search

WPIRoboticsProjects / GRIP

Program for rapidly developing computer vision applications
http://wpiroboticsprojects.github.io/GRIP
Other
373 stars 108 forks source link

PublishContoursReport only uses Rectangles, not Quadrilaterals #393

Open Poppeseed opened 8 years ago

Poppeseed commented 8 years ago

Since the Publish ContoursReport operation uses the boundingRect, any time you are not facing the target directly head on the report is a bit inaccurate. I'm not well-versed in OpenCV, but it seems the only way to currently get this sort of information is the Publish LinesReport operation, or a combination of the two.

Is it possible to add a new Publish ContoursReport operation that bounds using a quadrilateral instead of a rectangle?

Poppeseed commented 8 years ago

Looking into it a bit further, if you could publish a report of the actual points after a ConvexHull contour, it would likely solve this.

image

ThomasJClark commented 8 years ago

Yeah, I don't think there's a simple, efficient solution for finding bounding quadrilaterals. Publishing a list of points might be good, but I don't think NetworkTables supports 2D arrays, and even if it did array sizes are limited to 255 elements.

Maybe we can extract a single contour at a time and publish up to 255 x/y coords?

Poppeseed commented 8 years ago

Yeah, without 2D arrays you'd be pretty limited. Having a criteria selection available (min/max area/width/height/etc) would probably work.

ThadHouse commented 8 years ago

NetworkTables 3 can also send raw data, so GRIP could create their own raw data type for sending more complicated arrays. I would require code on the team end to decode, but at least the teams could get the data.