关键词: 行为识别, 门控循环单元, 注意力机制, 工件特征

Abstract: The online recognition of the worker's operation is important to improve the ability of digital management in workshop.The traditional recognition method only makes the judgment through the human information,ignoring the auxiliary function of work piece to the operation recognition.Owing to the fact that the disorder of working procedure or the illegal operation could be reflected by workpiece utilization,an online operation recognition method based on the attention of workpiece was proposed.The Gated Recurrent Unit (GRU) model was used as the base model to recognize the human joint sequence detected from the monitoring video.Then the feature of the workpiece was extracted from the monitoring video by the pre-trained instance segmentation model.The workpiece features were viewed as attention information integrated into the basic GRU model for joint training.The experiments were performed in both self-built data sets and public data sets and used intersection over union in time to compute the precision and recall.The experimental results showed that the proposed method could effectively monitor the operations of worker.

Key words: operation recognition, gated recurrent unit, attention mechanism, workpiece features