I've processed your video. The project file is attached.
Camera heights are very close to the measurements you've posted, but distance between 1 and 2 is somewhat larger than your number. However, the camera positions correspond to their LEDs visible on the video.

Indeed, this calibration was not easy to process. First, I've disabled "Auto detect initial positions", and reset positions to full circle counter-clockwise.
Then I've selected frames 125-3800 as ROI, ending with 4th ground point in the video. But, after calibrating within this whole range, the results were disapponting. The reason for this was a big rate of marker misdetections - several dozens percents for every camera. These are absolutely unacceptable numbers, this value should stay 10% at the very most.
Analyzing the video, I can say that the light marker is often occluded by the body, especially when marking ground points. I advice you to practice with the light marker, finding ways to maximize its visibility to cameras. Hold it away from the body, put in that side from the body where most of the cameras see it. E.g. when approaching a boundary of the capture area, stay outside and hold an arm with the marker inside.
So, I've adjusted the ROI to 125-800, before approaching the ground. In this range the marker is mostly visible to all the cameras. And this helped, I've got nice results. Marker misdetection was below 1%, only 4th and 5th cameras had greater values (but still below 10%). As the marker detection had been previously completed up to 3800th frame, I've then simply expanded the ROI back and marked ground points.

Thanks for your help. Since I've been doing this, I've never realized I could calibrate ROI in stages.

You can run the calibration procedure many times on the same project, adjusting the settings or ROI. Each time you run it, there are 2 steps involved: marker detection and camera positions calculation.
Marker detection is run only for those frames within ROI where it has not been run previously. Thus, if you are narrowing ROI, this step is skipped. Positions calculation considers only the frames within ROI, so you can avoid "bad" frames screwing up the results by choosing an appropriate ROI.
Dealing with ground points is totally independent from the calibration procedure itself, just a correction of its results - namely, the ground plane location. It is just the UI requirement that the ROI should include "ground" frames so you are able to mark them. Setting the scene scale is also a correction of the calibration results, not considering ROI at all.