This doesn't describe what traditional CV approach this uses, presumably descriptor based ones: A-KAZE, SIFT etc. do work well for such scenarios. So both Deep learning and traditional machine vision have their place.