This document proposes a new approach called temporal frequency probing to analyze light transport using time-of-flight (ToF) cameras and projectors. By projecting patterns at different modulation frequencies and capturing the returning light with a ToF camera, it can separate direct from indirect light transport and further separate caustic from diffuse indirect light. This allows it to reconstruct high temporal resolution "light-in-flight" videos depicting the propagation of light over time within a scene. The approach leverages decades of research on light transport analysis and provides a new paradigm for studying transient 5D light transport using commercially available ToF sensors and projectors.
Temporal Frequency Probing for 5D Transient Analysis of Global Light Transport
1. Temporal Frequency Probing for
5D Transient Analysis of
Global Light Transport
Matt O’Toole1 Felix Heide2 Lei Xiao2 Matthias Hullin3 Wolfgang Heidrich2,4 Kyros Kutulakos1
1University of Toronto 2University of British Columbia 3University of Bonn 4KAUST
http://www.dgp.toronto.edu/~motoole/temporalprobing.html
2. the time-of-flight (ToF) sensing revolution
estimating poses
[Schwarz et al. 11]
Kinect for Xbox One
femto-photography
[Velten et al. 2013]
looking-around-corners
[Velten et al. 2012]
gesture recognition
[Droeschel et al. 11]
Google’s self-driving car
3. temporal probing for scene analysis
capture depth
robust to indirect light
reconstruct “light-in-flight” video
with high temporal resolution
key insight: complex-valued image formation model
4. contributions
the transient-frequency transport matrix
unified theory for light transport analysis
analyze ToF images using classical techniques
ToF imaging using projectors & masks
capture depth robust to indirect light
reconstruct “light-in-flight” video with high temporal resolution
28. Techniques Reference(s)
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
29. Techniques Reference(s) Conventional Analysis
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
30. Techniques Reference(s) Time-of-Flight Analysis
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
31. Techniques Reference(s) Time-of-Flight Analysis
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
32. Techniques Reference(s) Time-of-Flight Analysis
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
33. Techniques Reference(s) Time-of-Flight Analysis
transport equation [Ng et al. 03]
dual photography [Sen et al. 05; Sen and Darabi 09]
radiometric compensation [Wetzstein and Bimber 07]
radiosity equation [Goral et al. 84]
radiosity solution [Goral et al. 84]
inverse light transport [Seitz et al. 05; Bai et al. 10]
transport eigenvectors [O’Toole and Kutulakos 10]
structured light transport [O’Toole et al. CVPR 2014]
fast direct/global
separation
[Nayar et al. SIGGRAPH 2006]
34. structured light transport [O’Toole et al. CVPR 2014]
acquire direct & indirect ToF images using epipolar constraints
fast direct/global separation [Nayar et al. SIGGRAPH 2006]
acquire caustic indirect & diffuse indirect ToF images using radiometric constraints
35. structured light transport [O’Toole et al. CVPR 2014]
acquire direct & indirect ToF images using epipolar constraints
fast direct/global separation [Nayar et al. SIGGRAPH 2006]
acquire caustic indirect & diffuse indirect ToF images using radiometric constraints
42. structured light transport for freq.
mirror
projection
pattern
mask
pattern
1. open electronic shutter
2. for i = 1 to N
use random epipolar mask &
project complementary pattern
3. close electronic shutter
ToF camera ToF projector
53. structured light transport [O’Toole et al. CVPR 2014]
acquire direct & indirect ToF images using epipolar constraints
fast direct/global separation [Nayar et al. SIGGRAPH 2006]
acquire caustic indirect & diffuse indirect ToF images using radiometric constraints
54. effect of indirect transport on spatial frequencies
mirror
ToF camera ToF projector
55. effect of indirect transport on spatial frequencies
mirror
caustic high frequency
non-caustic ToF camera low frequency ToF projector
56. effect of indirect transport on spatial frequencies
mirror
caustic high frequency
non-caustic low frequency
ToF camera ToF projector
57. effect of indirect transport on spatial frequencies
mirror
1. for i = 1 to N
project high spatial-freq. pattern
capture image
2. non-caustic = min of images
caustic high frequency
non-caustic low frequency
ToF camera ToF projector
64. “light-in-flight” imaging: prior work
femto-photography
[Velten et al. 2013]
very high cost
well-resolved wavefronts
PMD-based system
[Heide et al. 2013]
many modulation frequencies
strong scene priors
optical wavefronts not resolvable
our work
many modulation frequencies
weaker, transport-specific priors
resolvable optical wavefronts
67. piecewise “light-in-flight” video construction
caustic ToF for = 100 MHz light-in-flight video
direct (no regularization)
caustic indirect (no regularization)
68. piecewise “light-in-flight” video construction
non-caustic ToF for = 12 to 140 MHz light-in-flight video
direct (no regularization)
caustic indirect (no regularization)
non-caustic indirect (regularized
using [Heide et al. 2013])
69. comparison to [Heide et al. 2013]
light-in-flight video [Heide et al. light-in-flight video (ours)
13]
70. comparison to [Heide et al. 2013]
light-in-flight video [Heide et al. light-in-flight video (ours)
13]
71. conclusion
• a new paradigm for ToF sensing
• combines ToF sensors with projectors & masks
• can now leverage from decade of light transport research
• widespread implications for spatio-temporal transport analysis
72. come visit us at our E-Tech booth
visualizing light transport phenomena
73. Temporal Frequency Probing for
5D Transient Analysis of
Global Light Transport
Matt O’Toole1 Felix Heide2 Lei Xiao2 Matthias Hullin3 Wolfgang Heidrich2,4 Kyros Kutulakos1
1University of Toronto 2University of British Columbia 3University of Bonn 4KAUST
http://www.dgp.toronto.edu/~motoole/temporalprobing.html
We do this by building the light-in-flight video one component at a time. We only need a single direct time-of-flight image to recover the light-in-flight component for direct paths, as demonstrated here. Similarly, a single caustic image is sufficient to recover the temporal contribution of caustic light paths. For non-caustic indirect light paths, we capture time-of-flight images at 65 different frequencies and fit a superposition of gaussians and exponentials to the data using Heide et al.’s approach.