Send bounding box to robot code

[spacey-sooty]

Right now for object detection we only send

• image area
• yaw
• pitch
• confidence
  according to @samfreund

[samfreund]

Should work actually, cause we translate it to an OpenCV contour which I think we can grab corners from. Needs testing, leaving open until it's confirmed.

[amquake]

TrackedTarget constructor
TrackedTarget to PhotonTrackedTarget
getMinAreaRect() from a target's Contour

We currently send the detected corners and their minimum area rectangle for all detected targets (of all types). As object detection returns the bounding box corners, that's what we send as the detected corners.

Note that if we have an OBB (oriented bounding box) object detection model, we only compute the min-area rectangle (which would be the same as the detected corners) and not the (non-oriented) bounding box. Is this something we want to change? Is there still a usecase for the minimum area rectangle corners we send?

[samfreund]

Note that if we have an OBB (oriented bounding box) object detection model, we only compute the min-area rectangle (which would be the same as the detected corners) and not the (non-oriented) bounding box. Is this something we want to change? Is there still a usecase for the minimum area rectangle corners we send?

I don't think we'd need the non-oriented bounding box in the case of an OBB. If we're using an OBB, I think the whole point is to get the rotated detection, so there's no real benefit to returning anything else.