Activity Recognition in Industrial Environment using Two Layers Learning

Abstract

Action and activity recognition is essential in the world of cobots to ensure the best efficiency and a safety collaboration between a robot and the human-being. The approach of the article is the creation of a new activity dataset for an industrial context with cobots for recognition. We proposed to use LSTM (Long Short-Term Memory) to analyse and recognize the activities and we also proposed to model the action using the Principal Component Analysis (PCA) and then recognize the activity using LSTM. Using this two level approaches on the dataset we collected, we obtained high recognition level : 96.826 (+/-0.383) %.