Some simple programs for capturing images from Alkeria Necta line scan camera.
- Alkeria Necta USB3 camera with proprietary binary drivers version 1.11. If you are using Arch Linux, you can install them using the PKGBUILD in the
arch-linuxdirectory. If on Ubuntu, just install using the deb file that you download from the Alkeria Download Area. - imgui 1.91.9 and above
- SDL2
The raw speed-constraint editor displays the Bayer-binned capture and autosaves circle, equal-width, and sequence-cut annotations beside the raw data:
uv run python python/speed_constraints.py /path/to/captureAt startup it reads every raw file once and retains the complete uint8 display
image in memory. The detailed window is 4096 half-resolution columns wide by
default, independent of the raw file sizes. A uniformly scaled navigator is
built from that image at the bottom; click it to center the detailed window.
The navigator wraps into multiple rows as needed, and its divider can be
dragged vertically. Use Left/Right to slide by 256 columns, Ctrl+mouse-wheel to
zoom, F to enable persistent fitting, F8 to hide the navigator, and F9 to hide
the annotation panel. Fit mode is enabled initially and refits whenever the
window is resized; manual zoom or 100% mode disables it. The Rotate 180°
checkbox rotates both detailed and
navigator views without changing saved coordinates; it starts checked for
pre-2026 captures and unchecked for 2026-and-later captures so both camera
mounting eras appear upright. The default output is speed_constraints.json
in the capture directory; an existing file is loaded when the tool starts.
preview.py loads that file automatically. Circle and equal-width annotations
refine the correlation-derived speed spline without changing its direction,
using a fixed, uniformly spaced spline basis. Each equal-width group has a
jointly fitted common output width; all manual dimensions use a fixed 2.5-pixel
annotation uncertainty, so longer annotations carry tighter relative weight.
Sequence positions replace motion-region detection and define one output
DNG/JPEG per adjacent pair. The first and last positions also bound speed
fitting: only complete correlation windows inside those bounds are eligible.
Featureless background windows are rejected as undefined unless they contain
both real horizontal structure and an absolute, distinct correlation peak, so
padded background cannot pull the manual fit. Pass --no-speed-constraints
to render with the legacy automatic behavior for comparison. Pass
--equal-length-chunks to use only the first
and last annotated positions and split that corrected output range into nearly
equal chunks smaller than OUTPUT_CHUNK_TARGET, ignoring intermediate cut
positions. Pass --pad-output-to-equal-widths instead to retain every annotated
cut and pad the resulting slices to the width of the widest one.
Output orientation follows the camera mounting date: captures before 2026 are
not rotated, while captures from 2026 onward are rotated 180 degrees. Override
the automatic choice with --flip-output-180 or --no-flip-output-180.
Each capture is rendered into a temporary staging directory. Only after the
complete DNG/JPEG set and stack preview succeed does preview.py replace that
capture's prior outputs, including chunks that no longer exist. Existing
Darktable .dng.xmp sidecars are carried forward by NECTA chunk number and
renamed for the new DNG hash; if several sidecars exist for one chunk, the most
recently modified one is used. Choose a separate --output-dir when an older
render needs to be preserved.

