Human behavior data extension method for mine scene

At present, there is no dedicated large-scale video data set for the task of monitoring unsafe behavior in mine scenes. In this paper, the monitoring data set of miners’ unsafe behavior is built by ourselves. The data set is divided into extended group and real group. The extended group and the real group, the extended group of clear video samples from 18 public data sets, for each clear video in the extended group with a particle simulation mask, to achieve a similar dust masking effect as the mine scene. The real group comes from the site monitoring of mine operations. This paper uses the MCTN(Muti-scenic Cyclic Translation Network) to complete cross-scenario knowledge transfer, so as to verify the validity of the mine unsafe behavior monitoring dataset constructed in this paper.