3. Executables Usage
This project generates the following command line executable binaries:
fmdt-detect(and its variants),
fmdt-ellipse,
fmdt-log-parser,
fmdt-visu,
fmdt-check,
fmdt-reduce.
fmdt-detect
fmdt-detect is an efficient C/C++ code for meteors detection for ground,
plane or balloon observations. It produces text and video outputs. The main
results are the detected tracks that can be read from the standard output.
If the CMake -DFMDT_SPU_RUNTIME=ON option is used to compile the project,
then additional detection binaries are produced:
fmdt-detect-rt*-seq: this version comes with new performance measurement tools. However, this is a sequential version and the efficiency should be similar to the standardfmdt-detect*executable,fmdt-detect-rt*-pip: this version is multi-threaded. Thus, the throughput in term of FPS is higher than the standardfmdt-detect*executable.
Both fmdt-detect-rt-seq and fmdt-detect-rt-pip have the same level of
features than the standard fmdt-detect executable. The *-rt-* binaries
are based on the StreamPU runtime
[CTA+23].
fmdt-detect*-no-fail is a variant of fmdt-detect that is more robust to
errors. Indeed, when the number of CCs on a frame is higher than expected,
fmdt-detect will stop because it lacks allocated memory (can be tuned with
--cca-roi-max1 and --cca-roi-max2 parameters). Instead,
fmdt-detect*-no-fail will detect that the current frame cannot be processed
(too much CCs) and it will skip the labeling. Still, the tracking will be
processed so the objects can be extrapolated for later frames.
fmdt-detect*-opt is a variant that is feature equivalent to fmdt-detect
or fmdt-detect*-no-fail (depending on the executable binary name). opt
suffix means “optimized” version: these versions are the fastest in term of
latency and throughput. They are well suited for heavy computations like
real-time or big data processing.
fmdt-ellipse
fmdt-ellipse is an executable designed to detect meteors (like
fmdt-detect). Its design is based on a max-reduction plus a classification
of the meteors with ellipsoid features. At this time this tool is not fully
documented, it is still at the research level.
fmdt-log-parser
fmdt-log-parser is a Python script used to convert fmdt-detect log
output into text files used by fmdt-visu and fmdt-check.
fmdt-visu
fmdt-visu mainly uses the fmdt-detect text outputs (after conversion
with fmdt-log-parser) to generate highlighted video sequences. It can be
combined with ground truth to distinguish good detected tracks (true positive)
and bad detected tracks (false positive).
fmdt-check
fmdt-check compares detected tracks (fmdt-detect) with a given ground
truth. The results are shown on the standard output.
fmdt-reduce
fmdt-reduce performs a reduction from a video sequence into an image.
The produced image is in grayscale mode.
The next sections describe the command line parameters of these tools.