【摘要】 跌倒所带来的伤害威胁着老年人的生命安全，为实时监控造成人体严重损伤的跌倒行为，为减轻子女、养老机构等人员在养老看护方面的压力，对调频连续波(Frequency Modulated Continuous Wave,FMCW)毫米波雷达的跌倒检测算法进行研究，在跌倒检测方面，通过对雷达信号进行预处理，得到距离-速度和多普勒两种特征图像，并使用帧差法去除静态物体干扰，为充分提取距离-速度图像和多普勒图像的特征，提出一种双流融合特征提取网络算法。实验结果表明，所提双流融合特征提取网络算法较其他网络算法，出现误报、漏报情况的概率大大减小，检测准确率高达98.75%，并且系统的整体网络结构简单，在当前环境下具有较大的实用价值。更多还原

【Abstract】 Injuries caused by falls threaten the lives of the elderly. In order to monitor fall behavior that causes serious human injury in real time and reduce the pressure on their children, elderly care institutions, and other personnel in elderly care, this paper proposes to study the fall detection algorithm of FMCW(Frequency Modulated Continuous Wave)millimeter wave radar. In terms of fall detection, two feature images, range-velocity and Doppler, are obtained by preprocessing the radar signal, and frame difference method is used to remove static object interference. This paper proposes a two-stream fusion feature extraction network algorithm to fully extract features of range-velocity images and Doppler images. The experimental results show that the proposed two-stream fusion feature extraction network algorithm significantly reduces the probability of false positives and false negatives compared to other network algorithms, with a detection accuracy of 98.75%.Moreover, the overall network structure of the system is simple, and it has great practical value in the current environment.更多还原