Skeleton-based human action recognition for manufacturing in assembly task by deep learning

In the field of manufacturing assembly tasks, Human Action Recognition (HAR) plays a significant role in serving several crucial purposes, such as worker safety and support, production improvement, employee training, human robot collaboration, and so on. Recognizing assembly actions is a challenging task due to the similarity of actions and heterogeneity of operators. This paper presents a skeleton-based action recognition method using a CNN-based deep neural network architecture. Raw joint coordinates are used to represent human movements during an assembly task. The main focus is to explore different input feature arrangements to investigate the ability of the CNN-based model to capture spatial and temporal dependencies. The proposed approach is evaluated on the publicly available HA4M dataset, demonstrating high recognition rates and demonstrating the efficiency and effectiveness of the method for action recognition in smart manufacturing environments.