基于可穿戴设备的姿势识别方法研究

Pose recognition method based on wearable device

针对现有的基于图像以及毫米波雷达进行人体姿势识别在医疗保健领域存在的不足,提出一种基于人体区域通信技术的穿戴式姿势识别方法。该方法通过人体姿势导致的信道衰减进行姿势识别。文中首先建立了人体站、蹲、坐、跑、走5种姿势模型,并在人体上放置1个发射器、2个接收器,构成两条信道,计算5种姿势下的信道衰减作为数据集。然后,对两条信道分别使用多种单一分类器进行识别,获得94.33%的最高准确率。最后,通过Stacking方法融合两条信道的特征得到99.67%的准确率。结果表明,基于人体区域通信技术的穿戴式姿势识别方法是可行的,并利用Stacking特征融合方法使识别结果相比单一特征下的最高准确率提高了5.34%。

A pose recognition method using wearable devices and human body area communication technology is proposed to overcome the drawbacks of current methods based on image and millimeter wave radar and realize the recognition through the channel attenuation caused by human posture.Five postures of human standing,squatting,sitting,running and walking are modeled,and two channels are constructed with one transmitter and two receivers for calculating the channel attenuation with 5 postures as the data set.The highest recognition accuracy of multiple single classifiers for the two channels is 94.33%,and the accuracy is improved to 99.67%with feature fusion of the two channels using Stacking method.The results show that the pose recognition method using wearable devices and human area communication technology is feasible,and that the highest accuracy rate using Stacking feature fusion is improved by 5.34%as compared with single feature.