Improve target tracing when fields are crowded.

[simontorres]

Tracing target works well in targets with high signal to noise and when they are in low number and well separated. When they are crowded the functions are messed up. For instance there is a target that has about 30 targets but of course they are close together and if the targets are faint the tracing goes away and crosses over the other traces. There must be a way to fix this. From a parametric point of view I have no idea but a first approach will be to get all the parameters. All the traces should be approximately parallel but not exactly either. Goodman spectra are not perfectly alined with the columns/row of the detector (hence the polynomial of degree 2). if there is no way to restrict it from the parameters approach an evaluative approach should be taken. At the moment I have no idea.

[simontorres]

Here is a good example, this is an extreme case but if the pipeline handles this will handle anything.

[simontorres]

Sky lines seems to be the area where the tracing goes wrong. It should be possible to detect sky lines by getting the media in the spatial axis, identify them and ignore those areas.

[simontorres]

For reference
https://nbviewer.jupyter.org/github/simontorres/general_documentation/blob/master/jupyter-notebooks/experiments/Target_Trace_Improvements.ipynb?flush_cache=true