面向聋哑人的MR智能眼镜生命监护系统：仿真设计与决策仲裁算法验证

A Mixed Reality (MR) Smart Glasses-Based Life Monitoring System for Deaf and Mute Individuals: Simulation Design and Verification of Decision Arbitration Algorithm

摘要: 聋哑人群体的健康监护与日常沟通长期面临双重困境。本文提出并验证了一个面向聋哑人MR智能眼镜生命监护场景的多传感器融合与决策仲裁算法，旨在系统性解决生理数据与视觉识别结果之间的决策冲突。系统基于C++17多线程架构，融合IMU、PPG与摄像头数据，实现跌倒检测、生命体征监测与紧急手语识别三大功能。针对多源数据冲突，提出三级决策仲裁机制：置信度加权处理手势识别置信度不足，生理–视觉冲突解决生理数据与手语指令的矛盾，时序一致性校验防止决策抖动。仿真实验表明，各功能模块逻辑触发准确率均达100% (各10次)，混合异常场景分级警报逻辑准确率为96.7% (30次中29次正确)，响应时间小于500 ms，72小时连续运行无崩溃。上述准确率为仿真环境下算法逻辑验证的初步结果，不代表真实硬件部署性能。本研究处于仿真验证阶段，所提架构与仲裁机制可为后续面向聋哑人的MR健康监护应用提供算法层面的设计参考。

Abstract: The health monitoring and daily communication of deaf and mute individuals have long faced dual challenges. This paper proposes and validates a multi-sensor fusion and decision arbitration algorithm for MR smart glasses-based life monitoring in deaf and mute scenarios, aiming to systematically resolve decision conflicts between physiological data and visual recognition results. Built on a C++17 multithreaded architecture, the system integrates IMU, PPG, and camera data to achieve three core functions: fall detection, vital signs monitoring, and emergency sign language recognition. To address multi-source data conflicts, a three-tier decision arbitration mechanism is proposed: confidence weighting handles insufficient gesture recognition confidence, physiological-visual conflict resolution reconciles contradictions between physiological data and sign language commands, and temporal consistency verification prevents decision jitter. Simulation results demonstrate that each functional module achieves 100% logical trigger accuracy (10 tests each), with a hierarchical alert logical accuracy of 96.7% (29 out of 30 mixed abnormal scenarios correct), a response time under 500 ms, and 72-hour continuous operation without crashes. The above accuracies are preliminary results of algorithmic logic validation under simulation and do not represent actual hardware deployment performance. This research is in the simulation validation phase, and the proposed architecture and arbitration mechanism can provide an algorithmic design reference for future MR-based health monitoring applications targeting the deaf and mute population.