TAD-Net: An Approach for Realtime Action Detection Based on TCN and GCN in Digital Twin Shop-floor

We proposed a real-time detection approach for shop-floor production action, this approach took the sequence data of continuous human skeleton joints sequence as input, reconstructed the Joint Classification-Regression Recurrent Neural Networks (JCR-RNN) based on Temporal Convolution Network (TCN) and Graph Convolution Network (GCN), constructed our Temporal Action Detection Net (TAD-Net), realized real-time shop-floor production action detection.

本研究提出了一种面向车间生产动作的实时检测方法。该方法以连续人体骨骼关节序列数据作为输入，基于时间卷积网络（Temporal Convolution Network, TCN）与图卷积网络（Graph Convolution Network, GCN）对关节分类回归循环神经网络（Joint Classification-Regression Recurrent Neural Networks, JCR-RNN）进行了重构，构建了本文提出的时序动作检测网络（Temporal Action Detection Net, TAD-Net），最终实现了车间生产动作的实时检测。