Impact: Traditional magnetic tracking approaches based on mathematical models and optimization algorithms are computationally intensive, depend on initial guesses, and do not guarantee convergence to a global optimum. Here we propose an annular magnet pose estimation network (called AMagPoseNet) based on dual-domain few-shot learning from a prior mathematical model, featuring the following advantages:
1) Higher localization accuracy (1.87ยฑ1.14 mm, 1.89ยฑ0.81ยฐ), especially in the near field;
2) Enhanced robustness, as AMagPoseNet is just a single feed-forward neural network that does not rely on initial guesses and avoids the risk of falling into local optima;
3) Lower computational latency (2.08ยฑ0.02 ms) since the magnet pose is directly regressed from a single feed-forward network rather than iterative optimization;
4) Real-time estimation of 6-DoF pose if discriminative magnetic field features are provided.
Paper: https://lnkd.in/gXBSQ-hG
Dataset: https://lnkd.in/gXEFfwfA
Authors: Shijian Su; Sishen YUAN; Mengya Xu; Huxin Gao; Xiaoxiao Yang; and Prof Hongliang Ren
