localdevices / pyorc Goto Github PK

depths are calculated at multiple locations. Must be moved to a more centralized location and related methods simplified.

this will require some experimentation with diversity of profiles and velocities.

• handling of masked values in DataArrays
• remove poorly correlated velocities
• remove velocities outside of feasible range
• interpolate missing from surrounding values in space (limit to maximum of about 0.2m)
• filter in time on bi-modal distribution

remove isolated values
remove values too much deviating from moving window mean

sphinx documentation

Currently sert to remove values under a certian intensity threshold. Make this normalization between min/max

not providing a crs raises errors, modify so that if no crs is provided anywhere the code keeps on working, and raises are given if a user requests a function that requires "lat" and "lon" coordinates

Currently, sampling is only done at exactly the pixel under consideration. This should be relaxed to the surrounding pixels, exclusing pixels that are significantly off in terms of their deviation from mean or median

The manner of frames extraction can be different. For instance a single channel in rgb or hsv, or mean of rgb or hsv. Default can still be method="mean" for grayscale.

This requires a couple of things

• ensure that unreliable values are set to missing data, before doing any statistics
• filling in of points without significant sampling of velocities with depth-velocity profile estimates
• retrieving a statistic such as the 50% quantile

plotting is now under one single method. This issue is to split the method into an API like xarray.Dataset.plot

In Transect, angles per section is not yet working appropriately.

default end_frame=None gives an error if start_frame is provided.

• implement raw piv without any filtering considerations
• storing to a plottable NetCDF (should work with Qgis!)

Starting with API functionalities

Get the original imagery in a better shape for PIV by detecting backgrounds or pre-detect moving parts. Docs in opencv to be reviewed

Can be implemented as .frames.stabilize.

Use https://stackoverflow.com/questions/3431434/video-stabilization-with-opencv as a starting point

Current behaviour is that still time steps are given. It is much easier and cleaner if all methods of a Transect object are based on quantile as dimension.

Split code structure in api and functions

currently a user needs to select control points on frames that are already undistorted for lens distortion. This should be changed so that it can be selected from a raw frame. This makes it a lot easier for a user to pin point control points.

In many cases, users do only want to record one single video, and work with that instead of multiple videos with changing water levels. In this case, only a water level reading in the same reference as the bathymetry is needed. Practically this means no h_ref is needed in a CameraConfig gcps, and no h_a is needed when supplying a video. Currently the user receives assertion errors when this info is omitted. This should change so that a video is processed under thge assumption of a single water level.

currently the logarithmic profile fits are done on-the-fly but not stored in the resulting Dataset. Add this to the metadata.

After manipulation and reprojection, it can be useful to make an animation of the ds.frames object. ad ds.frames.to_ani

implement ds.velocimetry.plot.streamplot. Challenge is to make this work with non-local coordinates as streamplot expects equidistant and increasing coordinates only! Strategy could be:

• make a set of lines in local projection
• reproject these to camera or geographical
• plot in style of streamplot