dtof op• Data kinds: counts → measurement
• Call: import photoncount; photoncount.dtof_depth(hist, bin_ps=100.0, mode='peak', offset_ps=0.0, subtract_background=False) (or opsphoton.get("dtof_depth"))
Distance from a photon arrival-time histogram: `d = c*t/2`.
Direct time-of-flight. The light travels to the target and back, so the
one-way distance is half the round-trip time times the speed of light.
*bin_ps* is the width of one time bin (a 100 ps bin is 1.50 cm of depth).
Four estimators, from crudest to sharpest:
• `"peak"` — the centre of the fullest bin. Quantised to the bin grid;
the error is uniform in `+-half a bin (+-0.75 cm` at 100 ps).
• `"centroid"` — the first moment of the whole histogram. Exact for a
symmetric pulse *with no background*, and badly biased toward the middle
of the window with one — pass `subtract_background=True`.
• `"parabolic"` — a parabola through the peak bin and its two neighbours.
Sub-bin, cheap, and biased for a Gaussian pulse.
• `"gaussian"` — the same parabola fitted to the log of those three
samples, which is the exact vertex for a Gaussian pulse.
Measured on a noiseless simulated return at 2.4371 m (256 bins x 100 ps,
500 ps IRF), absolute error: `peak 1.29 mm, centroid` 4.4e-16 m,
`parabolic 0.067 mm, gaussian` 9.4e-9 m — three orders of magnitude
between the crudest and the sharpest.
With Poisson noise (200 signal + 200 ambient photons, seed 0) the same
four give 13.7 mm, 146.5 mm (with `subtract_background=True`), 8.5 mm and
8.0 mm. Two honest readings of that: once shot noise dominates the sub-bin
estimators buy about 1.6x, not three orders of magnitude, and the centroid
collapses because a median-subtracted ambient floor still leaves noise
across the whole window that drags the first moment toward the centre. Use
`"gaussian" or "parabolic" on noisy data; use "centroid"` only when
the background is genuinely gone.
*offset_ps* is a system delay to remove: ``t_flight = t_measured -
offset_ps``, so a positive offset makes the answer *closer*. Returns the
distance in metres as a float.
Raises `ValueError`: negative, non-finite, non-1-D or all-zero *hist*,
a non-positive *bin_ps*, an unknown *mode*, a non-finite *offset_ps*, a
flat histogram in a peak-based mode (`argmax` would silently pick bin 0
and report the first bin's depth), a peak in the first or last bin with a
sub-bin *mode* (there is no neighbour to fit to — use `"peak"`), a
degenerate three-sample fit, and — instead of returning a negative distance —
an *offset_ps* larger than the measured arrival time.
• photon_timeresolved family guide
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• photon_timeresolved — py -3.11 examples/photon_timeresolved.py
measurement as input)—
dtof)dtof_cube_simulate · dtof_cube_depth
*Provenance: photoncount.py — PHOTON operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.