Unsafe Action Recognition of Miners Based on Video Description

The Recognition of miners' unsafe action is of great significance to coal mine safety production. However, its bottleneck lies in the lack of effective model to recognize miners' actions, interactive objects of actions as well as spatio-temporal multidimensional information in video. This paper puts forward a novel and effective description model of miners' insecure actions, aiming to comprehensively describe the above multidimensional information, which can be used in video recognition of miners' actions and video annotation. The multi-dimensional information in video is automatically extracted by Real-Time Object Detection with Region Proposal Networks (faster RCNN) and converted into vector representation, and then the feature vector is input into the two-layer Long Short- Term Memory (LSTM) model to generate language description. Finally, the model is verified by experiments on the data set of miners' insecure actions collected by ourselves, and the model achieves high accuracy.