Kitware
burnin crosshair detection
There is nothing too special in this code that ties it to OpenCV or VXL. This could all be written fairly easily using only VXL or only OpenCV. For that matter, it could even be written without either OpenCV or VXL (KWIVER vital only) with just a little more effort. However, there will be a bit of work either way to implement some simple image processing utilities.
This class is called moving_burnin_detector, but it seems to handle only the cross-hair case. Is there another class that handle the corners or the moving "N"? We might want to consider the broader set of functionality that needs to be ported before determining which way to go with this.
There is nothing too special in this code that ties it to OpenCV or VXL. This could all be written fairly easily using only VXL or only OpenCV. For that matter, it could even be written without either OpenCV or VXL (KWIVER vital only) with just a little more effort. However, there will be a bit of work either way to implement some simple image processing utilities.
This class is called
moving_burnin_detector, but it seems to handle only the cross-hair case. Is there another class that handle the corners or the moving "N"? We might want to consider the broader set of functionality that needs to be ported before determining which way to go with this.
Good, I can start with removing opencv and we can go from there. The original code also has brackets and text detectors, I am just waiting to get the crosshairs working before adding those in.
@hdefazio Can you point me at the original code? I'm worried that some of the other code has stronger dependencies on OpenCV. We don't want to port part of this if we can't easily port the rest later in the same way.
I'm also concerned that it's not worth porting this code in its current form to use VXL instead of OpenCV. This code is way more complicated than it needs to be. It's probably better to just rewrite a much simpler version of it. Maybe we keep this version as is as a baseline for comparison. We should probably get this working in it's current form, using both VXL and OpenCV, first. So we can do that in the burnout arrow for now.
Longer term I think we should do all of this (cross-hairs, corners, and text) with template matching. I believe at least the text matching does this. There is no reason for this custom code to match the cross-hair when you can just render a cross-hair template and use it in the same template matching code.
@mleotta The original code lives here: https://kwgitlab.kitware.com/computer-vision/vidtk/-/blob/master/library/object_detectors/moving_burnin_detector_opencv.cxx I was under the impression that this was all template matching. But I may have misunderstood.