| 摘 要: :针对矢量网络分析仪(VNA)操作场景中动作识别率低的问题,提出一种手部交互物体引导的矢量网络分析仪操作动作识别方法。利用人体手部骨架序列提取手部姿态时空特征,针对手部相似动作难以通过姿态特征区分的问题,凭借隐含位置关系的 YOLOv8网络提取手部交互物体特征,对得到的手部姿态时空特征和交互物体特征进行融合,使用极限学习机(ELM)算法得到动作识别结果。实验结果表明,所提方法对9种典型矢量网络分析仪操作动作识别准确率均在95%以上,最大识别速度为21.3frame/s,满足对矢量网络分析仪操作动作识别的准确率和实时性要求。 |
| 关键词: 动态手势识别 矢量网络分析仪 人体手部骨架序列 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| Hand-Object Interaction Guided Recognition of Operation Actions for Vector Network Analyzer |
|
LI Qi, CHEN Yang
|
(College of Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, China)
liqidq@sust.edu.cn; darcy0116@163.com
|
| Abstract: To address the issue of low action recognition accuracy in Vector Network Analyzer (VNA) operation scenarios, a hand-object interaction guided method for recognizing VNA operation actions is proposed. Spatiotemporal features of hand poses are extracted from sequences of the human hand skeleton. To overcome the challenge of distinguishing similar hand actions based solely on pose features, hand-interacted object features are extracted using a YOLOv8 network that incorporates implicit positional relationships. The extracted spatiotemporal hand pose features and object interaction features are then fused, and action recognition results are obtained using an Extreme Learning Machine (ELM) algorithm. Experimental results demonstrate that the proposed method achieves an accuracy of over 95% for recognizing nine typical VNA operation actions, with a maximum recognition speed of 21.3 frame/s. This meets the accuracy and rea-l time requirements for recognizing VNA operation actions. |
| Keywords: dynamic gesture recognition Vector Network Analyzer human hand skeleton sequence |
