基于多特征尺度特征融合的航空指挥动作识别

Aviation Guide Gesture Recognition Based on Multi-Feature Scale Feature Fusion

摘要:

视觉辅助引导系统(Visual aided Guide System,VaGS)是未来航空运输必不可少的重要组成部分,可以大大节省地面指挥和引导资源,提高安全性。VaGS的核心技术是识别地面指挥员手势并将其转化为指令。介绍了一种新的高效手势识别体系结构,它主要包括 2个部分:①采用多尺度浅层结构进行特征学习,将全局身体姿态特征与局部手势特征提这两种尺度特征进行融合；②所提取的特征输入到超限学习机(Exteme Learning Machine, ELM)中进行分类,输出指令。实验结果表明,在自建的 40个类别的航空指挥动作数据集中,准确率达到 98.5％,单帧用时 0.13 ms。

Abstract:

Visual aided Guide System (VaGS) is an essential part of the future air transport , which can greatly save theground command and guide the resources and improve the security. The core technology of VaGS is to identify the groundcommander's gestures and translate them into instructions. In this paper, a new efficient gesture recognition architecturewas introduced and it mainly included two parts: ① Using multi-scale shallow structures for feature learning, extractingand integrating global body pose features and local gesture features; ② Importing fusion features into an Extremum Learn.ing Machine (ELM) for classification, output instructions. The experimental results show that in 40 self-built aeronauticalcommand movement data sets, the accuracy rate reaches 99.6%, and each frame uses 0.13 ms.