Abstract:

Compared with the traditional camera, the use of Ultra-Wideband（UWB） radar for human posture recognition is not only low environmental requirements, high recognition rate and can better solve the problems of blind spots and easy to leak privacy in the camera. Combined with the characteristics of the UWB radar system, the common ultra-wideband pulse signals are analyzed in detail. According to the current research frontier, the traditional machine learning methods and deep learning methods of human posture recognition by UWB radar are analyzed, and the principle and each model’s limitation of the representative Support Vector Machine（SVM）and Convolutional Neural Networks（CNN） are analyzed and discussed in combination with the specific literature. The general model of UWB radar human posture recognition is proposed, the problems that need to be solved urgently in human posture recognition by UWB radar are analyzed and its future development direction is prospected.

Key words: Ultra-Wideband（UWB） radar, human posture recognition, deep learning, Support Vector Machine（SVM）, Convolutional Neural Networks（CNN）

摘要：

与传统摄像头相比，利用超宽带雷达进行人体姿态识别不仅对环境要求低、识别率高且能较好地解决摄像头存在视角盲区和易泄露隐私等问题。结合超宽带雷达系统的特性，对常见的超宽带脉冲信号进行了分析；针对当前的研究前沿，对超宽带雷达人体姿态识别的传统机器学习方法和深度学习方法进行分析，结合具体文献对具有代表性的支持向量机（SVM）和卷积神经网络（CNN）进行原理分析和模型的局限性进行了分析；提出超宽带雷达人体姿态识别的通用模型，分析了超宽带雷达人体姿态识别亟需解决的问题并对其未来发展方向进行了展望。

关键词: 超宽带雷达, 人体姿态识别, 深度学习, 支持向量机, 卷积神经网络