19th Asia-Pacific Vibration Conference (APVC2021)

377 / 2021-11-24 16:28:19

Abnormal action recognition in pharmaceutical workshop based on human key points

deep learning,human key points estimation,abnormal action

The abnormal action such as smoking in pharmaceutical workshop is often accompanied by immense potential dangers. Therefore, it is of great significance to detect and warn abnormal action during normal operation. With the improvement of human pose estimation algorithms, action analysis approaches based on human key points have been proposed, which could be utilized to identify various actions with high accuracy. In this paper, the most advanced human pose estimation algorithm HRNet is adopted to extract human key points for modelling analysis, which is used to establish a library of different actions. Afterwards, the template matching method is used to detect abnormal actions. Three different actions namely hat removal, smoking and non-abnormal actions are take into consideration for model performance evaluation. The final results show that the abnormal action analysis based on human key points has higher stability and reaches satisfactory accuracy of 82.8%.