ITER-Tag: An Adaptable Framework for Robust Uncoded Fiducial Marker Detection and Identification

Fiducial marker-based tracking is an effective method for pose estimation in coffined environments, such as the International Thermonuclear Experimental Reactor. In this paper, we propose a novel framework for marker detection and identification that is moderately robust to occlusion, even while using a relatively small number of marks. The proposed approach (ITER-Tag) consists of a hybrid pipeline that extracts marker candidates from images using classical methods and identifies uncoded markers using a shallow feed forward neural network. The method requires minimal parameter tuning, data collection and annotation. The methods can be easily adapted to different use cases with varying number and positions of the marks. We test the robustness of our approach in three different use cases in ITER's divertor, using either retro reflective markers or laser engravings and achieve high detectability rates. We demonstrate how the proposed approach can be used to accurately and robustly retrieve the six-degree-of-freedom pose of the targets.