Research on Laboratory Unsafe Behavior Recognition Based on Robust Time Feature Learning Advancement
Xinran Li, Liang Chen, Wenhong Liu · 2024
In order to solve the problem of dangerous accidents caused by unsafe behaviors in university laboratories, We propose an improved method based on the RTFM model. This method combines Dilated convolution and multi-head attention mechanism to form a new feature aggregation module and improve the sensitivity to abnormal features. In addition, the Gated Recurrent Unit (GRU) is introduced to capture the timing information in the video, effectively solving the problem of false detection when RTFM detects anomalies. Experimental results show that compared with the original model, the method in this paper improves AUC by 4%. Effectively improves the accuracy of identifying abnormal behaviors of laboratory personnel.