基于3D残差卷积注意力网络的跨域手势识别

Cross-domain gesture recognition based on 3D residual convolutional attention network

针对现有WiFi单一场景下手势识别模型泛化能力差,跨域识别精度低的问题,提出了基于3D残差卷积注意力网络(RAN)的信道状态信息(CSI)跨场景手势识别方法,通过提取与场景无关的特征人体坐标速度谱(BVP),并结合3D RAN来实现跨场景手势识别。结果表明:所提出的方法具有较好的跨域手势识别效果,并且具有更好的泛化能力。

Aiming at the problem of poor generalization ability of gesture recognition model and low cross-domain recognition precision in existing WiFi single scene,a channel state information(CSI)cross-scene gesture recognition method based on 3D residual convolutional attention network(RAN)is proposed.By extracting scene independent feature body-coordinate velocity profile(BVP),and combining with 3D RAN,realize cross-scene gesture recognition.The results show that the proposed method has a better cross-domain gesture recognition effect and a better generalization ability.