A multi camera unsupervised domain adaptation pipeline for object detection in cultural sites through adversarial learning and self-training