you can use raw lidar data and intrinsic/extrinsic with images to reconstruct a point cloud.

Anyway this issue is resolved: https://developer.apple.com/documentation/roomplan/roomcaptureview/ismodelenabled I found this variable that can turn off the preview

I'm not very experienced with IOS development, so this solution might not be very elegant with Swift: you can just grab camera transform and camera intrinsic from arFrame within arSession, then write an algorithm to map it back manually.

for the most simple setup you can just calculate the smallest rotating bounding box, similar to OpenCV's rotated rect.