Skip to main content

Ground-plane intersection

GroundPlaneIntersection projects the bottom center of each detection bounding box onto a horizontal field ground plane. It uses calibrated intrinsics and a field-relative camera pose.

Inputs​

  • detections: Dictionaries with normalized bbox coordinates [x1, y1, x2, y2] from an undistorted image.
  • camera_pose: A finite 4 by 4 transform from camera coordinates to field coordinates.

Outputs​

positions_3d is a new list. Each retained detection has a field-relative position_3d in meters; other fields are preserved.

When to use​

Use this to estimate field positions for objects that touch the ground when a field-relative camera pose is available.

Configuration​

  • camera_bus_id: Required camera bus ID used to resolve intrinsics.
  • ground_level: Ground-plane Z coordinate in field meters. Default 0.0.

Limitations​

Missing detections or camera pose produce an empty list. The operation skips malformed detections, cameras at or below the plane, backward intersections, and rays angled downward by no more than 3 degrees. Intrinsics must match the undistorted image and include a valid camera matrix and image size.