Researchers have developed EventHOI, a novel system designed to efficiently log hand-object interactions (HOIs) from the perspective of smartglasses. This system addresses the limitations of continuous RGB video capture by employing an event-driven approach that selectively triggers RGB capture and compresses visual data. By integrating low-power event sensing with IMU-based egomotion cancellation, EventHOI significantly reduces power consumption and bandwidth requirements while maintaining high accuracy for HOI recognition. The system's effectiveness has been demonstrated through extensive evaluations on benchmark datasets and real-world scenarios, showing substantial energy and bandwidth savings that enable stable video streaming over Bluetooth Low Energy. AI
IMPACT Enables more efficient data logging for AI applications that rely on understanding human-object interactions from egocentric viewpoints.
RANK_REASON The item describes a novel system and its evaluation presented in an arXiv paper. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
