Activity Recognition for Attachments of Construction Machinery Using Decision Trees

Activity recognition in construction helps operators and managers by providing information about current and past use of machines and tools. At the same time, excavator attachments enable excavators to perform other physical tasks in addition to earthmoving like screening or compacting. This work addresses the prototype of an activity recognition for excavator attachments independent of the carrier machine. Activity recognition was implemented for buckets, compactors, grabs, and screeners. The recognition is based on acceleration data and angular velocities collected by an inertial measurement unit on the attachment. Decision trees are used for classification. The activity classes “Operation”, “Transport”, and “Rest” were defined as target classes for the activity recognition. The decision trees achieved accuracies comparable to other machine learning algorithms. The results indicate that activity recognition of attachments should distinguish between different attachment classes like buckets and crabs.