基于毫米波雷达稀疏点云的人体行为识别方法

Human Activity Recognition Method from Mmwave Radar Sparse Point Clouds

摘要:

目前利用毫米波雷达进行人体行为识别的方法在复杂场景下无法很好的区分相似动作，与此同时模型的鲁棒性和抗干扰能力也相对较差；针对以上两个问题，提出了一种通用的基于毫米波雷达稀疏点云的人体行为识别方法，该方法首先利用K-means++聚类算法对点云进行采样，然后使用基于注意力特征融合的点云活动分类网络进行人体行为特征的提取和识别，该网络可以兼顾点云的空间特征以及时序特征，对稀疏点云的运动有灵敏的感知能力；为了验证所提出方法的有效性和鲁棒性，分别在MMActivity数据集和MMGesture数据集上进行了实验，其在两个数据集上取得97.50%和94.10%的准确率，均优于其它方法；此外，进一步验证了K-means++点云采样方法的有效性，相较于随机采样，准确率提升了0.4个百分点，实验结果表明所提出方法能够有效的提升人体行为识别的准确率，且模型具有较好的泛化能力。

Abstract:

At present, the human behavior recognition methods based on millimeter wave radar cannot distinguish similar actions when facing complicated scenes.In addition, there is a low robustness and interference resistance among these methods. To address the above two issues, a generic human behavior recognition method based on millimeter wave radar sparse point clouds is proposed, the method first samples the point cloud using the K-means++ clustering algorithm, and then uses a point cloud activity classification network based on attentional feature fusion for the extraction and recognition of human behavior features, which can take into account both spatial and temporal features of point clouds and has a sensitive perception of the motion of sparse point clouds. In order to verify the effectiveness and robustness of the proposed method, experiments were conducted on the MMActivity dataset and MMGesture dataset, respectively, which achieved 97.50% and 94.10% accuracy on both datasets, outperforming other methods. Furthermore, the effectiveness of the K-means++ point cloud sampling method is further verified, and the accuracy is improved by 0.4 percentage points compared to random sampling.The experimental results show that the proposed method can effectively promote the accuracy of human behavior recognition, and the model possesses a strong generalization ability.