After watching the video, my conclusion doesn't change. So the model data is trained using optical data so it can correlate the WiFi data. The resolution is poor, has no color, cannot identify anyone in particular and is only useful to determine if a large animated object is moving in the space. Much the same way IR detection works. It would be simple to defeat by adding multiple WiFi signals to interfere with detection. The receiver would need to know how many signals were present. The technique is interesting, but easy to circumvent