burnin crosshair detection #1458
Conversation
|
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 |
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. |
hdefazio
force-pushed
the
dev/burnout_pattern_match
branch
from
be1a05a to
3398d2f
Compare
| // convert image to cv | ||
| cv::Mat cv_image = ocv::image_container::vital_to_ocv( | ||
| input_image->get_image(), | ||
| ocv::image_container::RGB_COLOR ); |
There was a problem hiding this comment.
I wonder if this ocv::image_container::RGB_COLOR and a bit later the ocv::image_container::BGR_COLOR might be giving you problems. First, they don't match, which is probably intentional but I figure I'd mention it anyway. Second, if you look at the implementation of these functions in arrows/ocv/image_container.cxx, ocv_to_vital doesn't use the ColorMode parameter at all, and vital_to_ocv only explicitly handles the BGR_COLOR case and lumps everything else into a bucket, not really doing anything with it. It looks to me whoever wrote these functions only really worked with BGR images, so I might investigate if they are actually doing what you hope they would be doing. Perhaps a cause of the pixel weirdness you were seeing?
No description provided.