面向月面探测的空间机器人-航天员手势交互方法的研究

Research on Space Robot-Astronaut Gestures Interactive Method in Lunar Exploration

针对月面探测机器人-航天员手势交互的问题,以航天员助手机器人(AAR-2)为实验平台,运用卷积神经网络模型,提出了一种静态手势识别方法。采用该方法对设计的8种空间人机交互手势进行了手势识别试验,并将试验结果与单通道的CNN等方法的结果进行对比,证明了该方法能在一定程度上提高静态手势识别率,识别准确率达到93.3%。并自制了一套小型空间人机交互的手势数据集,用以补充该方法在月面探测的实际任务中的不足。

To solve the problem of hand gesture interaction between the lunar surface detection robot and the astronaut,the Astronaut Assistant Robot(AAR-2)was used as the experiment platform and a static gesture recognition method was proposed based on the convolutional neural network model.Then,the proposed method was used to perform hand gesture recognition experiments on eight kinds of spatial human-robot interaction hand gestures.The test results were compared with the results of the single-channel CNNs and other methods.It was proved that the method could improve the recognition accuracy of static hand gestures to a certain extent with the accuracy rate reached 93??3%.In addition,a small set of spatial human-robot interaction hand gesture dataset was created to supplement the inadequacy of the method in the actual task of lunar surface exploration.