Abstract: With the aging population increasing in China, the number of elderly people living alone has risen, highlighting home safety concerns. A multimodal home safety monitoring system based on YOLOv8-pose was proposed. The system integrated hardware, software, and server components to achieve comprehensive monitoring and health management for elderly individuals through multimodal data collection, model detection, and big data analysis. Hardware based on Raspberry Pi 4B was used for data acquisition. A management platform was built using a front-end and back-end separated architecture. The server-side deployed a posture detection model trained on YOLOv8-pose, enabling accurate behavior recognition and real-time alerts. A dataset of daily activity patterns was established using big data analysis techniques. When abnormalities were detected, alerts were sent to family members and community staff via a mini-program and web interface in no time, addressing emergency response challenges. Experiments showed that the human pose recognition accuracy of the detection model reached as high as 96.3%. The system also demonstrated excellent performance in overall cost control, providing practical references for the improvement of China’s smart elderly care system.

Key words: YOLOv8-pose, multimodal, home safety monitoring system, Raspberry Pi

摘要： 针对我国老龄化加剧,独居老人日益增多而引发的居家安全问题,文中提出基于YOLOv8-pose构建的多模态居家安全监护系统。该系统包括硬件、软件和服务器端3部分,通过多模态数据采集、模型检测与大数据分析,实现对独居老人全面监护与健康管理。硬件以树莓派4B采集数据,软件采用前后端分离架构搭建管理平台,服务器端运行基于YOLOv8-pose训练的姿态检测模型,可精准识别老人行为并预警。同时,利用大数据分析技术建立老人生活规律数据集,当发现老人异常时,系统即刻通过小程序和Web端提醒子女与社区人员,解决老人突发状况救助问题。实验表明,检测模型的人体姿态识别准确率高达96.3%,且系统在整体成本控制上表现出色,可为我国智慧养老体系完善提供实践参考。

关键词: YOLOv8-pose, 多模态, 居家安全监护系统, 树莓派