The field of human activity recognition is rapidly advancing, with a focus on developing innovative methods for sensor-based activity recognition, robust inertial navigation, and effective sensor fusion. Recent studies have explored the use of wearable devices, such as smartwatches and earphones, to recognize human activities and track user behavior. The integration of multiple sensors and modalities has shown significant promise in improving the accuracy and robustness of activity recognition systems. Notably, the development of transformer-based architectures and attention mechanisms has enabled the effective modeling of complex temporal and spatial relationships in human activity data.

Some noteworthy papers in this area include: The CMD-HAR paper, which proposes a spatiotemporal attention modal decomposition alignment fusion strategy to tackle the problem of mixed distribution of sensor data. The Suite-IN++ paper, which presents a deep learning framework for flexiwear bodynet-based pedestrian localization, integrating motion data from wearable devices on different body parts. The MultiTSF paper, which proposes a transformer-based sensor fusion method for human-centric multi-view and multi-modal action recognition, leveraging a human detection module to generate pseudo-ground-truth labels.