Relevance of deep sequence models for recognising automated construction activities: a case study on a low-rise construction system

Abstract

Recognising activities of construction equipment is essential for monitoring productivity, construction progress, safety, and environmental impacts. While there have been many studies on activity recognition of earth excavation and moving equipment, activity identification of Automated Construction Systems (ACS) has been rarely attempted. Especially for low-rise ACS that offers energy-efficient, cost-effective solutions for urgent housing needs, and provides more affordable living options for a broader population. Deep learning methods have gained a lot of attention because of their ability to perform classification without manually extracting relevant features. This study evaluates the feasibility of deep sequence models for developing an activity recognition framework for low-rise automated construction equipment. Time series acceleration data was collected from the structure to identify major operation classes of an ACS. Long Short Term Memory Networks (LSTM) were applied for identifying the activity classes and the performance was compared with that of traditional machine learning classifiers. Diverse augmentation methods were adopted for generating datasets for training the deep learning classifiers. Several recently published literature seem to establish the superiority of complex deep learning techniques over traditional machine learning algorithms regardless of the application context. However, the results of this study show that all the conventional machine learning classifiers perform equivalently or better than deep learning classifiers in identifying activities of the ACS. The performance of the deep learning classifiers is affected by the lack of diversity in the initial dataset. If the augmented dataset significantly alters the characteristics of the original dataset, it may not deliver good recognition results.