It’s hard to get dimensional reproductions from a photograph. I’ve tried to do this on my CNC plasma, using cameras (photo, DSLR, etc) for a while. In the end, the parallax error usually gets you. Dimensions will be most accurate at the center of the photo, and diverge at the edges.
With a camera, you can mitigate this somewhat by putting a lot of distance between the camera and the part, and using a zoom to get the image. Essentially, the image is being taken through more of the “center” of the lens, so it has a bit less distortion. Not really an issue with the LB setup, though, as the camera is a fixed distance from the part.
Ultimately, for dimensional parts, I’ve found it’s best to 1) scan them on a flatbed scanner, then trace, or 2) use a specialized tracing setup. I use Scananything on my table (6’ x 12’) if I need to get big parts. It replaces my torch head with a camera, and then traces the part. It’s finicky, a bit of a PITA, and still requires cleanup (it’s essentially a collection of points/paths), but it’s a good way to get sufficient accuracy.