【摘要】 为了有效判别真实摔倒动作与疑似摔倒动作、提高动作识别准确度，提出基于希尔伯特黄变换(Hilbert-Huang transform, HHT)和改进概率神经网络(probabilistic neural networks, PNN)的信道状态信息(channel state information, CSI)人体动作识别算法。对CSI的幅度与相位融合信号进行数据预处理，利用HHT来提取区分人体动作信息的瞬时幅值和瞬时频率作为分类特征构建特征矩阵，在遗传算法(genetic algorithm, GA)优化的PNN神经网络中训练出能有效检测真实摔倒和疑似摔倒动作的GA-PNN人体动作识别模型；利用训练好的识别模型对输入的CSI数据进行摔倒动作的判别。仿真实验表明，提出的算法能有效地检测真实摔倒和疑似摔倒动作，其识别准确度可达到97.18%,且误报率较低。更多还原

【Abstract】 To discriminate real and suspected fall actions effectively and improve the accuracy of action recognition, this paper proposes a CSI human action recognition algorithm based on Hilbert-Huang Transform(HHT) and improved Probabilistic Neural Networks(PNN). In the train phase, firstly, the amplitude and phase fusion signals of the acquired Channel State Information(CSI) are pre-processed. Secondly, HHT is applied to extract the instantaneous amplitude and instantaneous frequency of the fused signals to distinguish the human action information as the classification features to build the feature matrix. Finally, the GA-PNN human action recognition model, which can effectively detect real and suspected fall actions, is trained in the Genetic Algorithm(GA) optimized PNN neural network. In the test phase, the input CSI data is used to discriminate the fall actions through the trained recognition model. Through simulation experiments, it is verified that the algorithm can effectively detect real and suspected fall actions, and its recognition accuracy can reach 97.18% with low false alarm rate.更多还原