Paper Accepted
Mr. Nakamura's paper as an M1 student was accepted by IEEE Access.
Journal2026
Multi-Granularity Behavior Modeling for Real-Time Lost-Hiker Detection in Low Mountains
IEEE Access, 2026
Motivated by rising hiking accidents on low mountains, we propose a method to detect disoriented hikers from GPS trajectories. Trail trajectories are converted into sequences of multi-granularity grids, and a Transformer-based behavior model learns to predict the next grid a hiker will enter. An adaptive beam search then uses this model to adjust its width dynamically, cutting computation while improving detection. On real GPS data from Mt. Daimonji, the method reaches an F1 of 0.426 versus 0.353 for an LSTM baseline, and adaptive beam search reduces search iterations by about 55% without lowering F1.

