Head-Mounted IMU Enables Behavioral Activity Recognition for AR Smart Glasses
A new approach has been created by researchers to identify behavioral activities using data from head-mounted Inertial Measurement Units (IMUs), extending beyond basic movements such as walking or standing. This study outlines five behavioral categories designed to meet the requirements of augmented reality (AR) applications and the observability of sensors. A dataset comprising 160,000 samples was generated from Ego4D, supported by a four-level quality assurance system across eight activity scenarios. The model introduced, HiT-HAR, which contains 703,000 parameters, surpasses previous head-mounted IMU models in recognizing five action classes and eight scenario classes. An analysis of per-class separability highlights the observability frontier, indicating that Locomotion is consistently observable, while Object Transfer and Task Operation benefit from temporal context, revealing scenario-specific limitations. This research aims to deliver ongoing behavioral context for proactive AR support.
Key facts
- Head-mounted IMU is the most practical always-on sensor for AR smart glasses.
- Five behavioral categories defined for AR application need and sensor observability.
- 160K-sample Ego4D dataset constructed with four-tier quality assurance.
- HiT-HAR model has 703K parameters.
- Outperforms prior head-mounted IMU models on five-class action and eight-class scenario recognition.
- Locomotion is reliably observable from head-mounted IMU.
- Object Transfer and Task Operation benefit from temporal context.
- Research aims to provide continuous behavioral context for proactive AR assistance.
Entities
Institutions
- Ego4D