenzis, Dominik L. Michels, Matthias B. Hullin
In Proceedings of British Machine Vision Conference (BMVC 2018), Northumbria University, Newcastle, UK, September 3-6, 2018, 2018. A Quantitative Platform for Non-Line-of-Sight Imaging Problems

In this paper, we present a reference database of time-resolved light echoes for non-line-of-sight sensing.