基于动态手势识别的学前教育机器人人机交互技术研究

Research on human-machine interaction technology for preschool education robots based on dynamic gesture recognition

为进一步优化应用于学前教育的交互型机器人的使用体验,针对学前教育场景,提出一种基于动态手势识别的人机交互系统。其中,构建包含手势识别模块、目标检测模块以及定位抓取模块三个主要部分的人机交互系统:以YOLOv7网络作为基本的手势识别方法,并引入混合注意力机制CBAM对网络进行改进,以进一步提升识别性能。实验结果表明,构建的基于CBAM改进YOLOv7网络的手势识别模型具有较低的训练损失,与其他类型和未改进的识别模型相比,该模型具有更高的识别精度,在平均精度均值、精确率、召回率和每秒处理帧数上分别达到了99.7%、99.1%、98.8%以及0.025;所构建的人机交互系统能够进行准确且稳定的实物抓取,整体性能良好。综上,所构建的基于动态手势识别的人机交互系统性能优良,能够进行效果良好的人机交互,将其应用于学前教育机器人中时,可在实际的学习教育场景下,与用户进行流畅且有效的互动,进而提升用户的使用体验。

To further optimize the user experience of interactive robots applied in preschool education,a human-machine interaction system based on dynamic gesture recognition is proposed for preschool education scenarios.Among them,a human-computer interaction system is constructed,which includes three main parts:gesture recognition module,object detection module,and positioning and grabbing module.YOLOv7 network is used as the basic gesture recognition method,and a mixed attention mechanism CBAM is introduced to improve the network and further enhance recognition performance.The experimental results show that the gesture recognition model based on CBAM improved YOLOv7 network has lower training loss.Compared with other types and unimproved recognition models,this model has higher recognition accuracy,achieving 99.7%,99.1%,98.8%,and 0.025%in average accuracy,accuracy,recall,and processing frames per second,respectively;The human-computer interaction system constructed can accurately and stably capture physical objects,with overall good performance.In summary,the human-computer interaction system based on dynamic gesture recognition has excellent performance and can perform well in human-computer interaction.When applied to preschool education robots,it can smoothly and effectively interact with users in actual learning and education scenarios,thereby improving the user experience.