Safety behavior detection of factory workers based on multi task model

This paper focuses on the safety problems in the industrial production process. Through the investigation of copper processing plant, we found that workers have little safety awareness, who often get into factories without helmets and enter non-safety areas without permission. To prevent safety accidents, we propose a multi-task model to detect some unsafe behaviors based on computer vision techniques. The model is composed of a shared encoder and two independent decoder branches. It can be used to detect the safety helmet and the factory landmark lines at the same time. According to the detection results, it can be judged whether the workers regularly wear the safety helmet or they are in the unsafe area in real time. Through the judgment of workers’ safety status, we can strengthen the safety supervision of industrial site, which has strong practical significance and application prospect.