AEGIS: 저가시성 구조 환경에서 효율적인 사람 탐지를 위한 mmWave 기반의 웨어러블 시스템

초록

Low-visibility environments caused by dense smoke severely limit rescuers' vision, increasing cognitive load and delaying life-saving missions. Existing visual sensors are vulnerable to smoke, and conventional mmWave radar systems often assume static environments or require Inertial Measurement Units (IMUs) for motion compensation, which suffer from drift issues, making them unsuitable for dynamic rescue scenes. We propose AEGIS, a fully wearable mmWave radar system that operates standalone without auxiliary sensors like cameras, LiDAR, or IMUs. Using only point cloud data from a chest-mounted 60GHz radar, AEGIS first estimates the wearer's complex ego-motion through “motion- invariant geometric structure matching". Subsequently, a “velocity-aware calibration mechanism" robustly separates humans from the static environment by analyzing deviations between predicted motion and observed velocities, ensuring high generalization performance in unseen environments. In controlled indoor experiments, AEGIS achieved a 98.1% F1-score and 0.9832 MOTA for human detection, maintaining over 95% performance on unseen people and environments. Furthermore, experiments under various changing physical conditions confirmed its robust operation, and ablation studies verified the effectiveness of the geometric encoding and velocity-aware calibration mechanisms. This research demonstrates that body-mounted mmWave sensing can achieve reliable human detection in dynamic conditions without auxiliary sensors, suggesting future extensions for multi-person and stationary victim detection.

키워드